Browse the corpus

INTRODUCTION Prostate cancer is the most common cancer diagnosed in men in the Western Hemisphere, with approximately 899,000 patients diagnosed and 258,000 deaths worldwide in 2008.1 Patients with locally advanced disease, defined as stage categories T3-4, N0, and M0, are still prevalent in regions where the use of prostate-specific antigen (PSA) screening is not widespread.2 Previous uncertainties about the roles of radiotherapy (RT) and androgen-deprivation therapy (ADT)3,4 have been greatly clarified after the publication of randomized trials demonstrating the benefits of ADT added to RT and the benefits of RT added to ADT.5–7 Three reported randomized trials compared ADT alone with to ADT plus RT. The present trial was the largest of these and was developed by the NCIC Clinical Trials Group in collaboration with the Medical Research Council and the National Cancer Institute US Cancer Therapy Evaluation Program. The interim analysis of this intergroup trial has been reported previously8 and showed a significant overall survival (OS) improvement for patients treated with ADT+RT (hazard ratio [HR], 0.77; 95% CI, 0.61 to 0.98; P = .033) and improvement in disease-specific survival (DSS). The final, preplanned analysis presented here reports on the longer-term survival outcomes and toxicity. Quality-of-life analyses are reported by Brundage et al.8a

) improvement for patients treated with ADT+RT (hazard ratio [HR], 0.77; 95% CI, 0.61 to 0.98; P = .033) and improvement in disease-specific survival (DSS). The final, preplanned analysis presented here reports on the longer-term survival outcomes and toxicity. Quality-of-life analyses are reported by Brundage et al.8a PATIENTS AND METHODS The study design has been previously described in detail.8 Patients were randomly assigned to ADT alone or to ADT+RT. Eligible patients had locally advanced disease, initially defined as T3-4, N0, M0. In 1999, the entry criteria were broadened to include patients with localized (T1-2) but high-risk disease, defined either as a PSA of more than 40 μg/L or PSA of 20 to 40 μg/L plus a Gleason score of 8 to 10. Pelvic node imaging was not mandatory unless only the prostate was to be irradiated, rather than the whole pelvis. Surgical lymph node staging before random assignment was permitted and had to be negative for nodal disease. No previous therapy for prostate cancer was allowed, but random assignment was permitted within a 12-week window after starting first-line ADT. The primary objective was to determine whether the addition of RT to ADT prolonged OS, defined as time from random assignment to death from any cause or censoring at last follow-up. Secondary end points were time to progression (TTP), DSS, quality of life, toxicity, and symptomatic local control (defined as surgical interventions for symptomatic local disease).

her the addition of RT to ADT prolonged OS, defined as time from random assignment to death from any cause or censoring at last follow-up. Secondary end points were time to progression (TTP), DSS, quality of life, toxicity, and symptomatic local control (defined as surgical interventions for symptomatic local disease). Disease progression was defined as the first of any of the following events: biochemical progression, local progression, development of metastatic disease, or death from prostate cancer. For the per-protocol analysis, biochemical progression was defined by two consecutive PSA readings of more than 10 ng/mL in patients whose PSA nadir was ≤ 4 ng/mL. In patients whose PSA nadir was greater than 4 ng/mL, biochemical progression was defined as a PSA level of more than 10 ng/mL and 20% above the baseline reading. In addition to this prespecified definition, we analyzed biochemical progression according to the American Society for Radiation Oncology Phoenix criteria.9 Local progression was defined either after histologic confirmation or after the development of ureteric obstruction. Distant progression was defined by imaging. Patients were randomly assigned using a straight minimization strategy,10 stratified by center, initial PSA level (< 20 v 20 to 50 v > 50 μg/L), choice of hormonal therapy (luteinizing hormone–releasing hormone [LHRH] agonist v orchiectomy), method of lymph node staging (clinical v surgical v not done), Gleason score (< 8 v 8 to 10), and prior hormonal therapy.

aight minimization strategy,10 stratified by center, initial PSA level (< 20 v 20 to 50 v > 50 μg/L), choice of hormonal therapy (luteinizing hormone–releasing hormone [LHRH] agonist v orchiectomy), method of lymph node staging (clinical v surgical v not done), Gleason score (< 8 v 8 to 10), and prior hormonal therapy. ADT consisted of either bilateral orchiectomy or LHRH agonists (plus 2 weeks of oral antiandrogen to cover flare), according to patient and physician preference, and was continued for life. RT was to start within 8 weeks after random assignment and was given to the whole pelvis, to a dose of 45 Gy in 25 fractions, followed by a further 20 to 24 Gy to the prostate alone in 10 to 12 fractions. The total treatment time was 7 to 7.5 weeks. If the treating physician felt that a patient was unsuitable for whole pelvic RT or if histologically negative lymph nodes had been demonstrated, RT to the prostate alone, to a dose of 64 to 69 Gy in 35 to 39 fractions, was permitted. Pelvic RT was delivered using a four-field box technique to cover the prostate, seminal vesicles, and internal and external iliac lymph nodes. The prostate volume encompassed the prostate and periprostatic tissues with a margin at the investigating physician's discretion. Patients in Canada underwent real-time review of their RT implementation. The dose distributions, treatment prescription sheets and simulator films for Canadian patients were reviewed by one author (C.H.) before treatment or at latest within 3 days of start of treatment, and recommendations for change, if necessary, were faxed back to the prescribing physician within 24 hours. After the completion of RT, copies of all completed prescription sheets were sent for review to ascertain that the treatment was delivered according to protocol. The dose was specified at the intersection of the beam axes, according to the guidelines of the International Commission on Radiation Units.11 The study received the required local and national ethics committee approvals; all patients signed an informed consent document.

e treatment was delivered according to protocol. The dose was specified at the intersection of the beam axes, according to the guidelines of the International Commission on Radiation Units.11 The study received the required local and national ethics committee approvals; all patients signed an informed consent document. The original study design mandated accrual of 650 patients to the trial, based on an assumed 10-year survival of 35% for patients treated with ADT only, to detect a 10% improvement in 10-year survival (HR of 0.76, 80% power using a one-sided 5% level test). In September 2002, after 688 patients had been recruited, the protocol was amended because of a low event rate. The revised statistical parameters assumed a 57% survival rate at 10 years in the control arm, 80% power, and an overall 2.5% level one-sided test to detect the same HR of 0.76, which would translate to an absolute 8.4% increase in 10-year survival. This required a total of 421 deaths for the final analysis, taking two planned interim analyses into account. On the basis of the type I error spending function as proposed by DeMets and Lan,12 the prespecified significance level for final analysis was P = .035 (two-sided) to maintain the overall significant level of two-sided P = .05. According to these specifications, the estimated sample size for the study was 1,200.

count. On the basis of the type I error spending function as proposed by DeMets and Lan,12 the prespecified significance level for final analysis was P = .035 (two-sided) to maintain the overall significant level of two-sided P = .05. According to these specifications, the estimated sample size for the study was 1,200. OS was determined using the Kaplan-Meier product-limit method13 and compared using a log-rank test stratified for initial PSA level, choice of hormonal therapy, method of lymph node staging, Gleason score, and prior hormonal therapy. Two Cox model analyses were used. First, HRs and CIs were estimated (Cox model 1).14 The DSS rates were estimated by cumulative incidence. The analyses were performed with the use of SAS software, version 9.2 (SAS Institute, Cary, NC). Second, a Cox proportional hazards model was used to assess the treatment effect while adjusting for known important prognostic factors in addition to the prespecified stratification factors (Cox model 2). These were region (North America v rest of world), initial PSA level (< 20 v 20 to 50 v > 50 μg/L v missing), Gleason score (< 8 v 8 to 10 v missing), prior hormone therapy (excluding orchiectomy; yes v no), choice of hormonal therapy (orchiectomy v LHRH agonist), method of lymph node staging (clinical [no computed tomography] v radiologic [computed tomography negative] v surgical dissection v not done or missing), age (< v ≥ 65 years), and orchiectomy versus LHRH. TTP was defined as the time from random assignment to the date of the first documented disease progression or death from prostate cancer.

taging (clinical [no computed tomography] v radiologic [computed tomography negative] v surgical dissection v not done or missing), age (< v ≥ 65 years), and orchiectomy versus LHRH. TTP was defined as the time from random assignment to the date of the first documented disease progression or death from prostate cancer. Toxicity was measured using the NCIC Clinical Trials Group Expanded Common Toxicity Criteria. Information on quality of life is presented by Brundage et al.8a RESULTS Between 1995 and 2005, 1,205 patients were randomly assigned (Fig 1). Trial participants were well matched in terms of their baseline characteristics (Table 1). Their median age was 70 years. Eighty-seven percent of patients had locally advanced (T3-4) disease, 63% of patients had a PSA of more than 20 μg/L, and 18% had a Gleason score of more than 8. The database contained data up to and including December 31, 2010, and included 465 reported deaths. The median follow-up time was 8 years (range, 0 to 15.2 years). Ninety-four percent of patients included in the analysis had data available in the 2 years preceding the clinical cutoff date. Fig 1. CONSORT diagram. ADT, androgen-deprivation therapy; RT, radiotherapy. Table 1. Baseline Demographics and Clinical Characteristics

RESULTS Between 1995 and 2005, 1,205 patients were randomly assigned (Fig 1). Trial participants were well matched in terms of their baseline characteristics (Table 1). Their median age was 70 years. Eighty-seven percent of patients had locally advanced (T3-4) disease, 63% of patients had a PSA of more than 20 μg/L, and 18% had a Gleason score of more than 8. The database contained data up to and including December 31, 2010, and included 465 reported deaths. The median follow-up time was 8 years (range, 0 to 15.2 years). Ninety-four percent of patients included in the analysis had data available in the 2 years preceding the clinical cutoff date. Fig 1. CONSORT diagram. ADT, androgen-deprivation therapy; RT, radiotherapy. Table 1. Baseline Demographics and Clinical Characteristics Characteristic ADT (n = 602) ADT+RT (n = 603) Total (N = 1,205) No. of Patients % No. of Patients % No. of Patients % Age, years < 65 134 22.3 132 21.9 266 22.1 ≥ 65 468 77.7 471 78.1 939 77.9 Median 69.7 69.7 69.7 Range 50-79.8 46.3-80.3 46.3-80.3 ECOG performance status 0 474 78.7 469 77.8 943 78.3 1 119 19.8 126 20.9 245 20.3 2 9 1.5 8 1.3 17 1.4 Lymph node staging Clinical 427 70.9 422 70.0 849 70.5 Radiologic 50 8.3 53 8.8 103 8.5 Surgical 12 2.0 17 2.8 29 2.4 Not done 113 18.8 111 18.4 224 18.6 T stage Missing 1 0.2 1 0.1 T2 76 12.6 71 11.8 147 12.2 T3 499 82.9 501 83.1 1000 83.0 T4 27 4.5 30 5.0 57 4.7 Rectal exam Done 586 97.3 588 97.5 1174 97.4 Not done 16 2.7 15 2.5 31 2.6 Results of rectal examination Abnormal 161 26.7 162 26.9 323 26.8 Normal 19 3.2 19 3.2 38 3.2 Unknown/missing 422 70.1 422 70.0 844 70.0 Region of patients Northern America 180 29.9 181 30.0 361 30.0 MRC 422 70.1 422 70.0 844 70.0 Initial PSA level, μg/L < 20 223 37.0 216 35.8 439 36.4 20-50 229 38.0 231 38.3 460 38.2 > 50 150 24.9 156 25.9 306 25.4 Gleason score Missing 6 1.0 3 0.5 9 0.7 < 8 380 63.1 381 63.2 761 63.2 8-10 216 35.9 219 36.3 435 36.1 ADT before random assignment No 314 52.2 315 52.2 629 52.2 Yes 288 47.8 288 47.8 576 47.8 Choice of hormonal therapy LHRH 562 93.4 562 93.2 1124 93.3 Bilateral orchiectomy 40 6.6 41 6.8 81 6.7 Abbreviations: ADT, androgen-deprivation therapy; ECOG, Eastern Cooperative Oncology Group; LHRH, luteinizing hormone–releasing hormone; MRC, Medical Research Council; PSA, prostate-specific antigen; RT, radiotherapy.

Choice of hormonal therapy LHRH 562 93.4 562 93.2 1124 93.3 Bilateral orchiectomy 40 6.6 41 6.8 81 6.7 Abbreviations: ADT, androgen-deprivation therapy; ECOG, Eastern Cooperative Oncology Group; LHRH, luteinizing hormone–releasing hormone; MRC, Medical Research Council; PSA, prostate-specific antigen; RT, radiotherapy. Of the 603 patients randomly assigned to ADT+RT, 586 (97%) received RT, and 13 did not receive RT; in four patients, it was unknown whether or not RT was received. Of the 586 patients known to have received RT, 43 received doses less than 65 Gy, and 10 received doses greater than 69 Gy. Thus, 88% of the patients allocated to the ADT+RT arm received doses between 65 and 69 Gy. Nine (1%) of 602 patients randomly assigned to ADT alone received RT, as defined by irradiation to the pelvis of more than 50 Gy within 1 year of random assignment and without evidence of disease progression. LHRH agonists were used in 1,105 patients (92%), and bilateral orchiectomy was performed in 93 patients (8%), with no evidence of differences in proportions between the two arms.

T, as defined by irradiation to the pelvis of more than 50 Gy within 1 year of random assignment and without evidence of disease progression. LHRH agonists were used in 1,105 patients (92%), and bilateral orchiectomy was performed in 93 patients (8%), with no evidence of differences in proportions between the two arms. OS There were 260 deaths reported in patients treated with ADT alone and 205 deaths in patients treated with ADT+RT. The addition of RT led to a 30% reduction in the risk of death (HR, 0.70, based on Cox model 1; 95% CI, 0.57 to 0.85; P < .001; Fig 2). The median OS time was 9.7 years (95% CI, 8.8 to 10.5 years) for patients on the ADT-alone arm, whereas it was 10.9 years (95% CI, 10.0 to 12.8 years) for patients on the ADT+RT arm. The 10-year OS rate was 49% (95% CI, 44% to 54%) for patients on the ADT arm, whereas it was 55% (95% CI, 49% to 60%) for patients on the ADT+RT arm. A multivariable Cox model confirmed the effect of treatment, independent from other variables, with a P = .0011 in favor of the ADT+RT arm. The adjusted HR of ADT+RT versus ADT alone was 0.74 (95% CI, 0.61 to 0.87, based on Cox model 2). Both PSA level (> 50 v < 20 μg/L) and Gleason score (8 to 10 v < 8) were significant prognostic factors for OS. Fig 2. Overall survival (OS). ADT, androgen-deprivation therapy; HR, hazard ratio; RT, radiotherapy.

OS There were 260 deaths reported in patients treated with ADT alone and 205 deaths in patients treated with ADT+RT. The addition of RT led to a 30% reduction in the risk of death (HR, 0.70, based on Cox model 1; 95% CI, 0.57 to 0.85; P < .001; Fig 2). The median OS time was 9.7 years (95% CI, 8.8 to 10.5 years) for patients on the ADT-alone arm, whereas it was 10.9 years (95% CI, 10.0 to 12.8 years) for patients on the ADT+RT arm. The 10-year OS rate was 49% (95% CI, 44% to 54%) for patients on the ADT arm, whereas it was 55% (95% CI, 49% to 60%) for patients on the ADT+RT arm. A multivariable Cox model confirmed the effect of treatment, independent from other variables, with a P = .0011 in favor of the ADT+RT arm. The adjusted HR of ADT+RT versus ADT alone was 0.74 (95% CI, 0.61 to 0.87, based on Cox model 2). Both PSA level (> 50 v < 20 μg/L) and Gleason score (8 to 10 v < 8) were significant prognostic factors for OS. Fig 2. Overall survival (OS). ADT, androgen-deprivation therapy; HR, hazard ratio; RT, radiotherapy. DSS Analysis of DSS indicated an excess of deaths caused by prostate cancer in patients treated with ADT alone (Table 2). A competing risks analysis indicated a significant reduction in the risk of death from prostate cancer in patients treated with ADT+RT (HR, 0.46; 95% CI, 0.34 to 0.61; P < .001; Fig 3). There was no evidence of any differences in deaths from other causes (P = .58; Table 2). Sensitivity analyses were performed to test the impact of potential inaccuracy in investigator assignment of cause of death. In each case, the reduction in risks of death from prostate cancer in RT-treated patients was confirmed, with P < .001

evidence of any differences in deaths from other causes (P = .58; Table 2). Sensitivity analyses were performed to test the impact of potential inaccuracy in investigator assignment of cause of death. In each case, the reduction in risks of death from prostate cancer in RT-treated patients was confirmed, with P < .001 Table 2. Causes of Death Cause of Death ADT (n = 260) ADT+RT (n = 205) Total (n = 465) No. of Patients % No. of Patients % No. of Patients % Prostate cancer 134 52 65 32 199 43 Cardiac/stroke 37 14 33 16 70 15 Other cancer 31 12 44 17 75 16 Pneumonia 11 4 11 9 22 5 Other 31 12 34 21 65 14 Unknown 16 6 18 5 34 7 Alive 342 398 740 Abbreviations: ADT, androgen-deprivation therapy; RT, radiotherapy. Fig 3. Deaths from prostate cancer. ADT, androgen-deprivation therapy; RT, radiotherapy. Nonfatal End Points Disease progression. Using the prespecified definition of biochemical progression, the 10-year disease progression–free rate was 46% (95% CI, 41% to 51%) for patients on the ADT-alone arm and 74% (95% CI, 68% to 78%) for patients on the ADT+RT arm. The HR for TTP on the ADT+RT arm versus the ADT-alone arm was 0.31 (95% CI, 0.25 to 0.39; Fig 4A). Fig 4. Time to disease progression (TTP; A) using prespecified definition of biochemical progression and (B) using American Society for Radiation Oncology Phoenix definition of biochemical progression. ADT, androgen-deprivation therapy; RT, radiotherapy.

Nonfatal End Points Disease progression. Using the prespecified definition of biochemical progression, the 10-year disease progression–free rate was 46% (95% CI, 41% to 51%) for patients on the ADT-alone arm and 74% (95% CI, 68% to 78%) for patients on the ADT+RT arm. The HR for TTP on the ADT+RT arm versus the ADT-alone arm was 0.31 (95% CI, 0.25 to 0.39; Fig 4A). Fig 4. Time to disease progression (TTP; A) using prespecified definition of biochemical progression and (B) using American Society for Radiation Oncology Phoenix definition of biochemical progression. ADT, androgen-deprivation therapy; RT, radiotherapy. Using the American Society for Radiation Oncology Phoenix definition of biochemical progression, the 10-year biochemical progression-free rate was 27% (95% CI, 23% to 32%) for patients on the ADT-alone arm, whereas it was 63% (95% CI, 57% to 68%) for patients on the ADT+RT arm. The HR for TTP on the ADT+RT arm versus the ADT arm was 0.31 (95% CI, 0.27 to 0.37).

cology Phoenix definition of biochemical progression, the 10-year biochemical progression-free rate was 27% (95% CI, 23% to 32%) for patients on the ADT-alone arm, whereas it was 63% (95% CI, 57% to 68%) for patients on the ADT+RT arm. The HR for TTP on the ADT+RT arm versus the ADT arm was 0.31 (95% CI, 0.27 to 0.37). Toxicity. The five most frequent grade 3 or higher treatment-related toxicities were impotence/libido (29% for ADT alone and 33% for ADT+RT; P = .17), hot flushes (8% for ADT alone and 5% for ADT+RT; P = .10), urinary frequency (4% for ADT alone and 7% for ADT+RT; P = .13), ischemia (3% for ADT alone and 5% for ADT+RT; P = .09), and hypertension (3% for ADT alone and 4% for ADT+RT; P = .54). There was no evidence of a difference in reported cardiovascular toxicities between the two arms. Bowel-related adverse events were more frequent in the ADT+RT arm compared with the ADT arm; most were grade 1 and 2. The rate of bowel adverse events greater than grade 3 at 24 months was negligible (two of 589 assessable patients in the ADT+RT arm (Table 3). Table 3. Bowel-Related Adverse Events at 24 Months Toxicity No. of Patients ADT (n = 606) ADT+RT (n = 589) Total Any Grade 1-2 Grade 3-5 Any Grade 1-2 Grade 3-5 Any Grade 1-2 Grade 3-5 Diarrhea 87 14 0 223 32 2 310 46 2 Flatulence 20 1 1 37 7 0 57 8 1 Bleeding 50 5 0 133 41 0 183 46 0 Pain 62 12 1 79 10 0 141 22 1 Proctitis 43 4 0 119 19 0 162 23 0 Abbreviations: ADT, androgen-deprivation therapy; RT, radiotherapy.

n = 589) Total Any Grade 1-2 Grade 3-5 Any Grade 1-2 Grade 3-5 Any Grade 1-2 Grade 3-5 Diarrhea 87 14 0 223 32 2 310 46 2 Flatulence 20 1 1 37 7 0 57 8 1 Bleeding 50 5 0 133 41 0 183 46 0 Pain 62 12 1 79 10 0 141 22 1 Proctitis 43 4 0 119 19 0 162 23 0 Abbreviations: ADT, androgen-deprivation therapy; RT, radiotherapy. Effect of RT Field Using data prospectively reported by investigators, we performed exploratory post hoc analyses (uncorrected for multiple comparisons) to examine the effects of RT field among patients allocated to the ADT+RT arm, stratified by region, PSA, Gleason score, choice of hormone therapy, lymph node staging, and prior hormone therapy. Patients planned for pelvic irradiation (n = 420, 72%) showed a trend toward improved OS compared with prostate-only RT (n = 166, 28%; HR, 0.70; 95% CI, 0.45 to 1.09; P = .12). The other efficacy analyses were as follows: for TTP, the HR was 0.77 (95% CI, 0.44 to 1.34; stratified log-rank P = .35), and for DSS, the HR was 0.53 (95% CI, 0.25 to 1.13; P = .098).

= 420, 72%) showed a trend toward improved OS compared with prostate-only RT (n = 166, 28%; HR, 0.70; 95% CI, 0.45 to 1.09; P = .12). The other efficacy analyses were as follows: for TTP, the HR was 0.77 (95% CI, 0.44 to 1.34; stratified log-rank P = .35), and for DSS, the HR was 0.53 (95% CI, 0.25 to 1.13; P = .098). DISCUSSION These results confirm the previous, interim findings from our randomized controlled trial,8 with RT improving both overall and prostate cancer–specific outcomes (DSS, TTP) when added to ADT in men with locally advanced disease. Furthermore, this was achieved without major detriment in terms of long-term toxicity. Our results concur with those from other trials in similar groups of patients. The Swedish Prostate Cancer Group (SPCG) study SPCG-715 showed a similar reduction in overall mortality (HR, 0.68) and disease-specific mortality (HR, 0.44) for the addition of RT to flutamide. A study (ClinicalTrials.gov identifier: NCT01122121) by the French collaborative group in 264 patients with T3-4 disease reported a significant improvement in progression-free survival with the addition of RT to ADT.16 Median OS times had not been reached in that study, and no improvement in OS was detected; however, this trial was relatively small and was powered only to detect a difference in 5-year progression-free survival and not a difference in OS.

icant improvement in progression-free survival with the addition of RT to ADT.16 Median OS times had not been reached in that study, and no improvement in OS was detected; however, this trial was relatively small and was powered only to detect a difference in 5-year progression-free survival and not a difference in OS. Our results are noteworthy for a number of reasons. First, they indicate that the benefits seen for RT in the context of ADT with flutamide (in the SPCG-7 study) are also seen with LHRH agonist therapy or surgical castration. Second, our patient population represents a higher risk group than in SPCG-7; the Swedish group had pathologic confirmation of N0 status if the PSA was greater than 11 ng/mL (2% of our study), 20% had T1-2 disease (10% in our study), 60% had a PSA of less than 20 ng/mL (37% of our study), and the maximum allowed PSA level was 70 ng/mL (unlimited in our study), although 75% of patients had WHO grade 1 or 2 tumors, (81% in our study had Gleason score < 7). Third, the median follow-up in our study was 8 years compared with 7.6 years in the SPCG-7 study.

% had a PSA of less than 20 ng/mL (37% of our study), and the maximum allowed PSA level was 70 ng/mL (unlimited in our study), although 75% of patients had WHO grade 1 or 2 tumors, (81% in our study had Gleason score < 7). Third, the median follow-up in our study was 8 years compared with 7.6 years in the SPCG-7 study. The RT technique used in our trial reflects the prevailing treatment philosophies of the time. The study predated outcome data from randomized trials of dose-escalated RT, and the RT doses used here are modest by modern standards.17–19 Whether dose escalation in this setting might achieve superior outcomes even to those reported here is a matter for speculation. Data from dose-escalation studies suggest that biochemical control might be as much as two-fold better for RT doses greater than 70 Gy,20 and there is no evidence that patients with T3-4 disease do not realize such benefits from dose escalation.17,20 A recent study using ADT plus dose-escalated RT (80 Gy in 40 fractions or equivalent) in 168 patients with high-risk prostate cancer reported a biochemical progression-free rate of 79% at 5 years.21

70 Gy,20 and there is no evidence that patients with T3-4 disease do not realize such benefits from dose escalation.17,20 A recent study using ADT plus dose-escalated RT (80 Gy in 40 fractions or equivalent) in 168 patients with high-risk prostate cancer reported a biochemical progression-free rate of 79% at 5 years.21 Another unanswered question relates to the optimum field of RT. This question is still unresolved despite several randomized trials.22–24 Our nonrandomized, post hoc exploratory subgroup analysis indicates a trend toward improved outcome with larger field size, a finding that requires confirmation in a rigorously conducted phase III randomized controlled trial. These results must be interpreted with caution because of biases inherent with selection of RT field in our study (ie, the influence of comorbidities or other patient factors that may confound the efficacy analyses) and the potential for more toxicity associated with a wider field of treatment delivery. Ongoing trials, such as the United Kingdom PIVOTAL trial (ClinicalTrials.gov identifier: NCT01685190; A Randomised Phase II Trial of Prostate and Pelvic Versus Prostate Alone Radiotherapy Treatment Volumes Using High-Dose IMRT for Locally Advanced Prostate Cancer), currently a randomized phase II study, or the Radiation Therapy Oncology Group 0924 phase III trial (ClinicalTrials.gov identifier: NCT01368588), will establish whether the addition of high-dose nodal irradiation using intensity-modulated RT improves outcomes compared with prostate-only RT in patients with N0M0 disease. Some patients, staged clinically as N0, would be found to have pathologic evidence of lymph node metastases if subjected to lymph node dissection, and it might be argued that our findings for pelvic RT suggest a benefit for such treatment in patients with clinical or pathologic node-positive disease. Alternatively, it is possible that the larger fields used for pelvic irradiation might actually have achieved better coverage of the prostate itself, compared with prostate-only fields as used in the pre–intensity-modulated RT era, and the trends seen could reflect this. However, our data on local control in the prostate do not permit further exploration of this possibility here.

pelvic irradiation might actually have achieved better coverage of the prostate itself, compared with prostate-only fields as used in the pre–intensity-modulated RT era, and the trends seen could reflect this. However, our data on local control in the prostate do not permit further exploration of this possibility here. As would be expected, our toxicity data indicate a detectable, although modest, impact of RT as administered in this trial. It is noteworthy that the grade ≥ 3 toxicity that we detected was short term only, and we would suggest that the toxicity of RT should not be regarded as a barrier to its routine use in this patient population. In this trial, the intended duration of ADT was lifelong. Data have emerged elsewhere on the long-term toxicities of ADT,25–27 and we are unable to provide data regarding optimal duration because of the symmetry and long-term nature of ADT in both treatment arms of our study. For high-risk disease, data from clinical studies support longer (> 2 years) rather than shorter durations (≤ 6 months).5 This trial underlines the benefits of achieving local control with RT in locally advanced prostate cancer. Surgery might also be an effective means of achieving local disease control,2,28 but given the Level I evidence presented here and by others,15 alternatives to ADT+RT should only be administered in the context of a prospective randomized controlled trial.

eving local control with RT in locally advanced prostate cancer. Surgery might also be an effective means of achieving local disease control,2,28 but given the Level I evidence presented here and by others,15 alternatives to ADT+RT should only be administered in the context of a prospective randomized controlled trial. Recent data suggest that some men with T3-4 disease are still being managed with ADT alone.29,30 Although there are undoubtedly patients for whom RT or indeed any curative therapy would be inappropriate because of age, comorbidity, or other factors, we conclude that patients with clinically node-negative, locally advanced prostate cancer who are suitable for additional RT should be offered that option, an opinion shared by European31 and North American32 guidelines. Supplementary Material Protocol Supported by National Cancer Institute Grant No. CA077202, Canadian Cancer Society Research Institute Grants No. 015469 and No. 021039, United Kingdom Medical Research Council Grant No. G9805643, and the United Kingdom National Cancer Research Network. Presented at the 48th Annual Meeting of the American Society of Clinical Oncology, June 1-5, 2012, Chicago, IL. Clinical trial information: NCT00002633 Terms in blue are defined in the glossary, found at the end of this article and online at www.jco.org. Authors' disclosures of potential conflicts of interest are found in the article online at www.jco.org. Author contributions are found at the end of this article. Written on behalf of the NCIC Clinical Trials Group PR.3/Medical Research Council PR07/Intergroup T94-0110 investigators.

Terms in blue are defined in the glossary, found at the end of this article and online at www.jco.org. Authors' disclosures of potential conflicts of interest are found in the article online at www.jco.org. Author contributions are found at the end of this article. Written on behalf of the NCIC Clinical Trials Group PR.3/Medical Research Council PR07/Intergroup T94-0110 investigators. AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST Disclosures provided by the authors are available with this article at www.jco.org. AUTHOR CONTRIBUTIONS Conception and design: Malcolm D. Mason, Wendy R. Parulekar, Matthew R. Sydes, Michael Brundage, Mary Gospodarowicz, Edmund C. Kostashuk, Mahesh K.B. Parmar, Bingshu E. Chen, Padraig Warde Administrative support: Wendy R. Parulekar Provision of study materials or patients: Wendy R. Parulekar, Richard Cowan, Gregory Swanson, P.D. John Hardman Collection and assembly of data: Malcolm D. Mason, Wendy R. Parulekar, Matthew R. Sydes, Michael Brundage, Peter Kirkbride, Mary Gospodarowicz, Richard Cowan, Edmund C. Kostashuk, John Anderson, Gregory Swanson, Charles Hayter, Gordana Jovic, Andrea Hiltz, John Hetherington, Jinka Sathya, James B.P. Barber, Michael McKenzie, Salah El-Sharkawi, Luis Souhami, Bingshu E. Chen, Padraig Warde Data analysis and interpretation: Malcolm D. Mason, Wendy R. Parulekar, Matthew R. Sydes, Michael Brundage, Peter Kirkbride, Mary Gospodarowicz, Richard Cowan, Mahesh K.B. Parmar, Andrea Hiltz, John Hetherington, James B.P. Barber, Michael McKenzie, Salah El-Sharkawi, Luis Souhami, P.D. John Hardman, Bingshu E. Chen, Padraig Warde

s and interpretation: Malcolm D. Mason, Wendy R. Parulekar, Matthew R. Sydes, Michael Brundage, Peter Kirkbride, Mary Gospodarowicz, Richard Cowan, Mahesh K.B. Parmar, Andrea Hiltz, John Hetherington, James B.P. Barber, Michael McKenzie, Salah El-Sharkawi, Luis Souhami, P.D. John Hardman, Bingshu E. Chen, Padraig Warde Manuscript writing: All authors Final approval of manuscript: All authors Glossary Terms Gleason score:pathologic description of prostate cancer grade on the basis of the degree of abnormality in the glandular architecture. Gleason patterns 3, 4, and 5 denote low, intermediate, and high levels of histologic abnormality and tumor aggressiveness, respectively. The score assigns primary and secondary numbers on the basis of the most common and second most common patterns identified.

of the degree of abnormality in the glandular architecture. Gleason patterns 3, 4, and 5 denote low, intermediate, and high levels of histologic abnormality and tumor aggressiveness, respectively. The score assigns primary and secondary numbers on the basis of the most common and second most common patterns identified. intensity-modulated radiation therapy:radiation treatment using beams with nonuniform fluence profiles that shape the dose distribution in the target volume and adjacent normal structures. Beam modulation is typically achieved via multileaf collimators or custom-milled compensators to achieve the appropriate fluence profiles calculated by inverse optimization algorithms. The radiation beam is divided into beamlets of varying intensity such that the sum from multiple beams via inverse planning results in improved tumor targeting and normal tissue sparing. A technique of radiation therapy delivery in which the intensity of each beamlet of radiation coming from a specific angle can be adjusted to provide a desired dose distribution when the doses delivered from all beamlets are added from a single angle and from all dose delivery angles. An advanced type of high-precision radiotherapy, which aims to improve the coverage of the radiotherapy target and/or minimize radiation dose to surrounding normal tissue. overall survival:the duration between random assignment and death.

intensity-modulated radiation therapy:radiation treatment using beams with nonuniform fluence profiles that shape the dose distribution in the target volume and adjacent normal structures. Beam modulation is typically achieved via multileaf collimators or custom-milled compensators to achieve the appropriate fluence profiles calculated by inverse optimization algorithms. The radiation beam is divided into beamlets of varying intensity such that the sum from multiple beams via inverse planning results in improved tumor targeting and normal tissue sparing. A technique of radiation therapy delivery in which the intensity of each beamlet of radiation coming from a specific angle can be adjusted to provide a desired dose distribution when the doses delivered from all beamlets are added from a single angle and from all dose delivery angles. An advanced type of high-precision radiotherapy, which aims to improve the coverage of the radiotherapy target and/or minimize radiation dose to surrounding normal tissue. overall survival:the duration between random assignment and death. AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST Final Report of the Intergroup Randomized Study of Combined Androgen-Deprivation Therapy Plus Radiotherapy Versus Androgen-Deprivation Therapy Alone in Locally Advanced Prostate Cancer The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or jco.ascopubs.org/site/ifc.

s are considered compensated. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or jco.ascopubs.org/site/ifc. Malcolm D. Mason Consulting or Advisory Role: Sanofi, Dendreon Speakers' Bureau: Sanofi Wendy R. Parulekar No relationship to disclose Matthew R. Sydes No relationship to disclose Michael Brundage No relationship to disclose Peter Kirkbride No relationship to disclose Mary Gospodarowicz No relationship to disclose Richard Cowan Travel, Accommodations, Expenses: MIOT Institute Chennai Edmund C. Kostashuk No relationship to disclose John Anderson No relationship to disclose Gregory Swanson No relationship to disclose Mahesh K.B. Parmar Research Funding: BiPar/sanofi-aventis (Inst), Novartis (Inst), Pfizer (Inst), Janssen Pharmaceuticals (Inst), Astellas Pharma (Inst), B&C (Inst) Charles Hayter No relationship to disclose Gordana Jovic No relationship to disclose Andrea Hiltz No relationship to disclose John Hetherington No relationship to disclose Jinka Sathya No relationship to disclose James B.P. Barber No relationship to disclose Michael McKenzie No relationship to disclose Salah El-Sharkawi No relationship to disclose Luis Souhami No relationship to disclose P.D. John Hardman No relationship to disclose Bingshu E. Chen No relationship to disclose Padraig Warde No relationship to disclose

Jinka Sathya No relationship to disclose James B.P. Barber No relationship to disclose Michael McKenzie No relationship to disclose Salah El-Sharkawi No relationship to disclose Luis Souhami No relationship to disclose P.D. John Hardman No relationship to disclose Bingshu E. Chen No relationship to disclose Padraig Warde No relationship to disclose Appendix The NCIC Clinical Trials Group (CTG) PR.3/Medical Research Council (MRC) UK PR07 investigators were as follows: Central Office staff: Canada—K. James, M. Bacon, A. LeMaitre, K. Ding (NCIC CTG, Kingston, Ontario, Canada). MRC Clinical Trials Unit staff: UK—M.R. Sydes, S.L. Naylor, N. Kelk, J. Latham, J. Nuttall, K. Sanders (trial managers), M.K.B. Parmar, M.R. Sydes, R.C. Morgan, G. Jovic (statisticians; MRC, London, United Kingdom). Collaborators: Canada—P. Joseph, D. Wilke (Queen Elizabeth II Health Sciences Centre, Halifax, Nova Scotia); M. Duclos, S. Faria, C. Freeman, L. Souhami (McGill University, Montreal, Quebec); J.P. Bahary, J.P. Guay (CHUM Hopital Notre-Dame, Montreal, Quebec); L. Eapen (Ottawa Health Research Institute General Division, Ottawa, Ontario); C. Hayter (Cancer Centre of South-Eastern Ontario at Kingston, Kingston, Ontario); T. Corbett, H. Lukka, M. Patel (Juravinski Centre at Hamilton Health Sciences, Hamilton, Ontario); L. Klotz (Odette Cancer Centre, Toronto, Ontario); A. Bayley, R. Bristow, C. Catton, J. Crook, M. McLean, M. Milosevic (University Health Network–Princess Margaret Hospital, Toronto, Ontario); J. Radwan, G. Rodrigues (London Regional Cancer Program, London, Ontario); C. Springer (Windsor Regional Cancer Centre, Windsor, Ontario); B. Lada, H. Prichard, R. Samant (Northeastern Ontario Cancer Centre, Sudbury, Ontario); H. Dhaliwal, S. Gulavita (Thunder Bay Regional Health Science Centre, Thunder Bay, Ontario); R. Halperin, D. McGowan, M. Parliament (Cross Cancer Institute, Edmonton, Alberta). United Kingdom—P. Kirkbride, J. Anderson (Weston Park and Royal Hallamshire Hospitals, Sheffield); M. Mason, J. Barber (Velindre Hospital, Cardiff); J. Heatherington, C. Bevis (Castle Hill Hospital, Hull); I. Pedley, R. McMenimen, T. Roberts (Newcastle General Hospital, Newcastle-upon-Tyne); A. El Sharkawi (Singleton Hospital, Swansea); J. Hardman, H. Van der Voet (The James Cook University Hospital, Middlesborough); D. Dearnaley, R. Huddart, A. Horwich, R. Eeles (Royal Marsden Hospitals, Sutton); I. Syndikus, J. Littler, D. Errington, A. Shenoy (Clatterbridge Centre for Oncology, Bebington); D. Sheehan, R.

. El Sharkawi (Singleton Hospital, Swansea); J. Hardman, H. Van der Voet (The James Cook University Hospital, Middlesborough); D. Dearnaley, R. Huddart, A. Horwich, R. Eeles (Royal Marsden Hospitals, Sutton); I. Syndikus, J. Littler, D. Errington, A. Shenoy (Clatterbridge Centre for Oncology, Bebington); D. Sheehan, R. Srinivasan (Royal Devon and Exeter, Exeter); A. Lydon (Torbay Hospital, Torbay); S. Sundar, M. Sokal (Nottingham City Hospital, Nottingham); C. Loughrey, S. Stranex (Belfast City and Belvoir Park Hospitals, Belfast); P. Rogers, A. Folkes (Royal Berkshire and Battle Hospitals, Berkshire); A. Stockdale, C. Irwin (Walsgrave Hospital, Coventry); R. Owen, P. Jenkins, A. Ritchie (Gloucestershire Hospitals, Cheltenham); R. Cowan, J. Wylie (Christie Hospital, Manchester); A. Al-Samarraie, S. Gollins, R. Russell (Glan Clwyd Hospital, Rhyl); N. Srihari, A. Cook, C. Beacock (Royal Shrewsbury Hospital, Shrewsbury); C. Parker, T. Porter (Yeovil District Hospital, Yeovil); P. Symmonds, M. Madden (Leicester Royal Infirmary, Leicester); S. Susnerwala (Blackpool Victoria Hospital, Blackpool); D. Bloomfield (Brighton Hospitals, Brighton); P. Whelan (St James's University Hospital, Leeds); D. Cole, E. Sugden (Churchill Hospital, Oxford); T. Sreenivasan (United Lincolnshire Trust, Lincoln); A. Harnett, T. Kumar, R. Mills (Norfolk and Norwich Hospital, Norwich); T. Sreenivasan, S. Dixit (Scunthorpe General Hospital, Scunthorpe); H. Newman (Royal United Hospital, Bath); C. Powell (Countess of Chester Hospital, Chester); H. Algurafi (Mayday Hospital, Croydon); S. Beesley (The Kent Cancer Centre, Maidstone); C. Elwell (Northampton Oncology Centre, Northampton); S. Sundaram, M. Murphy, P. Weston (Pinderfields Hospital, Wakefield); A. Al-Samarraie (Wrexham Maelor Hospital, Wrexham); J. Graham, S. Falk (Bristol Oncology Centre, Bristol); M. Russell (Beatson Oncology Centre, Glasgow); D. Bottomley, A. Kiltie (Cookridge Hospital, Leeds); A. Robinson, O. Koriech (Southend General Hospital, Southend); C. Woodward (West Suffolk Hospital, Bury St Edmunds); D. Wheatley, R. Ellis (Royal Cornwall Hospital, Truro); I. Syndikus (Warrington Hospital, Warrington); M. Stower, G. Urwin (York District Hospital, York); F. Bramble, J. Rundle, C. Carter (Bournemouth and Christchurch Hospitals); G. Sole (Hereford County Hospital, Hereford); C. Heath (Royal South Hants Hospital, Southampton); D. Bloomfield (Sussex Oncology Centre, Brighton); M. Pantelides (Royal Bolton Hospital, Bolton); H.

tower, G. Urwin (York District Hospital, York); F. Bramble, J. Rundle, C. Carter (Bournemouth and Christchurch Hospitals); G. Sole (Hereford County Hospital, Hereford); C. Heath (Royal South Hants Hospital, Southampton); D. Bloomfield (Sussex Oncology Centre, Brighton); M. Pantelides (Royal Bolton Hospital, Bolton); H. Patterson (Addenbrooke's Hospital, Cambridge); S. Dixit (Diana Princess of Wales Hospital, Grimsby); D. Fermont, R. Shah (Mount Vernon Hospital, Northwood); C. Read (Royal Preston Hospital, Preston); A. El-Modir (Sandwell General Hospital, West Bromwich); S. Beesley (Conquest Hospital, St Leonards-on-Sea); D. Cole (The Great Western Hospital, Swindon); R. Wilson (Taunton and Somerset Hospital, Taunton); C. Humber (Warwick Hospital, Warwick); A. Samanci, B. Waymont (New Cross Hospital, Wolverhampton); D. Bissett (Aberdeen Royal Infirmary, Aberdeen); P. Chakraborti (Queen's Hospital, Burton-upon-Trent); F. Mckinna (Eastbourne Hospital, Eastbourne); I. Syndikus (Southport and Formby District General Hospital, Southport); J. Glaholm (Good Hope Hospital, Sutton Coldfield). Russia—O. Kariakine (MRRC RAMS, Obninsk). New Zealand—C. Jose (Auckland Hospital, Auckland). South Africa—A. Abratt (Groote Schuur Hospital, Cape Town).

INTRODUCTION A key component of cancer control is screening, and significant research is underway to develop highly sensitive and specific tests that are minimally invasive. Circulating biomarkers have a major role in this effort. Many are not specific to the cancer because they are altered in other malignant or benign conditions. Therefore, it is essential to carefully define the cutoff for abnormality. Frequently, biomarker levels are interpreted by using a single-threshold rule developed in the context of differential diagnosis of clinically presenting cancers. Biomarker velocity, which can be significantly different in patients with cancer compared with controls1 is often ignored. Where it has been used, the data may be conflicting as they are for prostate-specific antigen velocity in prostate cancer2–4 or limited as they are for ovarian cancer.1,5,6 Modeling studies5,6 that use data from the Prostate, Lung, Colorectal and Ovarian (PLCO) Cancer Screening Trial7 suggest that up to a third of the ovarian cancer cases could have been detected earlier if cancer antigen 125 (CA-125) velocity had been used instead of a fixed cutoff.

d as they are for ovarian cancer.1,5,6 Modeling studies5,6 that use data from the Prostate, Lung, Colorectal and Ovarian (PLCO) Cancer Screening Trial7 suggest that up to a third of the ovarian cancer cases could have been detected earlier if cancer antigen 125 (CA-125) velocity had been used instead of a fixed cutoff. In the multimodal screening (MMS) arm of the United Kingdom Collaborative Trial of Ovarian Cancer Screening (UKCTOCS), women underwent serial serum CA-125 testing.8,9 CA-125 velocity was interpreted by using a risk of ovarian cancer algorithm (ROCA), which compares an individual's serial profile with that of cases and controls to estimate the risk of having ovarian cancer.10 We report here on the impact of using CA-125 velocity compared with a single-threshold rule on ovarian cancer detection during 296,911 woman-years of annual incidence screening. PATIENTS AND METHODS The trial was approved by the United Kingdom North West Multicentre Research Ethics Committee (International Standard Randomized Controlled Trial Number ISRCTN22488978 and ClinicalTrials.gov NCT00058032). Trial design, including details of recruitment and randomization, and the results of the initial (prevalence) screen have been described elsewhere.8,9 All women provided written informed consent.

ics Committee (International Standard Randomized Controlled Trial Number ISRCTN22488978 and ClinicalTrials.gov NCT00058032). Trial design, including details of recruitment and randomization, and the results of the initial (prevalence) screen have been described elsewhere.8,9 All women provided written informed consent. In brief, between 2001 and 2005, 202,638 women were randomly assigned, 50,640 of whom were allocated to the MMS group. Of those, 50,078 (98.9%) underwent a prevalence screen (Fig 1). The sensitivity, specificity, and positive predictive values (PPVs) for detection of invasive epithelial ovarian and/or tubal cancers (iEOCs) within 1 year of first screen (the prevalence screen) were 89.5%, 99.8%, and 35.1%, respectively.9 Fig 1. CONSORT diagram.

In brief, between 2001 and 2005, 202,638 women were randomly assigned, 50,640 of whom were allocated to the MMS group. Of those, 50,078 (98.9%) underwent a prevalence screen (Fig 1). The sensitivity, specificity, and positive predictive values (PPVs) for detection of invasive epithelial ovarian and/or tubal cancers (iEOCs) within 1 year of first screen (the prevalence screen) were 89.5%, 99.8%, and 35.1%, respectively.9 Fig 1. CONSORT diagram. MMS Strategy Following the initial prevalence screen, trial participants underwent an annual blood test on the anniversary of the randomization date. Serum CA-125 (level I screen) was measured by using an electrochemiluminescence sandwich immunoassay (Catalog No. 11776223 322; Roche Diagnostics, Mannheim, Germany).10 The screening protocol and management of screen-detected abnormalities have been previously described8,9 and are illustrated in Figure 2. In brief, at the annual screen, women were triaged as follows: risk of ovarian cancer (ROC) normal, return to annual screening; ROC intermediate, repeat CA-125 (repeat level I screen) in 12 weeks; and ROC elevated, repeat CA-125 and transvaginal scan (TVS; level II screen) in 6 weeks, with earlier screens arranged when results are suggestive of clinical disease. At level II screen, women with normal or intermediate ROC and a normal scan were returned to annual screening, whereas those with elevated ROC and a normal scan or an unsatisfactory scan irrespective of ROC had a repeat level II screen in 6 weeks. Those with abnormal scans irrespective of ROC were referred for clinical assessment. At repeat level II, women were again triaged to annual screening or clinical assessment (Fig 2). Women with an ROC of more than 1 in 5 (severe risk) were recommended to have surgery irrespective of scan findings.

II screen in 6 weeks. Those with abnormal scans irrespective of ROC were referred for clinical assessment. At repeat level II, women were again triaged to annual screening or clinical assessment (Fig 2). Women with an ROC of more than 1 in 5 (severe risk) were recommended to have surgery irrespective of scan findings. Fig 2. Multimodal screening algorithm and outcome of incidence screening. A, abnormal; CA-125, cancer antigen 125; CE, clinical evaluation; E, elevated; I, intermediate; LI, level I CA-125 test; LII, level II transvaginal scan (TVS) and CA-125 test; N, normal; ROC, risk of ovarian cancer; S, severe; SD, screening discontinued; U, unsatisfactory. The protocol was strictly enforced by using a custom-built Web-based trial management system with central classification of results, subsequent actions, and automated screening appointments.9 At study conception, the ROC cutoffs (< 1 in 1,818 and < 1 in 500) were set to allow approximately 15% and 2% of women to be triaged at annual screen to intermediate and elevated ROC groups, respectively. In April 2005, on the basis of data analysis on the performance of ROCA within UKCTOCS, the cutoffs were decreased to less than 1 in 3,500 and less than 1 in 1,000, respectively, to maintain the target proportions for triage.

of women to be triaged at annual screen to intermediate and elevated ROC groups, respectively. In April 2005, on the basis of data analysis on the performance of ROCA within UKCTOCS, the cutoffs were decreased to less than 1 in 3,500 and less than 1 in 1,000, respectively, to maintain the target proportions for triage. Clinical Assessment, Surgery, and Conservative Management Clinical assessment and appropriate investigations were undertaken locally by a designated clinician. The latter included repeat CA-125, imaging (TVS with Doppler ultrasound, computed tomography, and/or magnetic resonance imaging of abdomen and pelvis) and other tumor markers. In women with severe ROC, a chest computed tomography scan and mammogram were also requested. All women who were thought to have cancer were discussed at the local gynecologic oncology multidisciplinary team meeting. If surgery was recommended, laparoscopic bilateral salpingo-oophorectomy was performed unless the assessment was definitively suggestive of ovarian cancer or the procedure was inappropriate for other reasons, in which case laparotomy was preferred. Women who underwent bilateral salpingo-oophorectomy and were found to have ovarian and/or tubal cancer had completion surgery with staging. In those who did not have surgery, the coordinating center was informed of the follow-up plan, which usually involved repeat CA-125/TVS every 3 months. When clinicians felt reassured that the woman was unlikely to have ovarian cancer, she was returned to annual screening within UKCTOCS.

completion surgery with staging. In those who did not have surgery, the coordinating center was informed of the follow-up plan, which usually involved repeat CA-125/TVS every 3 months. When clinicians felt reassured that the woman was unlikely to have ovarian cancer, she was returned to annual screening within UKCTOCS. All center staff were asked to report intra- and postoperative complications, return to operating theater, and readmissions by using standard UKCTOCS forms and to report any serious adverse events to a designated safety officer. In addition, coordinating center staff reviewed medical notes and follow-up questionnaire responses to capture any additional complications. All were independently reviewed by a senior trial gynecologic oncologist blinded to the randomization group.

d to report any serious adverse events to a designated safety officer. In addition, coordinating center staff reviewed medical notes and follow-up questionnaire responses to capture any additional complications. All were independently reviewed by a senior trial gynecologic oncologist blinded to the randomization group. Confirmation of Diagnosis In all women who underwent screen-positive surgery, copies of medical records including surgery notes, discharge letters, and histopathology and/or cytology reports were obtained as previously described.9 For women who resided in England, additional information up to March 31, 2010, was obtained from the Hospital Episode Statistics.11 In women diagnosed with cancer, further information was obtained, including the discharge summary, multidisciplinary team meeting notes, and other correspondence. These reports were also obtained for all women when there was notification through cancer registry, death certificate, follow-up questionnaire, or personal communication of a possible ovarian or tubal cancer (International Classification of Diseases and Related Health Problems [10th revision; ICD-10] codes; Appendix Table A1, online only). The case notes of all these individuals were reviewed using a strict protocol by an Outcomes Review Committee (two pathologists and two gynecologic oncologists) who were blinded to the randomization group. They confirmed the final diagnosis, stage, and morphology of any cancer and, when possible, they classified iEOCs into type I (low-grade serous, low-grade endometrioid, mucinous, and clear cell cancers) or type II (high-grade serous, high-grade endometrioid, carcinosarcomas, and undifferentiated carcinoma) cancers.12 Where it was not possible to delineate whether the primary site was ovary, fallopian tube, or peritoneum, the diagnosis was classified as undesignated.13

de endometrioid, mucinous, and clear cell cancers) or type II (high-grade serous, high-grade endometrioid, carcinosarcomas, and undifferentiated carcinoma) cancers.12 Where it was not possible to delineate whether the primary site was ovary, fallopian tube, or peritoneum, the diagnosis was classified as undesignated.13 Follow-Up All volunteers were followed up through their National Health Service number by the appropriate national agencies for cancer registrations and/or deaths as well as by postal questionnaires 3 to 5 years after randomization and 2 years after the end of screening in the trial.9 The most recent cancer registrations for this analysis were received on June 17, 2014 (England and Wales), and July 2, 2014 (Northern Ireland).

ional agencies for cancer registrations and/or deaths as well as by postal questionnaires 3 to 5 years after randomization and 2 years after the end of screening in the trial.9 The most recent cancer registrations for this analysis were received on June 17, 2014 (England and Wales), and July 2, 2014 (Northern Ireland). Analysis A screen was defined as a single or series of serum CA-125 assays with or without scans culminating in surgery or return to annual screening. All women were censored at 1 year from last scan and/or CA-125 assay performed during their last incidence screen. The screen was considered positive (screen positive) if the woman had surgery or image-guided biopsy as a result of screening. Included in this category were women who were found to have ovarian lesions during imaging for other disease and who underwent surgery while awaiting repeat testing. The primary outcome for this analysis was primary iEOC diagnosed within 12 months of the last test in the incidence screen. Women with primary peritoneal cancer, borderline or nonepithelial ovarian cancers, and ovarian neoplasms of uncertain behavior were not included as true positives in the primary outcome analysis. A screen-detected cancer was one that resulted from screen-positive surgery and/or biopsy. A screen-negative/interval cancer was one diagnosed clinically within 12 months of the last test in the incidence screen in women returned to annual screening.

were not included as true positives in the primary outcome analysis. A screen-detected cancer was one that resulted from screen-positive surgery and/or biopsy. A screen-negative/interval cancer was one diagnosed clinically within 12 months of the last test in the incidence screen in women returned to annual screening. Overall sensitivity, specificity, PPV, and descriptive statistics for MMS were calculated for iEOCs and for all primary malignant ovarian and fallopian tube cancers (including borderline tumors and nonepithelial ovarian cancers).9 Receiver operating characteristic curves were constructed to compare the performance characteristics of annual serum CA-125 interpreted by using the ROCA with that of CA-125 interpreted by using several normal fixed cutoffs in this population, specifically, more than 35, more than 30, and more than 22 U/mL. A test for the difference in the area under the curves (AUCs) was performed as described by DeLong et al.14 RESULTS The CONSORT diagram (Fig 1) shows that 46,237 (91.3%) of the 50,640 women randomly assigned to the MMS arm participated in incidence screening. Between June 25, 2002, and December 21, 2011, 296,911 incidence screens were undertaken. Appendix Table A2 (online only) lists the reasons for screens that were not performed. The median number of incidence screens was seven (range, one to 10; interquartile range [IQR], six to eight). Median follow-up from the last incidence screen to latest cancer registration update was 3.1 years (IQR, 2.8 to 4.1 years).

Appendix Table A2 (online only) lists the reasons for screens that were not performed. The median number of incidence screens was seven (range, one to 10; interquartile range [IQR], six to eight). Median follow-up from the last incidence screen to latest cancer registration update was 3.1 years (IQR, 2.8 to 4.1 years). Figure 2 and Appendix Table A3 (online only) summarize the results. In all, 10.0% (29,584 of 296,911 involving 20,485 of 46,237 volunteers) of annual screens resulted in a recommendation for a repeat screen. Use of a single-threshold rule for CA-125 of more than 35, more than 30, or more than 22 U/mL would have resulted in 1.9% (5,597 of 296,911 involving 2,253 of 46,237 volunteers), 3.3% (9,699 of 296,911 involving 3,537 of 46,237 volunteers), and 9.7% (28,757 of 296,911 involving 8,596 of 46,237 volunteers), respectively, of screens needing to be repeated. Overall, 1,085 (0.4% of 296,911) of these screening episodes were not completed because women died (n = 95), changed their minds/moved away/did not attend repeat appointments (n = 689), were diagnosed with nonovarian cancer (n = 169) or other disease (n = 29), or had their ovaries removed as part of surgery for other conditions (n = 20).

% of 296,911) of these screening episodes were not completed because women died (n = 95), changed their minds/moved away/did not attend repeat appointments (n = 689), were diagnosed with nonovarian cancer (n = 169) or other disease (n = 29), or had their ovaries removed as part of surgery for other conditions (n = 20). Clinical evaluation was performed in 1.1% (3,329 of 296,911 involving 3,078 of 46,237 volunteers) of the screens (Fig 2). In 507 patients this was limited to assessment of screen results and return to annual screening. The remaining 2,822 screens resulted in clinical assessment; 3.6% (102 of 2,822) of the assessments were undertaken instead of protocol-mandated repeat testing. Reasons stated included CA-125 levels ≥ 50 U/mL (n = 50), elevated ROC (n = 68), or both (n = 38) and patient anxiety and/or symptoms suggestive of ovarian cancer (n = 22).

2,822 screens resulted in clinical assessment; 3.6% (102 of 2,822) of the assessments were undertaken instead of protocol-mandated repeat testing. Reasons stated included CA-125 levels ≥ 50 U/mL (n = 50), elevated ROC (n = 68), or both (n = 38) and patient anxiety and/or symptoms suggestive of ovarian cancer (n = 22). A proportion of the screens (0.2%; 640 of 296,911) resulted in women having screen-positive surgery, 64.8% (415 of 640) of which was laparoscopic. Primary ovarian and/or tubal malignancies were detected in 154 (24.1%) of the 640 women (Table 1). The latter included 133 iEOCs, 17 borderline, and four nonepithelial ovarian cancers. Two of the 154 women had incomplete screening episodes, and ovarian cancer (one iEOC, one nonepithelial microscopic granulosa tumor) was diagnosed in the course of imaging for renal disease and surgery for endometrial cancer, respectively, while awaiting repeat testing. Thirty-two interval ovarian or tubal cancers (22 iEOCs, nine borderline ovarian tumors, and one nonepithelial ovarian cancer) were diagnosed clinically within 12 months of the last incidence screen test. The 22 iEOCs include a protocol deviation in which the clinical team returned an asymptomatic woman estimated by ROCA to be at severe risk (one in four) to annual screening. Her CA-125 was 29 U/mL and pelvic magnetic resonance imaging was normal. Eight months later, she presented symptomatically with high-grade serous stage III iEOC (Table 1). In addition, a second woman was classified as intermediate risk by ROCA at both annual and first repeat screens but then classified as normal risk on her second repeat sample, at which point she was returned to annual screening. Eleven months later, she was diagnosed with stage IV high-grade serous cancer. An additional 21 iEOCs were diagnosed 12 to 24 months after the last annual screen.

at both annual and first repeat screens but then classified as normal risk on her second repeat sample, at which point she was returned to annual screening. Eleven months later, she was diagnosed with stage IV high-grade serous cancer. An additional 21 iEOCs were diagnosed 12 to 24 months after the last annual screen. Table 1. Pathologic Findings and CA-125 at Relevant Annual Screen (level I) in Screen-Positive Women and Screen-Negative Women (those with interval cancers)

at both annual and first repeat screens but then classified as normal risk on her second repeat sample, at which point she was returned to annual screening. Eleven months later, she was diagnosed with stage IV high-grade serous cancer. An additional 21 iEOCs were diagnosed 12 to 24 months after the last annual screen. Table 1. Pathologic Findings and CA-125 at Relevant Annual Screen (level I) in Screen-Positive Women and Screen-Negative Women (those with interval cancers) Outcome of Screen-Positive Surgery Total No. of Women Annual CA-125 < 35 U/mL ≥ 35 U/mL Total No. of women 640 455 185 Total No. of women with normal or benign pathology 441 344 97 Laparoscopy, ovaries normal, not removed* 13 12 1 Normal ovaries† 133 106 27 Benign ovarian pathology‡ 295 226 69 Total No. of nonovarian malignant neoplasms 45 24 21 Ovarian neoplasm of uncertain behavior (ICD D39.1) 2 2 0 Primary peritoneal cancer (ICD C48.2) 12 6 6 Other nonovarian and/or tubal cancer involving ovaries (secondary ovarian neoplasm)* 12 6 6 Other nonovarian and/or tubal cancer not involving ovaries§ 19 10 9 Total No. of screen-positive women diagnosed with malignant neoplasm of ovary (ICD C56) and fallopian tube (ICD C57.0) 154 87 67 Nonepithelial neoplasm of ovary (ICD C56) 4 3 1 Primary borderline epithelial neoplasm of ovary (ICD C56) 17 14 3 Primary invasive epithelial neoplasm of ovary (ICD C56) 113 56 57 Primary invasive epithelial neoplasm of fallopian tube (ICD C57.0) 11 8 3 Undesignated (unable to delineate whether primary site is ovary, fallopian tube, or peritoneum) 9 6 3 Total No. of women with screen-negative (interval) malignant neoplasm of ovary (ICD C56) or fallopian tube (ICD C57.0) diagnosed within 1 year of end of screen 32 31 1 Nonepithelial neoplasm of ovary (ICD C56) 1 1 0 Borderline epithelial neoplasm of ovary (ICD C56) 9 9 0 Primary invasive epithelial neoplasm of ovary (ICD C56) 18 17 1 Primary invasive epithelial neoplasm of fallopian tube (ICD C57.0) 1 1 0 Undesignated (unable to delineate whether the primary site is ovary, fallopian tube, or peritoneum) 3 3 0 Abbreviations: CA-125, cancer antigen 125; ICD, International Statistical Classification of Diseases and Related Health Problems (10th revision).

ary invasive epithelial neoplasm of fallopian tube (ICD C57.0) 1 1 0 Undesignated (unable to delineate whether the primary site is ovary, fallopian tube, or peritoneum) 3 3 0 Abbreviations: CA-125, cancer antigen 125; ICD, International Statistical Classification of Diseases and Related Health Problems (10th revision). * Includes a volunteer who had ultrasound-guided aspiration of ascites in her year 4 screen with normal cytology and was returned to annual screening. In her next screen, she had screen-positive laparotomy with a final diagnosis of colorectal primary metastatic to the ovaries. † Includes five women with para-tubal cysts, three with benign hydrosalpinx, one with mucinous cystadenoma of the appendix, and one with tumor-bearing endometrium. ‡ Includes one volunteer who had benign ovarian cysts at surgery. However, CA-125 continued to increase, and 1 year later, she was diagnosed with primary peritoneal cancer. § Includes six women who also had benign ovarian pathology.

† Includes five women with para-tubal cysts, three with benign hydrosalpinx, one with mucinous cystadenoma of the appendix, and one with tumor-bearing endometrium. ‡ Includes one volunteer who had benign ovarian cysts at surgery. However, CA-125 continued to increase, and 1 year later, she was diagnosed with primary peritoneal cancer. § Includes six women who also had benign ovarian pathology. At the relevant annual screen, median serum CA-125 in the 133 women with screen-detected iEOCs was 33.6 U/mL (IQR, 21.3 to 109.2). Seventy (52.6%) of 133 of these women had CA-125 levels within the normal range (≤ 35 U/mL; subgroup A), and the remaining 63 (47.4%) had increased CA-125 levels (> 35 U/mL; subgroup B; Table 2). Only one of the 22 women who had an interval iEOC had a CA-125 level more than 35 U/mL (36.9 U/mL). These results are shown graphically in Figure 3, in which the serial annual CA-125 levels of all screen-positive (n = 133) and screen-negative patients with iEOC (n = 22) are plotted with the annual CA-125 levels for all other women shown as a scatterplot. The ROCA had a significantly larger area under the curve (0.915) than the individual CA-125 measurements (0.869; P = .0027; Fig 4). The sensitivity of ROCA alone was 87.1% (95% CI, 80.8% to 91.9%) and that of using annual serum CA-125 cutoffs of more than 35, more than 30, and more than 22 U/mL were 41.3%, (95% CI, 33.5% to 49.5%), 48.4% (95% CI, 40.3% to 56.5%), and 66.5% (95% CI, 58.4% to 73.8%), respectively. The specificity of annual ROCA alone was 87.6%. At the same specificity, the sensitivity of the annual CA-125 cutoff (20.99 U/mL) was 68.4%.

utoffs of more than 35, more than 30, and more than 22 U/mL were 41.3%, (95% CI, 33.5% to 49.5%), 48.4% (95% CI, 40.3% to 56.5%), and 66.5% (95% CI, 58.4% to 73.8%), respectively. The specificity of annual ROCA alone was 87.6%. At the same specificity, the sensitivity of the annual CA-125 cutoff (20.99 U/mL) was 68.4%. Table 2. CA-125 at the Relevant Annual Screen by Stage and Type of Primary Invasive Epithelial Ovarian and Tubal Cancers

utoffs of more than 35, more than 30, and more than 22 U/mL were 41.3%, (95% CI, 33.5% to 49.5%), 48.4% (95% CI, 40.3% to 56.5%), and 66.5% (95% CI, 58.4% to 73.8%), respectively. The specificity of annual ROCA alone was 87.6%. At the same specificity, the sensitivity of the annual CA-125 cutoff (20.99 U/mL) was 68.4%. Table 2. CA-125 at the Relevant Annual Screen by Stage and Type of Primary Invasive Epithelial Ovarian and Tubal Cancers Characteristic Screen-Detected Status Positive Negative All Annual CA-125 < 35 U/mL (subgroup A) Annual CA-125 ≥ 35 U/mL (subgroup B) All No. % 95% CI No. % 95% CI No. % 95% CI No. % 95% CI Total No. of women 133 70 63 22 Serum CA-125 at corresponding annual screen, U/mL Median 33.6 21.8 112.1 13.6 IQR 21.3-109.2 16.5-26.3 66.4-375.4 11-20.8 ROC at corresponding annual screen Normal risk 0 0 0 0 0 0 20 90.9 Intermediate risk 37 27.8 33 47.1 4 6.3 1 4.5 Elevated risk 96 72.2 38 54.3 58 92.1 1 4.5 Stage I 35 22 13 4 II 20 12 8 2 III 68 32 36 11 IIIa 6 2 4 0 IIIb 16 11 5 3 IIIc 46 19 27 8 IV 10 4 6 5 Early stage (I or II) 41.4 32.9 to 50.2 48.6 36.4 to 60.8 33.3 22.0 to 46.3 27.3 10.7 to 50.2 Morphology Total type I iEOC 19 10 9 5 Low-grade serous 5 1 4 0 Low-grade endometrioid 8 4 4 1 Clear cell 5 4 1 4 Mucinous 1 1 0 0 Total type II iEOC 109 58 51 11 High-grade serous* 89 44 45 8 High-grade endometrioid 8 5 3 1 Unspecified adenocarcinoma 10 8 2 1 Carcinosarcoma (MMT) 2 1 1 1 Unclassified† 5 2 3 6 Abbreviations: CA-125, cancer antigen 125; iEOC, invasive epithelial ovarian and/or tubal cancer; IQR, interquartile range; MMT, malignant mesenchymal tumor; ROC, risk of ovarian cancer.

us* 89 44 45 8 High-grade endometrioid 8 5 3 1 Unspecified adenocarcinoma 10 8 2 1 Carcinosarcoma (MMT) 2 1 1 1 Unclassified† 5 2 3 6 Abbreviations: CA-125, cancer antigen 125; iEOC, invasive epithelial ovarian and/or tubal cancer; IQR, interquartile range; MMT, malignant mesenchymal tumor; ROC, risk of ovarian cancer. * Includes a case reported as mixed high-grade adenocarcinoma with serous and clear cell features and focal squamous differentiation. † Morphology could not be determined because only cytology was undertaken. Fig 3. Plot of all multimodal screening annual cancer antigen 125 (CA-125) measurements over time on a log scale, including truncation. Superimposed are the serial measurements for 155 invasive epithelial ovarian and/or tubal cancers with the large circles representing the final screen before diagnosis, either true positive (n = 133; gold lines and markers) or false negative (n = 22; blue lines and markers). The red line indicates one patient in whom the risk of ovarian cancer algorithm recommended surgery, but it was not performed following clinical evaluation. The black horizontal lines represent CA-125 cutoffs of 35, 30, and 22 U/mL. NOTE. 262 CA-125 values truncated above 100 U/mL and 174 CA-125 values truncated below 2 U/mL.

line indicates one patient in whom the risk of ovarian cancer algorithm recommended surgery, but it was not performed following clinical evaluation. The black horizontal lines represent CA-125 cutoffs of 35, 30, and 22 U/mL. NOTE. 262 CA-125 values truncated above 100 U/mL and 174 CA-125 values truncated below 2 U/mL. Fig 4. Risk of ovarian cancer (ROC) curves based on the performance characteristics of annual cancer antigen 125 (CA125) measurement alone and annual risk of ovarian cancer algorithm (ROCA) score alone. Overlaid points represent the actual characteristics of the multimodal screening strategy, hypothetical characteristics of annual ROCA classified as normal or abnormal (intermediate/elevated) risk, hypothetical characteristics of annual CA125 using the cutoff points of more than 35 U/mL (as in the Prostate, Lung, Colorectal and Ovarian Cancer Screening Trial), more than 30 U/mL (in clinical use), and more than 22 U/mL (as suggested by other groups), respectively. P value of .0027 is test of difference. Of the screen-detected iEOCs, 82.0% (109 of 133) were type II. The distributions of type I and type II cancers in the A and B subgroups were similar (Table 2). Fifty-five (41.4%) of 133 patients with iEOCs were diagnosed in stage I to II (Table 2). A greater proportion (P = .075) in subgroup A (48.6%; 34 of 70) were early-stage (stage I to II) cancers compared with subgroup B (33.3%; 21 of 63).

ons of type I and type II cancers in the A and B subgroups were similar (Table 2). Fifty-five (41.4%) of 133 patients with iEOCs were diagnosed in stage I to II (Table 2). A greater proportion (P = .075) in subgroup A (48.6%; 34 of 70) were early-stage (stage I to II) cancers compared with subgroup B (33.3%; 21 of 63). Overall, in women with screen-detected iEOCs, the median time from last screen test to surgery was 8 weeks (IQR, 4.9 to 13.7 weeks), and the median time from the start of the relevant annual screen (level I) to surgery was 20 weeks (IQR, 11 to 34 weeks). In subgroup A, the interval was significantly (P < .0001) longer (30 weeks; IQR, 18 to 43 weeks) compared with subgroup B (12 weeks; IQR, 7 to 19 weeks). This difference reflects the greater proportion of cases undergoing repeat screens following an intermediate ROC at annual screen in subgroup A (33 of 70) compared with subgroup B (four of 63; Table 2). The overall sensitivity and specificity of MMS for iEOCs were 85.8% (95% CI, 79.3% to 90.9%) and 99.8% (95% CI, 99.8% to 99.8%), respectively, with 4.8 surgeries per iEOC detected during incidence screening (Table 3). If the 12 screen-detected and three screen-negative primary peritoneal cancers (PPCs) were included, sensitivity, specificity, and PPV were 85.3% (95% CI, 79.1% to 90.3%), 99.8% (95% CI, 99.8% to 99.8%), and 22.7% (95% CI, 19.5% to 26.1%), respectively. If we extended performance characteristics to include iEOCs diagnosed up to 24 months from date of last scan/CA-125 assay performed during incidence screening, sensitivity for iEOCs was 74.4%.

and PPV were 85.3% (95% CI, 79.1% to 90.3%), 99.8% (95% CI, 99.8% to 99.8%), and 22.7% (95% CI, 19.5% to 26.1%), respectively. If we extended performance characteristics to include iEOCs diagnosed up to 24 months from date of last scan/CA-125 assay performed during incidence screening, sensitivity for iEOCs was 74.4%. Table 3. Performance Characteristics of MMS Incidence Screening for Malignant Ovarian (C56), Tubal (C57.0), and Primary Peritoneal (C48.2) Neoplasm Characteristic Ovarian and Fallopian Tube Cancers Ovarian, Fallopian Tube, and Primary Peritoneal Cancers No. 95% CI No. 95% CI No. of women-years 296,911 296,911 No. of surgeries 640 640 Primary ovarian (C56) and tubal (C57.0) malignancies and primary peritoneal cancer (C48.2) within 1 year of screen (includes borderline and ovarian neoplasm of uncertain behavior) Screen positive 154 166 Screen negative 32 35 Sensitivity 82.8 76.6 to 87.9 82.6 76.6 to 87.6 Specificity 99.8 99.8 to 99.9 99.8 99.8 to 99.9 PPV 24.1 20.8 to 27.6 25.9 22.6 to 29.5 No. of operations per screen positive 4.2 3.9 Primary invasive epithelial ovarian, tubal, and primary peritoneal malignancies within 1 year of screen (excludes borderline epithelial ovarian neoplasms) Screen positive 133 145 Screen negative 22 25 Sensitivity 85.8 79.3 to 90.9 85.3 79.1 to 90.3 Specificity 99.8 99.8 to 99.8 99.8 99.8 to 99.8 PPV 20.8 17.7 to 24.1 22.7 19.5 to 26.1 No. of operations per screen positive 4.8 4.4 NOTE. All codes are International Statistical Classification of Diseases and Related Health Problems, 10th revision (ICD-10).

gative 22 25 Sensitivity 85.8 79.3 to 90.9 85.3 79.1 to 90.3 Specificity 99.8 99.8 to 99.8 99.8 99.8 to 99.8 PPV 20.8 17.7 to 24.1 22.7 19.5 to 26.1 No. of operations per screen positive 4.8 4.4 NOTE. All codes are International Statistical Classification of Diseases and Related Health Problems, 10th revision (ICD-10). Abbreviations: MMS, multimodal strategy; PPV, positive predictive value. Of the 640 women who had screen-positive surgery, 31 had nonovarian cancers, and 441 had normal or benign pathology (Table 1; Appendix Table A4, online only). An intraoperative or early postoperative complication was reported in 20 of the 441 women (4.5%; 95% CI, 2.8% to 6.9%). Twelve of these women had a major complication or significant sequelae (Appendix Table A5, online only).

rian cancers, and 441 had normal or benign pathology (Table 1; Appendix Table A4, online only). An intraoperative or early postoperative complication was reported in 20 of the 441 women (4.5%; 95% CI, 2.8% to 6.9%). Twelve of these women had a major complication or significant sequelae (Appendix Table A5, online only). DISCUSSION In the largest ovarian cancer screening trial that we are aware of, a risk algorithm using serial biomarker measurement doubled the number of screen-detected cancers compared with a single-threshold rule. Of the 155 women with iEOCs, the ROCA detected 86.4% whereas using annual serum CA-125 fixed cutoffs of more than 35, more than 30, and more than 22 U/mL would have identified only 41.3%, 48.4%, and 66.5%, respectively. Our data provide prospective evidence of the improvement that CA-125 velocity analysis brings to iEOC detection compared with a predetermined cutoff. The impact of such screening on ovarian cancer mortality will be known later in 2015 when follow-up is complete. However, our current findings are of immediate importance because they highlight the need to examine serial change in biomarker levels in the context of screening and early detection of cancer. Reliance on predefined single-threshold rules may result in biomarkers of value being discarded.

5 when follow-up is complete. However, our current findings are of immediate importance because they highlight the need to examine serial change in biomarker levels in the context of screening and early detection of cancer. Reliance on predefined single-threshold rules may result in biomarkers of value being discarded. The encouraging sensitivity (85.8%) and specificity (99.8%) for detecting iEOCs in low-risk postmenopausal women noted during the prevalence screen persisted during incidence screening.9 The high sensitivity remained even when PPC was included as an outcome measure. This was reassuring given that PPC probably shares common origins with primary high-grade serous iEOCs.15 The ROCA increases sensitivity by personalizing the interpretation of serial biomarker values. This explains the higher sensitivity observed in our trial compared with other trials in which a single-threshold CA-125 rule was used—67% in the PLCO trial16 (four rounds of screening including prevalence) and 77% in the Shizuoka Cohort Study.17 Overall, 41.4% (55 of 133) of women were detected with stage I or II disease. A majority (82.0%) of screen-detected iEOCs were aggressive type II, which are associated with the highest mortality rates.18 This is reassuring, given the concern that screening detects more indolent cancers. In the Shizuoka Cohort Study, 48% of screen-detected cancers were type I mucinous and clear cell iEOCs.17

ase. A majority (82.0%) of screen-detected iEOCs were aggressive type II, which are associated with the highest mortality rates.18 This is reassuring, given the concern that screening detects more indolent cancers. In the Shizuoka Cohort Study, 48% of screen-detected cancers were type I mucinous and clear cell iEOCs.17 The strategy involved at least one repeat test such that the median time from annual screen to surgery was 20 weeks. The interval was significantly longer in subgroup A (30 weeks) compared with subgroup B (12 weeks) because women with annual CA-125 levels in the normal range required more repeat testing. Despite this, there was a higher proportion of stage I or II iEOCs in subgroup A. The latter coupled with the fact that ovarian cancers double every two and half months,19 suggests that modifications to the screening strategy that could decrease this interval may have an additional impact on tumor stage and volume. This could include decreasing the 3-month interval to repeat CA-125 testing following an intermediate ROC and measuring levels of a second blood biomarker such as HE420,21 in intermediate-risk annual samples. Although HE4 does not improve CA-125 lead time,22,23 it could help confirm ovarian cancer risk and reduce time to surgery. In the presence of an increasing CA-125, HE4 was increased in samples from 27 of 39 women with ovarian cancer in the PLCO trial.21 TVS does not seem to have the resolution to detect iEOC at low CA-125 levels. Twenty-nine (41%) of 70 women with iEOCs in subgroup A had no abnormality on the initial level II scan, and TVS was abnormal in only 17 of the 39 women in the study by Urban et al.21 The potential of newer technology such as contrast-enhanced TVS with targeted microbubbles warrants assessment in this context.24

levels. Twenty-nine (41%) of 70 women with iEOCs in subgroup A had no abnormality on the initial level II scan, and TVS was abnormal in only 17 of the 39 women in the study by Urban et al.21 The potential of newer technology such as contrast-enhanced TVS with targeted microbubbles warrants assessment in this context.24 For each iEOC detected, four additional women underwent surgery. These figures are slightly higher than previously reported in trials using the ROCA9,25,26 but lower than the 19.516 and 3317 surgeries undertaken for each cancer detected in trials using other screening strategies. Excess surgical morbidity in patients with false-positive diagnoses is a key concern, especially with increasing comorbidity in the older women. In our study, the rate of complications in women with benign or normal histology, most of whom underwent laparoscopic bilateral salpingo-oophorectomy, was 4.5%. Similar rates have been reported in women at high-risk of ovarian cancer undergoing risk-reducing salpingo-oophorectomy (3.9%).27

morbidity in the older women. In our study, the rate of complications in women with benign or normal histology, most of whom underwent laparoscopic bilateral salpingo-oophorectomy, was 4.5%. Similar rates have been reported in women at high-risk of ovarian cancer undergoing risk-reducing salpingo-oophorectomy (3.9%).27 Key strengths of our trial are the scale, the multicenter setting within the United Kingdom health service, detailed screening and management protocols implemented by a dedicated local and central team, Web-based bespoke trial management system, high compliance with screening, and independent blinded outcome review. Completeness of data on screen-negative and/or interval cancers in the year following the end of screening (2012) was ensured by postal follow-up of all women in April 2014, coupled with cancer registry updates in July 2014. The limitations relate mainly to the long duration, a necessary feature of randomized controlled trials with mortality as the primary end point, and the associated improvements in clinical management over that period. A healthy volunteer effect reduced the expected number of cancers in the control arm and thereby further lengthened the trial.28 However, although these issues are pertinent to this analysis, they will not affect the primary intention-to-treat mortality analysis. In conclusion, our data support use of velocity-based algorithms as opposed to a predefined single-threshold rule in cancer screening strategies that use blood biomarkers.

Key strengths of our trial are the scale, the multicenter setting within the United Kingdom health service, detailed screening and management protocols implemented by a dedicated local and central team, Web-based bespoke trial management system, high compliance with screening, and independent blinded outcome review. Completeness of data on screen-negative and/or interval cancers in the year following the end of screening (2012) was ensured by postal follow-up of all women in April 2014, coupled with cancer registry updates in July 2014. The limitations relate mainly to the long duration, a necessary feature of randomized controlled trials with mortality as the primary end point, and the associated improvements in clinical management over that period. A healthy volunteer effect reduced the expected number of cancers in the control arm and thereby further lengthened the trial.28 However, although these issues are pertinent to this analysis, they will not affect the primary intention-to-treat mortality analysis. In conclusion, our data support use of velocity-based algorithms as opposed to a predefined single-threshold rule in cancer screening strategies that use blood biomarkers. Supplementary Material Protocol Publisher's Note Supported by the Medical Research Council, Cancer Research United Kingdom, and the Department of Health, with additional support from the Eve Appeal, and by researchers at the National Institute for Health Research (NIHR) University College London Hospital (UCLH) Biomedical Research Centre. I.J. held an NIHR Senior Investigator Award. The United Kingdom Collaborative Trial of Ovarian Cancer Screening Biobank was funded in part by the Special Trustees of Bart's and the London and Special Trustees of UCLH. S.J.S. is supported by Grant No. CA152990 from the National Cancer Institute Early Detection Research Network.

d an NIHR Senior Investigator Award. The United Kingdom Collaborative Trial of Ovarian Cancer Screening Biobank was funded in part by the Special Trustees of Bart's and the London and Special Trustees of UCLH. S.J.S. is supported by Grant No. CA152990 from the National Cancer Institute Early Detection Research Network. Terms in blue are defined in the glossary, found at the end of this article and online at www.jco.org. Authors' disclosures of potential conflicts of interest are found in the article online at www.jco.org. Author contributions are found at the end of this article. Clinical trial information: ISRCTN22488978; NCT00058032. Acknowledgment We thank the women throughout the United Kingdom who are participating in the trial and the entire medical, nursing, and administrative staff who contributed to the implementation of United Kingdom Collaborative Trial of Ovarian Cancer Screening. We also thank the independent members of the Trial Steering Committee: David Luesley, MD (chair), City Hospital, Birmingham; Jack Cuzick, PhD, Queen Mary, University of London, London; Julietta Patnick, CBE, Public Health England, Sheffield; and Louise Bayne, RGN, Ovacome, London, United Kingdom; and the Data Monitoring and Ethics Committee: Peter Boyle, PhD (chair), International Prevention Research Institute, Lyon; Susanne Kjaer, MD, Danish Cancer Society Research Center, Copenhagen; A.P.M. Heintz, MD, PhD, University Hospital Utrecht, Utrecht; and Edward Trimble, MD, National Cancer Institute, Bethesda.

dom; and the Data Monitoring and Ethics Committee: Peter Boyle, PhD (chair), International Prevention Research Institute, Lyon; Susanne Kjaer, MD, Danish Cancer Society Research Center, Copenhagen; A.P.M. Heintz, MD, PhD, University Hospital Utrecht, Utrecht; and Edward Trimble, MD, National Cancer Institute, Bethesda. AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST Disclosures provided by the authors are available with this article at www.jco.org. AUTHOR CONTRIBUTIONS Conception and design: Usha Menon, David Oram, Jonathan Herod, Robert Woolas, Lesley Fallowfield, Alistair McGuire, Mahesh Parmar, Steven J. Skates, Ian Jacobs Provision of study materials or patients: Keith Godfrey, Alberto Lopes, David Oram, Jonathan Herod, Karin Williamson, Mourad W. Seif, Howard Jenkins, Tim Mould, Robert Woolas, John B. Murdoch, Stephen Dobbs, Nazar N. Amso, Simon Leeson, Derek Cruickshank, Ian Scott Collection and assembly of data: Usha Menon, Andy Ryan, Aleksandra Gentry-Maharaj, Anne Dawnay, Mariam Habib, Sophia Apostolidou, Naveena Singh, Matthew Burnell, Susan Davies, Richard Gunu, Keith Godfrey, Alberto Lopes, Karin Williamson, Mourad W. Seif, Howard Jenkins, Tim Mould, John B. Murdoch, Stephen Dobbs, Nazar N. Amso, Simon Leeson, Derek Cruickshank, Ian Scott, Martin Widschwendter, Stuart Campbell Data analysis and interpretation: Usha Menon, Andy Ryan, Jatinderpal Kalsi, Aleksandra Gentry-Maharaj, Elizabeth Benjamin, Matthew Burnell, Susan Davies, Aarti Sharma, Simon Leeson, Derek Cruickshank, Karina Reynolds, Mahesh Parmar, Steven J. Skates, Ian Jacobs Manuscript writing: All authors

Collection and assembly of data: Usha Menon, Andy Ryan, Aleksandra Gentry-Maharaj, Anne Dawnay, Mariam Habib, Sophia Apostolidou, Naveena Singh, Matthew Burnell, Susan Davies, Richard Gunu, Keith Godfrey, Alberto Lopes, Karin Williamson, Mourad W. Seif, Howard Jenkins, Tim Mould, John B. Murdoch, Stephen Dobbs, Nazar N. Amso, Simon Leeson, Derek Cruickshank, Ian Scott, Martin Widschwendter, Stuart Campbell Data analysis and interpretation: Usha Menon, Andy Ryan, Jatinderpal Kalsi, Aleksandra Gentry-Maharaj, Elizabeth Benjamin, Matthew Burnell, Susan Davies, Aarti Sharma, Simon Leeson, Derek Cruickshank, Karina Reynolds, Mahesh Parmar, Steven J. Skates, Ian Jacobs Manuscript writing: All authors Final approval of manuscript: All authors Glossary Terms biomarker:a functional biochemical or molecular indicator of a biologic or disease process that has predictive, diagnostic, and/or prognostic utility.

Data analysis and interpretation: Usha Menon, Andy Ryan, Jatinderpal Kalsi, Aleksandra Gentry-Maharaj, Elizabeth Benjamin, Matthew Burnell, Susan Davies, Aarti Sharma, Simon Leeson, Derek Cruickshank, Karina Reynolds, Mahesh Parmar, Steven J. Skates, Ian Jacobs Manuscript writing: All authors Final approval of manuscript: All authors Glossary Terms biomarker:a functional biochemical or molecular indicator of a biologic or disease process that has predictive, diagnostic, and/or prognostic utility. CA-125 (cancer antigen 125):a protein produced by the fallopian tubes, the endometrium, and the lining of the abdominal cavity (peritoneum). CA-125 is a tumor marker present in higher than normal amounts in the blood and urine of patients with certain cancers. Typically, women with ovarian cancer have high levels of CA-125. Other conditions associated with increased levels of CA-125 include endometriosis, pancreatitis, pregnancy, normal menstruation, and pelvic inflammatory disease. CA-125 levels may be used to help diagnose ovarian cancer and to determine whether these tumors are responding to therapy. The normal range for CA-125 is less than 35 U/mL and less than 20 U/mL for women who have been treated for ovarian cancer. Women with ovarian cancer may show values higher than 65 U/mL. positive predictive value:the probability of a positive test result being truly positive.

CA-125 (cancer antigen 125):a protein produced by the fallopian tubes, the endometrium, and the lining of the abdominal cavity (peritoneum). CA-125 is a tumor marker present in higher than normal amounts in the blood and urine of patients with certain cancers. Typically, women with ovarian cancer have high levels of CA-125. Other conditions associated with increased levels of CA-125 include endometriosis, pancreatitis, pregnancy, normal menstruation, and pelvic inflammatory disease. CA-125 levels may be used to help diagnose ovarian cancer and to determine whether these tumors are responding to therapy. The normal range for CA-125 is less than 35 U/mL and less than 20 U/mL for women who have been treated for ovarian cancer. Women with ovarian cancer may show values higher than 65 U/mL. positive predictive value:the probability of a positive test result being truly positive. AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST Risk Algorithm Using Serial Biomarker Measurements Doubles the Number of Screen-Detected Cancers Compared With a Single-Threshold Rule in the United Kingdom Collaborative Trial of Ovarian Cancer Screening The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or jco.ascopubs.org/site/ifc.

s are considered compensated. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or jco.ascopubs.org/site/ifc. Usha Menon Stock or Other Ownership: Abcodia Andy Ryan No relationship to disclose Jatinderpal Kalsi No relationship to disclose Aleksandra Gentry-Maharaj No relationship to disclose Anne Dawnay Employment: HCA International Travel, Accommodations, Expenses: Helena Biosciences Europe Mariam Habib No relationship to disclose Sophia Apostolidou No relationship to disclose Naveena Singh No relationship to disclose Elizabeth Benjamin Employment: HCA International Matthew Burnell No relationship to disclose Susan Davies No relationship to disclose Aarti Sharma No relationship to disclose Richard Gunu No relationship to disclose Keith Godfrey No relationship to disclose Alberto Lopes Consulting or Advisory Role: Sanofi Pasteur MSD, Roche David Oram No relationship to disclose Jonathan Herod No relationship to disclose Karin Williamson No relationship to disclose Mourad W. Seif Travel, Accommodations, Expenses: European Board and College of Obstetrics and Gynaecology, Assiut University, Alexandria University Howard Jenkins No relationship to disclose Tim Mould No relationship to disclose Robert Woolas No relationship to disclose John B. Murdoch Employment: J. B. Murdoch Leadership: J. B. Murdoch Stock or Other Ownership: J. B. Murdoch Stephen Dobbs No relationship to disclose Nazar N. Amso Stock or Other Ownership: MEDAPHORE

Mourad W. Seif Travel, Accommodations, Expenses: European Board and College of Obstetrics and Gynaecology, Assiut University, Alexandria University Howard Jenkins No relationship to disclose Tim Mould No relationship to disclose Robert Woolas No relationship to disclose John B. Murdoch Employment: J. B. Murdoch Leadership: J. B. Murdoch Stock or Other Ownership: J. B. Murdoch Stephen Dobbs No relationship to disclose Nazar N. Amso Stock or Other Ownership: MEDAPHORE Patents, Royalties, Other Intellectual Property: MEDAPHOR Simon Leeson Travel, Accommodations, Expenses: DySIS Medical Derek Cruickshank Consulting or Advisory Role: AstraZeneca Ian Scott No relationship to disclose Lesley Fallowfield Consulting or Advisory Role: Sanofi Oncology, Astellas Pharma, Bristol-Myers Squibb, Pfizer Speakers' Bureau: Teva Pharmaceuticals Research Funding: Boehringer Ingelheim, Roche Martin Widschwendter No relationship to disclose Karina Reynolds No relationship to disclose Alistair McGuire No relationship to disclose Stuart Campbell No relationship to disclose Mahesh Parmar No relationship to disclose Steven J. Skates Stock or Other Ownership: SISCAPA Assay Technologies Speakers' Bureau: AstraZeneca Patents, Royalties, Other Intellectual Property: Abcodia (Inst) Ian Jacobs Employment: Abcodia, Women's Health Specialists Patents, Royalties, Other Intellectual Property: ROC Algorithm Appendix Table A1. ICD-10 Codes for Notes Reviewed by the Outcomes Committee

Steven J. Skates Stock or Other Ownership: SISCAPA Assay Technologies Speakers' Bureau: AstraZeneca Patents, Royalties, Other Intellectual Property: Abcodia (Inst) Ian Jacobs Employment: Abcodia, Women's Health Specialists Patents, Royalties, Other Intellectual Property: ROC Algorithm Appendix Table A1. ICD-10 Codes for Notes Reviewed by the Outcomes Committee ICD-10 Code Description C56 Malignant neoplasm of ovary C57.0 Malignant neoplasm of fallopian tube C57.4 Uterine adnexa, unspecified C57.7 Other specified female genital organs C57.8 Malignant neoplasm of overlapping lesion of female genital organs C57.9 Malignant neoplasm of female genital organ, unspecified C48.0 Retroperitoneum C48.1 Specified parts of peritoneum C48.2 Malignant neoplasm of peritoneum, unspecified C48.8 Overlapping lesions of retroperitoneum and peritoneum C76.2 Malignant neoplasm of abdomen C76.3 Malignant neoplasm of pelvis C80 Malignant neoplasm without specification of site D07.3 Carcinoma-in-situ of other or unspecified female genital organ D28.2 Benign neoplasm of fallopian tube D28.9 Benign neoplasm of female genital organ, unspecified D36.9 Benign neoplasm of unspecified site D39.1 Neoplasm of uncertain or unknown behavior of ovary D39.9 Neoplasm of uncertain or unknown behavior of female genital organ, unspecified Abbreviation: ICD-10, International Statistical Classification of Diseases and Related Health Problems (10th revision). Table A2. Reasons Why Screens That Should Have Been Performed Were Not Undertaken

ICD-10 Code Description C56 Malignant neoplasm of ovary C57.0 Malignant neoplasm of fallopian tube C57.4 Uterine adnexa, unspecified C57.7 Other specified female genital organs C57.8 Malignant neoplasm of overlapping lesion of female genital organs C57.9 Malignant neoplasm of female genital organ, unspecified C48.0 Retroperitoneum C48.1 Specified parts of peritoneum C48.2 Malignant neoplasm of peritoneum, unspecified C48.8 Overlapping lesions of retroperitoneum and peritoneum C76.2 Malignant neoplasm of abdomen C76.3 Malignant neoplasm of pelvis C80 Malignant neoplasm without specification of site D07.3 Carcinoma-in-situ of other or unspecified female genital organ D28.2 Benign neoplasm of fallopian tube D28.9 Benign neoplasm of female genital organ, unspecified D36.9 Benign neoplasm of unspecified site D39.1 Neoplasm of uncertain or unknown behavior of ovary D39.9 Neoplasm of uncertain or unknown behavior of female genital organ, unspecified Abbreviation: ICD-10, International Statistical Classification of Diseases and Related Health Problems (10th revision). Table A2. Reasons Why Screens That Should Have Been Performed Were Not Undertaken Reason Screen Was Not Performed No. of Screens No. of Women Died 6,945 1,781 Decided to discontinue 73,632 16,393 Ovaries removed 4,505 1,060 Cancer diagnosed 3,425 953 Over-ran previous screen 3,105 2,874 Table A3. Results of Incidence Screens

Table A2. Reasons Why Screens That Should Have Been Performed Were Not Undertaken Reason Screen Was Not Performed No. of Screens No. of Women Died 6,945 1,781 Decided to discontinue 73,632 16,393 Ovaries removed 4,505 1,060 Cancer diagnosed 3,425 953 Over-ran previous screen 3,105 2,874 Table A3. Results of Incidence Screens Incidence Screens Woman-Years No. % Level 1 screen* 296,911 100 Normal ROC, returned to annual screening 267,327 90.0 Intermediate ROC, referred for repeat level 1 screen 25,133 8.5 Elevated ROC, referred for level 2 screen 4,451 1.5 Repeat level 1 CA-125† 24,788 8.3 Returned to annual screening 21,234 85.7 Referred for level 2 screen 3,355 13.5 Did not complete all repeat screens 199 0.8 Level 2 screen† 7,323 2.5 Returned to annual screening 2,988 40.8 Referred for clinical assessment 1,023 14.0 Referred for repeat level 2 screen 3,312 45.2 Repeat level 2 screen† 2,766 0.9 Returned to annual screening 960 34.7 Referred for clinical assessment 1,806 65.3 Clinical assessments‡ 3,329 1.1 Screen-positive surgery 640 0.2 Diagnostic laparoscopy 17 2.7 Operative laparoscopy 356 55.6 Combined laparoscopy and laparotomy 42 6.6 Laparotomy 206 32.2 Imaging-guided cytology and/or biopsy 15 2.3 Other 4 0.8 Abbreviations: CA-125, cancer antigen 125; ROC, risk of ovarian cancer. * Denominators for header rows are number of incidence screens. Denominators for subsequent rows are numbers of women who underwent a specific screen. † Difference in numbers between those who were recommended tests and number who underwent test is because of noncompliance.

Incidence Screens Woman-Years No. % Level 1 screen* 296,911 100 Normal ROC, returned to annual screening 267,327 90.0 Intermediate ROC, referred for repeat level 1 screen 25,133 8.5 Elevated ROC, referred for level 2 screen 4,451 1.5 Repeat level 1 CA-125† 24,788 8.3 Returned to annual screening 21,234 85.7 Referred for level 2 screen 3,355 13.5 Did not complete all repeat screens 199 0.8 Level 2 screen† 7,323 2.5 Returned to annual screening 2,988 40.8 Referred for clinical assessment 1,023 14.0 Referred for repeat level 2 screen 3,312 45.2 Repeat level 2 screen† 2,766 0.9 Returned to annual screening 960 34.7 Referred for clinical assessment 1,806 65.3 Clinical assessments‡ 3,329 1.1 Screen-positive surgery 640 0.2 Diagnostic laparoscopy 17 2.7 Operative laparoscopy 356 55.6 Combined laparoscopy and laparotomy 42 6.6 Laparotomy 206 32.2 Imaging-guided cytology and/or biopsy 15 2.3 Other 4 0.8 Abbreviations: CA-125, cancer antigen 125; ROC, risk of ovarian cancer. * Denominators for header rows are number of incidence screens. Denominators for subsequent rows are numbers of women who underwent a specific screen. † Difference in numbers between those who were recommended tests and number who underwent test is because of noncompliance. ‡ In all, 109 (29 + 35 + 45) women withdrew before a clinical assessment was performed and 609 (55 + 5 + 29 + 136 + 384) additional women were clinically evaluated before completing all protocol screens. Table A4. Screen-Detected Nonovarian, Tubal, or Primary Peritoneal Cancer

† Difference in numbers between those who were recommended tests and number who underwent test is because of noncompliance. ‡ In all, 109 (29 + 35 + 45) women withdrew before a clinical assessment was performed and 609 (55 + 5 + 29 + 136 + 384) additional women were clinically evaluated before completing all protocol screens. Table A4. Screen-Detected Nonovarian, Tubal, or Primary Peritoneal Cancer Cancer Type No. of Women Women with other nonovarian or tubal cancers not involving the ovaries (n = 19) Appendiceal 2 Endometrial 8 Lymphoma 3 Malignant neoplasm of unknown but not ovarian or tubal origin 1 Breast 1 Colorectal 1 Pancreatic 1 Liver 1 Renal 1 Women with other nonovarian or tubal cancers involving the ovaries (secondary ovarian neoplasm; n = 12) Appendiceal 1 Breast 3 Colorectal 3 GI 2 Lymphoma 1 Endometrial 2 Table A5. Details of the Complications in Women Who Had Normal Ovaries or Benign Pathology at Screen-Positive Surgery Intra- and Early Postoperative Complications Women No. % Major Intraoperative episode of severe tachycardia with asystole requiring cardiopulmonary resuscitation 1 Bowel obstruction* 4 Bowel injury 2 Hemorrhage† 3 Wound dehiscence requiring resuturing 1 Significant ileus 1 Minor Wound infection requiring antibiotics 5 Chest infection 1 Diarrhea and vomiting 1 Perforation of uterus, urinary retention, and UTI 1 Total number of benign surgeries with complications 20 Total number of benign surgeries 441 Complication rate 4.5 Abbreviation: UTI, urinary tract infection.

1 Significant ileus 1 Minor Wound infection requiring antibiotics 5 Chest infection 1 Diarrhea and vomiting 1 Perforation of uterus, urinary retention, and UTI 1 Total number of benign surgeries with complications 20 Total number of benign surgeries 441 Complication rate 4.5 Abbreviation: UTI, urinary tract infection. * One small bowel obstruction from port site hernia, one subacute bowel obstruction requiring readmission. † One from rectus sheet bleed, one from umbilical port site hematoma, one two-unit transfusion.

INTRODUCTION Tamoxifen and third-generation aromatase inhibitors (AIs), such as anastrozole, exemestane, and letrozole are established first-line endocrine therapies for the treatment of postmenopausal women with estrogen receptor (ER) –positive, advanced breast cancer.1–3 Given the high prevalence of resistance to AI therapy, multiple treatment options with distinct mechanisms of action are desirable.4 Fulvestrant, a 17β-estradiol analog, is a selective ER antagonist that suppresses estrogen signaling by binding to ER and inducing a conformational change.5,6 Dimerization is subsequently blocked, triggering accelerated degradation and downregulation of the ER protein.5 Fulvestrant exhibits lack of cross-reactivity with tamoxifen. Consequently, patients whose disease progresses on fulvestrant may retain sensitivity to treatment with further endocrine therapies.7,8 The clinical efficacy of fulvestrant was initially demonstrated in two phase III trials that compared fulvestrant 250 mg per month with anastrozole 1 mg daily as a second-line therapy for advanced breast cancer.9,10 A combined analysis of these trials demonstrated that time to progression (TTP) with fulvestrant 250 mg was noninferior to anastrozole.11

ant was initially demonstrated in two phase III trials that compared fulvestrant 250 mg per month with anastrozole 1 mg daily as a second-line therapy for advanced breast cancer.9,10 A combined analysis of these trials demonstrated that time to progression (TTP) with fulvestrant 250 mg was noninferior to anastrozole.11 Fulvestrant 250 mg was not proven to be superior to tamoxifen in a double-blind, randomized trial.12 This finding was unexpected given the superiority of anastrozole over tamoxifen13 and the comparable efficacy of anastrozole and fulvestrant 250 mg as second-line therapy.11 Pharmacokinetic modeling, as well as observations made during early clinical studies,11 suggested the efficacy of fulvestrant could be improved with use of a higher dose, which led to the development of a dosage regimen of fulvestrant 500 mg, including a loading dose component to reduce the time to reach steady-state plasma levels. Subsequently, the phase III Comparison of Faslodex in Recurrent or Metastatic Breast Cancer (CONFIRM) trial found that fulvestrant 500 mg was associated with improved progression-free survival (PFS) and overall survival (OS) compared with the 250-mg dose in patients who experienced disease recurrence or progression after previous endocrine therapy.14,15

son of Faslodex in Recurrent or Metastatic Breast Cancer (CONFIRM) trial found that fulvestrant 500 mg was associated with improved progression-free survival (PFS) and overall survival (OS) compared with the 250-mg dose in patients who experienced disease recurrence or progression after previous endocrine therapy.14,15 The Fulvestrant First-Line Study Comparing Endocrine Treatments (FIRST) was a phase II, randomized, open-label, multicenter trial that also used the fulvestrant 500-mg dose regimen, comparing efficacy and safety with anastrozole in the first-line setting. The primary end point of clinical benefit rate was noninferior for fulvestrant 500 mg compared with anastrozole,16 with both treatments demonstrating similar, well-tolerated safety profiles. A follow-up analysis, performed because only 35.6% of patients experienced disease progression at the time of the primary analysis, reported a hazard ratio (HR) of TTP for fulvestrant 500 mg versus anastrozole of 0.66 with a 95% CI of 0.47 to 0.92 (P = .01; median TTP, 23.4 months v 13.1 months). No additional safety issues were reported.17 Given the improvement in TTP observed during fulvestrant 500 mg treatment compared with anastrozole in this phase II trial, a subsequent protocol amendment was made to address whether this apparent extension in disease control would translate into an improvement in OS.

ths). No additional safety issues were reported.17 Given the improvement in TTP observed during fulvestrant 500 mg treatment compared with anastrozole in this phase II trial, a subsequent protocol amendment was made to address whether this apparent extension in disease control would translate into an improvement in OS. PATIENTS AND METHODS Study Design and Participants FIRST was a phase II, randomized, open-label, multicenter, parallel-group trial comparing fulvestrant 500 mg with anastrozole 1 mg. Postmenopausal women with ER-positive locally advanced or metastatic breast cancer who had not received any previous systemic therapy for locally advanced or metastatic disease were included. Patients were permitted to have received previous endocrine therapy for early disease, providing this had been completed more than 12 months before random assignment. This trial was conducted in accordance with the Declaration of Helsinki, was consistent with the International Conference on Harmonisation–Good Clinical Practice guidelines, and is registered with Clinicaltrials.gov. All patients provided written, informed consent. Full details of this trial have been reported previously.16,17

al was conducted in accordance with the Declaration of Helsinki, was consistent with the International Conference on Harmonisation–Good Clinical Practice guidelines, and is registered with Clinicaltrials.gov. All patients provided written, informed consent. Full details of this trial have been reported previously.16,17 Random Assignment and Procedures Eligible patients were randomly assigned sequentially 1:1 to either fulvestrant 500 mg (administered intramuscularly on days 0, 14, 28, and every 28 days thereafter) or anastrozole 1 mg (administered orally once per day). The data cutoff for the primary analysis was 6 months after the last patient was randomly assigned. On disease progression or after data cutoff for the primary analysis, all patients entered a follow-up phase after a protocol amendment for an analysis of TTP. The TTP follow-up required a questionnaire to be completed for each patient 12 months after the patient entered the follow-up phase and every 12 months thereafter for patients continuing to receive randomized treatment. After the TTP analysis was performed, a further protocol amendment was developed to enter patients into an optional follow-up phase to establish OS. To ensure sufficient maturity, the OS analysis was planned for when approximately 65% of patients had died. Patients who did not contribute additional data to the follow-up extension were right-censored at the last known date they were alive, and their data until this point were included in the analysis. Sites were invited to request written consent from patients for the collection of additional data. Patients were contacted every 3 months until the first of the following events: death, patient withdrawal, data cutoff was reached, or the patient was lost to follow-up. Patients with a last known survival status of alive were contacted within 2 weeks of data cutoff to ensure they were still alive.

of additional data. Patients were contacted every 3 months until the first of the following events: death, patient withdrawal, data cutoff was reached, or the patient was lost to follow-up. Patients with a last known survival status of alive were contacted within 2 weeks of data cutoff to ensure they were still alive. Outcomes The primary study end point was clinical benefit rate; secondary end points included objective response rate, TTP, duration of clinical benefit, and duration of response. These primary and secondary end points have been reported previously.16,17 The follow-up analysis assessed OS, defined as the time from being randomly assigned to death from any cause. A log-rank test (unadjusted model with treatment factor only) was performed for the primary analysis of OS. HRs with 95% CIs were used to compare fulvestrant 500 mg with anastrozole; no adjustments were made for multiplicity. A statistical significance level of .05 was used to indicate a difference in OS between the treatment groups. For patients for whom follow-up responses could not be obtained, data were censored at the date the patient was last known to be alive.

strant 500 mg with anastrozole; no adjustments were made for multiplicity. A statistical significance level of .05 was used to indicate a difference in OS between the treatment groups. For patients for whom follow-up responses could not be obtained, data were censored at the date the patient was last known to be alive. Exploratory subgroup analyses were conducted using the log-rank test to compare OS for the following prespecified patient subgroups: less than 65 years of age versus 65 years of age or greater; not positive for both ER and progesterone receptor versus positive for both ER and progesterone receptor; no visceral involvement versus visceral involvement; no previous chemotherapy versus previous adjuvant chemotherapy; no measurable disease versus measurable disease; and no previous endocrine therapy versus previous endocrine therapy. Two sensitivity analyses were performed to examine any potential impact of nonparticipation on OS results: a Kaplan-Meier OS analysis was performed in which the censoring indicator was reversed; and baseline covariates were assessed for patients censored greater than 3 months before data cutoff and for those censored 3 months or less before data cutoff, which corresponds to patients who did not participate in the OS follow-up and to those who did, respectively.

ed in which the censoring indicator was reversed; and baseline covariates were assessed for patients censored greater than 3 months before data cutoff and for those censored 3 months or less before data cutoff, which corresponds to patients who did not participate in the OS follow-up and to those who did, respectively. Tolerability was assessed by serious adverse event (SAE) monitoring. All SAEs were coded in compliance with the Medical Dictionary for Regulatory Activities and recorded in an internal AstraZeneca database for evaluation. SAEs were monitored for up to 8 weeks after the last dose of fulvestrant 500 mg or for 30 days after the last dose of anastrozole. RESULTS In total, 205 patients were randomly assigned to receive fulvestrant 500 mg (n = 102) or anastrozole 1 mg (n = 103) at 62 centers in nine countries (Brazil, Bulgaria, the Czech Republic, France, Italy, Poland, Spain, the United Kingdom, and the United States).

Tolerability was assessed by serious adverse event (SAE) monitoring. All SAEs were coded in compliance with the Medical Dictionary for Regulatory Activities and recorded in an internal AstraZeneca database for evaluation. SAEs were monitored for up to 8 weeks after the last dose of fulvestrant 500 mg or for 30 days after the last dose of anastrozole. RESULTS In total, 205 patients were randomly assigned to receive fulvestrant 500 mg (n = 102) or anastrozole 1 mg (n = 103) at 62 centers in nine countries (Brazil, Bulgaria, the Czech Republic, France, Italy, Poland, Spain, the United Kingdom, and the United States). Baseline characteristics and patient demographics were similar between the treatment groups as reported previously.16 The proportion of patients who had not received previous endocrine treatment for early disease was similar for the fulvestrant 500 mg and anastrozole treatment groups (71.6% and 77.7% of patients at baseline, respectively). Of those that did, almost all had received tamoxifen exclusively. Of the 205 randomly assigned patients, 35 (16 in the fulvestrant 500 mg group and 19 in the anastrozole group) did not participate in the OS follow-up phase and were censored at the date they were last known to be alive; for these patients, data until this time are included in the OS analysis, and thus all patients contributed data to the analysis. The majority of the nonparticipating patients (n = 20) did not contribute additional data because they attended centers that declined to contribute to the OS follow-up phase. An additional 15 individual patients from nine participating centers did not consent to follow-up. No patients participating in the OS phase were lost to follow-up, and the survival status at data cutoff was known for all patients consenting to the OS follow-up.

rs that declined to contribute to the OS follow-up phase. An additional 15 individual patients from nine participating centers did not consent to follow-up. No patients participating in the OS phase were lost to follow-up, and the survival status at data cutoff was known for all patients consenting to the OS follow-up. Efficacy At the time of the follow-up analysis for OS, 63 of 102 patients in the fulvestrant 500 mg group (61.8%) and 74 of 103 patients in the anastrozole group (71.8%) were known to have died (Fig 1). The primary analysis of OS was improved in the fulvestrant 500 mg group compared with anastrozole 1 mg; the HR was 0.70 (95% CI, 0.50 to 0.98; log-rank test P = .04; median OS, 54.1 months v 48.4 months; Fig 2). The HR for fulvestrant 500 mg versus anastrozole was found to be generally consistent across all subgroup analyses (Fig 3). At 3 years, 64% (fulvestrant 500 mg) and 58% (anastrozole) of patients were event free; at 5 years, the equivalent values were 47% and 38%. Fig 1. Study overview. (*) These patients were right censored at the time of their last known date alive, and data until this point were used in the overall survival (OS) analysis. Fig 2. Kaplan-Meier plot of overall survival. Fig 3. Overall survival subgroup analysis. ER+, estrogen receptor positive; NC, not calculable; PgR+, progesterone receptor positive.

Fig 1. Study overview. (*) These patients were right censored at the time of their last known date alive, and data until this point were used in the overall survival (OS) analysis. Fig 2. Kaplan-Meier plot of overall survival. Fig 3. Overall survival subgroup analysis. ER+, estrogen receptor positive; NC, not calculable; PgR+, progesterone receptor positive. Sensitivity Analyses There were no important differences between the treatment groups in time to censoring (data not shown). Furthermore, when key baseline covariates for patients censored within the last 3 months before data cutoff and for those censored more than 3 months before data cutoff were summarized, there were no important differences between treatment groups, indicating that the results were not caused by differences between patients who did and did not consent to OS follow-up (Table 1). Table 1. Baseline Covariates and Subgroups by Patients Censored ≥ 3 Months and ≤ 3 Months Before DCO

Sensitivity Analyses There were no important differences between the treatment groups in time to censoring (data not shown). Furthermore, when key baseline covariates for patients censored within the last 3 months before data cutoff and for those censored more than 3 months before data cutoff were summarized, there were no important differences between treatment groups, indicating that the results were not caused by differences between patients who did and did not consent to OS follow-up (Table 1). Table 1. Baseline Covariates and Subgroups by Patients Censored ≥ 3 Months and ≤ 3 Months Before DCO Subgroup No. of Patients (%) Censored > 3 Months Before DCO Censored ≤ 3 Months Before DCO Fulvestrant 500 mg (n = 16) Anastrozole 1 mg (n = 19) Fulvestrant 500 mg (n = 23) Anastrozole 1 mg (n = 10) Age, years < 65 5 (31.3) 7 (36.8) 11 (47.8) 4 (40.0) ≥ 65 11 (68.8) 12 (63.2) 12 (52.2) 6 (60.0) Receptor status at diagnosis Not both ER+ and PgR+ 6 (37.5) 5 (26.3) 4 (17.4) 2 (20.0) Both ER+ and PgR+ 10 (62.5) 14 (73.7) 19 (82.6) 8 (80.0) Visceral involvement No 9 (56.3) 11 (57.9) 16 (69.6) 8 (80.0) Yes 7 (43.8) 8 (42.1) 7 (30.4) 2 (20.0) Previous chemotherapy No 11 (68.8) 13 (68.4) 19 (82.6) 8 (80.0) Yes 5 (31.3) 6 (31.6) 4 (17.4) 2 (20.0) Measurable disease at diagnosis No 1 (6.3) 3 (15.8) 1 (4.3) 0 Yes 15 (93.8) 16 (84.2) 22 (95.7) 10 (100.0) Previous endocrine therapy No 11 (68.8) 13 (68.4) 18 (78.3) 8 (80.0) Yes 5 (31.3) 6 (31.6) 5 (21.7) 2 (20.0) Abbreviations: DCO, data cutoff; ER+, estrogen receptor–positive; PgR+, progesterone receptor–positive.

Measurable disease at diagnosis No 1 (6.3) 3 (15.8) 1 (4.3) 0 Yes 15 (93.8) 16 (84.2) 22 (95.7) 10 (100.0) Previous endocrine therapy No 11 (68.8) 13 (68.4) 18 (78.3) 8 (80.0) Yes 5 (31.3) 6 (31.6) 5 (21.7) 2 (20.0) Abbreviations: DCO, data cutoff; ER+, estrogen receptor–positive; PgR+, progesterone receptor–positive. Safety The occurrence of SAEs during the main study period and the follow-up period combined is detailed in Table 2. The majority of SAEs were considered by the investigator to be unrelated to the treatment. Two SAEs considered to be treatment related were documented (one case of hypertension and one case of pulmonary embolism, both in the fulvestrant 500 mg treatment group). Table 2. Incidence of SAEs and Deaths SAE No. of Patients (%) Fulvestrant 500 mg (n = 101) Anastrozole 1 mg (n = 103) Any SAE 24 (23.8) 22 (21.4) Any SAE related to death 3 (3.0) 5 (4.9) Any SAE with outcome other than death 21 (20.8) 18 (17.5) Any causally related SAE 2 (2.0) 0 Most commonly reported (≥ two patients) SAEs Atrial fibrillation 1 (1.0) 1 (1.0) Cardiac failure 2 (2.0) 0 Death 0 2 (1.9) Decreased appetite 2 (2.0) 0 Dehydration 2 (2.0) 0 Dyspnea 2 (2.0) 0 Femur fracture 1 (1.0) 2 (1.9) Neuralgia 1 (1.0) 1 (1.0) Transient ischemic attack 0 2 (1.9) Abbreviation: SAE, serious adverse event.

ommonly reported (≥ two patients) SAEs Atrial fibrillation 1 (1.0) 1 (1.0) Cardiac failure 2 (2.0) 0 Death 0 2 (1.9) Decreased appetite 2 (2.0) 0 Dehydration 2 (2.0) 0 Dyspnea 2 (2.0) 0 Femur fracture 1 (1.0) 2 (1.9) Neuralgia 1 (1.0) 1 (1.0) Transient ischemic attack 0 2 (1.9) Abbreviation: SAE, serious adverse event. DISCUSSION This study reports improved OS with fulvestrant 500 mg treatment compared with anastrozole in the first-line setting for ER-positive advanced breast cancer, with an approximately 30% reduction in mortality risk. The previously reported improvements in TTP have translated into an improvement in OS of approximately 6 months with fulvestrant 500 mg (54.1 months) compared with anastrozole (48.4 months). This OS advantage is consistent with the OS benefit for fulvestrant 500 mg versus 250 mg in the second-line setting in the CONFIRM trial.15 The effect of fulvestrant 500 mg on OS was generally consistent across all prespecified subgroups (Fig 3). Furthermore, no new safety or tolerability issues were reported from the OS follow-up phase of this study, consistent with previously reported safety data.16,17

g in the second-line setting in the CONFIRM trial.15 The effect of fulvestrant 500 mg on OS was generally consistent across all prespecified subgroups (Fig 3). Furthermore, no new safety or tolerability issues were reported from the OS follow-up phase of this study, consistent with previously reported safety data.16,17 The improved OS with fulvestrant 500 mg (54.1 months) relative to anastrozole (48.4 months) was observed although the median OS for the anastrozole group in this study was higher than has previously been reported. For example, OS of 39.2 months was reported for anastrozole as first-line endocrine therapy for advanced breast cancer in a combined analysis of two phase III studies,18 and OS of 41.3 months was reported for the anastrozole monotherapy arm of a phase III combination study.19 In addition, corresponding median OS values of 34.0 months (letrozole)20 and 37.2 months (exemestane)21 have been reported for other AIs. It is therefore unlikely that the present analysis overestimates the margin of improvement with fulvestrant 500 mg over anastrozole, which might have been possible had the control arm underperformed.

esponding median OS values of 34.0 months (letrozole)20 and 37.2 months (exemestane)21 have been reported for other AIs. It is therefore unlikely that the present analysis overestimates the margin of improvement with fulvestrant 500 mg over anastrozole, which might have been possible had the control arm underperformed. The role of fulvestrant 500 mg as first-line therapy will be further defined by the ongoing phase III, double-blind FALCON (Fulvestrant and Anastrozole Compared in Hormonal Therapy Naïve Advanced Breast Cancer) trial (ClinicalTrials.gov identifier: NCT01602380). The FALCON trial will assess the efficacy of fulvestrant 500 mg versus anastrozole in women with locally advanced or metastatic breast cancer with strict definitions of endocrine therapy–naïve disease, including restrictions on exposure to hormone replacement therapy.

trial (ClinicalTrials.gov identifier: NCT01602380). The FALCON trial will assess the efficacy of fulvestrant 500 mg versus anastrozole in women with locally advanced or metastatic breast cancer with strict definitions of endocrine therapy–naïve disease, including restrictions on exposure to hormone replacement therapy. Endocrine therapy–naïve advanced breast cancer is relatively uncommon in countries with advanced health care, but represents a numerically substantial patient population, given the high disease prevalence. Furthermore, in unscreened populations and in developing countries, metastatic disease at presentation is a significant problem. Recent clinical trials reporting on first-line endocrine therapy in patients with ER-positive breast cancer have contained a substantial proportion, and often a majority, of endocrine therapy–naïve patients.19,22–24 In FIRST, previous endocrine therapy had been received by 29 (28.4%) of the patients treated with fulvestrant 500 mg and 23 (22.3%) of the anastrozole-treated patients. Of these 52 patients, only 3 had received AI previously (2 in the anastrozole group and 1 in the fulvestrant 500 mg group); the remainder had received adjuvant tamoxifen. Therefore, AI resistance resulting from previous AI exposure cannot account for the observed OS difference. Indeed, hypothetically, previous exposure to tamoxifen may bias against fulvestrant as both agents are in the same therapeutic class. Upon disease progression, patients were treated according to the standard of care, and therefore, there could potentially be imbalances between the two treatment groups that could have affected the OS analysis. However, response to subsequent therapies (systemic chemotherapy or endocrine therapy) has previously been shown to be similar between the treatment groups, demonstrating that patients with disease progression on fulvestrant retain sensitivity to subsequent treatments.17 Differential second-line response, therefore, is also an unlikely explanation for the observed OS effect.

therapy or endocrine therapy) has previously been shown to be similar between the treatment groups, demonstrating that patients with disease progression on fulvestrant retain sensitivity to subsequent treatments.17 Differential second-line response, therefore, is also an unlikely explanation for the observed OS effect. There are significant limitations to this report. The sample size was relatively small, and the OS analysis was not specified in the original protocol but was added as a hypothesis in a protocol amendment after TTP results were known. Furthermore, 35 patients did not contribute additional data to the OS follow-up; the decision not to participate in the extended follow-up for OS was made solely by the patient or participating center and was known at the start of the OS follow-up and before the data were collected and analyzed. Data from these patients until the time of censoring were included in the OS analysis, and similar censoring patterns were seen in the two treatment groups. The sensitivity analyses support the main findings, that is, the differences in OS between treatment arms were unrelated to differences in censoring patterns. All-cause mortality was used to determine OS in this analysis because it is regarded as the most unbiased and objective end point used in oncology.25 This point is particularly relevant to an open-label study like FIRST. A final limitation was that the number of patients within subgroups was relatively small. Therefore, care should be taken when interpreting results.

his analysis because it is regarded as the most unbiased and objective end point used in oncology.25 This point is particularly relevant to an open-label study like FIRST. A final limitation was that the number of patients within subgroups was relatively small. Therefore, care should be taken when interpreting results. Recent results from several trials with the cyclin-dependent kinase 4/6 (CDK4/6) inhibitor palbociclib are also pertinent to the discussion. PALOMA-1 (Palbociclib Ongoing Trials in the Management of Breast Cancer), a phase II trial of letrozole plus palbociclib versus letrozole alone, provided provisional US Food and Drug Administration approval for palbociclib in the first-line setting on the basis of PFS.23 No positive OS data have been reported to date; the results of a phase III trial of this comparison are pending (PALOMA-2, NCT01740427). Data from the phase III PALOMA-3 trial, comparing fulvestrant 500 mg plus palbociclib versus fulvestrant 500 mg alone in the second-line or subsequent setting in postmenopausal women (or pre- or perimenopausal women receiving goserelin), reported a marked PFS advantage for the combination, but OS data were also pending at the time of publication.26 The median PFS for fulvestrant 500 mg alone was shorter in PALOMA-3 than in previous studies, indicative of the younger, higher-risk, and more heavily pretreated population recruited into the PALOMA-3 trial.

reported a marked PFS advantage for the combination, but OS data were also pending at the time of publication.26 The median PFS for fulvestrant 500 mg alone was shorter in PALOMA-3 than in previous studies, indicative of the younger, higher-risk, and more heavily pretreated population recruited into the PALOMA-3 trial. The treatment algorithm for ER-positive advanced breast cancer, therefore, is in a state of flux. Currently, it is rational to consider fulvestrant 500 mg as a first-line treatment option given the potential for survival benefits, particularly in settings where palbociclib is not available or palbociclib cost or adverse effects are a significant concern, and especially if these results are confirmed in FALCON. These data also suggest that a first-line study of fulvestrant 500 mg with a CDK4/6 inhibitor versus fulvestrant 500 mg alone is a logical proposition that could lead to further prolonged TTP. Recent preclinical data on the efficacy of an ER degrading agent with a CDK4/6 inhibitor in ESR1-mutant breast cancer provides further rationale for this population, because improvements in TTP or OS could be caused by suppression of ESR1-mutant AI-resistant clones.27

oposition that could lead to further prolonged TTP. Recent preclinical data on the efficacy of an ER degrading agent with a CDK4/6 inhibitor in ESR1-mutant breast cancer provides further rationale for this population, because improvements in TTP or OS could be caused by suppression of ESR1-mutant AI-resistant clones.27 In conclusion, we report that fulvestrant 500 mg may be associated with improved OS versus anastrozole in the first-line setting for ER-positive advanced breast cancer. To our knowledge, this represents the first time an endocrine monotherapy has demonstrated improved efficacy compared with a third-generation AI. The phase III FALCON trial may provide confirmation for these OS results; until then, the findings reported here should be regarded as preliminary, but clinically relevant. Supplementary Material Protocol Supported by AstraZeneca. Terms in blue are defined in the glossary, found at the end of this article and online at www.jco.org. Presented at the 2014 San Antonio Breast Cancer Symposium, San Antonio, TX, December 9-13, 2014. Authors' disclosures of potential conflicts of interest are found in the article online at www.jco.org. Author contributions are found at the end of this article. Clinical trial information: NCT00274469. Acknowledgment We thank Martin Bell, PhD, from Complete Medical Communications, who provided medical writing support, funded by AstraZeneca. AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST Disclosures provided by the authors are available with this article at www.jco.org.

Authors' disclosures of potential conflicts of interest are found in the article online at www.jco.org. Author contributions are found at the end of this article. Clinical trial information: NCT00274469. Acknowledgment We thank Martin Bell, PhD, from Complete Medical Communications, who provided medical writing support, funded by AstraZeneca. AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST Disclosures provided by the authors are available with this article at www.jco.org. AUTHOR CONTRIBUTIONS Conception and design: Matthew J. Ellis, John F.R. Robertson Provision of study materials or patients: Matthew J. Ellis, John F.R. Robertson Collection and assembly of data: Matthew J. Ellis, David Feltl, John F.R. Robertson Data analysis and interpretation: Matthew J. Ellis, Antonio Llombart-Cussac, John A. Dewar, Marek Jasiówka, Nicola Hewson, Yuri Rukazenkov, John F.R. Robertson Manuscript writing: All authors Final approval of manuscript: All authors Glossary Terms Anastrozole:a third-generation nonsteroidal aromatase inhibitor that prevents the conversion of androgen to estrogen in the peripheral tissues in postmenopausal women. Because hormone-dependent breast cancer progresses with estrogen, anastrozole has been used in the treatment of breast cancer in postmenopausal women. See aromatase inhibitors.

ration nonsteroidal aromatase inhibitor that prevents the conversion of androgen to estrogen in the peripheral tissues in postmenopausal women. Because hormone-dependent breast cancer progresses with estrogen, anastrozole has been used in the treatment of breast cancer in postmenopausal women. See aromatase inhibitors. Aromatase inhibitors:inhibitors used in treating breast cancer in postmenopausal women. Aromatase inhibitors inhibit the conversion of androgens to estrogens by the enzyme aromatase, thus depriving the tumor of estrogenic signals. Because of decreased production of estrogen, estrogen receptors, which are important in the progression of breast cancer, cannot be activated. Estrogen receptor (ER):ligand-activated nuclear proteins, belonging to the class of nuclear receptors, present in many breast cancer cells that are important in the progression of hormone-dependent cancers. After binding, the receptor-ligand complex activates gene transcription. There are two types of estrogen receptors (ERα and ERβ). ERα is one of the most important proteins controlling breast cancer function. ERβ is present in much lower levels in breast cancer, and its function is uncertain. Estrogen receptor status guides therapeutic decisions in breast cancer. Overall survival:the duration between random assignment and death.

Estrogen receptor (ER):ligand-activated nuclear proteins, belonging to the class of nuclear receptors, present in many breast cancer cells that are important in the progression of hormone-dependent cancers. After binding, the receptor-ligand complex activates gene transcription. There are two types of estrogen receptors (ERα and ERβ). ERα is one of the most important proteins controlling breast cancer function. ERβ is present in much lower levels in breast cancer, and its function is uncertain. Estrogen receptor status guides therapeutic decisions in breast cancer. Overall survival:the duration between random assignment and death. AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST Fulvestrant 500 mg Versus Anastrozole 1 mg for the First-Line Treatment of Advanced Breast Cancer: Overall Survival Analysis From the Phase II FIRST Study The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or jco.ascopubs.org/site/ifc. Matthew J. Ellis Employment: Bioclassifier Leadership: Bioclassifier Stock or Other Ownership: Bioclassifier Consulting or Advisory Role: AstraZeneca, Pfizer, Novartis, Celgene Patents, Royalties, Other Intellectual Property: Bioclassifier Antonio Llombart-Cussac Honoraria: Roche, Eli Lilly, Pfizer, Novartis, AstraZeneca

AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST Fulvestrant 500 mg Versus Anastrozole 1 mg for the First-Line Treatment of Advanced Breast Cancer: Overall Survival Analysis From the Phase II FIRST Study The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or jco.ascopubs.org/site/ifc. Matthew J. Ellis Employment: Bioclassifier Leadership: Bioclassifier Stock or Other Ownership: Bioclassifier Consulting or Advisory Role: AstraZeneca, Pfizer, Novartis, Celgene Patents, Royalties, Other Intellectual Property: Bioclassifier Antonio Llombart-Cussac Honoraria: Roche, Eli Lilly, Pfizer, Novartis, AstraZeneca Consulting or Advisory Role: Roche, AstraZeneca, Pfizer Research Funding: MedSIR David Feltl No relationship to disclose John A. Dewar No relationship to disclose Marek Jasiówka Honoraria: Roche, Amgen Research Funding: AstraZeneca, Roche Travel, Accommodations, Expenses: AstraZeneca, Roche, Janssen-Cilag Nicola Hewson Employment: AstraZeneca Yuri Rukazenkov Employment: AstraZeneca Stock or Other Ownership: AstraZeneca John F.R. Robertson Leadership: Oncimmune Stock or Other Ownership: Oncimmune Honoraria: AstraZeneca, Bayer AG Consulting or Advisory Role: AstraZeneca, Bayer AG, Oncimmune Research Funding: AstraZeneca (Inst), Bayer AG (Inst), Oncimmune (Inst), Novartis (Inst)

Yuri Rukazenkov Employment: AstraZeneca Stock or Other Ownership: AstraZeneca John F.R. Robertson Leadership: Oncimmune Stock or Other Ownership: Oncimmune Honoraria: AstraZeneca, Bayer AG Consulting or Advisory Role: AstraZeneca, Bayer AG, Oncimmune Research Funding: AstraZeneca (Inst), Bayer AG (Inst), Oncimmune (Inst), Novartis (Inst) Patents, Royalties, Other Intellectual Property: Oncimmune Travel, Accommodations, Expenses: AstraZeneca, Bayer AG, Oncimmune, Novartis

INTRODUCTION Informed consent remains a cornerstone of medical decision making, but, for myriad reasons,1 oncologists often struggle with whether and how best to convey sensitive information about prognoses to their patients with advanced cancer. Sometimes, oncologists do not share information with patients about a terminal cancer prognosis.2,3 However, information from physicians about end-of-life care is often wanted by patients4,5 and is related to patient receipt of higher-quality care near death.3,6 As highlighted in a recent Institute of Medicine report,7 patient-physician communication at the end of life is a promising target for improvement of the delivery of patient-centered end-of-life care.

end-of-life care is often wanted by patients4,5 and is related to patient receipt of higher-quality care near death.3,6 As highlighted in a recent Institute of Medicine report,7 patient-physician communication at the end of life is a promising target for improvement of the delivery of patient-centered end-of-life care. Often, patients with common advanced cancers have inaccurate illness understanding and would benefit from effective prognostic communication. Many patients mistakenly believe that their cancers are curable when they are not. A recent study of patients with advanced lung and colorectal cancer found that 69% and 81%, respectively, believed that the chemotherapy they received was intended to cure them.8 These data suggest a need for oncologists to share prognostic information that is relevant to the understanding of the intent of treatment. Nevertheless, many oncologists are reluctant to do so, because they worry that, by sharing prognostic information, they will make patients needlessly hopeless or upset3 and/or that patients will view them less favorably as a result.8 Some evidence suggests that, at least in the short run, physicians who share prognostic information are viewed less positively by their patients,8 but other findings suggest that the sharing of prognostic information does not damage the oncologist-patient relationship.4,6

nts will view them less favorably as a result.8 Some evidence suggests that, at least in the short run, physicians who share prognostic information are viewed less positively by their patients,8 but other findings suggest that the sharing of prognostic information does not damage the oncologist-patient relationship.4,6 Research has shown that accurate prognostic understanding is associated with anxiety and worse quality of life9 and that training clinicians to communicate about end-of-life care may actually result in higher patient depression scores.10 However, patients who report engagement in end-of-life discussions have not been shown to be more depressed or worried.3,4,6 Studies show that patients with serious illness do not lose hope,11 suffer,12 or die sooner13 as a result of end-of-life discussions. Bereaved caregivers also do not incur lasting psychological harms from such discussions.3 Thus, the effects of prognostic communication on care,3,6 patient mood, and patient relationships with their oncologists have been studied. However, research is needed to determine the effect of prognostic communication on illness understanding by patients. Specifically, the impact of prognostic discussions on patient understanding of disease status, curability, and life expectancy has not been examined with data designed explicitly for this purpose. Such data would provide guidance on how to communicate more effectively to promote illness understanding in patients.

ents. Specifically, the impact of prognostic discussions on patient understanding of disease status, curability, and life expectancy has not been examined with data designed explicitly for this purpose. Such data would provide guidance on how to communicate more effectively to promote illness understanding in patients. This study sought to evaluate the effects of recent and past clinical discussions about prognosis on changes in illness understanding by patients with advanced cancer. We hypothesized that recent and ongoing prognostic discussions would improve illness understanding in patients. METHODS Study Sample The analyzed patient sample (N = 178) was drawn from the Coping with Cancer II (CwC-II) study. CwC-II is a National Cancer Institute–funded, prospective, multi-institutional cohort study of patients with advanced cancer and their oncology providers designed to evaluate how end-of-life communication affects illness understanding by patients. Participants were recruited at nine US cancer centers: Dana-Farber/Harvard Cancer Center (DF/HCC; Dana-Farber Cancer Institute, Brigham and Women’s Hospital, and Massachusetts General Hospital, Boston, MA), Parkland Hospital (Dallas, TX), Simmons Comprehensive Cancer Center (Dallas, TX), Yale Cancer Center (New Haven, CT), Meyer Cancer Center at Weill Cornell Medical College (New York, NY), Memorial Sloan Kettering Cancer Center (New York, NY), Virginia Commonwealth University Massey Cancer Center (Richmond, VA), University of New Mexico Cancer Center (Albuquerque, NM), and Pomona Valley Hospital Medical Center (Pomona, CA).

Haven, CT), Meyer Cancer Center at Weill Cornell Medical College (New York, NY), Memorial Sloan Kettering Cancer Center (New York, NY), Virginia Commonwealth University Massey Cancer Center (Richmond, VA), University of New Mexico Cancer Center (Albuquerque, NM), and Pomona Valley Hospital Medical Center (Pomona, CA). Patients had to meet the following eligibility criteria: stage IV gastrointestinal, lung, or gynecologic cancer and select incurable and poor-prognosis stage III cancers (eg, pancreas and lung); oncologist-estimated life expectancy of 6 or fewer months; disease progression after at least one chemotherapy regimen or, in the case of advanced colorectal cancers, progression during treatment with two chemotherapy regimens. Additional eligibility criteria included age of at least 21 years and the ability to complete the study interviews. Patients with cognitive impairment (eg, rater-perceived inability to provide reliable responses and validly respond to the questions posed of them) were excluded. Institutional review boards of all participating institutions approved study procedures, and all participants provided written, informed consent. This study focused on changes in illness understanding by patients before and after a visit with his or her oncology provider to discuss scan results and to evaluate disease progression. A total of 178 CwC-II participants who completed both pre- and post-scan visit interviews between January 2011 and February 2015 were included in the analyses.

ges in illness understanding by patients before and after a visit with his or her oncology provider to discuss scan results and to evaluate disease progression. A total of 178 CwC-II participants who completed both pre- and post-scan visit interviews between January 2011 and February 2015 were included in the analyses. Measures Patient characteristics. Patients provided information about age, sex, race/ethnicity, education, marital status, and health insurance status. Changes in illness understanding by patients. In pre- and post-scan interviews, patients were asked four questions that assessed their terminal illness acknowledgment (TIA), recognition of their incurable disease status, knowledge of the advanced stage of their disease, and expectation to live months as opposed to years. These elements of illness understanding were deemed by us to be essential for patients to make informed decisions about end-of-life care. Responses were coded 1 or 0 to indicate the presence or absence, respectively, of each of these elements of illness understanding by patients. These four indicators were then added together to construct summary scores (possible range, 0 to 4) to reflect illness understanding at the times of both the pre- and post-scan visit interviews. Differences between pre- and post-scan visit illness understanding scores (possible range, −4 to 4) were used to define changes in illness understanding by a patient between te pre- and post-scan visit interviews.

ge, 0 to 4) to reflect illness understanding at the times of both the pre- and post-scan visit interviews. Differences between pre- and post-scan visit illness understanding scores (possible range, −4 to 4) were used to define changes in illness understanding by a patient between te pre- and post-scan visit interviews. TIA was assessed with the question “How would you describe your current health status?” Response options were (1) relatively healthy, (2) relatively healthy and terminally ill, (3) seriously ill but not terminally ill, (4) seriously ill and terminally ill, and (5) do not know. TIA was coded 1 for responses options 2 and 4 and 0 for response options 1, 3, and 5. Recognition of an incurable disease status was assessed with the question “Which of the following best represents what your oncology providers have told you about a cure for your cancer?” Response options were (1) my cancer will be cured, (2) my cancer may be cured if treatments are successful, (3) my cancer cannot be cured but we will try to control the cancer with treatment, (4) my cancer cannot be cured and I am not able to have any additional cancer treatment, and (5) do not know. Recognition of incurable disease was coded 1 for response options 3 and 4 and 0 for response options 1, 2, and 5.

successful, (3) my cancer cannot be cured but we will try to control the cancer with treatment, (4) my cancer cannot be cured and I am not able to have any additional cancer treatment, and (5) do not know. Recognition of incurable disease was coded 1 for response options 3 and 4 and 0 for response options 1, 2, and 5. Knowledge of advanced stage of cancer was assessed with the question “What stage is your cancer?” Responses were (1) no evidence of cancer, (2) early stage of cancer, (3) middle stage of cancer, (4) late stage of cancer, (5) end stage of cancer, and (6) do not know. Knowledge of advanced stage of cancer was coded 1 for responses options 4 and 5 and 0 for response options 1, 2, 3, and 6. Expectation to live months as opposed to years was assessed with the question “Many patients have thoughts about how having cancer might affect their life expectancy, either on the basis of what their doctors have told them, what they have read, or just their own sense about how long they might live with cancer. When you think about this, do you think in terms of (select response)?” Response options were (1) months, (2) years, and (3) do not know. Expectation to live months as opposed to years was coded 1 for response option 1 and 0 for response options 2 and 3.

just their own sense about how long they might live with cancer. When you think about this, do you think in terms of (select response)?” Response options were (1) months, (2) years, and (3) do not know. Expectation to live months as opposed to years was coded 1 for response option 1 and 0 for response options 2 and 3. Patient-oncologist discussions of prognosis/life expectancy. During the post-scan visit, patients were asked “At the last oncology visit, was there any discussion of your prognosis or life expectancy with this disease?” and “Have you discussed your prognosis/life expectancy with your oncology provider in past visits?” Response options for each of these questions were yes or no. Responses to these questions were used to identify patients who reported discussions at only recent (last visit), only past (prior to last visit), or both recent and past visits or never having discussions of prognosis/life expectancy with the oncologist.

s?” Response options for each of these questions were yes or no. Responses to these questions were used to identify patients who reported discussions at only recent (last visit), only past (prior to last visit), or both recent and past visits or never having discussions of prognosis/life expectancy with the oncologist. Statistical Analysis To evaluate potential sociodemographic confounders, differences in age and education between groups of patients who reported recent only, past only, both recent and past, and never having discussions of prognosis/life expectancy with oncologists were examined with means and standard deviations (SDs); between-group differences in sex, race/ethnicity, marital status, and insurance status were examined with distributions of absolute and relative frequencies. Correlations between age or education and pre- and post-scan visit illness understanding scores, and change in illness understanding score, were used to evaluate associations between age or education and illness understanding. Change in illness understanding and differences in pre-scan visit and post-scan visit understanding on the basis of sex, race/ethnicity, marital status, and insurance status were examined with means and standard deviations.

ng score, were used to evaluate associations between age or education and illness understanding. Change in illness understanding and differences in pre-scan visit and post-scan visit understanding on the basis of sex, race/ethnicity, marital status, and insurance status were examined with means and standard deviations. Mean changes in illness understanding scores for patients who reported recent only, past only, both recent and past, and never having discussions of prognosis/life expectancy with oncologists, adjusted for potential patient demographic confounders, were estimated via least-squares means in a generalized linear model of changes in illness understanding by patients. Minimum effect sizes for changes in illness understanding scores for tests with adequate (80%) statistical power for the full sample (N = 178) and for subgroups with 20 or 60 patients were 0.17, 0.53, and 0.29, respectively. Time between interviews was unrelated to changes in illness understanding and, therefore, was not considered a confounder in the analysis. Statistical analysis was conducted with SAS statistical software, version 9.4 (Cary, NC) and was based on two-sided tests. P < .05 was considered statistically significant.

Mean changes in illness understanding scores for patients who reported recent only, past only, both recent and past, and never having discussions of prognosis/life expectancy with oncologists, adjusted for potential patient demographic confounders, were estimated via least-squares means in a generalized linear model of changes in illness understanding by patients. Minimum effect sizes for changes in illness understanding scores for tests with adequate (80%) statistical power for the full sample (N = 178) and for subgroups with 20 or 60 patients were 0.17, 0.53, and 0.29, respectively. Time between interviews was unrelated to changes in illness understanding and, therefore, was not considered a confounder in the analysis. Statistical analysis was conducted with SAS statistical software, version 9.4 (Cary, NC) and was based on two-sided tests. P < .05 was considered statistically significant. RESULTS The median time between the pre- and post-scan interviews was 6 weeks (range, 1 to 32 weeks). At pre-scan interviews, 32 patients (18%) had an illness understanding score of 0; 48 (27%), of 1; 47 (26%), of 2; 42 (24%), of 3; and nine (5%), of 4. Post-scan interviews revealed that 26 (15%), 46 (26%), 49 (28%), 44 (25%), and 13 (7%) had illness understanding scores of 0, 1, 2, 3, and 4, respectively. Between interviews, three patients (2%) had a change in illness understanding score of −2; 29 (16%) had a change score of −1; 96 (54%) had a change score of 0; 41 (23%) had a change score of +1; and nine (5%) had a change score of +2. According to patient reports, 18 (10%) had only recent discussions of prognosis/life expectancy with the oncologist; 68 (38%) had only past discussions; 24 (13%) had both recent and past discussions; and 68 (38%) never had discussions of prognosis/life expectancy with the oncologist.

nine (5%) had a change score of +2. According to patient reports, 18 (10%) had only recent discussions of prognosis/life expectancy with the oncologist; 68 (38%) had only past discussions; 24 (13%) had both recent and past discussions; and 68 (38%) never had discussions of prognosis/life expectancy with the oncologist. As listed in Table 1, patients who reported only past (n = 68) or both recent and past (n = 24) discussions were more highly educated (mean years of education: 15.3; SD, 3.0 and 15.3; SD, 2.7, respectively) than patients who reported only recent (n = 18) or never having (n = 68) discussions (mean years of education: 13.4; SD, 2.9 and 13.3; SD, 4.2, respectively). Black patients were more highly represented in the recent only and never having discussions groups (4 [22%] of 18 patients and 15 [22%] of 68 patients, respectively) and were less highly represented in the past discussions only group (3[4%] of 68 patients). Married patients were more highly represented in the only past discussions group (47 [70%] of 67 patients) and were less represented in the both recent and past discussions group (10 [43%] of 23 patients). Table 1. Patient Characteristics and Their Associations With Reported Discussions of Prognosis/Life Expectancy

As listed in Table 1, patients who reported only past (n = 68) or both recent and past (n = 24) discussions were more highly educated (mean years of education: 15.3; SD, 3.0 and 15.3; SD, 2.7, respectively) than patients who reported only recent (n = 18) or never having (n = 68) discussions (mean years of education: 13.4; SD, 2.9 and 13.3; SD, 4.2, respectively). Black patients were more highly represented in the recent only and never having discussions groups (4 [22%] of 18 patients and 15 [22%] of 68 patients, respectively) and were less highly represented in the past discussions only group (3[4%] of 68 patients). Married patients were more highly represented in the only past discussions group (47 [70%] of 67 patients) and were less represented in the both recent and past discussions group (10 [43%] of 23 patients). Table 1. Patient Characteristics and Their Associations With Reported Discussions of Prognosis/Life Expectancy Patient Characteristic Full Sample (N = 178) Reported Discussions of Prognosis/Life Expectancy Recent Only (n = 18; 10.1%) Past Only (n = 68; 38.2%) Both (n = 24; 13.5%) Never (n = 68; 38.2%) No. % No. % No. % No. % No. % Age, years, mean (SD) 59.7 (9.9) 57.3 (12.2) 60.3 (9.8) 58.5 (7.9) 60.1 (10.0) Education, years, mean (SD) 14.4 (3.6) 13.4 (2.9) 15.3 (3.0) 15.3 (2.7) 13.3 (4.2) Sex Male 58 32.8 6 33.3 15 22.4 9 37.5 28 41.2 Female 119 67.2 12 66.7 52 77.6 15 62.5 40 58.8 Race/ethnicity Black 25 14.0 4 22.2 3 4.4 3 12.5 15 22.1 Latino 19 10.7 1 5.6 6 8.8 3 12.5 9 13.2 White 134 75.3 13 72.2 59 86.8 18 75.0 44 64.7 Marital status Married 101 57.7 10 55.6 47 70.1 10 43.5 34 50.7 Not married 74 42.3 8 44.4 20 29.9 13 56.5 33 49.3 Insurance status Insured 149 83.7 14 77.8 62 91.2 21 87.5 52 76.5 Not insured 29 16.3 4 22.2 6 8.8 3 12.5 16 23.5 NOTE: Missing data: age (n = 1), sex (n = 1), marital status (n = 3).

86.8 18 75.0 44 64.7 Marital status Married 101 57.7 10 55.6 47 70.1 10 43.5 34 50.7 Not married 74 42.3 8 44.4 20 29.9 13 56.5 33 49.3 Insurance status Insured 149 83.7 14 77.8 62 91.2 21 87.5 52 76.5 Not insured 29 16.3 4 22.2 6 8.8 3 12.5 16 23.5 NOTE: Missing data: age (n = 1), sex (n = 1), marital status (n = 3). Abbreviation: SD, standard deviation. Table 2 lists changes in illness understanding scores by characteristic, including the following: Black patients (n = 25) had negative changes in illness understanding scores (mean, −0.24; SD, 1.01), and Latino (n = 19) and white (n = 134) patients had positive changes in illness understanding scores (mean, 0.37; SD, 0.76 and mean, 0.17; SD, 0.75, respectively). Overall, patients (N = 178) had positive changes in illness understanding scores (mean, 0.13; SD, 0.81). Table 2. Associations Between Patient Characteristics and Illness Understanding Scores

Table 2 lists changes in illness understanding scores by characteristic, including the following: Black patients (n = 25) had negative changes in illness understanding scores (mean, −0.24; SD, 1.01), and Latino (n = 19) and white (n = 134) patients had positive changes in illness understanding scores (mean, 0.37; SD, 0.76 and mean, 0.17; SD, 0.75, respectively). Overall, patients (N = 178) had positive changes in illness understanding scores (mean, 0.13; SD, 0.81). Table 2. Associations Between Patient Characteristics and Illness Understanding Scores Patient Characteristic No. of Patients Illness Understanding Score Pre-Scan Visit Post-Scan Visit Change Mean SD r Mean SD r Mean SD r Overall 178 1.71 1.16 1.84 1.17 0.13 0.81 Age, years 177 0.00 0.04 0.05 Education, years 178 0.14 0.24 0.14 Sex Male 58 1.52 1.13 1.62 1.17 0.10 0.83 Female 119 1.78 1.16 1.94 1.16 0.16 0.79 Race/ethnicity Black 25 1.72 1.10 1.48 0.92 −0.24 1.01 Latino 19 1.63 1.30 2.00 1.05 0.37 0.76 White 134 1.72 1.16 1.89 1.22 0.17 0.75 Marital status Married 101 1.67 1.23 1.82 1.15 0.15 0.73 Not married 74 1.74 1.07 1.88 1.20 0.14 0.91 Insurance status Insured 149 1.68 1.14 1.81 1.21 0.13 0.82 Not insured 29 1.86 1.27 2.03 0.94 0.17 0.76 NOTE: Missing data: age (n = 1), sex (n = 1), marital status (n = 3). r indicates the correlation coefficient. Abbreviation: SD, standard deviation.

Patient Characteristic No. of Patients Illness Understanding Score Pre-Scan Visit Post-Scan Visit Change Mean SD r Mean SD r Mean SD r Overall 178 1.71 1.16 1.84 1.17 0.13 0.81 Age, years 177 0.00 0.04 0.05 Education, years 178 0.14 0.24 0.14 Sex Male 58 1.52 1.13 1.62 1.17 0.10 0.83 Female 119 1.78 1.16 1.94 1.16 0.16 0.79 Race/ethnicity Black 25 1.72 1.10 1.48 0.92 −0.24 1.01 Latino 19 1.63 1.30 2.00 1.05 0.37 0.76 White 134 1.72 1.16 1.89 1.22 0.17 0.75 Marital status Married 101 1.67 1.23 1.82 1.15 0.15 0.73 Not married 74 1.74 1.07 1.88 1.20 0.14 0.91 Insurance status Insured 149 1.68 1.14 1.81 1.21 0.13 0.82 Not insured 29 1.86 1.27 2.03 0.94 0.17 0.76 NOTE: Missing data: age (n = 1), sex (n = 1), marital status (n = 3). r indicates the correlation coefficient. Abbreviation: SD, standard deviation. Table 3 lists mean changes in elements of illness understanding, stratified by discussions of prognosis/life expectancy. The largest contributors to changes in illness understanding among patients who reported only recent discussions of prognosis/life expectancy (n = 18; mean change, 0.50; SD, 0.86) were changes in understanding of incurability (mean, 0.17; SD, 0.38) and late stage of disease (mean, 0.17; SD, 0.62). The largest contributors to changes in illness understanding among patients who reported both recent and past discussions (n = 24; mean change, 0.38; SD, 0.92) were changes in understanding of late stage of disease (mean, 0.21; SD, 0.51) and in expectations to live months, not years (mean, 0.21; SD, 0.41).

; SD, 0.62). The largest contributors to changes in illness understanding among patients who reported both recent and past discussions (n = 24; mean change, 0.38; SD, 0.92) were changes in understanding of late stage of disease (mean, 0.21; SD, 0.51) and in expectations to live months, not years (mean, 0.21; SD, 0.41). Table 3. Composition of Change in Illness Understanding Score, Stratified by Discussions of Prognosis/Life Expectancy Reported Discussions of Prognosis/Life Expectancy No. of Patients Mean Changes in Elements of Illness Understanding Total Score* TIA Incurable Late Stage Months to Live Mean SD Mean SD Mean SD Mean SD Mean SD Only recent 18 0.11 0.58 0.17 0.38 0.17 0.62 0.06 0.42 0.50 0.86 Only past 68 0.00 0.42 0.12 0.37 0.09 0.38 −0.04 0.27 0.16 0.64 Both recent and past 24 −0.13 0.45 0.08 0.50 0.21 0.51 0.21 0.41 0.38 0.92 Never 68 −0.07 0.50 0.00 0.39 0.03 0.46 −0.03 0.17 −0.07 0.85 NOTE. Changes refer to difference between pre- and post-scan illness understanding. Abbreviations: SD, standard deviation; TIA, terminal illness acknowledgment. * Total score is the summation of the four elements of illness understanding (ie, TIA, incurable, late state, months to live).

Reported Discussions of Prognosis/Life Expectancy No. of Patients Mean Changes in Elements of Illness Understanding Total Score* TIA Incurable Late Stage Months to Live Mean SD Mean SD Mean SD Mean SD Mean SD Only recent 18 0.11 0.58 0.17 0.38 0.17 0.62 0.06 0.42 0.50 0.86 Only past 68 0.00 0.42 0.12 0.37 0.09 0.38 −0.04 0.27 0.16 0.64 Both recent and past 24 −0.13 0.45 0.08 0.50 0.21 0.51 0.21 0.41 0.38 0.92 Never 68 −0.07 0.50 0.00 0.39 0.03 0.46 −0.03 0.17 −0.07 0.85 NOTE. Changes refer to difference between pre- and post-scan illness understanding. Abbreviations: SD, standard deviation; TIA, terminal illness acknowledgment. * Total score is the summation of the four elements of illness understanding (ie, TIA, incurable, late state, months to live). In Table 4, results adjusted for potential confounders (ie, patient years of education and race/ethnicity) are listed. Groups of patients who reported recent only and both recent and past discussions of prognosis/life expectancy with their oncologists had significant, positive changes in their illness understanding scores (least-squares mean change score: 0.62; 95% CI ,0.23 to 1.01; P = .002 and 0.37; 95% CI, 0.04 to 0.70; P = .028, respectively). Table 4. Changes in Patient Illness Understanding in Relation to Discussions of Prognosis/Life Expectancy

In Table 4, results adjusted for potential confounders (ie, patient years of education and race/ethnicity) are listed. Groups of patients who reported recent only and both recent and past discussions of prognosis/life expectancy with their oncologists had significant, positive changes in their illness understanding scores (least-squares mean change score: 0.62; 95% CI ,0.23 to 1.01; P = .002 and 0.37; 95% CI, 0.04 to 0.70; P = .028, respectively). Table 4. Changes in Patient Illness Understanding in Relation to Discussions of Prognosis/Life Expectancy Discussion of Prognosis/Life Expectancy No. of Patients Change in Illness Understanding Score P LS Mean 95% CI Only recent 18 0.62 0.23 to 1.01 .002 Only past 68 0.15 −0.09 to 0.39 .221 Both recent and past 24 0.37 0.04 to 0.70 .028 Never 68 0.02 −0.21 to 0.24 .885 Abbreviation: LS, least-squares mean score adjusted for patient race/ethnicity and years of education.

atients Change in Illness Understanding Score P LS Mean 95% CI Only recent 18 0.62 0.23 to 1.01 .002 Only past 68 0.15 −0.09 to 0.39 .221 Both recent and past 24 0.37 0.04 to 0.70 .028 Never 68 0.02 −0.21 to 0.24 .885 Abbreviation: LS, least-squares mean score adjusted for patient race/ethnicity and years of education. DISCUSSION Results of this study demonstrate how poorly patients with advanced cancer understand their prognoses and how effective recent prognostic discussions are to improve illness understanding by patients. All enrolled patients in this study had incurable cancer that was at an advanced stage (eg, late, stage IV gastrointestinal cancer) and a life expectancy of months, not years. A small minority of patients accurately, and completely, understood the gravity of their illnesses (eg, 5% endorsed each element of the terminal prognosis at study entry); approximately one in four (23%) reported only recent or recent and past discussion of prognosis with the oncologist. Patients who reported at least a recent discussion about prognosis with the oncology provider exhibited significant improvements in illness understanding. These results highlight the need for timely (ie, current) prognostic disclosures to terminally ill patients who meet the criteria used for this study. The results also suggest that oncologists should discuss prognosis on an ongoing basis, and as frequently as appropriate, with their terminally ill patients. If this occurred, patients would likely have better illness understanding and, thus, make more informed decisions about their end-of-life care.

ed for this study. The results also suggest that oncologists should discuss prognosis on an ongoing basis, and as frequently as appropriate, with their terminally ill patients. If this occurred, patients would likely have better illness understanding and, thus, make more informed decisions about their end-of-life care. These results are consistent with, and advance, the existing literature on illness understanding, prognostic disclosure, and advance care planning. The effect of recently reported prognostic discussions on improvements in illness understanding by patients is in line with the advance care planning strategy to regularly and dually address both dynamic medical situations and individual patient goals. This approach encourages medical decisions to be made in the moment14 instead of on the basis of advance directive documents, which can sometimes be nonspecific, outdated, or unavailable.15,16 Consideration of prognostic understanding as an evolving awareness of one’s changing health empowers patients, their loved ones, and their healthcare team to make informed decisions. Furthermore, recognition of the need to update patients frequently about a prognosis may help patients and families who struggle17 to come to terms with the terminal nature of a disease.

volving awareness of one’s changing health empowers patients, their loved ones, and their healthcare team to make informed decisions. Furthermore, recognition of the need to update patients frequently about a prognosis may help patients and families who struggle17 to come to terms with the terminal nature of a disease. In the delicate task of delivering prognoses, some have argued that the median informs the message,18 which argues for the use of a prognostic range such as months instead of communication of a specific time frame, such as 6 months to live. Outcomes research in strategies of communicating with both realism and hope18 for patients with serious illness is needed; statements, such as hoping for the best (eg, years of survival) while being prepared for the worst (eg, months left to live), during ongoing discussions of prognosis may be one way to achieve a balance. This report suggests that, regardless of the approach, the recency of the prognostic discussion matters for prognostic understanding by the patient. Future research is needed to identify the most effective ways to communicate prognostic information to ensure that patients have accurate illness understanding. Such insight seems to be a prerequisite for informed decision making.

recency of the prognostic discussion matters for prognostic understanding by the patient. Future research is needed to identify the most effective ways to communicate prognostic information to ensure that patients have accurate illness understanding. Such insight seems to be a prerequisite for informed decision making. There are strengths and limitations to our study. One strength is that our data were drawn from a large, prospective, observational cohort of patients from several centers and with cancers representative of common terminal illnesses, in a study explicitly designed to discern the effects of oncology provider communication on terminal illness understanding by patients. To our knowledge, our study is the first to directly address and demonstrate these associations between the timing of patient-reported prognostic discussions and improvements in illness understanding by patients. One weakness is that patients who do not know their cancer stages or prognoses might also inaccurately recall whether their doctors have talked with them about such topics. This notion of discord, because of optimism bias19 or misunderstanding between what physicians say and what patients19 or caregivers20 hear has been described in oncology settings.20 If patients misheard what was said, the effect was likely in the direction of underreporting prognostic discussions; this would suggest that prognostic discussions by oncologists would have less impact. We contend that, for informed decision making, how patients hear and understand what their oncology providers say about their illnesses matters the most.

the effect was likely in the direction of underreporting prognostic discussions; this would suggest that prognostic discussions by oncologists would have less impact. We contend that, for informed decision making, how patients hear and understand what their oncology providers say about their illnesses matters the most. Despite these limitations, future research directions include additional elucidation of the communication elements that are beneficial (or deleterious) to patient understanding and that promote advance care planning. Studies to investigate the combined use of booklets and audio recordings21 for education to patients with cancer about the chemotherapies they will receive have proven effective. Larger-scale validation of these approaches is warranted, as is ongoing research into other educational media, such as videos,22 for these and other topics of importance to patients with cancer and their loved ones. Communication decision aids for caregivers themselves have also been effective advance care planning strategies.23 Although the impact of communication skills training on patient outcomes has recently been called into question,10 other data show that communication skills training, whether as workshops24-26 or technologies for individual practice,27 can at least help clinicians acquire these important communication skills. Ongoing research into communication skills training needs to examine the interplay between cognitive information delivery (eg, communication about the prognostic realities of an advanced cancer as realistically and hopefully as possible) and response to emotion (eg, patient sadness, anxiety, anger) with empathic responses and the effects that such skills have on illness understanding by patients.28

y between cognitive information delivery (eg, communication about the prognostic realities of an advanced cancer as realistically and hopefully as possible) and response to emotion (eg, patient sadness, anxiety, anger) with empathic responses and the effects that such skills have on illness understanding by patients.28 In conclusion, patients with advanced cancer who acknowledge recent or ongoing discussions of prognosis/life expectancy with their oncology providers come to have a better understanding of the terminal nature of their illnesses and, thus, may be better prepared to make informed end-of-life care decisions. Supported in part by the National Cancer Institute (Grants No. CA106370 and CA197730 to H.G.P.), the National Palliative Care Research Center (2013 Career Development Award to A.S.E.), the National Institute of Minority Health Disparities (Grant No. MD007652 to H.G.P. and P.K.M.), and the National Cancer Institute Cancer Center Support Grant No. P30 CA008748. Presented in abstract form at the 51st Annual Meeting of the American Society of Clinical Oncology (ASCO), Chicago, IL, May 29-June 2, 2015, and at the ASCO Palliative Care in Oncology Symposium, Boston, MA, October 9-10, 2015. Authors’ disclosures of potential conflicts of interest are found in the article online at www.jco.org. Author contributions are found at the end of this article.

Presented in abstract form at the 51st Annual Meeting of the American Society of Clinical Oncology (ASCO), Chicago, IL, May 29-June 2, 2015, and at the ASCO Palliative Care in Oncology Symposium, Boston, MA, October 9-10, 2015. Authors’ disclosures of potential conflicts of interest are found in the article online at www.jco.org. Author contributions are found at the end of this article. Acknowledgment H.G.P., principal investigator of work funded by National Cancer Institute Grant No. CA106370, was awarded the competing continuation of this grant while on the faculty at the Dana-Farber Cancer Institute. Data collection occurred at this site, along with eight other sites, before and after the coordinating site for this study relocated to Weill Cornell Medicine. AUTHOR CONTRIBUTIONS Conception and design: Andrew S. Epstein, Holly G. Prigerson, Paul K. Maciejewski Financial support: Holly G. Prigerson Provision of study materials or patients: All authors Collection and assembly of data: All authors Data analysis and interpretation: All authors Manuscript writing: All authors Final approval of manuscript: All authors

AUTHOR CONTRIBUTIONS Conception and design: Andrew S. Epstein, Holly G. Prigerson, Paul K. Maciejewski Financial support: Holly G. Prigerson Provision of study materials or patients: All authors Collection and assembly of data: All authors Data analysis and interpretation: All authors Manuscript writing: All authors Final approval of manuscript: All authors AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST Discussions of Life Expectancy and Changes in Illness Understanding in Patients With Advanced Cancer The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or jco.ascopubs.org/site/ifc. Andrew S. Epstein No relationship to disclose Holly G. Prigerson No relationship to disclose Eileen M. O'Reilly Consulting or Advisory Role: Celgene, Hexal (I), Aduro Biotech, Astellas Pharma (I), Celsion (I), Cipla, Eli Lilly (I), Exelixis (I), IntegraGen (I), Jennerex (I), MedImmune, Novartis (I), Pharmacyclics, Sanofi, Silenseed, Vicus Therapeutics, Gilead Sciences, Merrimack Research Funding: Momenta Pharmaceuticals (Inst), Incyte (Inst), Immunomedics (Inst), Myriad Genetics (Inst), OncoMed (Inst), Clovis Oncology (Inst), Sanofi (Inst), Celgene (Inst), Polaris (Inst) Paul K. Maciejewski No relationship to disclose

Surgical cytoreduction followed by platinum-based chemotherapy has been the standard first-line treatment of patients with high-risk early-stage and advanced epithelial ovarian cancer for nearly two decades.1-4 Although the majority of women with advanced disease will respond to combined platinum/taxane therapy, most will ultimately experience recurrence and eventually die as a result of their ovarian cancer. Platinum-refractory and -resistant disease is defined as progression on first-line chemotherapy or within 6 months of platinum completion, respectively. These patients are treated with nonplatinum chemotherapy; however, anticipated response rates are low.5 Despite the urgent need for more effective treatments, few new agents have demonstrated sufficient efficacy to warrant approval by the Food and Drug Administration (FDA) in the last 10 years. In 2014, the Avastin Use in Platinum-Resistant Epithelial Ovarian Cancer (AURELIA) trial established that the addition of bevacizumab to nonplatinum chemotherapy increases progression-free survival (PFS) from 3.4 to 6.7 months, leading to FDA approval.6 In 2015, Kaufman et al7 showed that olaparib achieved a 31% response rate in heavily pretreated women with germline BRCA1/2 mutations, also leading to FDA approval in germline BRCA1/2 carriers. Despite this recent progress, there remains a significant unmet need for improved therapies in platinum-resistant and -refractory ovarian cancer.

al7 showed that olaparib achieved a 31% response rate in heavily pretreated women with germline BRCA1/2 mutations, also leading to FDA approval in germline BRCA1/2 carriers. Despite this recent progress, there remains a significant unmet need for improved therapies in platinum-resistant and -refractory ovarian cancer. In the article that accompanies this editorial, Liu et al8 report the results of a 233-patient open-label, randomized phase II trial of once-weekly paclitaxel with or without the human epidermal growth factor receptor 3 (HER3) antibody, seribantumab, in platinum-resistant/refractory epithelial ovarian cancer. Patients were randomly assigned 2:1 in favor of seribantumab and enrolled without prospective biomarker selection or stratification. Unfortunately, the study did not reach its primary end point, showing no difference in PFS between the two arms (3.8 months with the combination compared with 3.7 months with paclitaxel alone). Despite this disappointing result, the study fortunately mandated the collection of both archival and fresh tumor biopsies, and these were used to conduct an extensive retrospective biomarker analysis to determine if a subset of patients may have benefited from the addition of seribantumab. By evaluating multiple biomarkers relevant to the mechanism of action of seribantumab, Liu et al8 identified a tumor-based bivariate signature defined by high heregulin (HRG, also named neuregulin 1), the ligand of HER3, and low human epidermal growth factor receptor 2 (HER2) that was predictive of seribantumab benefit. Specifically, in the 38% of evaluable patients who were biomarker positive, the median PFS was 5.7 months with the combination compared with 3.5 months with paclitaxel alone. In addition to being predictive of benefit to seribantumab, this signature seemed to be prognostic in the paclitaxel monotherapy arm, with biomarker-positive patients experiencing more rapid disease progression (PFS, 3.5 months v 5.4 months). Together, these observations suggest that seribantumab overcomes the negative prognosis associated with high HRG and low HER2 levels in platinum-resistant/refractory ovarian cancer. Distressingly, the biomarker-negative patients did worse when exposed to seribantumab, although the mechanism underlying this apparent harm is not understood.

rvations suggest that seribantumab overcomes the negative prognosis associated with high HRG and low HER2 levels in platinum-resistant/refractory ovarian cancer. Distressingly, the biomarker-negative patients did worse when exposed to seribantumab, although the mechanism underlying this apparent harm is not understood. HER3, the protein encoded by ErbB3, is a member of the human epidermal growth factor (EGFR) family and the only one that lacks catalytic kinase function. Instead, HER3 mediates its effects on signaling through heterodimerization with, and allosteric activation of, other EGFR family members, leading to downstream activation of the phosphatidylinositol 3-kinase/AKT pathway.9-11 Pertuzumab, an anti-HER2 monoclonal antibody, is believed to act by preventing dimerization with HER3 and has been approved for treatment of HER2-positive breast cancer, further credentialing HER3 as a therapeutic target in cancer.12 In ovarian cancer, HER3 is highly expressed in a subset of patients and is associated with a worse prognosis.13 Autocrine signaling loops between HER3 and its ligand, HRG, promote growth in patient-derived ovarian cancer models and cell lines.14 Treatment of ovarian cancer cell lines with certain chemotherapies increases activation of HER3, suggesting HER3 may be one of several mechanisms responsible for chemotherapy resistance in ovarian cancer.15 Seribantumab is a fully humanized monoclonal antibody that blocks binding of HRG to HER3 and has been shown to cause tumor growth arrest in ovarian cancer xenograph models.16 In tumors with high HRG expression, seribantumab is believed to block ligand-dependent activation of HER2/HER3 dimers. Conversely, high levels of HER2, leading to a greater presence of HER2/HER3 dimers, may mitigate therapeutic benefit by promoting ligand-independent signaling. Thus, the findings by Liu et al8 that the combination of high HRG and low HER2 levels were associated with benefit to seribantumab is consistent with preclinical predictions. In distinction, the finding that biomarker-negative patients fared worse when treated with seribantumab is not explained by this proposed mechanism and therefore raises important unanswered questions.

f high HRG and low HER2 levels were associated with benefit to seribantumab is consistent with preclinical predictions. In distinction, the finding that biomarker-negative patients fared worse when treated with seribantumab is not explained by this proposed mechanism and therefore raises important unanswered questions. Despite this, preliminary reports from studies in non–small-cell lung cancer and breast cancer have provided additional clinical corroboration of this predictive biomarker.17,18 There is currently an ongoing, potentially registration-enabling randomized phase II trial of chemotherapy with or without seribantumab in HRG-positive non–small-cell lung cancer.19

studies in non–small-cell lung cancer and breast cancer have provided additional clinical corroboration of this predictive biomarker.17,18 There is currently an ongoing, potentially registration-enabling randomized phase II trial of chemotherapy with or without seribantumab in HRG-positive non–small-cell lung cancer.19 Although both the general association with and directionality of HRG and HER2 levels with benefit of seribantumab match preclinical expectations, some caution is appropriate when interpreting the results of the current study. Tumor levels of HRG, HER3, HER2, EGFR, and betacellulin (an EGF family ligand) were measured using four orthogonal techniques (reverse transcriptase quantitative polymerase chain reaction, RNA–in situ hybridization, fluorescence-based quantitative immunohistochemistry, and chromogenic RNA-in situ hybridization). In the resulting analysis, 13 unique marker/measurement combinations were tested for association with treatment outcome in both archived and fresh tumor biopsies. Because of the exploratory nature of the analysis, no modifications were used to account for multiple hypothesis testing, and therefore the possibility of false discovery cannot be excluded. On univariate analysis, all four analytes were significantly associated with outcome, although associations were not concordant across the different measurement techniques used for each analyte. Although this finding may be caused by differences in the performance of various quantification techniques when using limited quantity or degraded tumor material, the observation provides some cause for concern. Moreover, although the markers evaluated were prespecified by the protocol, the specifics of this analysis itself and cut points used were not. Finally, only 57 patients meeting criteria for biomarker positivity were enrolled in the current study, limiting our ability to estimate the true effect size of seribantumab in this population with precision.

valuated were prespecified by the protocol, the specifics of this analysis itself and cut points used were not. Finally, only 57 patients meeting criteria for biomarker positivity were enrolled in the current study, limiting our ability to estimate the true effect size of seribantumab in this population with precision. Given these various considerations, our next steps as a field should be guided by how confident we are that this analysis has identified the optimal patient selection strategy for HER3-targeted therapy in platinum-resistant ovarian cancer. A definitive phase III superiority study pursuing these preliminary findings would require approximately 250 biomarker-positive patients if targeting a hazard ratio of 0.65 for the combination versus paclitaxel alone, assuming the reported PFS difference and using a two-sided type I error of 5% with 90% power. The prospect that biomarker-negative patients may be harmed by seribantumab would necessitate prospective selection of only biomarker-positive patients for enrollment. Assuming a biomarker positivity rate of 40%, this study would require screening of more than 600 patients. Given the discordant biomarker status defined using archival and pretreatment tumor biopsies, patients may require fresh biopsies for screening. Thus, a definitive phase III study would be a major undertaking for both investigators and patients. Does the current rigorous but ultimately exploratory biomarker analysis support moving forward in this manner without further clinical and analytic validation of this selection strategy in ovarian cancer? A more conservative alternative approach could be to conduct a smaller follow-up study using prospective biomarker selection and incorporating less stringent statistical controls to provide additional support for these preliminary observations and refine end points for a future definitive study.

tegy in ovarian cancer? A more conservative alternative approach could be to conduct a smaller follow-up study using prospective biomarker selection and incorporating less stringent statistical controls to provide additional support for these preliminary observations and refine end points for a future definitive study. In conclusion, we congratulate Liu et al8 for not only conducting a well-designed trial but also having the foresight to collect the biospecimens necessary to conduct a rigorous biomarker evaluation. Their efforts have identified a potential path forward for this drug in ovarian cancer, salvaging what otherwise would have been a negative study. Moving forward, it is critical that we follow in the example of Liu et al8 and ensure that we have the opportunity to learn from our failures to ultimately improve the outcome for our patients. See accompanying article on page 4345 ACKNOWLEDGMENT Dr Schram’s training and education is supported in part by the NCI MSK T32 Investigational Cancer Therapeutics Training Program Grant (T32-CA009207). Supported by the National Institutes of Health Grant No. P30 CA008748 and T32-CA009207. AUTHOR CONTRIBUTIONS Administrative support: David M. Hyman Manuscript writing: All authors Final approval of manuscript: All authors

ACKNOWLEDGMENT Dr Schram’s training and education is supported in part by the NCI MSK T32 Investigational Cancer Therapeutics Training Program Grant (T32-CA009207). Supported by the National Institutes of Health Grant No. P30 CA008748 and T32-CA009207. AUTHOR CONTRIBUTIONS Administrative support: David M. Hyman Manuscript writing: All authors Final approval of manuscript: All authors AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST Picking the Right Patient for Human Epidermal Growth Factor Receptor 3–Targeted Therapy in Platinum-Resistant Ovarian Cancer The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or jco.ascopubs.org/site/ifc. Alison M. Schram No relationship to disclose Alexia Iasonos No relationship to disclose David M. Hyman Consulting or Advisory Role: Atara Biotherapeutics, CytomX Therapeutics Research Funding: Puma Biotechnology, Loxo Oncology, AstraZeneca

INTRODUCTION Inherited mutations in BRCA1 and BRCA2 and Lynch syndrome (LS) account for a significant minority (15% to 25%) of ovarian cancers (OCs)1,2 and confer a high risk for OC: 11% to 37% by age 70 years in BRCA2 carriers and 39% to 65% in BRCA1 carriers.3,4 Other lower-penetrance homologous repair pathway genes have been implicated in familial OC.5,6 Although medium-term survival with BRCA-associated OC exceeds that of sporadic OC,7,8 the long-term outlook remains poor.9 Risk-reducing salpingo-oophorectomy (RRSO) for women older than 35 years of age to prevent OC or fallopian tube cancer (FTC) and to detect occult neoplasia is recommended as the only proven mortality-reducing intervention.10,11 Although effective when used premenopausally,10,11 RRSO causes infertility and premature menopause, with associated cardiovascular risks,12 osteoporosis,13 and neurologic risks14 (although premature menopause can be treated with hormone replacement therapy). Some women decline RRSO regardless of OC risk, and others prefer to defer it (eg, until menopause). Effective OC screening would be a welcome option for such women.

h associated cardiovascular risks,12 osteoporosis,13 and neurologic risks14 (although premature menopause can be treated with hormone replacement therapy). Some women decline RRSO regardless of OC risk, and others prefer to defer it (eg, until menopause). Effective OC screening would be a welcome option for such women. Annual OC screening in the general population that uses a cutoff for the serum tumor marker cancer antigen 125 (CA-125) was associated with improved survival.15 In the high-risk population, we16 and others17-20 have reported annual screening using a CA-125 cutoff and transvaginal sonography (TVS). Although we demonstrated high sensitivity (> 80%) and positive predictive value (PPV; 25%), two symptomatic interval cancers occurred, and 69% of detected cancers were stage III to IV.16 This annual screening interval has been associated with a poor 10-year survival rate of 36% in BRCA1/2 carriers.21 Multimodal screening with the risk of ovarian cancer algorithm (ROCA) to interpret serial CA-125 results, and TVS as a second-line test, in the randomized general-population United Kingdom Collaborative Trial of Ovarian Cancer Screening (UKCTOCS) achieved high sensitivity and specificity.22,23 Significantly more (40%) low-volume (stages I, II, or IIIa) invasive epithelial ovarian/peritoneal cancers were identified compared with unscreened controls (26%) on an intention-to-screen analysis, and the trial provided an encouraging, though not definitive, mortality benefit.24

high sensitivity and specificity.22,23 Significantly more (40%) low-volume (stages I, II, or IIIa) invasive epithelial ovarian/peritoneal cancers were identified compared with unscreened controls (26%) on an intention-to-screen analysis, and the trial provided an encouraging, though not definitive, mortality benefit.24 Random assignment to a nonscreening arm is thought to be unacceptable to high-risk women and clinicians.16 Even if ethical, it would likely be unfeasible, so research screening in this population is limited to prospective cohort studies. To our knowledge, this is the first published study to use ROCA-based screening to define sensitivity in the high-risk population. PATIENTS AND METHODS A prospective multicenter cohort screening study was undertaken within the United Kingdom (UK) National Health Service (NHS). Ethical approval was given by the Eastern Multicentre Research Ethics Committee (Reference No. 97/5/007). The protocol can be found online.25 Entry Criteria High-risk women at an estimated minimum 10% lifetime risk of OC were recruited; inclusion criteria (Data Supplement, online only) depended on family history or predisposing mutations. Documentation (death certificates and/or histopathology reports) of relevant cancers was required, and eligibility was confirmed by the coordinating center (CC). Clinical genetic testing was performed by accredited NHS laboratories. After screening ended, 45.2% of the study population underwent BRCA1/2 next-generation sequencing research testing.26

tes and/or histopathology reports) of relevant cancers was required, and eligibility was confirmed by the coordinating center (CC). Clinical genetic testing was performed by accredited NHS laboratories. After screening ended, 45.2% of the study population underwent BRCA1/2 next-generation sequencing research testing.26 Recruitment Participants were recruited at 42 centers in the UK by specialist nurses, clinical geneticists, or gynecologists. In December 2006, participants in the UK Familial Ovarian Cancer Screening Study (UK FOCSS) Phase I (annual screening using a CA-125 cutoff and TVS)16 were invited to join this study—UK FOCSS Phase II. Other participants were recruited de novo. Women were counseled about RRSO and screening limitations. All participants provided written consent. Screening The trial database16 scheduled serum CA-125 tests every 4 months and TVS annually. Venipuncture packs were mailed to participants for use in primary care and blood samples returned to the CC laboratory for CA-125 assay.23 Results were uploaded to the database, which calculated OC risk using the high-risk ROCA, which also incorporated the higher a priori risk in our population and different reference levels for risk stratification for postmenopausal compared with premenopausal women, because of the higher baseline CA-125 and variability in premenopausal women.27 Menopausal status was determined by the database by using the age of participants and their responses to questions about gynecologic history and/or symptoms, which were returned with serum samples (Data Supplement).

ith premenopausal women, because of the higher baseline CA-125 and variability in premenopausal women.27 Menopausal status was determined by the database by using the age of participants and their responses to questions about gynecologic history and/or symptoms, which were returned with serum samples (Data Supplement). Initial risk of ovarian cancer (ROC) was based on initial CA-125 level and estimated age-specific OC incidence. Subsequently, ROC was based on absolute CA-125 level and rate of change. Initially high or increasing CA-125 levels (even < 30 iU/ml) generated a high ROC, whereas initially low, stable-high (even > 30 iU/ml), or decreasing levels generated low ROCs. ROCA results were used for triage, including expedition of repeat CA-125 tests and/or TVS after abnormal results (Data Supplement). Collaborating centers performed scans and completed proformas (Data Supplement), which were classified by the database according to predetermined criteria (Data Supplement).16 When indicated, women were referred to a gynecologist for clinical assessment, with a view to surgical removal of the fallopian tubes and ovaries. The final decision about surgery was made after additional investigation and discussion with the patient.

e database according to predetermined criteria (Data Supplement).16 When indicated, women were referred to a gynecologist for clinical assessment, with a view to surgical removal of the fallopian tubes and ovaries. The final decision about surgery was made after additional investigation and discussion with the patient. Follow-Up Participants were flagged (by their unique NHS number) with relevant cancer registries, which provided cancer and/or death data.16 Collaborators notified the CC when women withdrew before routine screening ended (June 30, 2011). Women were observed through cancer registries with censorship that was based on date of death, last notification from the registry, or last contact if they were lost to registry follow-up. Participants were sent health questionnaires in January 2011 and April 2013 specifically asking about surgery that involved removal of fallopian tubes/ovaries and cancer diagnosis.

h censorship that was based on date of death, last notification from the registry, or last contact if they were lost to registry follow-up. Participants were sent health questionnaires in January 2011 and April 2013 specifically asking about surgery that involved removal of fallopian tubes/ovaries and cancer diagnosis. Diagnostic Documentation Whenever women underwent salpingo-oophorectomy, the CC obtained documentation of indication, operation notes, and histopathology/cytopathology reports. These were reviewed by a gynecologic oncologist (A.N.R.) and pathologist (E.B./N.S.) and were classified according to the International Statistical Classification of Diseases and Related Health Problems, 10th revision (ICD-10). Trial surgery was defined as either screen-positive or screen-related (nonconcerning abnormal results, such as simple cysts and/or transient/stable abnormal ROC results that contributed to the participant’s decision to undergo surgery).15 Centers were provided with an RRSO protocol, which advocated serial sectioning of fallopian tubes/ovaries (Data Supplement). A surgical complexity score was assigned using recognized criteria (Data Supplement).28

transient/stable abnormal ROC results that contributed to the participant’s decision to undergo surgery).15 Centers were provided with an RRSO protocol, which advocated serial sectioning of fallopian tubes/ovaries (Data Supplement). A surgical complexity score was assigned using recognized criteria (Data Supplement).28 Statistical Analysis For performance analyses, data were censored 365 days after the last UK FOCSS screen. Invasive OC, FTC, or primary peritoneal cancer (PPC) diagnosed < 365 days after the last screen were included. Cancers that occurred after censoring and diagnosed before February 28, 2016 were reported but not included in the performance analyses. The study was powered to estimate sensitivity within 10% (expected 95% CI), given an annual OC incidence of 0.5%. Analyses were done with Stata (version 14; STATA, College Station, TX). Compliance with blood tests and scans was defined as the proportion of requested tests received by the CC. These were analyzed separately and according to screen type (eg, routine, protocol-indicated repeat).

Statistical Analysis For performance analyses, data were censored 365 days after the last UK FOCSS screen. Invasive OC, FTC, or primary peritoneal cancer (PPC) diagnosed < 365 days after the last screen were included. Cancers that occurred after censoring and diagnosed before February 28, 2016 were reported but not included in the performance analyses. The study was powered to estimate sensitivity within 10% (expected 95% CI), given an annual OC incidence of 0.5%. Analyses were done with Stata (version 14; STATA, College Station, TX). Compliance with blood tests and scans was defined as the proportion of requested tests received by the CC. These were analyzed separately and according to screen type (eg, routine, protocol-indicated repeat). Women who underwent salpingo-oophorectomy were only classified as having undergone RRSO if they were asymptomatic, they had normal results at prior screen, and the recruiting center indicated RRSO as the reason for withdrawal. Cases in which abnormal results prompted surgery were true positive (TP) if invasive epithelial OC/FTC was diagnosed. All other diagnoses (including borderline/benign tumors) that resulted from surgery that was prompted by abnormal results were false positive (FP). Cases in which nonconcerning test results (simple cysts/transiently elevated CA-125) contributed to the decision for surgery were classified as screen-related surgery, to provide estimates of likely additional surgeries in any future screening program. True-negative (TN) designations were for those patients in whom the last screen was normal and no OC/FTC was diagnosed < 365 days. Patients who presented with clinically diagnosed interval cancers between screens or < 365 days after the final screen were considered false negative (FN). Prevalent cases were those diagnosed at first screen. Incident cases were those diagnosed subsequently. For women who transferred from Phase I (annual CA-125 cutoff and scan) to Phase II (ROCA every 4 months and annual TVS), their first Phase II screen was classified as incident.

en were considered false negative (FN). Prevalent cases were those diagnosed at first screen. Incident cases were those diagnosed subsequently. For women who transferred from Phase I (annual CA-125 cutoff and scan) to Phase II (ROCA every 4 months and annual TVS), their first Phase II screen was classified as incident. We reported performance according to whether occult cancers diagnosed < 365 days after a prior screen were classified as FN or TP.16 In an attempt to estimate true sensitivity, we assumed that the proportion of occult cancers identified at RRSO, which would have been screen detected had women not undergone surgery, would be identical to that observed in those who continued screening. We then used the lower confidence limit of observed sensitivity in women who did not undergo RRSO as a conservative estimate of occult cancer detection sensitivity, and we rounded the predicted number of occult cancers detected to the nearest integer. Because the protocol required parallel CA-125 and TVS, the results of which influenced each other’s timing, it was not possible to calculate performance characteristics per test. Therefore, we calculated these metrics per woman-screen year (WSY) for the protocol overall.

We reported performance according to whether occult cancers diagnosed < 365 days after a prior screen were classified as FN or TP.16 In an attempt to estimate true sensitivity, we assumed that the proportion of occult cancers identified at RRSO, which would have been screen detected had women not undergone surgery, would be identical to that observed in those who continued screening. We then used the lower confidence limit of observed sensitivity in women who did not undergo RRSO as a conservative estimate of occult cancer detection sensitivity, and we rounded the predicted number of occult cancers detected to the nearest integer. Because the protocol required parallel CA-125 and TVS, the results of which influenced each other’s timing, it was not possible to calculate performance characteristics per test. Therefore, we calculated these metrics per woman-screen year (WSY) for the protocol overall. To allocate WSYs to correct outcomes we applied the following rules; for TP and FP detection screens, the WSY that commenced with that screen was classified as TP or FP, respectively. WSYs before the detection screen were TN. For occult cancers diagnosed < 365 days after prior screen, the WSY that commenced with that screen was classified as FN or TP (dependent on analysis type), and prior WSYs were TN. For TN cases, all WSYs were classified TN.

that screen was classified as TP or FP, respectively. WSYs before the detection screen were TN. For occult cancers diagnosed < 365 days after prior screen, the WSY that commenced with that screen was classified as FN or TP (dependent on analysis type), and prior WSYs were TN. For TN cases, all WSYs were classified TN. To investigate potentially avoidable delays, we analyzed screening and screen-to-surgery intervals.16 Detection screens were defined as an abnormal TVS and/or abnormal ROC that led to a surgery/biopsy that diagnosed OC/FTC. Delayed screens were defined as any detection screen performed after the protocol-indicated date. Delay was calculated as the detection screen date minus the protocol-indicated date. The interval from screen date to diagnosis was calculated to the date of surgery/biopsy. We compared International Federation of Gynecology and Obstetrics stage and postsurgery zero residual disease rates in OC/FTC diagnosed during and < 365 days from the end of UK FOCSS screening with those diagnosed > 365 days after screening ended. We also compared stage-distribution and zero residual disease rates in incident screen-detected cancers in Phases I and II of the study. No survival analysis was performed because of the low number of events observed.

d < 365 days from the end of UK FOCSS screening with those diagnosed > 365 days after screening ended. We also compared stage-distribution and zero residual disease rates in incident screen-detected cancers in Phases I and II of the study. No survival analysis was performed because of the low number of events observed. RESULTS Between June 14, 2007, and May 15, 2012, 4,531 women were recruited. This included 2,362 (66.3%) of 3,563 eligible women from UK FOCSS Phase I (Fig 1). Table 1 lists inclusion indications. A total of 183 (4.0%) women withdrew before screening (Fig 1). The outcome of the remaining 4,348 women (96.0%) were analyzed. The median age at recruitment was 45.5 years (range, 34.2 to 84.8 years). Of the eligible women, 1,278 women (29.4% of participants) underwent mutation testing, and 1,965 (45.2%) subsequently underwent next-generation sequencing.24 Overall, 924 (21.3%) women were known mutation carriers (further demographics in Data Supplement).

age at recruitment was 45.5 years (range, 34.2 to 84.8 years). Of the eligible women, 1,278 women (29.4% of participants) underwent mutation testing, and 1,965 (45.2%) subsequently underwent next-generation sequencing.24 Overall, 924 (21.3%) women were known mutation carriers (further demographics in Data Supplement). Fig 1. CONSORT diagram. Percentages refer to the proportion of the total in preceding box. (*)Ineligible due to new information about family cancer history, tested negative for family mutation, or already undergoing investigation for abnormal screening results during UK Familial Ovarian Cancer Screening Study Phase I. (†)Unable to establish current whereabouts or nonresponder despite correct address. (‡)Defined as either screen-positive or screen-related (nonconcerning abnormal results, such as simple cysts and/or transient/stable abnormal ROC results that contributed to the participant’s decision to undergo surgery). Includes volunteers who underwent unilateral salpingo-oophorectomy or diagnostic laparoscopy only who returned to screening. (§)Insufficient data to determine indication (all had normal final screen results, none had cancer). RRSO, risk-reducing salpingo-oophorectomy. Table 1. Inclusion Criteria and Mutation Status in Screened Participants (N = 4,348)

Fig 1. CONSORT diagram. Percentages refer to the proportion of the total in preceding box. (*)Ineligible due to new information about family cancer history, tested negative for family mutation, or already undergoing investigation for abnormal screening results during UK Familial Ovarian Cancer Screening Study Phase I. (†)Unable to establish current whereabouts or nonresponder despite correct address. (‡)Defined as either screen-positive or screen-related (nonconcerning abnormal results, such as simple cysts and/or transient/stable abnormal ROC results that contributed to the participant’s decision to undergo surgery). Includes volunteers who underwent unilateral salpingo-oophorectomy or diagnostic laparoscopy only who returned to screening. (§)Insufficient data to determine indication (all had normal final screen results, none had cancer). RRSO, risk-reducing salpingo-oophorectomy. Table 1. Inclusion Criteria and Mutation Status in Screened Participants (N = 4,348) The last cancer notifications from NHS Digital were received on February 28, 2016 (England/Wales), May 15, 2016 (Scotland), and April 19, 2016 (Northern Ireland); the last death notifications were received on March 14, 2016 (all countries). Follow-up was possible for 4,046 women (93.1%). Median follow-up beyond last screen/withdrawal was 4.7 years (range, 0 to 8.7 years).

ceived on February 28, 2016 (England/Wales), May 15, 2016 (Scotland), and April 19, 2016 (Northern Ireland); the last death notifications were received on March 14, 2016 (all countries). Follow-up was possible for 4,046 women (93.1%). Median follow-up beyond last screen/withdrawal was 4.7 years (range, 0 to 8.7 years). Screening/Compliance The 4,348 screened participants underwent 13,728 WSY (median, 3.26 screen-years per woman; range, 1.00 to 5.94 screen-years per woman). A total of 189 women (4.3%) ceased screening by choice. Five hundred fifty-eight (12.8%) ceased because of surgical removal of both fallopian tubes/ovaries for RRSO (n = 534) or indeterminate reasons (n = 24). A total of 377 women (8.7%) whose last screen was abnormal continued screening until May 15, 2012, by which time 315 had normal results and did not undergo surgery. Care was transferred to local gynecologists for the 62 women who still had abnormal results. Three of these 62 women underwent surgery; none had cancer. Compliance with requested routine CA-125 tests and scans were 92.1% (27,138 of 29,450 CA-125 tests) and 94.6% (9,100 of 9,619 scans), respectively (Data Supplement). Compliance for scans was based on reports received, not scans undertaken, so it is likely an underestimate. Protocol-indicated repeat test compliance was higher: 97.4% (4,716 of 4,834) of blood tests, and 98.8% (2,792 of 2,825 ) of scans requested were received.

% (9,100 of 9,619 scans), respectively (Data Supplement). Compliance for scans was based on reports received, not scans undertaken, so it is likely an underestimate. Protocol-indicated repeat test compliance was higher: 97.4% (4,716 of 4,834) of blood tests, and 98.8% (2,792 of 2,825 ) of scans requested were received. Of the 32,587 blood samples received, routine tests comprised 83.3% (27,138 of 32,587), protocol-triggered repeats comprised 14.5% (4,716 of 32,587), and 2.2% (733 of 32,587) were requested by study clinicians (eg, because CA-125 levels had increased by ≥ 50%, despite a normal ROC). A total of 2,233 (6.9%) of 32,587 blood samples were discarded because they arrived more than 56 hours after venipuncture. Of the 12,038 scan results, 75.6% (9,100 of 12,038) were annual, 23.2% (2,792 of 12,038) were triggered early by abnormal ROC results and/or previous abnormal scans, and 1.2% (146 of 12,038) were repeated because of a poor view of the ovaries.

s were discarded because they arrived more than 56 hours after venipuncture. Of the 12,038 scan results, 75.6% (9,100 of 12,038) were annual, 23.2% (2,792 of 12,038) were triggered early by abnormal ROC results and/or previous abnormal scans, and 1.2% (146 of 12,038) were repeated because of a poor view of the ovaries. Overall, 162 (3.7%) of 4,348 women underwent screen-positive trial surgery. Thirteen of these 162 women had screen-detected cancers. The remaining 149 (3.4%) of the 4,348 women underwent false-positive surgery prompted by abnormal results (Table 2). Of these 149 women who underwent false-positive surgery, 46 (30.9%) had an abnormal ROC alone, 62 (41.6%) had an abnormal scan alone, and 41 (27.5%) had abnormal results for both tests. Overall, 95 (63.8%) of the 149 women who underwent false-positive surgery had benign ovarian pathology, two (1.3%) had borderline ovarian tumors, and 52 (35.0%) had other/no pathology identified. An additional 37 (0.9%) of the 4,348 women underwent screen-related trial surgery. Table 2. Diagnoses in Women Who Underwent False-Positive Surgery to Rule Out Ovarian Cancer As a Result of Abnormal Screening Tests (n = 149)

Overall, 162 (3.7%) of 4,348 women underwent screen-positive trial surgery. Thirteen of these 162 women had screen-detected cancers. The remaining 149 (3.4%) of the 4,348 women underwent false-positive surgery prompted by abnormal results (Table 2). Of these 149 women who underwent false-positive surgery, 46 (30.9%) had an abnormal ROC alone, 62 (41.6%) had an abnormal scan alone, and 41 (27.5%) had abnormal results for both tests. Overall, 95 (63.8%) of the 149 women who underwent false-positive surgery had benign ovarian pathology, two (1.3%) had borderline ovarian tumors, and 52 (35.0%) had other/no pathology identified. An additional 37 (0.9%) of the 4,348 women underwent screen-related trial surgery. Table 2. Diagnoses in Women Who Underwent False-Positive Surgery to Rule Out Ovarian Cancer As a Result of Abnormal Screening Tests (n = 149) Invasive OC/FTC/PPC Thirty-seven women were diagnosed with invasive cancer before February 28, 2016 (Table 3); nineteen occurred during 13,728 WSY < 365 days after prior screen and/or withdrawal (annual incidence 0.14%). In addition, 18 women were diagnosed > 365 days after their last UK FOCSS screen (median, 666 days; range, 400 to 2,159 days). The median age at diagnosis in the 37 women diagnosed with OC/FTC/PPC was 50 years (range, 37 to 79 years). All diagnoses occurred in families with hereditary breast-ovarian cancer. Thirty-four (91.2%) of the 37 women were diagnosed with high-grade serous carcinoma. Cancers in 31 (83.8%) of the 37 women occurred in mutation carriers—24 (64.9%) were BRCA1 carriers and seven (18.9%) were BRCA2 carriers. Three (8.1%) of the 37 women were BRCA1/2 negative; one (2.7%) of the 37 women had a BRCA2 variant of unknown significance; two (5.4%) of the 37 women were untested. Of the 37 women diagnosed with OC/FT/PPC, 23 (62.2%) knew they carried pathogenic mutations and 14 (37.8%) had a history of breast cancer. No OC occurred in women with a family history of LS or those who were mutation carriers for the syndrome (n = 192; 558 WSY).

significance; two (5.4%) of the 37 women were untested. Of the 37 women diagnosed with OC/FT/PPC, 23 (62.2%) knew they carried pathogenic mutations and 14 (37.8%) had a history of breast cancer. No OC occurred in women with a family history of LS or those who were mutation carriers for the syndrome (n = 192; 558 WSY). Table 3. Invasive Ovarian, Tubal, and Peritoneal Cancers That Occurred During Screening and Follow-Up The 19 invasive OC/FTCs diagnosed within 365 days of prior screen included one prevalent screen-positive OC (International Federation of Gynecology and Obstetrics stage IIIc) and 18 incident cancers. Twelve of the 18 incident OC/FTCs were screen detected and six were occult cancers identified at RRSO. Of the 12 patients with incident screen-detected cancer, 11 (91.7%) had an abnormal ROC and 5 (41.7%) had a normal TVS at detection (compared with zero of 13 patients who had normal TVS at detection in UK FOCSS Phase I; P = .015). The median CA-125 level at detection was 53.8 iU/ml (range, 11.7 to 219.2 iU/ml) in UK FOCSS Phase II (< 30 iU/ml in four of 12 patients) compared with 73 iU/ml (range, 4 to 3,874 iU/ml) in Phase I, which did not mandate assay type and recommended premenopausal and postmenopausal cutoffs of 35 and 30 iU/ml, respectively, rather than according to the ROCA. Five (38.5%) of the 13 screen-detected OC/FTCs (CI, 13.9% to 68.4%) and 5 (83.3%) of the six occult OC/FTCs (CI, 35.9% to 99.6%) were stage I to II. Overall, 10 (52.6%) of the 19 cancers diagnosed within 365 days of prior screen were stage I to II (CI, 28.9% to 75.6%).

, respectively, rather than according to the ROCA. Five (38.5%) of the 13 screen-detected OC/FTCs (CI, 13.9% to 68.4%) and 5 (83.3%) of the six occult OC/FTCs (CI, 35.9% to 99.6%) were stage I to II. Overall, 10 (52.6%) of the 19 cancers diagnosed within 365 days of prior screen were stage I to II (CI, 28.9% to 75.6%). Eighteen cancers were diagnosed > 365 days after the end of UK FOCSS screening. Two occult cancers were detected at RRSO, three cancers were detected at annual screening performed locally, and 13 were detected when women presented with symptoms. Only one (5.6%) of the 18 cancers was diagnosed at stage I to II (CI, 0.2% to 27.3%).

ncers were diagnosed > 365 days after the end of UK FOCSS screening. Two occult cancers were detected at RRSO, three cancers were detected at annual screening performed locally, and 13 were detected when women presented with symptoms. Only one (5.6%) of the 18 cancers was diagnosed at stage I to II (CI, 0.2% to 27.3%). Women were significantly less likely to be diagnosed with stage IIIb to IV OC during UK FOCSS Phase II screening (seven [36.8%] of 19; CI, 16.3% to 61.6%) compared with those diagnosed subsequently (17 [94.4%] of 18; CI, 72.7% to 99.9%; P < .001). Twelve (92.3%) of 13 women who had screen-detected cancers had zero postsurgical residual disease (CI; 64.0% to 99.8%). Overall, 18 (94.8%) of 19 women diagnosed with OC during UK FOCSS had zero postsurgical residual (CI, 74.0% to 99.9%) compared with 13 (72.2%) of 18 women who were diagnosed subsequently (CI, 46.5% to 90.3%; P = .09). None of the women diagnosed during UK FOCSS required complex surgery, one had interval surgery. Three of the subsequently diagnosed women required complex surgery, seven had interval surgery, and two had no debulking (Table 3). The proportion of women diagnosed with OC during UK FOCSS who had neoadjuvant chemotherapy (1 [5.3%] of 19 women; CI, 0.1% to 26.0%) was significantly lower than in the women diagnosed subsequently (eight [44.4%] of 18 women; CI, 21.5% to 69.2%; P = .008). The mean surgical complexity score28 in women diagnosed during screening or less than 365 days after the final screen was 2.7 compared with 4.3 in those diagnosed subsequently (Mann-Whitney U test, P = .16).

cantly lower than in the women diagnosed subsequently (eight [44.4%] of 18 women; CI, 21.5% to 69.2%; P = .008). The mean surgical complexity score28 in women diagnosed during screening or less than 365 days after the final screen was 2.7 compared with 4.3 in those diagnosed subsequently (Mann-Whitney U test, P = .16). Screening/Surgical Intervals The median delay in incident detection screens in this Phase II study was 6 days (range, 0 to 87 days) compared with 88 days (range, 6 to 737 days) in Phase I (gamma generalized linear model, P = .004). The median interval between detection screen and diagnosis in this Phase II study was 82 days (range, 9 to 209 days) compared with 79 days (range, 15 to 184 days) in Phase I (P = not significant). Reasons for the delay included falsely reassuring scans and reluctance to undergo surgery (Table 3). Screening Performance All 13 cancers (100%) in women who did not undergo RRSO were screen detected (CI, 75.3% to 100%). Hence, for modeled sensitivity, the lower confidence limit of 75.3% was used to conservatively estimate the proportion of occult cancers which would have been screen detected had women not undergone RRSO. Modeled sensitivity, PPV, and negative predictive value (NPV) for the detection of OC/FTC at 1 year for the whole population were 94.7% (CI, 74.0% to 99.9%), 10.8% (CI, 6.5% to 16.5%), and 100% (CI, 100% to 100%), respectively. PPV was significantly better in BRCA1/2 carriers than in women who had an unknown mutation status (Table 4).

ity, PPV, and negative predictive value (NPV) for the detection of OC/FTC at 1 year for the whole population were 94.7% (CI, 74.0% to 99.9%), 10.8% (CI, 6.5% to 16.5%), and 100% (CI, 100% to 100%), respectively. PPV was significantly better in BRCA1/2 carriers than in women who had an unknown mutation status (Table 4). Table 4. Overall Prevalence and Incidence Screening Performance Characteristics According To Population Screened Comparison of Phase I With Phase II Key comparisons of UK FOCSS Phase I and Phase II are listed in Table 5. Rates of clinically presenting interval cancers, zero residual disease after surgery, modeled sensitivity, proportions of women diagnosed with cancer stage less than IIIb, screening delays, and proportions with normal scans at referral were all better in Phase II, but only the comparisons of screening delays and proportions with normal scans at referral were significant. Table 5. Key Comparisons Between UK FOCSS Phase I and Phase II DISCUSSION These UK FOCSS Phase II results suggest that in a high-risk population, ROCA-based multimodal screening every 4 months, alongside reminders of the effectiveness of RRSO, is associated with high sensitivity, significantly lower high-volume disease, and a high zero residual disease rate after surgery compared with women from the same cohort in whom cancer was diagnosed > 1 year after screening ended. Two similar US studies were published while this paper was in press as a combined analysis, with three of six incident cancers found at stage I/II.29

-volume disease, and a high zero residual disease rate after surgery compared with women from the same cohort in whom cancer was diagnosed > 1 year after screening ended. Two similar US studies were published while this paper was in press as a combined analysis, with three of six incident cancers found at stage I/II.29 The strengths of this study are its size, multicenter setting, centralized screening with a validated algorithm, and reliable multiple-source follow-up. A limitation is the nonrandomized design. However, data about OC diagnosed after screening ended allowed comparisons in the absence of a nonscreening arm. Other limitations include the unknown mutation status of many participants and the small number of incident cancers, which limited power. Although screening delays were effectively eliminated (median, 6 days), the median interval between abnormal results and surgery continued to be > 2 months (82 days in this Phase II study v 79 days in Phase I). There were still some long intervals as a result of patient reluctance to undergo surgery and falsely reassuring imaging associated with an abnormal ROC at referral, as seen in UKCTOCS.23

dian interval between abnormal results and surgery continued to be > 2 months (82 days in this Phase II study v 79 days in Phase I). There were still some long intervals as a result of patient reluctance to undergo surgery and falsely reassuring imaging associated with an abnormal ROC at referral, as seen in UKCTOCS.23 During and within 1 year of UK FOCSS screening, no patients with interval cancers presented with symptoms, sensitivity was high, and there was a significant stage shift compared with patients who had cancers diagnosed more than 1 year after screening ended. The significantly higher proportion of cancers diagnosed at stage IIIa or lower (ie, microscopic abdominal disease at worst) during the study was associated with more primary surgery and with higher zero residual disease achieved with less complex surgery. Published complete cytoreduction rates in clinically presenting BRCA1/2 carriers ranged from 28% to 30%.2,30 The overall findings suggest a screening-mediated reduction in disease volume. It is likely this would translate into reduced surgical morbidity and fewer incomplete resections. It remains unknown whether this would improve survival in screened high-risk women. We were unable to analyze survival, because there were only three deaths in the 37 women with invasive OC/FTC/PPCs at censorship on March 14, 2016. Although this is encouraging, medium-term survival of OC in BRCA1/2 carriers is better than that of BRCA1/2-negative patients.7,8

rove survival in screened high-risk women. We were unable to analyze survival, because there were only three deaths in the 37 women with invasive OC/FTC/PPCs at censorship on March 14, 2016. Although this is encouraging, medium-term survival of OC in BRCA1/2 carriers is better than that of BRCA1/2-negative patients.7,8 The performance characteristics of screening every 4 months were encouraging; overall incident sensitivity was 94.7%, with occult cancer detection modeled, and PPV was 10.8% (ie, greater than the suggested 10% level for general-population screening).16 However, PPV is less relevant in high-risk populations for whom RRSO is already recommended as optimal management. As expected, PPV was better in BRCA1/2 carriers (42.9%) than in women who had an unknown mutation status (7.7%) because of the lower cancer incidence in women with an unknown status. The high compliance with blood tests and TVS suggests that the protocol is feasible and acceptable. However, compliance might not be maintained outside a trial. A parallel psychological study found moderate cancer distress at 1 week in women with abnormal ROCA and/or scan results, which led to higher withdrawal from screening.31 However, there was no significant effect on general anxiety and/or depression on return to routine screening or at 9 months.

intained outside a trial. A parallel psychological study found moderate cancer distress at 1 week in women with abnormal ROCA and/or scan results, which led to higher withdrawal from screening.31 However, there was no significant effect on general anxiety and/or depression on return to routine screening or at 9 months. In conclusion, our protocol achieves encouraging performance characteristics, is associated with a low rate of high-volume disease at primary surgery, and had a high zero residual disease rate at low levels of surgical complexity. RRSO remains the treatment of choice for women at high-risk of OC/FTC. In those not ready or willing to undergo surgery, multimodal screening using ROCA every 4 months and TVS (at an interval determined by the ROCA), with regular discussions about the effectiveness of RRSO, appears to be a better option than symptom awareness alone. Such screening should not be viewed as an alternative to surgery, but it does seem to offer a better chance of avoiding a diagnosis of advanced incompletely resectable OC/FTC in the interim. Supported by Cancer Research UK (C1005/A6383, C1005/A7749), The Eve Appeal, and the United Kingdom National Institute for Health Research (NIHR). Researchers at University College London were supported by the NIHR Biomedical Research Centre at University College London Hospitals National Health Service Foundation Trust and University College London.

(C1005/A6383, C1005/A7749), The Eve Appeal, and the United Kingdom National Institute for Health Research (NIHR). Researchers at University College London were supported by the NIHR Biomedical Research Centre at University College London Hospitals National Health Service Foundation Trust and University College London. Previously presented in part at the Annual Meeting of the American Society of Clinical Oncology, Chicago, IL, May 31-June 4, 2013; the British Gynecological Cancer Society Annual Meeting, Belfast, Northern Ireland, June 20-21, 2013; and the 18th Annual Meeting of the European Society of Gynecological Oncology, Liverpool, United Kingdom, October 19-22, 2013. This study reports about off-license use of cancer antigen 125 and sonography for screening for ovarian cancer in a registered clinical trial. Clinical trial information: ISRCTN32794457 See accompanying Editorial on page 1384 ACKNOWLEDGMENT We thank the participants, without whom the study would not have been possible. We thank the lead clinicians at collaborating centers and their research, administrative, and sonography colleagues for their hard work and support over so many years. We thank the independent data monitoring committee (Max Parmar [chair], Shehla Mohammed, Karina Reynolds) and trial steering committee (Julietta Patnick [chair], Diana Eccles, Stephen Duffy, Derek Cruikshank, Louise Bayne). We also thank departmental colleagues Lisa Hinton, Michelle Johnson, Sarah Chamberlain, Jessica Mozersky, Lisa Perreault, Lesley Hague, and Tracy Pearmain and the laboratory team.

mmed, Karina Reynolds) and trial steering committee (Julietta Patnick [chair], Diana Eccles, Stephen Duffy, Derek Cruikshank, Louise Bayne). We also thank departmental colleagues Lisa Hinton, Michelle Johnson, Sarah Chamberlain, Jessica Mozersky, Lisa Perreault, Lesley Hague, and Tracy Pearmain and the laboratory team. AUTHOR CONTRIBUTIONS Conception and design: Adam N. Rosenthal, James Mackay, Steven J. Skates, Usha Menon, Ian J. Jacobs Collection and assembly of data: Adam N. Rosenthal, Lindsay S.M. Fraser, Sue Philpott, Ranjit Manchanda, Philip Badman, Richard Hadwin, Ivana Rizzuto, Elizabeth Benjamin, Naveena Singh, D. Gareth Evans, Diana M. Eccles, Andy Ryan, Robert Liston, Anne Dawnay, Jeremy Ford, Richard Gunu, James Mackay, Usha Menon, Ian J. Jacobs Data analysis and interpretation: Adam N. Rosenthal, Lindsay S.M. Fraser, Sue Philpott, Ranjit Manchanda, Matthew Burnell, Richard Hadwin, Ivana Rizzuto, Elizabeth Benjamin, Naveena Singh, D. Gareth Evans, Diana M. Eccles, James Mackay, Steven J. Skates, Usha Menon, Ian J. Jacobs Manuscript writing: All authors Final approval of manuscript: All authors Accountable for all aspects of the work: All authors

Data analysis and interpretation: Adam N. Rosenthal, Lindsay S.M. Fraser, Sue Philpott, Ranjit Manchanda, Matthew Burnell, Richard Hadwin, Ivana Rizzuto, Elizabeth Benjamin, Naveena Singh, D. Gareth Evans, Diana M. Eccles, James Mackay, Steven J. Skates, Usha Menon, Ian J. Jacobs Manuscript writing: All authors Final approval of manuscript: All authors Accountable for all aspects of the work: All authors AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST Evidence of Stage Shift in Women Diagnosed With Ovarian Cancer During Phase II of the United Kingdom Familial Ovarian Cancer Screening Study The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/jco/site/ifc. Adam N. Rosenthal Honoraria: Roche Consulting or Advisory Role: Myriad Genetics, Abcodia Speakers' Bureau: Roche Travel, Accommodations, Expenses: Roche Lindsay S.M. Fraser No relationship to disclose Sue Philpott No relationship to disclose Ranjit Manchanda Other Relationship: Barts and the London Charity; The Eve Appeal; Cancer Research UK; Israel National Institute for Health Policy Research Matthew Burnell No relationship to disclose Philip Badman No relationship to disclose Richard Hadwin No relationship to disclose Ivana Rizzuto No relationship to disclose Elizabeth Benjamin No relationship to disclose

Ranjit Manchanda Other Relationship: Barts and the London Charity; The Eve Appeal; Cancer Research UK; Israel National Institute for Health Policy Research Matthew Burnell No relationship to disclose Philip Badman No relationship to disclose Richard Hadwin No relationship to disclose Ivana Rizzuto No relationship to disclose Elizabeth Benjamin No relationship to disclose Naveena Singh No relationship to disclose D. Gareth Evans Honoraria: AstraZeneca Diana M. Eccles Consulting or Advisory Role: AstraZeneca Andy Ryan No relationship to disclose Robert Liston No relationship to disclose Anne Dawnay Consulting or Advisory Role: Abcodia Travel, Accommodations, Expenses: Roche Jeremy Ford No relationship to disclose Richard Gunu No relationship to disclose James Mackay Employment: Myriad Genetics, CIS Oncology Honoraria: Myriad Genetics International Consulting or Advisory Role: Company: Myriad Genetics, Abcodia, CIS Oncology Steven J. Skates Stock or Other Ownership: SISCAPA Assay Technologies Consulting or Advisory Role: Abcodia Patents, Royalties, Other Intellectual Property: Abcodia Usha Menon Stock or Other Ownership: Abcodia Research Funding: Abcodia (Inst) Patents, Royalties, Other Intellectual Property: European patent No. EP10178345.4: Breast Cancer Diagnostics, which describes the implications of receptor activator of nuclear factor κ-B ligand inhibition in the prevention of hormone-driven breast cancer. Ian J. Jacobs Employment: Women's Health Specialists Leadership: Abcodia Stock or Other Ownership: Abcodia Consulting or Advisory Role: Abcodia Patents, Royalties, Other Intellectual Property: ROC Algorithm

INTRODUCTION The number of breast cancer (BC) survivors is increasing as a result of rising incidence, earlier diagnosis, and better treatment results.1,2 Although adjuvant radiotherapy (RT) after surgery for BC improves locoregional control and overall survival, incidental exposure of the heart to radiation increases the risk of RT-induced cardiac toxicity.3-5 Consequently, the prevalence of BC survivors at risk for long-term RT-induced cardiac toxicity is increasing and may have a significant impact on health-related quality of life. Darby et al6 demonstrated a dose-effect relationship based on the mean heart dose (MHD) to the whole heart. They found a relative increase of 7.4% per Gy of MHD in the rate of major acute coronary events (ACEs) for the entire follow-up period. Confining the analysis to the first 9 years after radiation exposure, a relative increase of approximately 16% per Gy was found. However, the study had some limitations: its design (case-control study), use of outdated RT technologies, and use of reconstructed MHDs derived from two-dimensional data. Therefore, the first aim of our study was to validate the findings of Darby et al6 with an independent cohort of consecutive patients with BC based on individual three-dimensional (3D) dose distributions derived from computed tomography (CT) planning scans. The second aim of this cohort study was to investigate whether other dose-distribution parameters could better predict the excess risk of ACEs after RT in individual patients with BC compared with MHD.

BC based on individual three-dimensional (3D) dose distributions derived from computed tomography (CT) planning scans. The second aim of this cohort study was to investigate whether other dose-distribution parameters could better predict the excess risk of ACEs after RT in individual patients with BC compared with MHD. PATIENTS AND METHODS Study Population This study population was composed of a consecutive series of female patients with BC treated with RT after breast-conserving surgery for stage I to III invasive adenocarcinoma or carcinoma in situ from January 2005 to December 2008 in our hospital (Appendix Fig A1, online only). Patients with BC were eligible for inclusion only if CT-based RT planning data were available. Patients were excluded if they had a history of other malignancies or had received prior RT or treatment with neoadjuvant chemotherapy. The primary end point was an ACE, defined as a diagnosis of myocardial infarction (International Classification of Diseases, 10th Revision, codes 121 to 124), coronary revascularization, or death resulting from ischemic heart disease (codes 120 to 125) after completion of treatment. Pretreatment risk factors for ACEs that were taken into account included history of ischemic heart disease, any other cardiac disease, hypertension, chronic obstructive pulmonary disease, pulmonary embolism, diabetes, current smoker status, and body mass index ≥ 30 kg/m2. Both the end point and pretreatment risk factors were similar to those defined by Darby et al.6

en into account included history of ischemic heart disease, any other cardiac disease, hypertension, chronic obstructive pulmonary disease, pulmonary embolism, diabetes, current smoker status, and body mass index ≥ 30 kg/m2. Both the end point and pretreatment risk factors were similar to those defined by Darby et al.6 Data Collection Patient characteristics, treatment plans, follow-up data, and information on cardiac risk factors and cardiac end points were retrospectively extracted from patient records of the Department of Radiation Oncology (University Medical Center Groningen, University of Groningen, Groningen, the Netherlands). Incomplete patient data were supplemented with information derived from general practitioners’ (GPs’) records. To this end, surviving patients were informed about the study by letter and asked for their written informed consent. GPs of deceased patients were allowed to provide relevant information directly, because GPs have legal governance over deceased patients’ records in the Netherlands. The aforementioned procedure was approved by the medical ethical committee of the University Medical Center Groningen.

for their written informed consent. GPs of deceased patients were allowed to provide relevant information directly, because GPs have legal governance over deceased patients’ records in the Netherlands. The aforementioned procedure was approved by the medical ethical committee of the University Medical Center Groningen. Data Definitions The baseline date was defined as the first day of breast irradiation. Patient event times were censored in cases where a new radiation treatment was delivered in the follow-up period, in cases of death, or at the end of follow-up time. The follow-up interval was defined as the time between baseline and censoring date or date of event. Patient information was collected until the last known date of medical follow-up or last known information obtained from the GP.

was delivered in the follow-up period, in cases of death, or at the end of follow-up time. The follow-up interval was defined as the time between baseline and censoring date or date of event. Patient information was collected until the last known date of medical follow-up or last known information obtained from the GP. Radiation Dosimetry Irradiation of the breast for all patients was performed with 3D conformal RT using CT-based planning, as described previously.7 All treatment plans were calculated using heterogeneity corrections. Beam configuration comprised tangential fields and additional beams for optimization of planning target volume coverage, as well as for minimization of the dose to the heart, lungs, and contralateral breast. A dose of 50.4 Gy was prescribed for the whole breast in 28 fractions, with a simultaneous integrated boost dose of 14 or 16.8 Gy in the same 28 fractions, depending on pathologic risk factors. The heart and its substructures, including the left ventricle (LV), left atrium, right ventricle, and right atrium, were recontoured with a multiatlas automatic segmentation tool of the heart developed in house based on the atlas by Feng et al8 (Mirada RTx [version 1.6]; Mirada Medical, Oxford, United Kingdom).9 Automatic segmentation reduces interobserver variability in contouring organs at risk and therefore generates more consistent data to create normal tissue complication probability (NTCP) models.10,11 With the delineated volumes, it was possible to calculate the exact planned radiation dose to the different volumes. This so-called dose-volume histogram showed the relationship between the dose in Gy to the volume percentage of the structure of interest.12,13 With the dose of the individual patients, the dose-effect relationship could be calculated independently of RT technique or treatment volume. Finally, the planned dose-distribution parameters for the whole heart and its substructures were extracted from our treatment planning system (Pinnacle [version 9.1]; Philips Radiation Oncology, Fitchburg, WI).

patients, the dose-effect relationship could be calculated independently of RT technique or treatment volume. Finally, the planned dose-distribution parameters for the whole heart and its substructures were extracted from our treatment planning system (Pinnacle [version 9.1]; Philips Radiation Oncology, Fitchburg, WI). Statistical Analysis The cumulative incidence of ACEs was analyzed using the Kaplan-Meier method. To validate the model of Darby et al,6 a multivariable Cox regression analysis was used, including the same risk factors and end point. Model performance was tested for calibration using the Hosmer-Lemeshow (HL) test, and discrimination was tested using the c-statistic.

ACEs was analyzed using the Kaplan-Meier method. To validate the model of Darby et al,6 a multivariable Cox regression analysis was used, including the same risk factors and end point. Model performance was tested for calibration using the Hosmer-Lemeshow (HL) test, and discrimination was tested using the c-statistic. The most relevant dose-distribution parameters for the different cardiac substructures were identified by comparing the mean dose-distribution parameters of patient cases (patients who experienced an ACE) with noncases (patients who did not). To this end, we calculated the mean V(x) in bins of 5 Gy for both patient cases and noncases, where V(x) refers to the relative volume (in percentage) of the heart or cardiac substructure that received x Gy. Differences between the two groups regarding all mean dose-distribution parameters were tested with a t test or Wilcoxon rank sum test whenever appropriate. The dichotomous variable (no risk factor v one or more risk factors) was replaced by a weighted ACE risk score per patient. To this end, we first investigated which risk factors were significantly associated with the incidence of ACEs by using univariable Cox regression analysis and then performed a multivariable analysis taking into account only the significant cardiac risk factors. For the risk factors that were significantly associated with ACEs in the multivariable analysis, the regression coefficients were calculated and used for the weighted sum of the risk factor(s) per patient. In correspondence with Darby et al,6 age was entered into the model as well. Because the number of events was limited, we decided not to add more than these three factors to the model to prevent overfitting.14,15 For internal validation and adjustment for possible internal optimism for both the c-statistics and some estimators, bootstrapping was performed by using 1,000 random subsets. Model performance was tested for calibration using the HL test. Finally, the excess risk of an ACE resulting from RT was calculated via the individual patient risk based on the model minus the individual patient risk assuming the LV receiving 5 Gy (LV-V5) received 0%. Calculations were performed SPSS software (version 22; SPSS, Chicago, IL).

ested for calibration using the HL test. Finally, the excess risk of an ACE resulting from RT was calculated via the individual patient risk based on the model minus the individual patient risk assuming the LV receiving 5 Gy (LV-V5) received 0%. Calculations were performed SPSS software (version 22; SPSS, Chicago, IL). RESULTS Patient Characteristics A total of 910 patients were included in this study. The characteristics of these patients are summarized in Table 1. The median age of all patients was 59 years (range, 26 to 84 years). At baseline, more than half of the patients had one or more risk factors for ACEs. The median follow-up time was 7.6 years (range, 0.1 to 10.1 years). Table 1. Patient Clinical Characteristics at Baseline (N = 910) More detailed information about the distribution of MHD and the univariable analysis between MHD and the end point ACE is provided in Appendix Table A1 (online only), Appendix Figures A2 to A4 (online only), and Appendix Figure A5 (online only), and information about patients experiencing an event is listed in Appendix Table A2 (online only). In total, 30 patients (3.3%) developed an ACE during follow-up, 10 of whom died as a result of ischemic heart disease. In the first 5 years, 17 patients were diagnosed with ACEs. The 5- and 9-year cumulative incidences of ACEs were 1.9% (95% CI, 0.9% to 2.9%) and 3.9% (95% CI, 2.3% to 5.5%), respectively (Appendix Fig A6, online only).

ents (3.3%) developed an ACE during follow-up, 10 of whom died as a result of ischemic heart disease. In the first 5 years, 17 patients were diagnosed with ACEs. The 5- and 9-year cumulative incidences of ACEs were 1.9% (95% CI, 0.9% to 2.9%) and 3.9% (95% CI, 2.3% to 5.5%), respectively (Appendix Fig A6, online only). Validation To validate the model of Darby et al,6 a multivariable Cox regression model was created using the same prognostic factors (ie, age, MHD, and presence of pretreatment risk factors for ACEs, categorized as either none or one or more at baseline). The cumulative incidence of ACEs increased by 16.5% per Gy (P = .042) within 9 years of RT (Table 2). Table 2. Multivariable NTCP Model for Cumulative Incidence of ACEs On the basis of this model, the 9-year excess cumulative risk (CER9y) can be calculated using the following equations: The linear predictor LPMHD-model = (0.153 × MHD) + (0.087 × AGE) + (1.821 × RISK), in which MHD = mean heart dose in Gy, AGE = age in years, and RISK = 0 when no risk factors for ACEs are present at baseline and RISK = 1 if one or more risk factors at baseline are present. The cumulative incidence for each individual patient at 9 years (CI9y) can then be calculated using the following equation: CI9y = 1 – [EXP(−0.000025 × EXP(LPMHD-model))]. The 9-year excess cumulative risk (CER9y) can then be calculated by using Equation 2 minus the CI9y assuming an MHD of 0 Gy (CI9y-0Gy): CER9y = CI9y – CI9y-0Gy.

The cumulative incidence for each individual patient at 9 years (CI9y) can then be calculated using the following equation: CI9y = 1 – [EXP(−0.000025 × EXP(LPMHD-model))]. The 9-year excess cumulative risk (CER9y) can then be calculated by using Equation 2 minus the CI9y assuming an MHD of 0 Gy (CI9y-0Gy): CER9y = CI9y – CI9y-0Gy. The HL test showed no significant difference between expected and observed rates of ACEs (P = .406), indicating good calibration. Model discrimination was good, with a c-statistic of 0.79 (95% CI, 0.71 to 0.87). The mean predicted CI9y for the entire population was 4.0%, which was in agreement with the CI9y actually observed: 3.9%. To get an impression of the early risk of ACEs, a model for the first 5 years after RT (Table 2) was tested separately. Using the same risk factors and end point as those of Darby et al,6 an increase of 24.6% in the rate of ACEs per Gy of MHD was found for the complete follow-up period of 5 years.

The HL test showed no significant difference between expected and observed rates of ACEs (P = .406), indicating good calibration. Model discrimination was good, with a c-statistic of 0.79 (95% CI, 0.71 to 0.87). The mean predicted CI9y for the entire population was 4.0%, which was in agreement with the CI9y actually observed: 3.9%. To get an impression of the early risk of ACEs, a model for the first 5 years after RT (Table 2) was tested separately. Using the same risk factors and end point as those of Darby et al,6 an increase of 24.6% in the rate of ACEs per Gy of MHD was found for the complete follow-up period of 5 years. Model Optimization To identify the most relevant dose-distribution parameters, we compared the mean dose parameters of the patient cases (patients who experienced an ACE) with noncases (patients who did not). Figure 1 shows the differences between the mean dose-distribution parameters per cardiac substructure that were tested for significance. The largest difference was found for LV-V5. In the univariable Cox regression analysis, summarized in Table 3, LV-V5 was significantly associated with the cumulative incidence of ACEs, with a hazard ratio of 1.016 (95% CI, 1.002 to 1.030; P = .016). Because of this strong association, we chose to include LV-V5 in the model. Replacement of MHD with LV-V5 resulted in an improvement of the c-statistic of the NTCP model to 0.80 (95% CI, 0.72 to 0.88). We also tested the relationship of the maximum dose to the heart with the cumulative incidence of ACEs using a univariable Cox regression analysis and found it was not significantly associated with ACEs (data not shown).

-V5 resulted in an improvement of the c-statistic of the NTCP model to 0.80 (95% CI, 0.72 to 0.88). We also tested the relationship of the maximum dose to the heart with the cumulative incidence of ACEs using a univariable Cox regression analysis and found it was not significantly associated with ACEs (data not shown). Fig 1. Comparison of the mean dose distribution parameters of patient cases (patients who experienced an acute coronary event [ACE]) and noncases (those who did not) and calculation of the differences. NOTE. All data are given as the relative volumes (%) of the cardiac substructures that received (x) Gy or more in bins of 5 Gy. LA, left atrium; LV, left ventricle; RA, right atrium; RV, right ventricle. Table 3. Univariable and Multivariable NTCP Models for Cumulative Incidence of ACEs Within First 9 Years After RT After Correction for Overfitting To further optimize the NTCP model based on LV-V5, the dichotomous variable (no risk factor v one or more risk factors) was replaced with a weighted ACE risk score per patient. Because there were only 30 events, LV-V5, age, and weighted ACE risk score per patient based on the regression coefficient of the significant risk factors for ACEs (0.8 for diabetes, 1.4 for hypertension, and 1.8 for history of ischemic cardiac events) were entered into the multivariable model. The final multivariable NTCP model summarized in Table 3 is corrected for optimism. On the basis of this model, the 9-year excess cumulative risk (CER9y) can be calculated using the following equations:

To further optimize the NTCP model based on LV-V5, the dichotomous variable (no risk factor v one or more risk factors) was replaced with a weighted ACE risk score per patient. Because there were only 30 events, LV-V5, age, and weighted ACE risk score per patient based on the regression coefficient of the significant risk factors for ACEs (0.8 for diabetes, 1.4 for hypertension, and 1.8 for history of ischemic cardiac events) were entered into the multivariable model. The final multivariable NTCP model summarized in Table 3 is corrected for optimism. On the basis of this model, the 9-year excess cumulative risk (CER9y) can be calculated using the following equations: The linear predictor LPLV-V5-model = (0.017 × LV-V5) + (0.063 × AGE) + (0.711 × RISKSCORE), in which LV-V5 = LV-V5 in %, AGE = age in years, and RISKSCORE = weighted ACE risk score (0 for no risk factors; add 0.8 in case of diabetes, add 1.4 in case of hypertension, and add 1.8 in case of ischemic cardiac events before RT). The cumulative incidence for each individual patient at 9 years (CI9y) can then be calculated using the following equation: CI9y = 1 – [EXP(−0.000223 × EXP(LPLV-V5-model))]. The 9-year excess cumulative risk (CER9y) can then be calculated by using Equation 2 minus the CI9y assuming an LV-V5 of 0% (CI9y-0%): CER9y = CI9y – CI9y-0Gy.

The cumulative incidence for each individual patient at 9 years (CI9y) can then be calculated using the following equation: CI9y = 1 – [EXP(−0.000223 × EXP(LPLV-V5-model))]. The 9-year excess cumulative risk (CER9y) can then be calculated by using Equation 2 minus the CI9y assuming an LV-V5 of 0% (CI9y-0%): CER9y = CI9y – CI9y-0Gy. The mean predicted CI9y for the entire population was 3.5%, which was in agreement with the CI9y actually observed: 3.9%. This modified model showed good agreement between expected and observed rates of ACEs (HL test P = .380). Discrimination of the final model in terms of the c-statistic showed good results at 0.83 (95% CI, 0.75 to 0.91), which was significantly better than that in the MHD model (P = .042). The excess cumulative risk related to RT was 1.13% within 9 years of follow-up, indicating that approximately 10 patients in this BC cohort experienced an ACE that could be attributed to RT. The excess risk for the occurrence of an ACE, depending on the mean dose, is shown in Figure 2 and based on the LV-V5 in Figure 3. Fig 2. Excess risk of an acute coronary event (ACE) depending on the mean heart dose (MHD) in volume percentage calculated per age category and (A) absence or (B) presence of cardiac risk factors.

The excess cumulative risk related to RT was 1.13% within 9 years of follow-up, indicating that approximately 10 patients in this BC cohort experienced an ACE that could be attributed to RT. The excess risk for the occurrence of an ACE, depending on the mean dose, is shown in Figure 2 and based on the LV-V5 in Figure 3. Fig 2. Excess risk of an acute coronary event (ACE) depending on the mean heart dose (MHD) in volume percentage calculated per age category and (A) absence or (B) presence of cardiac risk factors. Fig 3. Excess risk of an acute coronary event (ACE) depending on the mean V5 of the left ventricle (LV-V5) in volume percentage calculated per age category and risk factor: (A) no cardiac risk factors, (B) diabetes, (C) hypertension, and (D) ischemic cardiac event. For example, a patient age 70 years with an LV-V5 of 50% and no cardiac risk factors has an excess risk of 2.52% of developing an ACE within 9 years after radiotherapy. If the same patient had a history of ischemic heart disease, with a similar value for LV-V5, the excess risk would increase to 8.42%.

ic cardiac event. For example, a patient age 70 years with an LV-V5 of 50% and no cardiac risk factors has an excess risk of 2.52% of developing an ACE within 9 years after radiotherapy. If the same patient had a history of ischemic heart disease, with a similar value for LV-V5, the excess risk would increase to 8.42%. DISCUSSION To our knowledge, this is the first study to validate the model published by Darby et al6 in an independent cohort using individual 3D CT planning data. Using exactly the same risk factors and end point as Darby et al, we found an increase of 16.5% (95% CI, 0.6 to 35.0) in the cumulative incidence of ACEs per Gy of radiation to the whole heart in the first 9 years after treatment. These results are consistent with the hazard ratios of 16.3% increase per Gy, as observed by Darby et al in the first 4 years of follow-up, and 15.5% increase in the next 5 to 9 years after RT. Furthermore, our study suggests that the NTCP model for ACEs could be improved by using LV-V5 instead of MHD. Model performance showed good results in terms of calibration and discrimination.

ease per Gy, as observed by Darby et al in the first 4 years of follow-up, and 15.5% increase in the next 5 to 9 years after RT. Furthermore, our study suggests that the NTCP model for ACEs could be improved by using LV-V5 instead of MHD. Model performance showed good results in terms of calibration and discrimination. An NTCP model is a term generally used in radiation oncology, which refers to any prediction model describing the relationship between 3D dose-distribution parameters of normal tissues and a complication end point. In radiotherapy, NTCP models are generally used to estimate the risks of adverse effects, as well as to optimize dose distributions for individual patients by minimizing the most relevant dose metrics derived from NTCP models.16 To enhance the clinical utility of prediction models, it is highly recommended that the performance of the model be evaluated in an independent data set.17 Despite differences with regard to study design (case-control v cohort study), irradiation technique, estimated dose distributions (reconstructed MHD v 3D planning CT based), timeframe, and nationality, the results found in our study are in line with those reported by Darby et al.6 Therefore, the model summarized in Table 2 can be considered as a TRIPOD (Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis) type IV prediction model, the performance of which has been evaluated in an independent data set.17 The results of case-control studies, as reported by Darby et al, provide only relative risk against baseline risk, which requires other prediction models to assess these baseline risks. Because our multivariable externally validated model (Table 2) was based on a cohort study, it allows for a direct risk estimation of ACEs for individual patients with BC. However, because we were not able to externally validate the LV-V5 model, this model should be regarded as TRIPOD type Ib, which requires external validation first before it can be used in routine clinical practice.

based on a cohort study, it allows for a direct risk estimation of ACEs for individual patients with BC. However, because we were not able to externally validate the LV-V5 model, this model should be regarded as TRIPOD type Ib, which requires external validation first before it can be used in routine clinical practice. Our dose-distribution analysis (Fig 1) showed that the LV received the highest dose of all cardiac structures, which is mainly because of the anatomic location of the LV in relation to the breasts and treatment technique, which may increase statistical power. The analysis comparing the dose-distribution parameters between patient cases and noncases also revealed large differences, even for lower dose levels (eg, LV-V2 to -V4; data not shown). LV-V5 was eventually chosen because this dose-distribution parameter has been widely used in many other recent reports.18-22 As shown in a recent study, heart doses from RT for BC vary widely, even among seemingly similar regimens.23 Therefore, we chose to use an automatic delineation tool to exclude interobserver variability.8,24 Furthermore, we used individual dose-volume data, which account for differences in anatomy and treatment volume.

As shown in a recent study, heart doses from RT for BC vary widely, even among seemingly similar regimens.23 Therefore, we chose to use an automatic delineation tool to exclude interobserver variability.8,24 Furthermore, we used individual dose-volume data, which account for differences in anatomy and treatment volume. It has long been assumed that the clinical events of incidental cardiac irradiation occur after more than 10 years.25-29 One of the biologic mechanisms leading to radiation-induced ACEs is accelerated atherosclerosis.30-32 However, in our analysis, a dose-effect relationship was found for events occurring within the first 5 years after radiation exposure. This early risk is consistent with that reported by Darby et al6 and that seen in other research in patients with Hodgkin lymphoma.33 However, other studies found only a small effect in 6 to 10 years after treatment, when the internal mammary nodes were not treated.34,35 When these nodes were treated, the occurrence of cardiac damage was found within 5 years.36 Given these results, and setting aside the relatively slowly progressing phenomenon of atherosclerosis, other biologic mechanisms are most likely responsible for the relatively early cardiac events occurring after RT (eg, microvascular damage, impairment in myocardial perfusion and/or fatty acid metabolism, and many more).37-41 Studies investigating these underlying mechanisms for early RT-induced cardiac damage using modern imaging techniques are currently under way.42

ble for the relatively early cardiac events occurring after RT (eg, microvascular damage, impairment in myocardial perfusion and/or fatty acid metabolism, and many more).37-41 Studies investigating these underlying mechanisms for early RT-induced cardiac damage using modern imaging techniques are currently under way.42 A limitation of our study is the relatively small number of ACEs. Because 3D conformal RT at our hospital was clinically introduced in 2005, the follow-up time was relatively short. To prevent overfitting by using too many candidate variables in relation to the number of events, we included only two other prognostic factors, besides the dose-distribution parameter: clinical risk factors for ACEs and age, based on the fact that these are considered the most important predictors for ACEs.43 Consequently, the effects of other potential confounders could not be taken into account, such as the addition of systemic agents that could also cause cardiac toxicity.44,45 In conclusion, the MHD-based NTCP model for ACEs has been independently validated using 3D dose-distribution data among patients with BC treated with postoperative RT. Radiation dose to the heart is an important risk factor for ACEs in BC survivors. Model performance was significantly improved by replacing MHD with LV-V5 and using the weighted ACE risk score, but this optimized model requires further external validation in an independent data set. Presented orally at the Annual Meeting of the American Society for Radiation Oncology, Boston, MA, September 25-28, 2016. Clinical trial information: NCT02471079.

In conclusion, the MHD-based NTCP model for ACEs has been independently validated using 3D dose-distribution data among patients with BC treated with postoperative RT. Radiation dose to the heart is an important risk factor for ACEs in BC survivors. Model performance was significantly improved by replacing MHD with LV-V5 and using the weighted ACE risk score, but this optimized model requires further external validation in an independent data set. Presented orally at the Annual Meeting of the American Society for Radiation Oncology, Boston, MA, September 25-28, 2016. Clinical trial information: NCT02471079. See accompanying Editorial on page 1146 AUTHOR CONTRIBUTIONS Conception and design: Veerle A.B. van den Bogaard, Bastiaan D.P. Ta, Arjen van der Schaaf, Angelique B. Bouma, Astrid M.H. Middag, Enja J. Bantema-Joppe, Femke B.J. van Dijk-Peters, Johannes A. Langendijk, John H. Maduro, Anne P.G. Crijns Collection and assembly of data: Veerle A.B. van den Bogaard, Bastiaan D.P. Ta, Angelique B. Bouma, Astrid M.H. Middag, Enja J. Bantema-Joppe, Lisanne V. van Dijk, Laurens A.W. Marteijn Data analysis and interpretation: Veerle A.B. van den Bogaard, Arjen van der Schaaf, Astrid M.H. Middag, Gertruida H. de Bock, Johannes G.M. Burgerhof, Jourik A. Gietema, Johannes A. Langendijk, John H. Maduro, Anne P.G. Crijns Manuscript writing: All authors Final approval of manuscript: All authors Accountable for all aspects of the work: All authors

Data analysis and interpretation: Veerle A.B. van den Bogaard, Arjen van der Schaaf, Astrid M.H. Middag, Gertruida H. de Bock, Johannes G.M. Burgerhof, Jourik A. Gietema, Johannes A. Langendijk, John H. Maduro, Anne P.G. Crijns Manuscript writing: All authors Final approval of manuscript: All authors Accountable for all aspects of the work: All authors AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST Validation and Modification of a Prediction Model for Acute Cardiac Events in Patients With Breast Cancer Treated With Radiotherapy Based on Three-Dimensional Dose Distributions to Cardiac Substructures The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO’s conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/jco/site/ifc. Veerle A.B. van den Bogaard No relationship to disclose Bastiaan D.P. Ta No relationship to disclose Arjen van der Schaaf No relationship to disclose Angelique B. Bouma No relationship to disclose Astrid M.H. Middag No relationship to disclose Enja J. Bantema-Joppe No relationship to disclose Lisanne V. van Dijk No relationship to disclose Femke B.J. van Dijk-Peters No relationship to disclose Laurens A.W. Marteijn No relationship to disclose Gertruida H. de Bock No relationship to disclose Johannes G.M. Burgerhof No relationship to disclose

Astrid M.H. Middag No relationship to disclose Enja J. Bantema-Joppe No relationship to disclose Lisanne V. van Dijk No relationship to disclose Femke B.J. van Dijk-Peters No relationship to disclose Laurens A.W. Marteijn No relationship to disclose Gertruida H. de Bock No relationship to disclose Johannes G.M. Burgerhof No relationship to disclose Jourik A. Gietema Research Funding: Roche (Inst), AbbVie (Inst), Siemens (Inst) Johannes A. Langendijk Honoraria: IBA Consulting or Advisory Role: IBA Speakers’ Bureau: IBA Research Funding: RaySearch Laboratories (Inst), IBA (Inst), Philips Healthcare (Inst), Mirada (Inst) John H. Maduro No relationship to disclose Anne P.G. Crijns No relationship to disclose Appendix Table A1. MHD in Relation to Patient Clinical Characteristics at Baseline Table A2. Detailed Information on All Patients With BC With ACE at Baseline Fig A1. Study population flowchart. ACE, acute coronary event; BC, breast cancer; CT, computed tomography; RT, radiotherapy. Fig A2. Distribution of mean heart dose (MHD) for the entire population (N = 910; median MHD, 2.37; standard deviation, 2.26; range, 0.51 to 15.25). Fig A3. Distribution of mean heart dose (MHD) for left-sided breast cancer (n = 451; median MHD, 4.44; standard deviation, 2.12; range, 0.99 to 15.25). Fig A4. Distribution of mean heart dose (MHD) for right-sided breast cancer (n = 459; median MHD, 1.31; standard deviation, 0.72; range, 0.51 to 6.87).

Fig A2. Distribution of mean heart dose (MHD) for the entire population (N = 910; median MHD, 2.37; standard deviation, 2.26; range, 0.51 to 15.25). Fig A3. Distribution of mean heart dose (MHD) for left-sided breast cancer (n = 451; median MHD, 4.44; standard deviation, 2.12; range, 0.99 to 15.25). Fig A4. Distribution of mean heart dose (MHD) for right-sided breast cancer (n = 459; median MHD, 1.31; standard deviation, 0.72; range, 0.51 to 6.87). Fig A5. Relationship between mean heart dose (MHD) and percentage of acute coronary events (ACEs) based on univariable analysis (ie, not corrected for age or presence of cardiovascular risk factors). The linear trend line crosses the y-axis, indicating ACEs not related to radiotherapy. Vertical bars indicate 95% CIs. Fig A6. Cumulative incidence of acute coronary events (ACEs) in the entire population; vertical bars indicate 95% CIs. RT, radiotherapy.

INTRODUCTION Systemic Therapy for Advanced or Metastatic Prostate Cancer: Evaluation of Drug Efficacy (STAMPEDE) is a multiarm, multistage (MAMS), platform, randomized controlled trial. Its novel design1,2 allowed simultaneous assessment of adding various therapies to standard of care (SOC; androgen deprivation). The trial recruited patients commencing long-term hormone therapy (HT) for high-risk, locally advanced, or metastatic prostate cancer (CaP), either newly diagnosed or after failure of previous local therapy. Results from STAMPEDE’s docetaxel comparison showed major improvements in survival.3 A companion meta-analysis4 combining data from other major international trials5,6 confirmed the usefulness of that combination, changing world-wide practice.7,8 Here, we report outcomes after SOC plus either celecoxib (Cel) or Cel and zoledronic acid (ZA; Data Supplement). Cox-2 inhibition is associated with inhibition of carcinogenesis,9-12 and case-control studies have shown a reduced risk of CaP.13-15 ZA has known anti-CaP effects, demonstrated both clinically in later-stage disease16 and in vitro.17 The first-generation bisphosphonate clodronate improved survival when used concurrently with long-term HT for metastatic CaP.18 The anticipated mechanisms of action of Cox-2 inhibitors such as Cel and bisphosphonates such as ZA were considered complementary, allowing targeting of both bone progression and the underlying molecular changes that lead to progression.

e improved survival when used concurrently with long-term HT for metastatic CaP.18 The anticipated mechanisms of action of Cox-2 inhibitors such as Cel and bisphosphonates such as ZA were considered complementary, allowing targeting of both bone progression and the underlying molecular changes that lead to progression. The MAMS design uses increasingly stringent hurdles at interim analyses to determine whether recruitment to a comparison should continue through to fully powered survival analysis. Interim analysis was performed on failure-free survival (FFS), primarily driven by rising prostate-specific antigen (PSA). In 2011, at the second, preplanned activity analysis, the Independent Data Monitoring Committee reviewed data, including those on toxicity and FFS. The observed safety of Cox-2 inhibition ± ZA was not questioned; closure to recruitment to both Cel-containing arms was recommended because of insufficient activity on FFS, guided by a protocol-defined activity target of a hazard ratio (HR) of 0.92. The Trial Steering Committee agreed that Cel should be stopped in both arms. The ZA was continued because the ZA comparison was continuing. On the committee’s recommendation, comparative FFS data for the Cel-only arm were published; follow-up continued as planned.19

fined activity target of a hazard ratio (HR) of 0.92. The Trial Steering Committee agreed that Cel should be stopped in both arms. The ZA was continued because the ZA comparison was continuing. On the committee’s recommendation, comparative FFS data for the Cel-only arm were published; follow-up continued as planned.19 Release of survival data was intended to follow publication of survival data for the docetaxel, ZA, and docetaxel + ZA comparisons, which started recruitment simultaneously and passed through all intermediate analyses.3 For the purposes of understanding, we also include some information on contemporaneously randomly assigned patients allocated to SOC + ZA; this is updated information on a subset of patients reported previously.3 We also contextualize our findings on Cel + ZA with summary information from the docetaxel arms in STAMPEDE, because that treatment is now an increasingly used SOC. PATIENTS AND METHODS The trial, detailed previously,2,3,19-20 was run according to Good Clinical Practice guidelines and the Declaration of Helsinki, with relevant regulatory and ethics approvals.

Release of survival data was intended to follow publication of survival data for the docetaxel, ZA, and docetaxel + ZA comparisons, which started recruitment simultaneously and passed through all intermediate analyses.3 For the purposes of understanding, we also include some information on contemporaneously randomly assigned patients allocated to SOC + ZA; this is updated information on a subset of patients reported previously.3 We also contextualize our findings on Cel + ZA with summary information from the docetaxel arms in STAMPEDE, because that treatment is now an increasingly used SOC. PATIENTS AND METHODS The trial, detailed previously,2,3,19-20 was run according to Good Clinical Practice guidelines and the Declaration of Helsinki, with relevant regulatory and ethics approvals. Study Design and Participants A MAMS platform approach was used.1,21 Eligible patients had CaP that was newly diagnosed and either metastatic, node positive, or high-risk locally advanced (with ≥ 2 of T3/4, Gleason 8 to 10, and PSA ≥ 40 ng/mL). Eligibility criteria also included that the patients had been treated previously with radical surgery or radiotherapy (RT) now relapsing with high-risk features. All were initiating long-term HT within 12 weeks before random assignment. Patients were required to be fit for chemotherapy with no history of severe cardiovascular disease. All gave written informed consent.

s had been treated previously with radical surgery or radiotherapy (RT) now relapsing with high-risk features. All were initiating long-term HT within 12 weeks before random assignment. Patients were required to be fit for chemotherapy with no history of severe cardiovascular disease. All gave written informed consent. Random Assignment and Masking A computerized algorithm implemented minimization-based random assignment (random element, 20%), stratifying for hospital, age, presence of metastases, planned RT use, nodal involvement, WHO performance status, planned HT type, and regular aspirin and/or nonsteroidal anti-inflammatory drug (NSAID) use. Allocation was 2:1:1 for open-label SOC-only, SOC + Cel, and SOC + ZA + Cel. Treatment HT was lifelong for patients with metastatic disease and ≥ 2years for patients with nonmetastatic disease. HT was with gonadotropin-releasing hormone agonists or antagonists or orchidectomy; patients with nonmetastatic disease could receive oral anti-androgens alone; patients with metastatic disease could undergo orchidectomy. RT was encouraged for patients with nonmetastatic disease 6 to 9 months after random assignment.22

HT was with gonadotropin-releasing hormone agonists or antagonists or orchidectomy; patients with nonmetastatic disease could receive oral anti-androgens alone; patients with metastatic disease could undergo orchidectomy. RT was encouraged for patients with nonmetastatic disease 6 to 9 months after random assignment.22 Cel was given orally 400 mg twice a day for 1 year after regulatory authority advice after withdrawal of another Cox-2 inhibitor, rofecoxib. ZA was planned for ≤ 2 years, given as outpatient infusions at 4 mg/15 min on approximately 66 occasions, starting once every 3 weeks for six cycles then once every 4 weeks. The protocol described modifications for adverse events (AEs). Either treatment was to stop for intolerable AEs or an FFS event. Follow-Up Follow-up, including PSA tests, was every 6 weeks for 6 months, then every 12 weeks to 2 years, every 6 months to 5 years, then annually. Additional tests were at the investigators’ discretion. Common Toxicity Criteria version 3.01 was used to grade AEs; serious AEs were reported promptly by sites.

Cel was given orally 400 mg twice a day for 1 year after regulatory authority advice after withdrawal of another Cox-2 inhibitor, rofecoxib. ZA was planned for ≤ 2 years, given as outpatient infusions at 4 mg/15 min on approximately 66 occasions, starting once every 3 weeks for six cycles then once every 4 weeks. The protocol described modifications for adverse events (AEs). Either treatment was to stop for intolerable AEs or an FFS event. Follow-Up Follow-up, including PSA tests, was every 6 weeks for 6 months, then every 12 weeks to 2 years, every 6 months to 5 years, then annually. Additional tests were at the investigators’ discretion. Common Toxicity Criteria version 3.01 was used to grade AEs; serious AEs were reported promptly by sites. Primary and Secondary Outcomes The primary outcome measure, survival, was time from random assignment to death from any cause. The intermediate primary outcome measure, FFS, was time from random assignment to first evidence of either biochemical failure; local, lymph node, or distant metastatic progression; or death as a result of CaP. Biochemical failure was determined as a rise in PSA ≥ 4 ng/mL and 50% above the lowest reported PSA within 24 weeks after random assignment, or failure to decrease by 50% from the starting PSA during this time. When possible, the cause of death was determined by blinded central review. Death as a result of CaP was recorded when classified by the reviewer as definitely or probably CaP; the site investigator's determination was used for 11 of 584 deaths (2%), with insufficient data available for central review; deaths without reported cause were classified as non-CaP.

ermined by blinded central review. Death as a result of CaP was recorded when classified by the reviewer as definitely or probably CaP; the site investigator's determination was used for 11 of 584 deaths (2%), with insufficient data available for central review; deaths without reported cause were classified as non-CaP. Statistical Design and Analysis We assumed a median FFS of 2 years and a survival of 4 to 5 years for control subjects and targeted a 25% relative improvement (HR, 0.75) in FFS and overall survival (OS) for each pairwise comparison of research arm to control arm. The Stata program nstage (Stata, College Station, TX) allowed the MAMS design with three intermediate activity analyses on the basis of FFS and an efficacy analysis on the basis of survival. The latter had 90% power and a 2.5% one-sided α, requiring approximately 400 control arm deaths; the former each had 95% power and increasingly stringent one-sided αs of 50%, 25%, and 10%, requiring approximately 114, 216, and 334 FFS events, respectively, and expressed as lack-of-benefit stopping guidelines.19 Only patients randomly assigned contemporaneously were compared head to head for each pairwise comparison. Patients were included in the efficacy analyses according to allocated treatment on an intention-to-treat (ITT) basis, unless stated.

S events, respectively, and expressed as lack-of-benefit stopping guidelines.19 Only patients randomly assigned contemporaneously were compared head to head for each pairwise comparison. Patients were included in the efficacy analyses according to allocated treatment on an intention-to-treat (ITT) basis, unless stated. Accumulating data were reviewed by the Independent Data Monitoring Committee. Patients allocated to the Cel arm who still met the eligibility criteria when accrual stopped prematurely were offered complete withdrawal and re-random assignment to an ongoing arm (this is not one of the primary comparisons reported here); five accepted and contribute here only a short period of data from random assignment to withdrawal from their original allocation. Standard survival analysis methods were used for analyses of time-to-event data in STATA version 14. Patients without the relevant event were censored when last reported event free. Cox proportional hazards regression models, adjusted for stratification factors (with the exception of hospital and planned HT), were used to estimate most relative treatment effects, with HR < 1.00 favoring the research arm. Flexible parametric models23 with five degrees of freedom for the baseline hazard function and five degrees of freedom for the time-dependent effect and adjusted for stratification factors were used to estimate medians, 5-year event rates, and restricted mean survival time. Restricted mean survival time took primacy if there was evidence of nonproportional hazards. The likelihood ratio test was used to test for the presence of treatment-subgroup interactions. Fine and Gray regression models were used for competing risk analysis of CaP-specific survival. All confidence intervals are at 95%. Prespecified analyses examined consistency of treatment effect by stratification factors, categorized Gleason score (≤ 7, ≥ 8, or unknown), timing of random assignment (newly diagnosed or recurrent after previous local therapy), and, as a continuous variable, pre-HT PSA. A P value < .10 was taken to be indicative of possible heterogeneity. A preplanned subgroup analysis of M1 patients at random assignment was included in the statistical analysis plan and the complementary M0 group. Sensitivity analyses dividing patients by whether they could have received the maximal duration of Cel (earlier), or not (later), were considered (Data Supplement).

rogeneity. A preplanned subgroup analysis of M1 patients at random assignment was included in the statistical analysis plan and the complementary M0 group. Sensitivity analyses dividing patients by whether they could have received the maximal duration of Cel (earlier), or not (later), were considered (Data Supplement). Median follow-up was estimated by reverse censoring on death, in which survival is treated as the event and death as censoring. Preplanned exploratory factorial analyses of ZA and Cel, with and without an interaction term, drew in data from those patients randomly assigned contemporaneously to SOC + ZA. The safety population for analysis of AEs grouped patients according to treatment started, with sensitivity analysis on an ITT basis. Data on first reported symptomatic skeletal event (SSE) and osteonecrosis of the jaw are also presented. RESULTS Accrual and Patient Characteristics Between October 5, 2005, and April 6, 2011, 1,245 men were randomly assigned 2:1:1 from 80 sites in the United Kingdom and two sites in Switzerland: 622 to SOC-only, 312 to SOC + Cel, and 311 to SOC + ZA + Cel. Data were frozen on December 15, 2015. Figure 1 presents the CONSORT flow diagram and Table 1, baseline characteristics; the Data Supplement contextualizes these arms overall. Median follow-up was 69 months. Median age was 65 years and median PSA was 66 ng/mL, and 65% had a Gleason sum score of 8 to 10. One thousand one hundred sixty-six patients (94%) were newly diagnosed, and 717 (61%) were metastatic at entry; 38 of 79 (48%) who were recurrent after local treatment had metastases.

verall. Median follow-up was 69 months. Median age was 65 years and median PSA was 66 ng/mL, and 65% had a Gleason sum score of 8 to 10. One thousand one hundred sixty-six patients (94%) were newly diagnosed, and 717 (61%) were metastatic at entry; 38 of 79 (48%) who were recurrent after local treatment had metastases. Fig 1. CONSORT flow diagram depicting the flow of patients who joined the STAMPEDE trial while these specific comparisons were open to recruitment. Further context is given in the Data Supplement. A, standard of care (SOC); AE, adverse event; Cel, celecoxib; D, SOC + Cel; F, ZA + SOC + Cel; ZA, zoledronic acid. Table 1. Baseline Characteristics, by Arm Treatment For SOC + Cel and SOC + ZA + Cel, respectively, median time to starting Cel was 0.9 and 1.7 weeks after random assignment, and 6.5 and 7.1 weeks after starting HT; 14 and 17 patients, respectively, did not report starting Cel. Median duration of Cel was 8.3 months for SOC + Cel and 8.0 months for SOC + ZA + Cel (Data Supplement). The most common reason for stopping was treatment completion: 120 of 298 (40%) and 116 of 293 (40%), respectively (Data Supplement). Median ZA starting time was 1.7 weeks after random assignment, and 7.1 weeks from starting HT. ZA was not reported as having started in 12 patients. Median ZA duration was 15 months (Data Supplement). Progression was the most common reason (130 of 297) for stopping treatment early; 100 of 297 (34%) completed the planned 2 years (Data Supplement).

1.7 weeks after random assignment, and 7.1 weeks from starting HT. ZA was not reported as having started in 12 patients. Median ZA duration was 15 months (Data Supplement). Progression was the most common reason (130 of 297) for stopping treatment early; 100 of 297 (34%) completed the planned 2 years (Data Supplement). SOC RT was reported in 158 of 622 SOC-only (25%), 70 of 312 SOC + Cel (22%), and 56 of 311 SOC + ZA + Cel (18%). In patients with nonmetastatic disease, 140 of 245 (57%), 54 of 124 (44%), and 51 of 121 (42%), respectively, reported primary site RT (Data Supplement). Survival There were 303 deaths (251 deaths related to CaP; 83%) in the SOC-only group; median survival was 66 months; and 5-year survival was 53%. There was no evidence of a survival advantage for SOC + Cel (HR, 0.98; 95% CI, 0.80 to 1.20; P = .847: 143 deaths [117 deaths related to CaP; 82%]); median survival was 70 months; and 5-year survival was 54%. Nor was there evidence of a survival advantage for SOC + ZA + Cel (HR, 0.86; 95% CI, 0.70 to 1.05; P = .130: 138 deaths [103 deaths related to CaP; 75%]); median survival was 76 months; and 5-year survival was 59% (Figs 2B and 2D). There was no evidence of nonproportional hazards data.

ths; and 5-year survival was 54%. Nor was there evidence of a survival advantage for SOC + ZA + Cel (HR, 0.86; 95% CI, 0.70 to 1.05; P = .130: 138 deaths [103 deaths related to CaP; 75%]); median survival was 76 months; and 5-year survival was 59% (Figs 2B and 2D). There was no evidence of nonproportional hazards data. Fig 2. Failure-free and overall survival, by research comparison. Kaplan-Meier plots showing time to event for the definitive primary outcome measure (overall survival) and intermediate primary outcome measure (failure-free survival). (A) Failure-free survival in the celecoxib comparison. (B) Overall survival in the celecoxib comparison. (C) Failure-free survival in the ZA + celecoxib comparison. (D) Overall survival in the ZA + Cel comparison. Cel, celecoxib; SOC, standard of care; ZA, zoledronic acid.

imary outcome measure (failure-free survival). (A) Failure-free survival in the celecoxib comparison. (B) Overall survival in the celecoxib comparison. (C) Failure-free survival in the ZA + celecoxib comparison. (D) Overall survival in the ZA + Cel comparison. Cel, celecoxib; SOC, standard of care; ZA, zoledronic acid. Preplanned subgroup analyses in 755 M1 patients included 355 and 349 deaths for the two comparisons. This included 245 deaths in SOC-only; median survival was 43 months; and 5-year survival was 37%. There were 110 deaths in the SOC + Cel group (HR, 0.94; 95% CI, 0.75 to 1.18; P = .602); median survival was 43 months; and 5-year survival was 40%. There were 104 deaths in the SOC + ZA + Cel group (HR, 0.78; 95% CI, 0.62 to 0.98; P = .033); median survival was 55 months; and 5-year was survival 47% (Fig 3). Similar comparisons in M0 patients are relatively immature, with < 100 deaths per comparison. However, we found some indication of possible heterogeneity of treatment effect by metastatic status at random assignment for SOC + ZA + Cel (P = .072; Fig 4). Apart from nodal status (P = .061), there was no other evidence of heterogeneity of treatment effect for either comparison (Fig 4). Fig 3. Overall survival for SOC + Cel + ZA versus SOC in patients with metastatic disease. Kaplan-Meier plot showing overall survival for the ZA + Cel comparison in patients who presented with metastatic disease at random assignment. Cel, celecoxib; SOC, standard of care; ZA, zoledronic acid.

Preplanned subgroup analyses in 755 M1 patients included 355 and 349 deaths for the two comparisons. This included 245 deaths in SOC-only; median survival was 43 months; and 5-year survival was 37%. There were 110 deaths in the SOC + Cel group (HR, 0.94; 95% CI, 0.75 to 1.18; P = .602); median survival was 43 months; and 5-year survival was 40%. There were 104 deaths in the SOC + ZA + Cel group (HR, 0.78; 95% CI, 0.62 to 0.98; P = .033); median survival was 55 months; and 5-year was survival 47% (Fig 3). Similar comparisons in M0 patients are relatively immature, with < 100 deaths per comparison. However, we found some indication of possible heterogeneity of treatment effect by metastatic status at random assignment for SOC + ZA + Cel (P = .072; Fig 4). Apart from nodal status (P = .061), there was no other evidence of heterogeneity of treatment effect for either comparison (Fig 4). Fig 3. Overall survival for SOC + Cel + ZA versus SOC in patients with metastatic disease. Kaplan-Meier plot showing overall survival for the ZA + Cel comparison in patients who presented with metastatic disease at random assignment. Cel, celecoxib; SOC, standard of care; ZA, zoledronic acid. Fig 4. Forest plots of treatment effect on survival within subgroups, by research comparison, showing assessment of consistency of the treatment effect on overall survival in preplanned subgroups for (A) the Cel comparison and (B) the ZA + Cel comparison. The number of deaths and the number of patients are shown by arm for each treatment level, together with an adjusted hazard ratio and a test for heterogeneity of the treatment effect. Cel, celecoxib; Mets, metastases; NSAID, nonsteroidal anti-inflammatory drug; PS, performance status; RT, radiotherapy; SOC, standard of care; ZA, zoledronic acid.

e number of patients are shown by arm for each treatment level, together with an adjusted hazard ratio and a test for heterogeneity of the treatment effect. Cel, celecoxib; Mets, metastases; NSAID, nonsteroidal anti-inflammatory drug; PS, performance status; RT, radiotherapy; SOC, standard of care; ZA, zoledronic acid. Exploratory analysis of the main effects of Cel and ZA individually in a single factorial model without a treatment-interaction term did not associate either drug with a survival advantage (Cel HR, 0.97; 95% CI, 0.83 to 1.13; P = .670; and ZA HR, 0.89; 95% CI, 0.77 to 1.04; P = .150). The factorial model, including a treatment-interaction term, found no evidence of treatment interaction (P = .788). In patients with metastatic disease, the single factorial model without a treatment-interaction term showed no advantage to Cel (HR, 0.92; 95% CI, 0.77 to 1.09; P = .341) or ZA (HR, 0.86; 95% CI, 0.72 to 1.02; P = .083). A further exploratory factorial model, including a treatment-interaction term, found no evidence of treatment interaction (P = .748).

e, the single factorial model without a treatment-interaction term showed no advantage to Cel (HR, 0.92; 95% CI, 0.77 to 1.09; P = .341) or ZA (HR, 0.86; 95% CI, 0.72 to 1.02; P = .083). A further exploratory factorial model, including a treatment-interaction term, found no evidence of treatment interaction (P = .748). FFS Figures 2A and 2C present an FFS plot for each comparison. PSA failure was the most common for each arm (Data Supplement). There were 457 FFS events for SOC-only; median FFS was 20 months; and 5-year FFS was 26%. SOC + Cel had 213 events with no evidence of improved FFS (HR, 0.87; 95% CI, 0.74 to 1.03; P = .102); median FFS was 22 months; and 5-year FFS was 29%. SOC + ZA + Cel had 213 events and some evidence of a difference compared with SOC (HR, 0.84; 95% CI, 0.72 to 0.99; P = .043); median FFS was 24 months; and 5-year FFS was 30%. There was no evidence of nonproportional hazards. There was evidence of heterogeneity of treatment effect by predefined subgroups including performance status and baseline NSAID and/or aspirin use for both comparisons, in addition to recurrent disease for SOC + Cel and Gleason score and age at random assignment for SOC + ZA + Cel (Data Supplement). In the subgroup analyses by baseline metastatic status, the estimates for FFS in SOC + ZA + Cel were HR, 0.77 (95% CI, 0.63 to 0.93) in metastatic disease and HR, 1.02 (95% CI, 0.75 to 1.39) in nonmetastatic disease, but there was no evidence of heterogeneity (P = .119).

assignment for SOC + ZA + Cel (Data Supplement). In the subgroup analyses by baseline metastatic status, the estimates for FFS in SOC + ZA + Cel were HR, 0.77 (95% CI, 0.63 to 0.93) in metastatic disease and HR, 1.02 (95% CI, 0.75 to 1.39) in nonmetastatic disease, but there was no evidence of heterogeneity (P = .119). Factorial analyses without an interaction term in the 643 earlier patients suggested effects on FFS in patients with metastatic disease from both Cel (HR, 0.85; 95% CI, 0.71 to 1.01; P = .069) and ZA (HR, 0.84; 95% CI, 0.70 to 1.00; P = .052), but no evidence of interaction between treatments in a further model (P = .942). CaP-Specific Survival Of 584 deaths, 471 (81%) were a result of CaP; a higher proportion of deaths was attributed to CaP in patients with metastatic disease (381 of 459 deaths [83%] in 755 M1 patients, and 90 of 125 deaths [72%] in 490 patients with nonmetastatic disease). Adjusted competing risk regression for CaP-specific survival showed no evidence of advantage over SOC-only for SOC + Cel (sub-HR, 0.97; 95% CI, 0.77 to 1.23; P = .782) but an advantage for SOC + ZA + Cel (sub-HR, 0.74; 95% CI, 0.59 to 0.94; P = .014). For patients with metastatic disease, the sub-HR for SOC + Cel was 0.91 (95% CI, 0.71 to 1.18), and the sub-HR for SOC + ZA + Cel was 0.64 (95% CI, 0.49 to 0.83); for patients with nonmetastatic disease, the sub-HR for SOC + Cel was 1.44 (95% CI, 0.86 to 2.40), and the sub-HR for SOC + ZA + Cel was 1.31 (95% CI, 0.78 to 2.18).

ents with metastatic disease, the sub-HR for SOC + Cel was 0.91 (95% CI, 0.71 to 1.18), and the sub-HR for SOC + ZA + Cel was 0.64 (95% CI, 0.49 to 0.83); for patients with nonmetastatic disease, the sub-HR for SOC + Cel was 1.44 (95% CI, 0.86 to 2.40), and the sub-HR for SOC + ZA + Cel was 1.31 (95% CI, 0.78 to 2.18). Of deaths not related to CaP, 23 of 113 (20%) were classified as cardiovascular disease: nine of 52 (17%) SOC-only, three of 26 (12%) SOC + Cel, and 11 of 35 (31%) SOC + ZA + Cel. SSEs A total of 207 of 622 SOC-only patients reported one or more SSEs. There was no evidence that the time to first SSE was improved with SOC + Cel (90 of 312 patients reported SSE: HR, 0.85; 95% CI, 0.66 to 1.08; P = .186) or with SOC + ZA + Cel (95 of 311 patients reported SSE: HR, 0.84; 95% CI, 0.66 to1.07; P = .162). AEs In per protocol analyses of safety, one third reported worst AE ever as grade ≥ 3: 222 of 625 (36%) SOC-only, 98 of 296 (33%) SOC + Cel, and 95 of 293 (32%) SOC + ZA + Cel (Table 2). In 799 patients with AE assessment at approximately 1 year after random assignment, the proportions with grade ≥ 3 AE were 43 of 398 (11%) SOC-only, 16 of 200 (8%) SOC + Cel, and 13 of 196 (7%) SOC + ZA + Cel, mostly related to SOC with androgen deprivation therapy. Patterns and levels of AEs were similar in the ITT population. There were six reported cases of osteonecrosis of the jaw, all in the SOC + ZA + Cel group. Table 2. Worst AE (grade) Reported Over Entire Time on Trial, by Treatment Reported

AEs In per protocol analyses of safety, one third reported worst AE ever as grade ≥ 3: 222 of 625 (36%) SOC-only, 98 of 296 (33%) SOC + Cel, and 95 of 293 (32%) SOC + ZA + Cel (Table 2). In 799 patients with AE assessment at approximately 1 year after random assignment, the proportions with grade ≥ 3 AE were 43 of 398 (11%) SOC-only, 16 of 200 (8%) SOC + Cel, and 13 of 196 (7%) SOC + ZA + Cel, mostly related to SOC with androgen deprivation therapy. Patterns and levels of AEs were similar in the ITT population. There were six reported cases of osteonecrosis of the jaw, all in the SOC + ZA + Cel group. Table 2. Worst AE (grade) Reported Over Entire Time on Trial, by Treatment Reported Second-Line Treatment The Data Supplement lists time from FFS event to next treatment and time to any of the five life-extending therapies in castrate-refractory prostate cancer (available agents with proven survival gain). There was no evidence among arms of a difference in time to either any therapy or life-extending therapies (Data Supplement). There were no reports of patients allocated to SOC-only switching to Cel after progression; 82 and 22 patients in the SOC-only group and the SOC + Cel group, respectively, reported using ZA after progression.

vidence among arms of a difference in time to either any therapy or life-extending therapies (Data Supplement). There were no reports of patients allocated to SOC-only switching to Cel after progression; 82 and 22 patients in the SOC-only group and the SOC + Cel group, respectively, reported using ZA after progression. DISCUSSION We report the findings of two randomized comparisons from STAMPEDE, a MAMS-platform trial, in 1,245 patients starting long-term first-line HT, with a median follow-up of > 5 years. These two Cel comparisons show no overall evidence that Cel, alone or combined with ZA, improved survival or FFS compared with SOC. However, a preplanned subgroup analysis suggested the possibility of benefit in terms of both survival and FFS for SOC + Cel + ZA over SOC alone in M1 patients at random assignment, although the test for interaction was not significant at a traditional 5% significance level. Importantly, follow-up continued after accrual to Cel stopped. There is no bias in the reporting time of these comparisons; we report them, as planned, straight after survival results from the other arms that started contemporaneously. Data return rates are good, with most patients known to be alive having reported data in the past year. Although recruitment to these comparisons was terminated early, there are still three quarters the number of control arm deaths that would have triggered this survival analysis had the comparisons passed all three interim analyses (n = approximately 300 of 400).

Importantly, follow-up continued after accrual to Cel stopped. There is no bias in the reporting time of these comparisons; we report them, as planned, straight after survival results from the other arms that started contemporaneously. Data return rates are good, with most patients known to be alive having reported data in the past year. Although recruitment to these comparisons was terminated early, there are still three quarters the number of control arm deaths that would have triggered this survival analysis had the comparisons passed all three interim analyses (n = approximately 300 of 400). Results for the combination of Cel and ZA are intriguing and unexpected, given that we observed no evidence of improvement in OS or FFS with the addition of either Cel or ZA alone. Unfortunately, no further comparative data that would help directly in interpretation are expected. No other powered randomized controlled trial combining bisphosphonates and Cox-2 inhibitors are listed on trial registers; studies listed are nonrandomized or small or they terminated early. Our data fall short of definite evidence that this is a real effect; tests for interaction typically lack power, particularly in subgroups and where accrual is terminated early. Rather, they are hypothesis generating and provoke further research. Biologically, an effect by directly targeting tumor and/or host cells, particularly of stromal and immune lineage, is not implausible.24-26

t; tests for interaction typically lack power, particularly in subgroups and where accrual is terminated early. Rather, they are hypothesis generating and provoke further research. Biologically, an effect by directly targeting tumor and/or host cells, particularly of stromal and immune lineage, is not implausible.24-26 The observed effect on survival of combined Cel and ZA in the metastatic setting is of similar magnitude to that reported for docetaxel in the STOPCaP (Systemic Treatment Options for Prostate Cancer) meta-analysis,4 although the observed effect on FFS is much less pronounced (Table 3 and Data Supplement). There are reports of anticancer activity with the nonsteroidal anti-inflammatory agent diclofenac,24 and STOPCaP suggests an overall benefit of bisphosphonates in metastatic disease, albeit driven by one trial of sodium clodronate.4 Table 3. Summary Data From this Article and for Doc and ZA From Meta-Analysis

The observed effect on survival of combined Cel and ZA in the metastatic setting is of similar magnitude to that reported for docetaxel in the STOPCaP (Systemic Treatment Options for Prostate Cancer) meta-analysis,4 although the observed effect on FFS is much less pronounced (Table 3 and Data Supplement). There are reports of anticancer activity with the nonsteroidal anti-inflammatory agent diclofenac,24 and STOPCaP suggests an overall benefit of bisphosphonates in metastatic disease, albeit driven by one trial of sodium clodronate.4 Table 3. Summary Data From this Article and for Doc and ZA From Meta-Analysis We defined survival as the definitive outcome measure, D, and FFS as the intermediate outcome measure, I. When this trial was first designed, some fundamental principles and assumptions were adopted, following the process for MAMS designs: (1) a treatment that does not improve I is unlikely to sufficiently improve D; (2) an improvement in I may not translate into an improvement in D; (3) therefore, a failure to improve I by a prespecified amount can be used as a triage for D, to allow early stopping of recruitment, but success in improving I does not remove the need to continue recruitment to reliably assess D; (4) I does not need to be a surrogate for D in the strictest sense, but only on the causal pathway; and (5) arms whose recruitment is stopped because of failure to improve I will continue to be followed up, and D will be reported soon after that for contemporaneously initiated arms.

ue recruitment to reliably assess D; (4) I does not need to be a surrogate for D in the strictest sense, but only on the causal pathway; and (5) arms whose recruitment is stopped because of failure to improve I will continue to be followed up, and D will be reported soon after that for contemporaneously initiated arms. When the trial launched in 2005, these assumptions were still thought to be reasonable with our choices for I and D. With the hindsight afforded by a decade of new knowledge and new data, we now realize that in CaP and in some other cancers, and with some interventions, there might be more discordance between FFS and survival in some circumstances.

5, these assumptions were still thought to be reasonable with our choices for I and D. With the hindsight afforded by a decade of new knowledge and new data, we now realize that in CaP and in some other cancers, and with some interventions, there might be more discordance between FFS and survival in some circumstances. FFS was the interim outcome measure for Cel and ZA. The vast majority of patients respond to HT with a fall in PSA values; we were specifically looking for an enhancement of the time to treatment failure by the addition of these agents, which legitimizes FFS as an outcome measure because response was not an appropriate outcome. The primary outcome measure is OS, which is not dependent directly on PSA values, although PSA values will have affected treatment decisions for clinicians along the way. The multistage approach in a MAMS trial does not require a perfect surrogate outcome measure, but one on the causal pathway, as FFS is for survival. Overall, there was minimal effect on FFS, and this translated into minimal effect on OS. In this regard, FFS was an acceptable choice. For the comparisons added later in STAMPEDE, we used the same definition of FFS treatment in both the abiraterone comparison and the enzalutamide plus abiraterone comparison in STAMPEDE, but allowed treatment to continue using all three types of progression (biochemical, clinical, and radiologic); for the new metformin comparison that opened in STAMPEDE in September 2016, we have chosen not to use a PSA-driven outcome measure as the intermediate outcome measure.27,28

abiraterone comparison in STAMPEDE, but allowed treatment to continue using all three types of progression (biochemical, clinical, and radiologic); for the new metformin comparison that opened in STAMPEDE in September 2016, we have chosen not to use a PSA-driven outcome measure as the intermediate outcome measure.27,28 Interpretation is further complicated by early cessation of Cel treatment when recruitment stopped at the second activity analysis (Data Supplement), which meant that fewer patients could receive the planned duration of their allocated treatment. Long-term estimation of the treatment effect on FFS of SOC + Cel versus SOC-only in the Cel comparison is consistent with our previous publication,19 when Cel failed to pass its second intermediate activity threshold. At random assignment, RT was stated as part of the treatment plan for approximately one quarter of M0 patients across all arms. Around the time recruitment to these comparisons was stopped, data were emerging from NCIC-CTG-PR.3/MRC-PR07 and SPCG-7 that RT improved survival for men starting long-term HT for M0 CaP.29,30 Herein, the reported use of RT in the control group met expectations, but it was lower in both research arms. This preferential omission of RT from the Cel-containing arms complicates the findings, and a more favorable signal may have been observed had RT not been preferentially omitted for unrecorded reasons.

aP.29,30 Herein, the reported use of RT in the control group met expectations, but it was lower in both research arms. This preferential omission of RT from the Cel-containing arms complicates the findings, and a more favorable signal may have been observed had RT not been preferentially omitted for unrecorded reasons. Approximately one half of the patients died, mostly as a result of CaP. There were several deaths with cardiovascular causes, but these occurred proportionally more in the combination arm. One stratification factor at random assignment was the use of NSAIDs and/or aspirin, included only because of these Cel comparisons. Both comparisons had much improved FFS in patients receiving NSAIDs and/or aspirin at baseline, but this did not translate into evidence of a difference in survival. Baseline use of NSAIDs and/or aspirin likely related to underlying, pre-existing comorbidities, but this effect deserves further exploration. Our data show no evidence of a survival advantage in adding Cel alone for all men starting long-term HT for the first time. We previously also showed no evidence of a survival advantage in adding ZA alone for the same patient group. Overall, the combination of Cel and ZA had no effect. Preplanned subgroup analyses may provide a hypothesis for future studies to investigate adding Cel in settings in which ZA is already part of the SOC.

time. We previously also showed no evidence of a survival advantage in adding ZA alone for the same patient group. Overall, the combination of Cel and ZA had no effect. Preplanned subgroup analyses may provide a hypothesis for future studies to investigate adding Cel in settings in which ZA is already part of the SOC. Supported by the Cancer Research UK, Medical Research Council, Novartis, Sanofi, Pfizer, Janssen Pharmaceuticals, Astellas Pharma, National Institute of Health Research Clinical Research Network (formerly National Cancer Research Network), and the Swiss Group for Clinical Cancer Research. Clinical trial information: NCT00268476 and ISRCTN78818544. See accompanying Editorial on page 1501 ACKNOWLEDGMENT D.P.D., C.C.P., and G.A. acknowledge support from the NIHR to the Royal Marsden NHS Trust and The Institute of Cancer Research Biomedical Research Centre AUTHOR CONTRIBUTIONS Conception and design: Malcolm D. Mason, Noel W. Clarke, Nicholas D. James, David P. Dearnaley, Melissa R. Spears, Alastair W.S. Ritchie, Gerhardt Attard, Christopher C. Parker, George N. Thalmann, Estelle Cassoly, David Matheson, Robin Millman, Jim Barber, Azman Ibrahim, Anna Lydon, Ashok D. Nikapota, John Wagstaff, Jan Wallace, Mahesh K.B. Parmar, Matthew R. Sydes Administrative support: Francesca Schiavone, Matthew R. Sydes

AUTHOR CONTRIBUTIONS Conception and design: Malcolm D. Mason, Noel W. Clarke, Nicholas D. James, David P. Dearnaley, Melissa R. Spears, Alastair W.S. Ritchie, Gerhardt Attard, Christopher C. Parker, George N. Thalmann, Estelle Cassoly, David Matheson, Robin Millman, Jim Barber, Azman Ibrahim, Anna Lydon, Ashok D. Nikapota, John Wagstaff, Jan Wallace, Mahesh K.B. Parmar, Matthew R. Sydes Administrative support: Francesca Schiavone, Matthew R. Sydes Provision of study materials or patients: Malcolm D. Mason, Noel W. Clarke, Nicholas D. James, David P. Dearnaley, Alastair W.S. Ritchie, Gerhardt Attard, William Cross, Rob J. Jones, Christopher C. Parker, J. Martin Russell, George N. Thalmann, Estelle Cassoly, David Matheson, Robin Millman, Jim Barber, Azman Ibrahim, John Logue, Anna Lydon, Ashok D. Nikapota, Joe M. O'Sullivan, Emilio Porfiri, Andrew Protheroe, Narayanan Nair Srihari, John Wagstaff, Jan Wallace, Mahesh K.B. Parmar, Matthew R. Sydes Collection and assembly of data: Malcolm D. Mason, Noel W. Clarke, Nicholas D. James, David P. Dearnaley, Melissa R. Spears, Alastair W.S. Ritchie, Gerhardt Attard, Rob J. Jones, Christopher C. Parker, J. Martin Russell, George N. Thalmann, Francesca Schiavone, Estelle Cassoly, Robin Millman, Cyrill A. Rentsch, Jim Barber, Azman Ibrahim, John Logue, Anna Lydon, Ashok D. Nikapota, Joe M. O'Sullivan, Narayanan Nair Srihari, David Tsang, John Wagstaff, Jan Wallace, Catherine Walmsley, Mahesh K.B. Parmar, Matthew R. Sydes

. Martin Russell, George N. Thalmann, Francesca Schiavone, Estelle Cassoly, Robin Millman, Cyrill A. Rentsch, Jim Barber, Azman Ibrahim, John Logue, Anna Lydon, Ashok D. Nikapota, Joe M. O'Sullivan, Narayanan Nair Srihari, David Tsang, John Wagstaff, Jan Wallace, Catherine Walmsley, Mahesh K.B. Parmar, Matthew R. Sydes Data analysis and interpretation: Malcolm D. Mason, Noel W. Clarke, Nicholas D. James, David P. Dearnaley, Melissa R. Spears, Gerhardt Attard, William Cross, Christopher C. Parker, J. Martin Russell, George N. Thalmann, Estelle Cassoly, Robin Millman, Cyrill A. Rentsch, Jim Barber, Clare Gilson, Azman Ibrahim, Anna Lydon, Ashok D. Nikapota, Emilio Porfiri, Andrew Protheroe, Narayanan Nair Srihari, John Wagstaff, Jan Wallace, Mahesh K.B. Parmar, Matthew R. Sydes Manuscript writing: All authors Final approval of manuscript: All authors Accountable for all aspects of the work: All authors AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST Adding Celecoxib With or Without Zoledronic Acid for Hormone-Naïve Prostate Cancer: Long-Term Survival Results From an Adaptive, Multiarm, Multistage, Platform, Randomized Controlled Trial The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/jco/site/ifc.

ancer is the relative risk (RR) of cancer in HL survivors, SIRfh is the RR associated with having an affected FDR, and SIRcancerxfh is the RR of cancer in HL survivors having an affected FDR. MII > 1 signifies greater than multiplicative interaction, and ICR > 0 signifies a positive interaction or more than additivity. The relative survival rate was calculated as the ratio of the observed survival rate to the expected survival rate in Sweden, matched by age, sex, and calendar year.26,27 Statistical analyses were performed using Stata version 14 (STATA, College State, TX) and R version 3.3.1 software.28 A P value ≤ .05 (two-sided) was considered statistically significant, although we acknowledge we have presented the results of many statistical tests, and therefore caution against the overinterpretation of our findings, especially when they are based on P values > .001.

s are considered compensated. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/jco/site/ifc. Malcolm D. Mason Consulting or Advisory Role: Sanofi, Bayer AG Speakers’ Bureau: Bayer AG, Janssen Pharmaceuticals Noel W. Clarke No relationship to disclose Nicholas D. James Honoraria: Sanofi, Bayer AG, Oncogenex, Janssen Pharmaceuticals, Astellas Pharma, Pierre Fabre Consulting or Advisory Role: Sanofi, Bayer AG, Merck, Astellas Pharma, Janssen Pharmaceuticals Speakers' Bureau: Pierre Fabre, Ferring Pharmaceuticals, Sanofi, Astellas Pharma Research Funding: Janssen (Inst), Astellas Pharma (Inst), Pfizer (Inst), Sanofi (Inst), Novartis (Inst) Travel, Accommodations, Expenses: Sanofi David P. Dearnaley Honoraria: Takeda Pharmaceuticals, Sandoz, Janssen Pharmaceuticals Consulting or Advisory Role: Takeda Pharmaceuticals, Janssen Research and Development, Sandoz, Cadence Research and Consulting, Janssen Pharmaceuticals Speakers' Bureau: Janssen Pharmaceuticals, Janssen-Cilag Research Funding: Takeda Pharmaceuticals Patents, Royalties, Other Intellectual Property: Abiraterone acetate was developed at The Institute of Cancer Research, which therefore has a commercial interest in the development of this agent. David P. Dearnaley is on the Institute’s Rewards to Inventors list for abiraterone acetate (Inst) Expert Testimony: Vitality Life

Patents, Royalties, Other Intellectual Property: Abiraterone acetate was developed at The Institute of Cancer Research, which therefore has a commercial interest in the development of this agent. David P. Dearnaley is on the Institute’s Rewards to Inventors list for abiraterone acetate (Inst) Expert Testimony: Vitality Life Travel, Accommodations, Expenses: Takeda Pharmaceuticals, Janssen Pharmaceuticals, Sandoz Melissa R. Spears Research Funding: Sanofi, Novartis, Pfizer, Janssen Pharmaceuticals, Astellas Pharma Alastair W.S. Ritchie No relationship to disclose Gerhardt Attard Honoraria: Janssen Pharmaceuticals, Astellas Pharma Consulting or Advisory Role: Janssen-Cilag, Veridex, Ventana Medical Systems, Astellas Pharma, Medivation, Novartis, Millennium Pharmaceuticals, Abbott Laboratories, ESSA Pharma, Bayer AG Speakers' Bureau: Janssen Pharmaceuticals, Astellas Pharma, Takeda Pharmaceuticals, Sanofi, Ventana Medical Systems Research Funding: Janssen Pharmaceuticals (Inst), AstraZeneca (Inst), Arno Therapeutics (Inst), Innocrin Pharma (Inst) Patents, Royalties, Other Intellectual Property: I am on The ICR rewards to inventors list of abiraterone acetate. Travel, Accommodations, Expenses: Janssen Pharmaceuticals, Astellas Pharma, Medivation, Ventana Medical Systems, Abbott Laboratories, Bayer AG, ESSA Pharma, Janssen Pharmaceuticals (I), Astellas Pharma (I) Other Relationship: Institute of Cancer Research William Cross Consulting or Advisory Role: Takeda Pharmaceuticals Speakers' Bureau: Takeda Pharmaceuticals, Janssen Pharmaceuticals, Bayer AG Travel, Accommodations, Expenses: Janssen Pharmaceuticals

Travel, Accommodations, Expenses: Janssen Pharmaceuticals, Astellas Pharma, Medivation, Ventana Medical Systems, Abbott Laboratories, Bayer AG, ESSA Pharma, Janssen Pharmaceuticals (I), Astellas Pharma (I) Other Relationship: Institute of Cancer Research William Cross Consulting or Advisory Role: Takeda Pharmaceuticals Speakers' Bureau: Takeda Pharmaceuticals, Janssen Pharmaceuticals, Bayer AG Travel, Accommodations, Expenses: Janssen Pharmaceuticals Rob J. Jones Honoraria: Novartis, Pfizer Consulting or Advisory Role: Novartis, Pfizer Research Funding: Novartis, Pfizer (Ins) Christopher C. Parker Honoraria: Bayer AG, Janssen Pharmaceuticals Consulting or Advisory Role: Bayer Schering Pharma, AAA Research Funding: Bayer Schering Pharma (Inst) Travel, Accommodations, Expenses: Bayer Schering Pharma, Janssen Pharmaceuticals J. Martin Russell Travel, Accommodations, Expenses: Janssen-Cilag George N. Thalmann No relationship to disclose Francesca Schiavone No relationship to disclose Estelle Cassoly Research Funding: Pfizer (Inst) David Matheson No relationship to disclose Robin Millman No relationship to disclose Cyrill A. Rentsch No relationship to disclose Jim Barber No relationship to disclose Clare Gilson No relationship to disclose Azman Ibrahim No relationship to disclose John Logue Travel, Accommodations, Expenses: Bayer AG Anna Lydon Speakers' Bureau: Astellas Pharma Travel, Accommodations, Expenses: Sanofi, Pfizer, Janssen Pharmaceuticals Ashok D. Nikapota No relationship to disclose Joe M. O'Sullivan Honoraria: Bayer AG Consulting or Advisory Role: Bayer AG, Janssen Pharmaceuticals, Astellas Pharma, Sanofi

John Logue Travel, Accommodations, Expenses: Bayer AG Anna Lydon Speakers' Bureau: Astellas Pharma Travel, Accommodations, Expenses: Sanofi, Pfizer, Janssen Pharmaceuticals Ashok D. Nikapota No relationship to disclose Joe M. O'Sullivan Honoraria: Bayer AG Consulting or Advisory Role: Bayer AG, Janssen Pharmaceuticals, Astellas Pharma, Sanofi Speakers' Bureau: Bayer AG, Janssen Pharmaceuticals Research Funding: Bayer AG Emilio Porfiri Consulting or Advisory Role: BMS, Novartis, Pfizer Speakers' Bureau: Bristol-Myers Squibb, Novartis, Pfizer Research Funding: Novartis Travel, Accommodations, Expenses: Astellas Pharma Andrew Protheroe No relationship to disclose Narayanan Nair Srihari Research Funding: Novartis (Inst) Travel, Accommodations, Expenses: Janssen Pharmaceuticals, Pfizer, Sanofi David Tsang No relationship to disclose John Wagstaff Consulting or Advisory Role: Bristol-Myers Squib, Novartis/Ipsen, Roche, Pfizer, Merck Speakers' Bureau: Sanofi, Bristol-Myers Squibb Travel, Accommodations, Expenses: Merck, Bristol-Myers Squibb Jan Wallace Travel, Accommodations, Expenses: Astellas Pharma Catherine Walmsley No relationship to disclose Mahesh K.B. Parmar Research Funding: Novartis, Sanofi, Pfizer, Janssen Pharmaceuticals, Astellas Pharma Matthew R. Sydes Research Funding: Astellas Pharma, Janssen-Cilag, Pfizer, Novartis, Sanofi, Clovis Oncology

llege State, TX) and R version 3.3.1 software.28 A P value ≤ .05 (two-sided) was considered statistically significant, although we acknowledge we have presented the results of many statistical tests, and therefore caution against the overinterpretation of our findings, especially when they are based on P values > .001. RESULTS Patients and Record Linkage From the Swedish Family-Cancer Project Database, data on 9,522 patients with a primary diagnosis of HL between 1965 and 2013 were analyzed. Of the 9,522 patients 5,488 were male and 4,034 were female, with a mean age at diagnosis of 49 years (Table 1). Five thousand seven hundred twenty-one were deceased and 129 had emigrated before the end of the study period. The median follow-up was 12.6 years, with the longest being 48 years. Of patients with HL in whom tumor with histology had been recorded, 1,839 (54%) were nodular sclerosis HL (NSHL) and 539 (16%) were of mixed cellularity HL. Among the 3,917 individuals who died > 1 year after the diagnosis of HL, 842 (9%) died with the occurrence of a subsequent cancer during the follow-up. Table 1. Clinicopathologic Characteristics of Patients With Hodgkin Lymphoma Diagnosed Between 1965 and 2013 in Sweden

INTRODUCTION Advances in the management of Hodgkin lymphoma (HL) over the past 40 years have led to improved disease-free survival in patients.1 However, this comes at the cost of an increased risk of second cancers, cardiovascular disease, and other treatment-related complications.2-13 The risk of second cancers in HL survivors, which persists for many decades after treatment, has been reported to be influenced by various factors, including age at treatment,3 site and dose of radiotherapy,14 chemotherapy,5 and smoking.15 The use of treatment regimens based on a reduction in the field and dose of radiotherapy and alkylating chemotherapy has been introduced to reduce rates of long-term complications while maintaining a high cure rate.1 Despite such modifications, a recent study from the Netherlands showed that this has not affected the risk of second cancers in patients with HL12. A family history of breast cancer was first suggested to be a risk factor for second cancer in HL nearly 20 years ago,16 and it has long been postulated that a subset of patients with cancer display a high sensitivity to mutational agents because of a genetic predisposition.17 Evidence for such an assertion in the context of HL is provided by an analysis of a Swedish cohort of patients with HL, but the power of the study did not allow the impact of family history to be studied in detail.18 Moreover, the analysis included HL survivors with a prior history of cancer other than HL, which potentially biased the conclusions.

n in the context of HL is provided by an analysis of a Swedish cohort of patients with HL, but the power of the study did not allow the impact of family history to be studied in detail.18 Moreover, the analysis included HL survivors with a prior history of cancer other than HL, which potentially biased the conclusions. To gain insight into the risk of second cancer after a diagnosis of HL and its relationship to temporal changes in treatment regimens, we analyzed data on a cohort of 9,522 Swedish patients. In addition, through the use of the Swedish Family-Cancer Project Database, we performed an updated analysis of the influence of family history, as a surrogate for genetic susceptibility, on the risk of second cancer in patients with HL.

ges in treatment regimens, we analyzed data on a cohort of 9,522 Swedish patients. In addition, through the use of the Swedish Family-Cancer Project Database, we performed an updated analysis of the influence of family history, as a surrogate for genetic susceptibility, on the risk of second cancer in patients with HL. PATIENTS AND METHODS Patients The Swedish Family-Cancer Project Database was created by linking information from the multigeneration register, national censuses, the Swedish Cancer Registry, and death notifications.19 The Swedish Cancer Registry, established in 1958, is based on the compulsory reporting of all diagnosed patients, thereby providing near-complete coverage of all cancer registrations in Sweden.20 There is an under-representation of individuals in the first generation in some families; however, this has not been shown to adversely bias estimates of familial risk.21,22 In this study, we analyzed all incident cases of HL between 1965 and 2012. Individuals with a diagnosis of malignancy before HL were excluded. Individuals with HL were observed until December 31, 2012, time of migration from Sweden, or death. Data regarding HL histologic subtype were available for all individuals diagnosed since 1993. The database includes the date and site of occurrence of up to four subsequent new malignancies after diagnosis of HL and dates and causes of death. Cancers in the first year after HL diagnosis were omitted from our analysis because of the likelihood of excess cases as a result of increased surveillance.23 First-degree relatives (FDRs) of individuals with HL, as well as the dates and sites of cancer diagnoses in the FDRs, were identified. The study was undertaken with approval from the ethics committee at Lund University, Sweden, and was conducted in accordance with the tenets of the Declaration of Helsinki.

veillance.23 First-degree relatives (FDRs) of individuals with HL, as well as the dates and sites of cancer diagnoses in the FDRs, were identified. The study was undertaken with approval from the ethics committee at Lund University, Sweden, and was conducted in accordance with the tenets of the Declaration of Helsinki. Statistical Analysis Expected numbers of cancers were computed using 5-year age, sex, and calendar period incidence rates for Sweden. Observed numbers were compared with expected numbers by means of the standardized incidence ratio (SIR) assuming a Poisson distribution. The risk of a second malignancy was estimated for different time intervals after treatment of HL. The absolute excess risk (AER) was calculated as the observed number of second cancers in our cohort minus that expected, divided by the number of person-years at risk, multiplied by 10,000. SIRs were calculated in the HL cohort and were stratified by patient characteristics. Tests for trend in SIRs were performed by evaluating the likelihood function in collapsed person-time additive Poisson regression models with and without the inclusion of the variable. Patients in whom multiple second cancers were diagnosed were counted only once in the analysis of all second cancers; in this analysis, the time at risk ended on the date on which a second cancer was diagnosed. For the site-specific cancer analyses, the time at risk ended on the date on which the site-specific cancer was diagnosed, regardless of whether this was preceded by another cancer. The cumulative incidence of second cancer was estimated with death treated as a competing risk.24 Interaction contrast ratios (ICRs) and multiplicative interaction indexes (MIIs) were used to investigate the possible interaction between HL treatment and family history of cancer:25 ICR = SIRcancerxfh − SIRcancer − SIRfh + 1 and MII = SIRcancerxfh/(SIRcancer × SIRfh), where SIRcancer is the relative risk (RR) of cancer in HL survivors, SIRfh is the RR associated with having an affected FDR, and SIRcancerxfh is the RR of cancer in HL survivors having an affected FDR. MII > 1 signifies greater than multiplicative interaction, and ICR > 0 signifies a positive interaction or more than additivity.

RESULTS Patients and Record Linkage From the Swedish Family-Cancer Project Database, data on 9,522 patients with a primary diagnosis of HL between 1965 and 2013 were analyzed. Of the 9,522 patients 5,488 were male and 4,034 were female, with a mean age at diagnosis of 49 years (Table 1). Five thousand seven hundred twenty-one were deceased and 129 had emigrated before the end of the study period. The median follow-up was 12.6 years, with the longest being 48 years. Of patients with HL in whom tumor with histology had been recorded, 1,839 (54%) were nodular sclerosis HL (NSHL) and 539 (16%) were of mixed cellularity HL. Among the 3,917 individuals who died > 1 year after the diagnosis of HL, 842 (9%) died with the occurrence of a subsequent cancer during the follow-up. Table 1. Clinicopathologic Characteristics of Patients With Hodgkin Lymphoma Diagnosed Between 1965 and 2013 in Sweden Risk of Second Cancer in Patients With HL A total of 1,215 second cancers were observed in 1,121 patients (12% of patients with HL). The risk of all second cancers was elevated significantly after HL diagnosis, with a SIR of 2.39 (95% CI, 2.25 to 2.53), translating to an AER of 71.2 cases per 10,000 person-years (Table 2). In the nonstratified analysis, non-Hodgkin lymphoma (NHL) contributed the most to the AER (16.2% of the excess cancer risk), followed by lung cancer (14.5% of the excess cancer risk), breast cancer (12.9% of the excess cancer risk), nonmelanoma skin cancers (11.4% of the excess cancer risk), leukemia (9.7% of the excess cancer risk), and colorectal cancer (7.6% of the excess cancer risk). The SIR for all second cancers remained high > 30 years after treatment of HL, although the patterns of excess risk observed at the different intervals differed depending on the cancer site (Table 2).

excess cancer risk), leukemia (9.7% of the excess cancer risk), and colorectal cancer (7.6% of the excess cancer risk). The SIR for all second cancers remained high > 30 years after treatment of HL, although the patterns of excess risk observed at the different intervals differed depending on the cancer site (Table 2). Table 2. Risk of Second Cancer at 1 to 9, 10 to 19, 20 to 30, and > 30 Years After Diagnosis of Hodgkin Lymphoma

excess cancer risk), leukemia (9.7% of the excess cancer risk), and colorectal cancer (7.6% of the excess cancer risk). The SIR for all second cancers remained high > 30 years after treatment of HL, although the patterns of excess risk observed at the different intervals differed depending on the cancer site (Table 2). Table 2. Risk of Second Cancer at 1 to 9, 10 to 19, 20 to 30, and > 30 Years After Diagnosis of Hodgkin Lymphoma Influence of Sex, Age, and Tumor Subtype on Second Cancer Risk For men diagnosed with HL before the age of 35 years, the SIRs for all second cancers, colorectal cancer, lung cancer, NHL, and leukemia were higher when compared with those diagnosed with HL after the age of 35 years. Similarly, for women diagnosed with HL before the age of 35 years, the SIRs for all second cancers, breast cancer, lung cancer, and NHL were higher when compared with women diagnosed with HL after the age of 35 years (Table 3). In women diagnosed with HL at younger than 35 years of age, the 30-year cumulative risk of breast cancer was 13.8% (95% CI, 11.1 to 16.9), which accounted for > 50% of the AER in this age group of women. This contrasted with women diagnosed with HL at older than 35 years of age for whom the 30-year cumulative incidence of breast cancer was only 3.3% (95% CI, 2.2 to 3.9), and accounted for < 3% of the AER (Fig 1 and Data Supplement). Given the difference in cause and tumor biology of HL histologic subtypes, we investigated whether second cancer risk might also differ. We observed similar SIRs for cancer overall and the common site-specific cancers for the most common subtype, NSHL (Data Supplement). We do acknowledge, however, that this observation should be interpreted with caution because numbers are small and the unclassified HL cases include a proportion of NSHL.

ht also differ. We observed similar SIRs for cancer overall and the common site-specific cancers for the most common subtype, NSHL (Data Supplement). We do acknowledge, however, that this observation should be interpreted with caution because numbers are small and the unclassified HL cases include a proportion of NSHL. Table 3. Risk of Second Cancer After Diagnosis of Hodgkin Lymphoma by Sex and Age at HL Diagnosis Fig 1. Cumulative incidence of breast cancer in female survivors of Hodgkin lymphoma, by age at Hodgkin lymphoma diagnosis, with death treated as a competing risk. The solid blue line represents women diagnosed with Hodgkin lymphoma at younger than 35 years of age, and the solid gold line represents women diagnosed with Hodgkin lymphoma at 35 years of age or older. The dashed blue line represents women in the population younger than 35 years of age, and the dashed gold line represents women in the population 35 years of age or older.

gkin lymphoma at younger than 35 years of age, and the solid gold line represents women diagnosed with Hodgkin lymphoma at 35 years of age or older. The dashed blue line represents women in the population younger than 35 years of age, and the dashed gold line represents women in the population 35 years of age or older. Temporal Effects on Risk of Second Cancer Although details of individual patient therapy are not registered by the Swedish Cancer Registry, the treatment principles for HL in Sweden are broadly similar to those of other Western countries.29 Briefly, extended field irradiation, mainly mantle field, was the standard treatment of patients with HL during the early phase of our analysis. Patients received radiotherapy, chemotherapy, or a combined modality treatment. For those patients treated after 1990, less toxic regimens were being introduced.29,30 Partitioning data, we analyzed the second cancer risk for patients diagnosed with HL in the time periods of 1965 to 1977, 1978 to 1988, and 1989 to 2000. We found little evidence of a change in SIRs for overall cancer and for the most common sites of cancer, including hematopoietic malignancy (Data Supplement).

d.29,30 Partitioning data, we analyzed the second cancer risk for patients diagnosed with HL in the time periods of 1965 to 1977, 1978 to 1988, and 1989 to 2000. We found little evidence of a change in SIRs for overall cancer and for the most common sites of cancer, including hematopoietic malignancy (Data Supplement). Impact of Family History on Risk of Second Cancer and Survival From the Multigenerational Register, a total of 28,277 FDRs of the 9,522 patients with HL were identified. The SIR for cancer risk in FDRs was 1.02 (95% CI, 0.99 to 1.06). In the HL survivors, 2,785 individuals (29%) had one or more FDRs with a family history of cancer (Data Supplement). We found an increase in second cancer risk in HL survivors who had an FDR with cancer, when compared with HL survivors with no FDR with cancer (P < .001) with SIRs of 2.83 (95% CI, 2.58 to 3.10) and 2.16 (95% CI, 2.00 to 2.33), respectively. Moreover, the increased risk of second cancer was correlated with the number of FDRs affected with cancer, respective SIRs being 2.67 (95% CI, 2.40 to 2.97) and 3.40 (95% CI, 2.85 to 4.09) for patients with one and two or more affected FDRs (P < .001; Data Supplement).

10) and 2.16 (95% CI, 2.00 to 2.33), respectively. Moreover, the increased risk of second cancer was correlated with the number of FDRs affected with cancer, respective SIRs being 2.67 (95% CI, 2.40 to 2.97) and 3.40 (95% CI, 2.85 to 4.09) for patients with one and two or more affected FDRs (P < .001; Data Supplement). In an analysis of colorectal, breast, and lung cancer, we observed elevated risks of second cancers in HL survivors with an FDR with the corresponding site-specific cancer. For lung cancer, SIRs were 11.24 (FDR with lung cancer; 95% CI, 6.38 to 19.79) and 3.39 (no FDR with lung cancer; 95% CI, 2.85 to 4.03; P < .001). For colorectal cancer, SIRs were 3.71 (FDR with colorectal cancer; 95% CI, 2.05 to 6.70) and 1.76 (no FDR with colorectal cancer; 95% CI, 1.45 to 2.14; P = .03). Finally, for breast cancer, SIRs were 4.36 (FDR with breast cancer; 95% CI, 2.60 to 6.55) and 2.36 (no FDR with breast cancer; 95% CI, 1.98 to 2.81; P = .04; Table 4). Moreover, for lung cancer, a more than additive interaction between family history and HL treatment was shown (P = .03; Table 5). Overall, having a family history of cancer was shown not to influence survival from a second cancer in patients with HL (Data Supplement). Table 4. Risk of All Second Cancers and Site-Specific Cancers in Survivors of Hodgkin Lymphoma, by Family History Table 5. Interaction Between Risk of Second Cancer in Survivors of Hodgkin Lymphoma and Family History of Concordant Cancer in the Swedish Population

In an analysis of colorectal, breast, and lung cancer, we observed elevated risks of second cancers in HL survivors with an FDR with the corresponding site-specific cancer. For lung cancer, SIRs were 11.24 (FDR with lung cancer; 95% CI, 6.38 to 19.79) and 3.39 (no FDR with lung cancer; 95% CI, 2.85 to 4.03; P < .001). For colorectal cancer, SIRs were 3.71 (FDR with colorectal cancer; 95% CI, 2.05 to 6.70) and 1.76 (no FDR with colorectal cancer; 95% CI, 1.45 to 2.14; P = .03). Finally, for breast cancer, SIRs were 4.36 (FDR with breast cancer; 95% CI, 2.60 to 6.55) and 2.36 (no FDR with breast cancer; 95% CI, 1.98 to 2.81; P = .04; Table 4). Moreover, for lung cancer, a more than additive interaction between family history and HL treatment was shown (P = .03; Table 5). Overall, having a family history of cancer was shown not to influence survival from a second cancer in patients with HL (Data Supplement). Table 4. Risk of All Second Cancers and Site-Specific Cancers in Survivors of Hodgkin Lymphoma, by Family History Table 5. Interaction Between Risk of Second Cancer in Survivors of Hodgkin Lymphoma and Family History of Concordant Cancer in the Swedish Population DISCUSSION This analysis provides further evidence that survivorship from HL is associated with a significant risk of a second cancer. Furthermore, we confirm the previous findings of a relationship between age at diagnosis and sex, and the risk of second cancers. Our analysis also shows that differences in patient management over successive decades have not led to a lessening of risk of second cancers in HL survivors, an observation consistent with recent data from the Netherlands.12 Possible reasons for this observation include the impact of screening or that the risk is maintained because of an interaction between less toxic chemotherapy and second cancer risk. For example, higher doses of alkylating agents, which are more likely to cause premature menopause, have been reported to reduce radiation-induced breast cancer risk.31 In addition, although patients diagnosed with HL are likely to have received lower doses of alkylating agents in recent years, the omission of radiotherapy from such treatment regimens is likely to have led to an increased proportion of all patients receiving alkylating agents.12 Finally, we cannot exclude entirely the possibility that the study periods we analyzed were too short or early to be able to demonstrate a difference.

n recent years, the omission of radiotherapy from such treatment regimens is likely to have led to an increased proportion of all patients receiving alkylating agents.12 Finally, we cannot exclude entirely the possibility that the study periods we analyzed were too short or early to be able to demonstrate a difference. In this study, HL survivors with a family history of colorectal, lung, or breast cancer showed an increased risk of concordant second cancers when compared with HL survivors without a family history. To our knowledge, this is the first population study to demonstrate site-specific second cancer risk after HL being influenced by family history. Our findings support the notion of familial determinants of second cancer risk being consistent with inherited genetic predisposition. Direct evidence for such a model is provided by the example of retinoblastoma, in which individuals with hereditary retinoblastoma have a much higher risk of radiotherapy-induced second malignancy compared with those with sporadic disease.32,33 Thus far, no similar high-impact mutation has been identified for HL, and sequencing of TP53, BRCA1, BRCA2, and ATM in HL survivors with a second malignancy has not substantiated a role for mutations in these genes as a cause of subsequent cancer risk.16,34 Polygenic susceptibility provides an alternative explanation for genetic susceptibility, whereby the elevated risk is enshrined in common genetic variants, which, in isolation, exert small effects but act in concert to have a relatively profound impact. Such an assertion is supported by the recent findings that variation in FGFR2 and PRDM1 influences the risk of second cancer in HL survivors.35,36 For lung cancer, we were able to demonstrate a greater than additive interaction between family history of lung cancer and HL treatment, which may be the consequence of additional shared nongenetic risk factors, most probably a propensity to smoke. This interaction is of significant importance because it may explain, in part, why no significant change in second lung cancer has been observed despite modifications in treatment regimens. The number of cases of breast and colorectal cancer did not permit the rejection of any interaction model.

a propensity to smoke. This interaction is of significant importance because it may explain, in part, why no significant change in second lung cancer has been observed despite modifications in treatment regimens. The number of cases of breast and colorectal cancer did not permit the rejection of any interaction model. Survival of patients with HL after diagnosis of a second cancer was not significantly worse in those with an affected FDR. Moreover, noninferiority of survival in individuals with a family history of cancer has been described in colorectal, breast, and prostate cancer.37-39 Although it remains to be established, reasons for this may include increased cancer surveillance, resulting in presentation at earlier stages of disease,40,41 improved health-related behavior,42 or differences in tumor biology and response to therapy.

has been described in colorectal, breast, and prostate cancer.37-39 Although it remains to be established, reasons for this may include increased cancer surveillance, resulting in presentation at earlier stages of disease,40,41 improved health-related behavior,42 or differences in tumor biology and response to therapy. Our study has major strengths. First, we have avoided ascertainment bias in patient selection because our cohort analysis was based on the Swedish population, for which there is near complete case registration with long-term follow-up.20 Second, the Swedish Family-Cancer Project Database includes a large number of individuals linked to family members, and this has allowed us to uniquely study the influence of family history on cancer risk.19 We do acknowledge, however, that a limitation of our study is the reliance on year of treatment as an effective surrogate for type of treatment and that we did not have the opportunity to incorporate information on risk factors such as smoking. However, most of the familial risk of lung cancer is a result of a predisposition to smoke. Hence, by performing an interactive analysis, we addressed this area.

treatment as an effective surrogate for type of treatment and that we did not have the opportunity to incorporate information on risk factors such as smoking. However, most of the familial risk of lung cancer is a result of a predisposition to smoke. Hence, by performing an interactive analysis, we addressed this area. Despite these caveats, our findings further substantiate the significant cancer risks associated with survivorship from HL and that these are modified by a family history of cancer. In addition, our findings are of importance in a primary health care setting where many surviving patients with HL are treated (for other symptoms and diseases) after the first 5 years of follow-up. Long-term understanding of the biologic basis of these associations offers the prospect of personalizing therapy in HL. However, such information has current value clinically, when planning risk-adapted therapy for patients with HL. Furthermore, our findings with respect to lung cancer emphasize the importance of instigating programs to reduce smoking in patients with HL. Finally, as well as offering breast cancer screening to women who have received supradiaphragmatic radiotherapy,43,44 obtaining family history information has a place in informing the long-term follow-up screening of all patients with HL. Supported by the German Cancer Aid, the Swedish Research Council, FORTE, and an ALF project grant, Region Skåne/Lund University, Sweden, and by a clinical fellowship from Cancer Research UK (A.S.). AUTHOR CONTRIBUTIONS Conception and design: Amit Sud, Richard S. Houlston, Kari Hemminki

Despite these caveats, our findings further substantiate the significant cancer risks associated with survivorship from HL and that these are modified by a family history of cancer. In addition, our findings are of importance in a primary health care setting where many surviving patients with HL are treated (for other symptoms and diseases) after the first 5 years of follow-up. Long-term understanding of the biologic basis of these associations offers the prospect of personalizing therapy in HL. However, such information has current value clinically, when planning risk-adapted therapy for patients with HL. Furthermore, our findings with respect to lung cancer emphasize the importance of instigating programs to reduce smoking in patients with HL. Finally, as well as offering breast cancer screening to women who have received supradiaphragmatic radiotherapy,43,44 obtaining family history information has a place in informing the long-term follow-up screening of all patients with HL. Supported by the German Cancer Aid, the Swedish Research Council, FORTE, and an ALF project grant, Region Skåne/Lund University, Sweden, and by a clinical fellowship from Cancer Research UK (A.S.). AUTHOR CONTRIBUTIONS Conception and design: Amit Sud, Richard S. Houlston, Kari Hemminki Collection and assembly of data: Amit Sud, Hauke Thomsen, Kristina Sundquist, Kari Hemminki Data analysis and interpretation: Amit Sud, Hauke Thomsen Manuscript writing: All authors Final approval of manuscript: All authors Accountable for all aspects of the work: All authors

AUTHOR CONTRIBUTIONS Conception and design: Amit Sud, Richard S. Houlston, Kari Hemminki Collection and assembly of data: Amit Sud, Hauke Thomsen, Kristina Sundquist, Kari Hemminki Data analysis and interpretation: Amit Sud, Hauke Thomsen Manuscript writing: All authors Final approval of manuscript: All authors Accountable for all aspects of the work: All authors AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST Risk of Second Cancer in Hodgkin Lymphoma Survivors and Influence of Family History The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/jco/site/ifc. Amit Sud No relationship to disclose Hauke Thomsen No relationship to disclose Kristina Sundquist No relationship to disclose Richard S. Houlton No relationship to disclose Kari Hemminki No relationship to disclose

We thank Tomao et al1 for their comments on our recent report on patient-reported outcomes (PROs) in the Suppression of Ovarian Function Suppression Trial (SOFT).2 When SOFT was designed more than a decade ago to investigate the role of ovarian function suppression (OFS) and the role of exemestane in premenopausal women with endocrine-responsive early breast cancer either after completion of (neo)adjuvant chemotherapy or after surgery alone,3 quality of life (QoL) was integrated to assess the patients’ perspective. Our results provide complementary information to the adverse event reporting in SOFT2 and add novel information to existing evidence on the effect of OFS on PROs and QoL. In the Zoladex in Premenopausal Patients4,5 trial, chemotherapy was given concurrently with adjuvant endocrine therapy, and the observation period was restricted to the first 2 years of treatment. Eastern Cooperative Oncology Group E-31936 and SOFT are the only two randomized phase III trials presenting long-term PROs in this setting. Our article complements the findings of E-3193, but with major differences. The SOFT population is international, and the sample size is larger, with 1,722 patients available for the primary QoL analysis. The E-3193 trial presented data from a summary score of patient-reported symptoms. We found that OFS added to tamoxifen results in variable magnitudes of treatment differences for individual symptoms. The E-3193 trial only included patients without chemotherapy, whereas SOFT enrolled two distinct cohorts of patients: those with and those without prior chemotherapy. The prior chemotherapy cohort had higher-risk disease than the E-3193 trial population, which is the current target population for OFS. Cognitive function was assessed as part of a substudy.7 The gold standard of cognitive assessment, a comprehensive neuropsychological testing,8 was not feasible in the entire study population, with more than 500 centers and many different languages.

Two ASTRO consensus guidelines have addressed technical issues in the setting of BCT. While largely focusing on invasive breast carcinoma, the ASTRO statement on HWBRT concluded there was insufficient evidence to recommend for or against HWBRT in the setting of DCIS.31 In the ASTRO statement on APBI, DCIS was placed into the “cautionary” group based on the lack of evidence from randomized trials, while noting that DCIS patients have been included in some retrospective cohort studies.32 Therefore, there is no evidence that margin width, in isolation, should determine radiation delivery technique, fractionation of WBRT, or use/dose of a boost. The MP considered the evidence base insufficient to address optimal margin width in APBI. DCIS in the Presence of Invasive Breast Cancer DCIS with microinvasion, defined as no invasive focus > 1 mm in size, should be considered as DCIS when considering the optimal margin width.

3193 trial population, which is the current target population for OFS. Cognitive function was assessed as part of a substudy.7 The gold standard of cognitive assessment, a comprehensive neuropsychological testing,8 was not feasible in the entire study population, with more than 500 centers and many different languages. Tomao et al1 suggest that we did not present several clinically relevant data. With regard to treatment, in the cohort of patients with prior chemotherapy, the majority received either anthracycline-based (38%), or anthracycline plus taxane–based (53%) chemotherapy. The impact of these two different types of chemotherapy on patients’ symptom experience is relevant primarily in the short term. The number of patients who were treated with human epidermal growth factor receptor 2–directed therapy (ie, trastuzumab, 8% overall) were reported in Table A1 (Data Supplement) of Ribi et al.2 This low percentage would not have changed our results in a meaningful way. Oral endocrine therapy before randomization was allowed while premenopausal status was established or re-established. Details on duration were included in our report (Table 1).2 In the overall population, three patients received an aromatase inhibitor before randomization.3 The number of patients with irregular menstruation or persistent amenorrhea at baseline (ie, after chemotherapy or previous endocrine therapy) were reported in Table A1 in the data supplement.2 We controlled for menstruation status in the mixed-effect models. During the 5 years of assigned endocrine treatment, it was not possible to accurately determine menopausal status, as we did not routinely measure hormone levels during treatment and amenorrhea during tamoxifen is not an accurate determinant of menopause.

e controlled for menstruation status in the mixed-effect models. During the 5 years of assigned endocrine treatment, it was not possible to accurately determine menopausal status, as we did not routinely measure hormone levels during treatment and amenorrhea during tamoxifen is not an accurate determinant of menopause. A further criticism relates to the inclusion of patients who both did and did not receive chemotherapy before enrollment. The authors are correct that patients in the chemotherapy cohort had higher-risk disease characteristics and were younger (Table 1 in our report and Table A1 in the Data Supplement).2 However, the occurrence of symptoms (eg, vasomotor, gynecologic, and sexual symptoms) was not lower in this cohort. On the contrary, patients with prior chemotherapy reported worse baseline scores for these symptoms compared with the no-chemotherapy cohort, possibly caused by chemotherapy and by the receipt of tamoxifen before enrollment in half of these patients. Thus, changes over time in these symptoms were smaller (ie, less worsening) in the chemotherapy compared with the no-chemotherapy cohort, leading to the interpretation that chemotherapy did not exacerbate adverse effects. Presenting our results not only for the overall population but also separately for the two chemotherapy cohorts in our report (Appendix Figs A2A and A2B)2 is a strength of our study.

hemotherapy compared with the no-chemotherapy cohort, leading to the interpretation that chemotherapy did not exacerbate adverse effects. Presenting our results not only for the overall population but also separately for the two chemotherapy cohorts in our report (Appendix Figs A2A and A2B)2 is a strength of our study. The efficacy results from SOFT,3 in conjunction with those from the SOFT plus TEXT combined analysis,9 are practice changing.10 PROs comparing exemestane versus tamoxifen in patients who received OFS were published earlier this year.11 The PROs of the comparison of tamoxifen plus OFS versus tamoxifen alone for the cohorts with and without chemotherapy provide physicians and patients with a comprehensive picture of the risks and benefits when choosing the best adjuvant treatment for these relatively young women.

OFS were published earlier this year.11 The PROs of the comparison of tamoxifen plus OFS versus tamoxifen alone for the cohorts with and without chemotherapy provide physicians and patients with a comprehensive picture of the risks and benefits when choosing the best adjuvant treatment for these relatively young women. ACKNOWLEDGMENT The Suppression of Ovarian Function Suppression Trial receives financial support for trial conduct from Pfizer, the International Breast Cancer Study Group, and the US National Cancer Institute. Pfizer and Ipsen provide drug supply. Support for the coordinating group International Breast Cancer Study Group is provided by Frontier Science and Technology Research Foundation, Swiss Group for Clinical Cancer Research, US National Cancer Institute Grant No. CA75362 (M.M.R.), Cancer Research Switzerland/Oncosuisse, and the Foundation for Clinical Cancer Research of Eastern Switzerland. Grant support for cooperative groups: Australia and New Zealand Breast Cancer Trials Group: National Health and Medical Research Council Grants No. 351161 and 510788; Southwest Oncology Group: US National Institutes of Health (NIH) Grant No. CA32102; Alliance: US NIH Grant No. U10-CA180821; Eastern Cooperative Oncology Group–American College of Radiology Imaging Network: US NIH Grants No. CA21115 and CA16116; National Surgical Adjuvant Breast and Bowel Project/NRG: US NIH Grants No. U10-CA-12027, U10-CA-69651, U10-CA-37377, and U10-CA-69974; National Cancer Institute of Canada: NIH Grant No. CA077202 and Canadian Cancer Society Research Institute Grants No. 015469 and 021039.

Network: US NIH Grants No. CA21115 and CA16116; National Surgical Adjuvant Breast and Bowel Project/NRG: US NIH Grants No. U10-CA-12027, U10-CA-69651, U10-CA-37377, and U10-CA-69974; National Cancer Institute of Canada: NIH Grant No. CA077202 and Canadian Cancer Society Research Institute Grants No. 015469 and 021039. AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST Reply to F. Tomao et al The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or jco.ascopubs.org/site/ifc. Karin Ribi No relationship to disclose Jürg Bernhard No relationship to disclose Weixiu Luo No relationship to disclose Meredith M. Regan Consulting or Advisory Role: Merck, Ipsen (Inst) Research Funding: Veridex (Inst), OncoGeneX (Inst), Pfizer (Inst), Ipsen (Inst), Novartis (Inst), Merck (Inst), Ferring Pharmaceuticals (Inst), Celgene (Inst), AstraZeneca (Inst) Gini F. Fleming Research Funding: Corcept Therapeutics (Inst) Prudence A. Francis Travel, Accommodations, Expenses: Roche, Pfizer

INTRODUCTION Breast-conserving therapy (BCT), defined as surgical excision of the primary tumor with a margin of surrounding normal tissue followed by whole-breast radiation therapy (WBRT), results in long-term cause-specific survival rates of greater than 95% for women with ductal carcinoma in situ (DCIS) as demonstrated in both randomized trials1 and observational studies.2,3 Although the addition of WBRT to surgical excision does not improve survival, it substantially reduces rates of ipsilateral breast tumor recurrence (IBTR), even among patients with small, non-high–grade DCIS.1,4 In the four early randomized trials of WBRT for DCIS, microscopically clear margins defined as no ink on tumor were required in three studies,5-7 but not in the fourth.8 These studies provide no information on whether more widely clear margins than no ink on tumor reduce rates of IBTR in patients having BCT.

e DCIS.1,4 In the four early randomized trials of WBRT for DCIS, microscopically clear margins defined as no ink on tumor were required in three studies,5-7 but not in the fourth.8 These studies provide no information on whether more widely clear margins than no ink on tumor reduce rates of IBTR in patients having BCT. Retrospective single-institution studies have suggested that a negative margin width of 1 cm or more may eliminate the reduction in IBTR seen with WBRT,9 leading some to conclude that larger margins are also beneficial in patients receiving WBRT. Despite the widespread use of BCT for DCIS, there is still no consensus on what constitutes an optimal negative margin width.10 As a consequence, approximately one in three women attempting BCT for DCIS undergo a re-excision.11 Re-excisions have the potential for added discomfort, surgical complications, compromise in cosmetic outcome, additional stress for patients and families, and increased health care costs, and have been associated with conversion to bilateral mastectomy.12

in three women attempting BCT for DCIS undergo a re-excision.11 Re-excisions have the potential for added discomfort, surgical complications, compromise in cosmetic outcome, additional stress for patients and families, and increased health care costs, and have been associated with conversion to bilateral mastectomy.12 Since BCT was established, the landscape of DCIS management has evolved with advances in imaging and pathologic evaluation, and the availability of adjuvant endocrine therapy, resulting in a decline in IBTR rates.13 In view of these changes and the lack of consensus on what represents adequate negative margins in DCIS, the Society of Surgical Oncology (SSO), American Society for Radiation Oncology (ASTRO), and the American Society of Clinical Oncology (ASCO) convened a multidisciplinary margins panel (MP) to evaluate IBTR in relation to margin width. The primary question addressed was “what margin width minimizes the risk of IBTR in patients with DCIS receiving breast-conserving surgery?” The guideline developed from this consensus panel is intended to assist treating physicians and patients in the clinical decision-making process based on the best available evidence. The key findings of the guideline are summarized in Table 1. Table 1. Summary of Clinical Practice Guideline Recommendations

Since BCT was established, the landscape of DCIS management has evolved with advances in imaging and pathologic evaluation, and the availability of adjuvant endocrine therapy, resulting in a decline in IBTR rates.13 In view of these changes and the lack of consensus on what represents adequate negative margins in DCIS, the Society of Surgical Oncology (SSO), American Society for Radiation Oncology (ASTRO), and the American Society of Clinical Oncology (ASCO) convened a multidisciplinary margins panel (MP) to evaluate IBTR in relation to margin width. The primary question addressed was “what margin width minimizes the risk of IBTR in patients with DCIS receiving breast-conserving surgery?” The guideline developed from this consensus panel is intended to assist treating physicians and patients in the clinical decision-making process based on the best available evidence. The key findings of the guideline are summarized in Table 1. Table 1. Summary of Clinical Practice Guideline Recommendations METHODS The guideline development process was funded by a Susan G. Komen grant. Committee members were chosen by their respective organizations based upon interest and expertise in DCIS management (Table 2). Processes recommended in the Institute of Medicine report “Clinical Practice Guidelines We Can Trust”14 which were followed as part of the guideline development process included: (1) the development of a systematic review/study-level meta-analysis based on questions to be addressed by the MP to serve as the primary evidence base, with additional topic-specific literature reviews conducted by participants for questions not addressed in the meta-analysis; (2) the provision for each recommendation of a rating of the strength of the evidence and the strength of the recommendation; (3) the ascertainment of the level of agreement of panel members with each recommendation by vote, and the revision of recommendations to achieve greater than 90% consensus; and (4) the declaration by MP candidates of potential conflicts of interest before convening, and the obtaining of written disclosures at the consensus meeting. (The co-chairs deemed no MP members had conflicts that could influence the process/development of specific recommendations.)

chieve greater than 90% consensus; and (4) the declaration by MP candidates of potential conflicts of interest before convening, and the obtaining of written disclosures at the consensus meeting. (The co-chairs deemed no MP members had conflicts that could influence the process/development of specific recommendations.) Table 2. Expert Panel Members The MP convened in November 2015; the resulting manuscript was approved by all panel members and externally reviewed, and feedback was incorporated. The final manuscript was approved by the SSO Executive Council, the ASTRO Board of Directors, and the ASCO Board of Directors, and endorsed by the Board of Directors of the American Society of Breast Surgeons. Patient-related materials will be available on the Susan G. Komen website (komen.org).

dback was incorporated. The final manuscript was approved by the SSO Executive Council, the ASTRO Board of Directors, and the ASCO Board of Directors, and endorsed by the Board of Directors of the American Society of Breast Surgeons. Patient-related materials will be available on the Susan G. Komen website (komen.org). Meta-Analysis The methodology for the systematic review/meta-analysis has been published elsewhere.15 Briefly, using Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) and Institute of Medicine guidelines, EMBASE, MEDLINE, PREMEDLINE, and evidence-based medicine databases were searched in October 2014 for eligible studies. A summary providing details of the methodology and statistical approaches is provided in the Appendix. Analysis was performed using two different statistical approaches. In the frequentist approach, multiple margin cut points within studies, if reported, were condensed into a single cut point, while the Bayesian approach allowed for the use of multiple cut points.16 All reported odds ratio (ORs) were adjusted for study-specific median follow up time (to account for the inherent increased risk of IBTR with longer follow up) and are reported relative to positive (or positive/close) margins, or to a minimal negative margin (no ink on tumor or margin > 1 mm).15 Inclusion/Exclusion Criteria Studies that included a minimum of 50 patients with DCIS treated with local excision and reported IBTR in relation to microscopic margin widths with a minimum median follow up of 4 years were eligible.15

Meta-Analysis The methodology for the systematic review/meta-analysis has been published elsewhere.15 Briefly, using Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) and Institute of Medicine guidelines, EMBASE, MEDLINE, PREMEDLINE, and evidence-based medicine databases were searched in October 2014 for eligible studies. A summary providing details of the methodology and statistical approaches is provided in the Appendix. Analysis was performed using two different statistical approaches. In the frequentist approach, multiple margin cut points within studies, if reported, were condensed into a single cut point, while the Bayesian approach allowed for the use of multiple cut points.16 All reported odds ratio (ORs) were adjusted for study-specific median follow up time (to account for the inherent increased risk of IBTR with longer follow up) and are reported relative to positive (or positive/close) margins, or to a minimal negative margin (no ink on tumor or margin > 1 mm).15 Inclusion/Exclusion Criteria Studies that included a minimum of 50 patients with DCIS treated with local excision and reported IBTR in relation to microscopic margin widths with a minimum median follow up of 4 years were eligible.15 Study Quality/Literature Limitations All publications in the meta-analysis (except for two) were retrospective and provided observational data at the study level. The characteristics of these studies have been reported elsewhere.15

Inclusion/Exclusion Criteria Studies that included a minimum of 50 patients with DCIS treated with local excision and reported IBTR in relation to microscopic margin widths with a minimum median follow up of 4 years were eligible.15 Study Quality/Literature Limitations All publications in the meta-analysis (except for two) were retrospective and provided observational data at the study level. The characteristics of these studies have been reported elsewhere.15 RESULTS The meta-analysis included 20 studies, 7883 DCIS patients with known margin status, and 865 IBTRs.15 The median proportion of patients receiving WBRT was 100% (interquartile range [IQR] 53.3%-100.0%), and the median proportion receiving endocrine therapy was 20.8% (IQR 0.0%-31.4%). The median follow up was 78.3 months, and the median incidence of IBTR was 8.3% (IQR 5.0%-11.9%). Due to heterogeneity in classification and reporting of margins data, both a frequentist analysis and a Bayesian network meta-analysis were conducted with sensitivity analyses. Characteristics of patients included in the studies are summarized in Table 3.15 Table 3. Summary of Study Characteristics Included in Meta-Analysis15 GUIDELINE RECOMMENDATIONS Positive Margins A positive margin, defined as ink on DCIS, is associated with a significant increase in IBTR. This increased risk is not nullified by the use of WBRT.

RESULTS The meta-analysis included 20 studies, 7883 DCIS patients with known margin status, and 865 IBTRs.15 The median proportion of patients receiving WBRT was 100% (interquartile range [IQR] 53.3%-100.0%), and the median proportion receiving endocrine therapy was 20.8% (IQR 0.0%-31.4%). The median follow up was 78.3 months, and the median incidence of IBTR was 8.3% (IQR 5.0%-11.9%). Due to heterogeneity in classification and reporting of margins data, both a frequentist analysis and a Bayesian network meta-analysis were conducted with sensitivity analyses. Characteristics of patients included in the studies are summarized in Table 3.15 Table 3. Summary of Study Characteristics Included in Meta-Analysis15 GUIDELINE RECOMMENDATIONS Positive Margins A positive margin, defined as ink on DCIS, is associated with a significant increase in IBTR. This increased risk is not nullified by the use of WBRT. There is no debate that a positive margin, defined as the presence of ink from the specimen surface on ducts containing DCIS, implies a potentially incomplete resection and is associated with a higher rate of IBTR. In the Early Breast Cancer Trialists’ Collaborative Group (EBCTCG) meta-analysis of randomized DCIS trials,1 patients with positive margins had a twofold higher IBTR risk compared with patients with negative margins despite receiving WBRT (10-year IBTR rate 24% vs 12%), and approximately 50% were invasive recurrences. The relationship between margin status and WBRT was examined in a subset analysis of the National Surgical Adjuvant Breast and Bowel Project (NSABP) B-17 trial by central pathology review of 573 of 818 participants.17 The annual hazard rate for IBTR after lumpectomy alone was 8.1% for those with positive margins compared with 3.3% for patients with negative margins, reduced by WBRT to 2.7% and 1.2%, respectively. Positive margins were significantly associated with IBTR in a multivariate analysis of the long-term results of the European Organization for Research and Treatment of Cancer (EORTC) 10853 trial.18 In the meta-analysis of Marinovich et al using the Bayesian analytic approach, similar results were obtained.15 After adjustment for study-level follow up, patients with negative margins were significantly less likely to experience IBTR than patients with positive margins (OR 0.45, 95% credible interval [CrI] 0.30-0.62). Similar findings were observed in the frequentist analysis which necessitated combining positive and close margins (OR 0.53, 95% CI 0.45-0.62; P < .001). This result persisted after study-level adjustment for age, median recruitment year, grade of DCIS, use of WBRT, and use of endocrine therapy.

erval [CrI] 0.30-0.62). Similar findings were observed in the frequentist analysis which necessitated combining positive and close margins (OR 0.53, 95% CI 0.45-0.62; P < .001). This result persisted after study-level adjustment for age, median recruitment year, grade of DCIS, use of WBRT, and use of endocrine therapy. Negative Margin Widths Margins of at least 2 mm are associated with a reduced risk of IBTR relative to narrower negative margin widths in patients receiving WBRT. The routine practice of obtaining negative margin widths wider than 2 mm is not supported by the evidence. To address the question of optimal negative margin width, the MP considered data on the distribution of DCIS in the breast. Studies of mastectomy specimens using whole organ sectioning and radiologic-pathologic correlation have demonstrated that while most cases of DCIS are unicentric, the involvement of the segment may be multifocal, with “gaps” of uninvolved tissue between foci of DCIS.19 Given this, a “negative margin” does not guarantee the absence of residual DCIS in the breast.

organ sectioning and radiologic-pathologic correlation have demonstrated that while most cases of DCIS are unicentric, the involvement of the segment may be multifocal, with “gaps” of uninvolved tissue between foci of DCIS.19 Given this, a “negative margin” does not guarantee the absence of residual DCIS in the breast. There are also technical limitations to margin assessment which impact the relationship between margin width and IBTR. For example, margins are artifactually narrower ex-vivo when specimens become flattened from lack of surrounding supportive tissue, a phenomenon exaggerated by compression for specimen radiography. Additionally, surface ink can track into deeper portions of the specimen, posing significant challenges in determining true margin location. Finally, tumor-to-ink distance on any single slide may not be representative of the entire specimen; an “adequate” margin on one section may become positive if additional or deeper sections are examined. Two common methods for margin evaluation include sectioning perpendicular to ink (to determine tumor-to-ink width) or en-face examination of shaved margins (where any residual tumor in the shaved specimen is considered a positive margin). While an advantage of the shaved method is greater surface-area examination, a known disadvantage is the higher frequency of margins categorized as positive that are, in comparison, negative by the perpendicular method, which may in turn result in unnecessary re-excision or even mastectomy.20 Specimen sampling is also highly variable, and even total sequential embedding results in only a small proportion (< 1%) of the specimen margins being examined.21 Together, these studies highlight the substantial variability in margin assessment irrespective of the technique used.

sary re-excision or even mastectomy.20 Specimen sampling is also highly variable, and even total sequential embedding results in only a small proportion (< 1%) of the specimen margins being examined.21 Together, these studies highlight the substantial variability in margin assessment irrespective of the technique used. Despite variability in margin assessment, great emphasis has been placed on achieving specific negative margin widths. In the Marinovich frequentist meta-analysis, comparison of specific margin width thresholds (2 mm, 3 or 5 mm, and 10 mm) relative to negative margins defined as > 0 mm or 1 mm included 7883 patients with a median follow up of 6.5 years. The ORs for 2 mm (0.51 [95% CI 0.31-0.85], P = .01), 3 or 5 mm (0.42 [95% CI 0.18-0.97], P = .04), and 10 mm (0.60 [0.33-1.08], P = .09) showed comparable reductions in the odds of IBTR compared with > 0 mm or 1 mm, and pairwise comparisons found no significant differences in the odds of IBTR between the 2 mm, 3 or 5 mm, and 10 mm margin thresholds (all P > 0.40). In this model, the predicted 10-year IBTR probability for 2 mm negative margins was 10.1% (95% CI 6.3%-16.0%) compared with 8.5% for 3 or 5 mm (95% CI 3.6%-18.9%) and 11.7% (95% CI 6.7%-19.4%) for 10 mm margins. In the Bayesian network meta-analysis (Table 4),15 the ORs of incrementally wider negative margins relative to the positive margin category were 0.45 (95% CrI 0.32-0.61) for > 0 or 1 mm, 0.32 (95% CrI 0.21-0.48) for 2 mm, 0.30 (95% CrI 0.12-0.76) for 3 mm, and 0.32 (95% CrI 0.19-0.49) for 10 mm. Adjustments for clinically relevant covariates, including sensitivity analysis limited to studies using radiation therapy (RT), did not alter these mean OR estimates (Table 4). In this analysis, the relative odds ratio (ROR) of IBTR between the 10 mm and 2 mm threshold groups compared with positive margins was 0.99 (95% CrI 0.61-1.64), indicating no statistically meaningful difference.

y analysis limited to studies using radiation therapy (RT), did not alter these mean OR estimates (Table 4). In this analysis, the relative odds ratio (ROR) of IBTR between the 10 mm and 2 mm threshold groups compared with positive margins was 0.99 (95% CrI 0.61-1.64), indicating no statistically meaningful difference. Table 4. Margin Threshold and Ipsilateral Breast Tumor Recurrence: Bayesian Network Meta-Analysis15

y analysis limited to studies using radiation therapy (RT), did not alter these mean OR estimates (Table 4). In this analysis, the relative odds ratio (ROR) of IBTR between the 10 mm and 2 mm threshold groups compared with positive margins was 0.99 (95% CrI 0.61-1.64), indicating no statistically meaningful difference. Table 4. Margin Threshold and Ipsilateral Breast Tumor Recurrence: Bayesian Network Meta-Analysis15 The choice of the 2 mm threshold rather than > 0 (no ink on tumor) or 1 mm was based upon evidence of a statistically significant decrease in IBTR for 2 mm compared with 0 or 1 mm in the frequentist analysis (OR 0.51, 95% CI 0.31-0.85; P = .01) coupled with weak evidence in the Bayesian model of a reduction in IBTR with the 2 mm distance compared with smaller distances (ROR 0.72, 95% CrI 0.47-1.08). However, while the MP felt that there was evidence that the 2 mm margin optimized local control, clinical judgment must be used in determining whether patients with smaller negative margin widths (> 0 or 1 mm) require re-excision. Factors felt to be important to consider include assessment of IBTR risk (residual calcifications on postexcision mammography, extent of DCIS in proximity to margin, which margin is close [i.e., anterior excised to skin or posterior excised to pectoral fascia v margins associated with residual breast tissue]), cosmetic impact of re-excision, and overall life expectancy. The conclusion that re-excision could be selectively used with margins smaller than 2 mm was influenced by the high long-term rates of local control reported in the NSABP DCIS trials which required a margin of no ink on tumor7 as well as the study of Van Zee et al which, after adjusting for multiple covariates, found no difference in risk between margins of ≤ 2 mm and more widely clear margins in patients receiving WBRT.22

the high long-term rates of local control reported in the NSABP DCIS trials which required a margin of no ink on tumor7 as well as the study of Van Zee et al which, after adjusting for multiple covariates, found no difference in risk between margins of ≤ 2 mm and more widely clear margins in patients receiving WBRT.22 Treatment With Excision Alone Treatment with excision alone, regardless of margin width, is associated with substantially higher rates of IBTR than treatment with excision and WBRT, even in predefined low-risk patients. The optimal margin width for treatment with excision alone is unknown, but should be at least 2 mm. Some evidence suggests lower rates of IBTR with margin widths wider than 2 mm.

gin width, is associated with substantially higher rates of IBTR than treatment with excision and WBRT, even in predefined low-risk patients. The optimal margin width for treatment with excision alone is unknown, but should be at least 2 mm. Some evidence suggests lower rates of IBTR with margin widths wider than 2 mm. The EBCTCG DCIS meta-analysis showed that the 10-year IBTR rate for patients treated with excision alone was higher than with excision and WBRT, both for those with negative margins (26.0% vs 12.0%, P < .00001) and positive margins (48.3% vs 24.2%; P = .00004).1 The same proportional benefit for WBRT was seen in women treated with local excision and those having large sector resections. In the Radiation Therapy Oncology Group (RTOG) 9804 trial where patients with small, mammographically detected low-to-intermediate grade DCIS and margins ≥ 3 mm were randomized to excision alone or excision plus WBRT, 7-year IBTR rates were 6.7% and 0.9% (P = .0003), respectively.4 The prospective, multi-institutional Eastern Cooperative Oncology Group (ECOG) E5194 study of patients with low-risk DCIS treated with excision alone (negative margin width ≥ 3 mm) reported 12-year rates of IBTR of 14.4% for nonhigh grade DCIS ≤ 2.5 cm in size and 24.6% for high-grade DCIS ≤ 1 cm in size. However, IBTR rates did not differ significantly for margins < 5 mm, 5-9 mm, or ≥ 10 mm (P = .85).23 A prospective single-arm study of patients with mammographically detected DCIS ≤ 2.5 cm in size reported a 10-year IBTR rate of 15.6%24 despite requiring margins of ≥ 1 cm.4 In contrast, Van Zee et al found in 1266 patients treated with excision alone that 10-year IBTR rates were 16% for margins > 10 mm, and increased to 23% for margins between 2.1 and 10 mm, 27% for > 0-2 mm, and 41% for positive margins. After adjustment for multiple factors, margin width was a more highly significant predictor of IBTR (P < .0001).22 The MP felt that, overall, the heterogeneity of the evidence between the above-reported studies did not allow for a definitive recommendation for margin widths greater than 2 mm in patients foregoing RT.

. After adjustment for multiple factors, margin width was a more highly significant predictor of IBTR (P < .0001).22 The MP felt that, overall, the heterogeneity of the evidence between the above-reported studies did not allow for a definitive recommendation for margin widths greater than 2 mm in patients foregoing RT. Endocrine Therapy Rates of IBTR are reduced with endocrine therapy, but there is no evidence of an association between endocrine therapy and negative margin width. Tamoxifen reduces the incidence of both IBTR and contralateral breast cancer, but the absolute benefit is relatively small.7,25 In the NSABP B-24 trial, patients treated with lumpectomy and WBRT were randomized to tamoxifen or placebo; 25% of the population had positive or unknown margins. The 15-year IBTR rate for the placebo group was 17.4% in those with positive margins compared with 7.4% for clear margins. Adjuvant tamoxifen lowered IBTR rates among those with positive margins to levels similar to those in the negative margin cohort (17.4% placebo, 11.5% tamoxifen); conversely, there was little impact of tamoxifen in the negative margin cohort (IBTR 7.4% placebo, 7.5% tamoxifen).7 Hence, the MP felt that while tamoxifen decreases IBTR in patients with positive margins, there was no evidence to suggest an association between negative margin width and the benefit of endocrine therapy.

nversely, there was little impact of tamoxifen in the negative margin cohort (IBTR 7.4% placebo, 7.5% tamoxifen).7 Hence, the MP felt that while tamoxifen decreases IBTR in patients with positive margins, there was no evidence to suggest an association between negative margin width and the benefit of endocrine therapy. Patient and Tumor Features Multiple factors have been shown to be associated with the risk of IBTR in patients treated with and without WBRT, but there are no data addressing whether margin widths should be influenced by these factors. Young patient age has consistently been associated with IBTR, and tumor factors such as histologic pattern, comedo necrosis, and nuclear grade and size of DCIS also modify the risk of IBTR.17,26,27 More recently, unfavorable gene profile scores have also been associated with IBTR.28,29 However, there are no data addressing whether margin widths should be influenced by these factors, and this represents an appropriate area for further study. Radiation Delivery Choice of WBRT delivery technique, fractionation, and boost dose should not be dependent upon negative margin width. There is insufficient evidence to address optimal margin widths for accelerated partial breast irradiation (APBI).

Young patient age has consistently been associated with IBTR, and tumor factors such as histologic pattern, comedo necrosis, and nuclear grade and size of DCIS also modify the risk of IBTR.17,26,27 More recently, unfavorable gene profile scores have also been associated with IBTR.28,29 However, there are no data addressing whether margin widths should be influenced by these factors, and this represents an appropriate area for further study. Radiation Delivery Choice of WBRT delivery technique, fractionation, and boost dose should not be dependent upon negative margin width. There is insufficient evidence to address optimal margin widths for accelerated partial breast irradiation (APBI). The vast majority of patients treated in the five prospective randomized DCIS trials of excision with or without WBRT received conventionally fractionated WBRT without a boost. Only one of the trials allowed the option of hypofractionated whole-breast RT (HWBRT) in addition to standard WBRT,4 and ≤ 10% of the patients in three of the trials received a boost.6-8 None of the randomized trials varied RT technique according to margin status, and intensity-modulated RT (IMRT) and accelerated partial breast irradiation (APBI) were not used. There is no direct evidence from randomized trials to support the use of a boost to the primary tumor site for patients with DCIS, although in patients with invasive breast carcinoma, the long-term value of a boost in reducing IBTR has been demonstrated.30

The vast majority of patients treated in the five prospective randomized DCIS trials of excision with or without WBRT received conventionally fractionated WBRT without a boost. Only one of the trials allowed the option of hypofractionated whole-breast RT (HWBRT) in addition to standard WBRT,4 and ≤ 10% of the patients in three of the trials received a boost.6-8 None of the randomized trials varied RT technique according to margin status, and intensity-modulated RT (IMRT) and accelerated partial breast irradiation (APBI) were not used. There is no direct evidence from randomized trials to support the use of a boost to the primary tumor site for patients with DCIS, although in patients with invasive breast carcinoma, the long-term value of a boost in reducing IBTR has been demonstrated.30 Two ASTRO consensus guidelines have addressed technical issues in the setting of BCT. While largely focusing on invasive breast carcinoma, the ASTRO statement on HWBRT concluded there was insufficient evidence to recommend for or against HWBRT in the setting of DCIS.31 In the ASTRO statement on APBI, DCIS was placed into the “cautionary” group based on the lack of evidence from randomized trials, while noting that DCIS patients have been included in some retrospective cohort studies.32 Therefore, there is no evidence that margin width, in isolation, should determine radiation delivery technique, fractionation of WBRT, or use/dose of a boost. The MP considered the evidence base insufficient to address optimal margin width in APBI.

Therefore, there is no evidence that margin width, in isolation, should determine radiation delivery technique, fractionation of WBRT, or use/dose of a boost. The MP considered the evidence base insufficient to address optimal margin width in APBI. DCIS in the Presence of Invasive Breast Cancer DCIS with microinvasion, defined as no invasive focus > 1 mm in size, should be considered as DCIS when considering the optimal margin width. There are two diagnoses for which there is overlap between the DCIS Margin Guideline and the Invasive Cancer Margin Guideline33: DCIS with microinvasion (DCIS-M) and invasive carcinoma associated with DCIS (extensive intraductal component [EIC] or lesser amounts of scattered DCIS). In DCIS-M, defined by the American Joint Committee on Cancer (AJCC) as the extension of cancer cells beyond the basement membrane with no focus more than 0.1 cm in greatest dimension,34 small retrospective studies suggest that rates of IBTR are similar to those seen with pure DCIS.35,36 In the absence of specific data to address margin width in DCIS-M, the MP, based on expert opinion, felt that DCIS-M should be considered as DCIS when considering the optimal margin width, given that the majority of the lesion is comprised of DCIS and that systemic therapy utilization for DCIS-M more closely reflects the treatment pattern for DCIS than for invasive carcinoma.

DCIS-M, the MP, based on expert opinion, felt that DCIS-M should be considered as DCIS when considering the optimal margin width, given that the majority of the lesion is comprised of DCIS and that systemic therapy utilization for DCIS-M more closely reflects the treatment pattern for DCIS than for invasive carcinoma. In contrast, when considering margin width for an invasive cancer with a DCIS component, regardless of extent, the MP felt that the invasive cancer guideline33 was applicable, primarily because the natural history and treatment of these lesions is more similar to invasive cancer than DCIS, even when the close margin contains DCIS. In particular, the vast majority of patients with invasive cancer receive systemic therapy, which remains less common for pure DCIS. The invasive cancer guideline33 did note that an EIC is a marker for a potential heavy burden of residual DCIS and that postexcision mammography, the presence of multiple close margins, and young patient age can be used to select patients who will benefit from re-excision. These statements echo the discussion of the MP regarding the use of re-excision in pure DCIS with margins < 2 mm discussed previously, and thus we believe that the guidelines are compatible.

, the presence of multiple close margins, and young patient age can be used to select patients who will benefit from re-excision. These statements echo the discussion of the MP regarding the use of re-excision in pure DCIS with margins < 2 mm discussed previously, and thus we believe that the guidelines are compatible. DISCUSSION There are limitations to this guideline. It applies to patients with DCIS and DCIS-M treated with WBRT. The findings should not be extrapolated to DCIS patients treated with APBI or those with invasive carcinoma for whom a separate guideline has been developed.33 While studies including patients treated with and without WBRT were included in the meta-analysis, a meta-analysis of studies of treatment with excision alone was not conducted. Additionally, all of the studies included in the meta-analysis were retrospective. However, in the absence of any planned prospective randomized trials addressing the question of margin width and local recurrence, these studies represent the best available evidence for clinical decision making. Supported by a grant from Susan G. Komen; by a National Breast Cancer Foundation (NBCF; NBCF Australia) Breast Cancer Research Leadership Fellowship (N.H.); and by a Cancer Institute New South Wales Fellowship (M.L.M.).

DISCUSSION There are limitations to this guideline. It applies to patients with DCIS and DCIS-M treated with WBRT. The findings should not be extrapolated to DCIS patients treated with APBI or those with invasive carcinoma for whom a separate guideline has been developed.33 While studies including patients treated with and without WBRT were included in the meta-analysis, a meta-analysis of studies of treatment with excision alone was not conducted. Additionally, all of the studies included in the meta-analysis were retrospective. However, in the absence of any planned prospective randomized trials addressing the question of margin width and local recurrence, these studies represent the best available evidence for clinical decision making. Supported by a grant from Susan G. Komen; by a National Breast Cancer Foundation (NBCF; NBCF Australia) Breast Cancer Research Leadership Fellowship (N.H.); and by a Cancer Institute New South Wales Fellowship (M.L.M.). This guideline was developed through collaboration between the Society of Surgical Oncology, the American Society of Clinical Oncology, and the American Society for Radiation Oncology, and is published jointly by invitation and consent in the Annals of Surgical Oncology, Journal of Clinical Oncology, and Practical Radiation Oncology. This statement has been endorsed by the Board of Directors of the American Society of Breast Surgeons. Authors’ disclosures of potential conflicts of interest are found in the article online at www.jco.org. Author contributions are found at the end of this article.

This guideline was developed through collaboration between the Society of Surgical Oncology, the American Society of Clinical Oncology, and the American Society for Radiation Oncology, and is published jointly by invitation and consent in the Annals of Surgical Oncology, Journal of Clinical Oncology, and Practical Radiation Oncology. This statement has been endorsed by the Board of Directors of the American Society of Breast Surgeons. Authors’ disclosures of potential conflicts of interest are found in the article online at www.jco.org. Author contributions are found at the end of this article. AUTHOR CONTRIBUTIONS Conception and design: All authors Administrative support: Monica Morrow Provision of study materials or patients: All authors Collection and assembly of data: Monica Morrow, Kimberly J. Van Zee, Lawrence J. Solin, Nehmat Houssami, Mariana Chavez-MacGregor, Jay R. Harris, Janet Horton, Shelley Hwang, M. Luke Marinovich, Stuart J. Schnitt, Irene Wapnir, Meena S. Moran Data analysis and interpretation: Monica Morrow, Kimberly J. Van Zee, Lawrence J. Solin, Nehmat Houssami, Mariana Chavez-MacGregor, Jay R. Harris, Janet Horton, Shelley Hwang, M. Luke Marinovich, Stuart J. Schnitt, Irene Wapnir, Meena S. Moran Manuscript writing: All authors Final approval of manuscript: All authors

Collection and assembly of data: Monica Morrow, Kimberly J. Van Zee, Lawrence J. Solin, Nehmat Houssami, Mariana Chavez-MacGregor, Jay R. Harris, Janet Horton, Shelley Hwang, M. Luke Marinovich, Stuart J. Schnitt, Irene Wapnir, Meena S. Moran Data analysis and interpretation: Monica Morrow, Kimberly J. Van Zee, Lawrence J. Solin, Nehmat Houssami, Mariana Chavez-MacGregor, Jay R. Harris, Janet Horton, Shelley Hwang, M. Luke Marinovich, Stuart J. Schnitt, Irene Wapnir, Meena S. Moran Manuscript writing: All authors Final approval of manuscript: All authors AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST Society of Surgical Oncology–American Society for Radiation Oncology–American Society of Clinical Oncology Consensus Guideline on Margins for Breast-Conserving Surgery With Whole-Breast Irradiation in Ductal Carcinoma In Situ The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or jco.ascopubs.org/site/ifc. Monica Morrow No relationship to disclose Kimberly J. Van Zee No relationship to disclose Lawrence J. Solin No relationship to disclose Nehmat Houssami No relationship to disclose Mariana Chavez-MacGregor No relationship to disclose Jay R. Harris No relationship to disclose Janet Horton No relationship to disclose Shelley Hwang No relationship to disclose

Monica Morrow No relationship to disclose Kimberly J. Van Zee No relationship to disclose Lawrence J. Solin No relationship to disclose Nehmat Houssami No relationship to disclose Mariana Chavez-MacGregor No relationship to disclose Jay R. Harris No relationship to disclose Janet Horton No relationship to disclose Shelley Hwang No relationship to disclose Peggy L. Johnson No relationship to disclose M. Luke Marinovich No relationship to disclose Stuart J. Schnitt No relationship to disclose Irene Wapnir No relationship to disclose Meena S. Moran No relationship to disclose Appendix Summary of Methodology for the Margins Meta-Analysis The methods for the meta-analysis are described in full in Marinovich et al,15 and are summarized briefly below. Study Eligibility Criteria Eligible studies enrolled ≥ 50 women with ductal carcinoma in situ (DCIS) undergoing breast conserving surgery (BCS); allowed calculation of the crude local recurrence (LR) rate by microscopic margin status; defined negative margins by a numeric threshold; reported mean or median age; and presented mean or median follow up of ≥ 48 months. Literature Search and Data Extraction MEDLINE, PREMEDLINE, EMBASE, and ALL EBM REVIEWS were searched in October 2014. One investigator screened citations, with a sample independently screened by a second. Two investigators independently extracted data; disagreements were arbitrated by a third investigator.

Study Eligibility Criteria Eligible studies enrolled ≥ 50 women with ductal carcinoma in situ (DCIS) undergoing breast conserving surgery (BCS); allowed calculation of the crude local recurrence (LR) rate by microscopic margin status; defined negative margins by a numeric threshold; reported mean or median age; and presented mean or median follow up of ≥ 48 months. Literature Search and Data Extraction MEDLINE, PREMEDLINE, EMBASE, and ALL EBM REVIEWS were searched in October 2014. One investigator screened citations, with a sample independently screened by a second. Two investigators independently extracted data; disagreements were arbitrated by a third investigator. Statistical Analysis Frequentist models (random effects logistic meta-regression). Margins were dichotomised into positive/close versus negative margin status using one distance threshold per study (> 0 or 1 mm; 2 mm; 3 or 5 mm; 10 mm). The association between LR and margin status and distance was modeled using random effects logistic meta-regression. Odds ratios (ORs) are presented for negative relative to positive/close margins, and threshold distances relative to > 0 or 1 mm. Bayesian models (network meta-analysis). Network meta-analysis using Mixed Treatment Comparisons used data from single or multiple thresholds within studies (when presented) to compare directly (within study) and indirectly (between studies) the probability of LR between margins categories (positive; > 0 or 1 mm; 2 mm; 3 mm; 10 mm). ORs compare negative versus positive margins, and distance categories relative to positive margins.

ngle or multiple thresholds within studies (when presented) to compare directly (within study) and indirectly (between studies) the probability of LR between margins categories (positive; > 0 or 1 mm; 2 mm; 3 mm; 10 mm). ORs compare negative versus positive margins, and distance categories relative to positive margins. Assessment of covariates. All models were adjusted for study-level follow up time. Other covariates were assessed for their effect on model estimates (age; median year of recruitment; proportion of patients who received endocrine therapy; proportion with high-grade DCIS; proportion of patients receiving whole breast radiation).

Patients with HIV infection are at an increased risk for certain cancers. With 36.7 million individuals estimated to be living with HIV worldwide, only 46% receive antiretroviral treatment.1 Human papillomavirus (HPV) –associated malignancies occur in excess among patients with HIV, with cervical cancer designated as an AIDS-defining condition.2 Of note, although the introduction of highly active antiretroviral treatment (HAART) has reduced the incidence of some cancer types in those living with HIV, such as Kaposi’s sarcoma and CNS non-Hodgkin lymphoma, the incidence of cervical cancer has not decreased. The link among HPV, HIV, and cervical cancer is becoming better understood and attributed to enhanced HPV carcinogenesis in the setting of HIV-related immunosuppression as well as more frequent infections, with multiple and/or high-risk HPV subtypes in women with HIV.3 In a study of > 309,000 US patients with HIV from 5 years before the date of HIV onset to 5 years after, the incidence of cervical cancer was documented to be significantly increased in women with HIV infection (relative risk, 5.4; 95% CI, 3.9 to 7.2).4

s, with multiple and/or high-risk HPV subtypes in women with HIV.3 In a study of > 309,000 US patients with HIV from 5 years before the date of HIV onset to 5 years after, the incidence of cervical cancer was documented to be significantly increased in women with HIV infection (relative risk, 5.4; 95% CI, 3.9 to 7.2).4 Some evidence also suggests that cervical cancer is more aggressive in women with HIV, who are more likely to present with more advanced-stage disease and respond less well to therapy. Cervical cancer remains the fourth leading cause of cancer death in women worldwide and is the leading cause of cancer death in women with HIV in sub-Saharan Africa.5,6 Because the majority of the disease burden is seen in low- and middle-income countries, with regions of a high prevalence of cervical cancer corresponding with regions of a high prevalence of HIV infection, our understanding of the impact of concurrent HIV on cervical cancer and its response to treatment is paramount.7 The implementation of routine Papanicolaou smears in developed countries has led to the detection of early cervical cancers curable with surgery alone. Similarly, the development of the cervical cancer vaccine is expected to significantly reduce the incidence of cervical cancer over the next 60 years.8 Despite this, women in the developing world and even disadvantaged areas of developed countries continue to have low vaccination and Papanicolaou test rates, which result in more locally advanced stages of cervical cancer at diagnosis.9-12

o significantly reduce the incidence of cervical cancer over the next 60 years.8 Despite this, women in the developing world and even disadvantaged areas of developed countries continue to have low vaccination and Papanicolaou test rates, which result in more locally advanced stages of cervical cancer at diagnosis.9-12 For example, in Botswana, a country in sub-Saharan Africa, cervical cancer is the leading cause of cancer death in women, with more than two thirds of cases occurring in women with HIV infection. The national prevalence of HIV in Botswana is 17% to 24%.13 In South Africa, cervical cancer is the most common cancer; however, despite this, the country’s cancer screening policy is only able to offer asymptomatic women three free cervical smears in a lifetime, which begin at age 30 years and are performed 10 years apart. For women with HIV and CD4 T-cell counts < 350 cells/μL, screening is done annually.14 The problem is compounded by the lack of knowledge about the survival interplay between HIV and cervical cancer. However, should these dual diagnoses predict worse outcomes, then a modified treatment approach may be required. Most concerning is that the last major improvement in cervical cancer survival was based on reports published 17 years ago and later confirmed in a large meta-analysis in 2008,15-17 which demonstrated the superiority of cisplatin-based chemoradiation to radiation therapy alone and resulted in a low 58% versus 50% 5-year disease-free survival. No subgroup analysis of the potential confounding effects of a concurrent HIV infection on outcome was possible because patients with HIV are commonly excluded from clinical trials in oncology.

f cisplatin-based chemoradiation to radiation therapy alone and resulted in a low 58% versus 50% 5-year disease-free survival. No subgroup analysis of the potential confounding effects of a concurrent HIV infection on outcome was possible because patients with HIV are commonly excluded from clinical trials in oncology. In the article that accompanies this editorial, Dryden-Peterson et al18 present results of a prospective cohort of 348 women with cervical cancer treated in Botswana from 2010 to 2015. The primary objective was to determine the impact of concurrent HIV infection, which was present in two thirds of the cohort, on survival. Their conclusion is that despite good access to and use of antiretroviral treatment in 82% of the women before a cancer diagnosis, HIV infection significantly decreases survival from cervical cancer. Specifically, 3-year survival for the group with HIV infection was only 35% (95% CI, 27% to 44%) compared with 48% in the group without HIV infection (95% CI, 35% to 60%), with the majority of the deaths attributable to the cancer rather than to HIV. This occurred despite the fact that patients with HIV were younger, (median age, 42 years) compared with patients without HIV (median age, 58 years), had significantly greater access to education, and higher measures of wealth than those without HIV.

majority of the deaths attributable to the cancer rather than to HIV. This occurred despite the fact that patients with HIV were younger, (median age, 42 years) compared with patients without HIV (median age, 58 years), had significantly greater access to education, and higher measures of wealth than those without HIV. Another concerning finding is that although the majority (82.9%) of patients were considered candidates for potentially curative therapy, the radiotherapy completion rates were far from ideal. Specifically, 30.6% of the cohort were considered to have received an inadequate radiotherapy dose, with only 61.2% completing the planned brachytherapy, some of which had to be administered in another country. Furthermore, only 80.8% received at least one dose of concurrent cisplatin. However, these findings were not specifically different between the study groups and highlight the challenges of delivering a complex treatment like chemoradiation for cervical cancer in a developing nation and the need to better understand barriers to treatment completion in all women with cervical cancer. Of note, rates of radiation toxicity did not seem to be different between patients with and without HIV or necessarily the reason for the low rates of completion of optimal treatment, which the study by Dryden-Peterson et al18 was not able to explain fully. This finding is different from the small number of other studies in the literature reviewed in a recent Cochrane analysis, which suggested that toxicity from radiotherapy for cervical cancer generally is higher in patients with HIV.19,20 Unfortunately, no guidelines for the treatment of locally advanced cervical cancer in women with HIV have been published in the past decade. However, the Cochrane review found that patients who were started on HAART early had higher rates of treatment completion, and HAART has been recommended to be initiated as soon as possible in patients with HIV and newly diagnosed cervical cancer to reduce treatment toxicity.

HIV have been published in the past decade. However, the Cochrane review found that patients who were started on HAART early had higher rates of treatment completion, and HAART has been recommended to be initiated as soon as possible in patients with HIV and newly diagnosed cervical cancer to reduce treatment toxicity. Dryden-Peterson et al18 address important questions given the burden of both HIV and cervical cancer in less-developed countries and emerging initiatives to establish chemoradiation treatment in these countries. However, several methodological concerns limit the generalizability of the findings. The cohort included a mix of patients treated with palliative and curative intent; the median follow-up of 15.1 months is relatively short, with limited follow-up information available because the treating team did not follow-up with patients after treatment completion. Despite this, an analysis restricted to women who received curative intent therapy and guideline-concordant curative intent treatment was presented and suggests that HIV infection nearly doubled the risk of death.

nformation available because the treating team did not follow-up with patients after treatment completion. Despite this, an analysis restricted to women who received curative intent therapy and guideline-concordant curative intent treatment was presented and suggests that HIV infection nearly doubled the risk of death. Another concern is that the diagnostic work-up consisted of a clinical examination, chest x-ray, and abdominal ultrasound with no use of positron emission tomography scanning or magnetic resonance imaging. Hence, no information is available about tumor volume or nodal status, which are considered some of the most important factors in predicting prognosis in locally advanced cervical cancer. Without the inclusion of these baseline characteristics in stratification, imbalances between cohorts are likely to exist and potentially affect the validity and reproducibility of the findings. In addition, Dryden-Peterson et al18 had limited information about the amount or number of cycles of concurrent cisplatin received during radiation, and although the study defined receipt of at least one cycle as guideline-concordant therapy, this is not consistent with the international standard recommendation to deliver cisplatin 40 mg/m2 once per week during pelvic radiotherapy. This issue is a significant confounder in the analysis given the established survival benefit from concurrent cisplatin therapy in the treatment of locally advanced disease.

apy, this is not consistent with the international standard recommendation to deliver cisplatin 40 mg/m2 once per week during pelvic radiotherapy. This issue is a significant confounder in the analysis given the established survival benefit from concurrent cisplatin therapy in the treatment of locally advanced disease. Despite methodological concerns and the absence of detailed data about certain aspects, this study describes the importance of further research and investment into improving the outcomes of women with both HIV and cervical cancer. International collaboration through research networks, such as the recently established Cervix Cancer Research Network,21 that could provide infrastructure, quality assurance, financial support, data sharing, education, and expertise in conducting clinical trials can potentially help establish future initiatives to answer outstanding questions such as those highlighted by Dryden-Peterson et al.18 We commend these authors for their dedication to improve treatment for women with HIV and cervical cancer in this rising burden of disease that has had limited research to date.

trials can potentially help establish future initiatives to answer outstanding questions such as those highlighted by Dryden-Peterson et al.18 We commend these authors for their dedication to improve treatment for women with HIV and cervical cancer in this rising burden of disease that has had limited research to date. One clear major barrier to optimal treatment of cervical cancer, and indeed many cancers, is access to radiation therapy. A recent analysis of radiation therapy infrastructure in 139 low- and middle-income countries found that only four (2.87%) have the requisite number of teletherapy units to manage the estimated burden of cancer in 2020 and that 55 (39.5%) have no radiation facilities.22 Another analysis of radiotherapy resources found that brachytherapy is available in only 20 of 52 African countries.23 If we are to reduce the number of deaths from cervical cancer in women with and without HIV, it is paramount that access to screening, vaccination, and treatment facilities in those areas of the world with the greatest burden of disease is addressed. See accompanying article on page 3749 ACKNOWLEDGMENT L.R.M. is supported by a Victorian Cancer Agency Clinical Research Fellowship.

One clear major barrier to optimal treatment of cervical cancer, and indeed many cancers, is access to radiation therapy. A recent analysis of radiation therapy infrastructure in 139 low- and middle-income countries found that only four (2.87%) have the requisite number of teletherapy units to manage the estimated burden of cancer in 2020 and that 55 (39.5%) have no radiation facilities.22 Another analysis of radiotherapy resources found that brachytherapy is available in only 20 of 52 African countries.23 If we are to reduce the number of deaths from cervical cancer in women with and without HIV, it is paramount that access to screening, vaccination, and treatment facilities in those areas of the world with the greatest burden of disease is addressed. See accompanying article on page 3749 ACKNOWLEDGMENT L.R.M. is supported by a Victorian Cancer Agency Clinical Research Fellowship. AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST Improvement of Outcomes for Women With HIV Infection and Cervical Cancer The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or jco.ascopubs.org/site/ifc. Linda R. Mileshkin Research Funding: Hospira Travel, Accommodations, Expenses: Merck Sharp & Dohme, Roche

AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST Improvement of Outcomes for Women With HIV Infection and Cervical Cancer The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or jco.ascopubs.org/site/ifc. Linda R. Mileshkin Research Funding: Hospira Travel, Accommodations, Expenses: Merck Sharp & Dohme, Roche Alison E. Freimund Travel, Accommodations, Expenses: Roche, Bristol-Meyers Squibb

To the Editor: Li et al1 reported the results of a randomized, double-blind, placebo-controlled phase III trial of apatinib, which showed significant survival benefits in patients with chemotherapy-refractory advanced or metastatic adenocarcinoma of the stomach or gastroesophageal junction. These results validate the role of vascular endothelial growth factor receptor (VEGFR) -2 signaling as an important therapeutic target; however, the clinical effects of apatinib are modest, with limited survival prolongation compared with placebo—median overall survival, 6.5 months versus 4.7 months; median progression-free survival, 2.6 months versus 1.8 months, respectively—and a low objective response rate of 1.70% as assessed by an independent response evaluation committee.1 Therefore, it is critically challenging to identify suitable predictive biomarkers that could be used to select patients who will benefit most from VEGFR-2 signal-inhibiting agents, such as apatinib, which would thereby improve efficacy and avoid unnecessary toxicity and high cost. These biomarkers might come from the cellular or molecular level using biospecimens collected from patients. Alternatively, occurrence of adverse events might act as surrogate biomarkers of drug activity, enabling the prediction of outcome during treatment because the occurrence of treatment-emergent toxic effects is associated with a pharmacodynamic effect of the drug.2-4 Recently, it has been suggested that the occurrence of specific adverse events, such as hypertension, hand-foot syndrome, and proteinuria, during antiangiogenic therapy might be associated with improved efficacy.4-7 Regarding apatinib, in particular, it was reported that hypertension and hand-foot skin reaction were significantly related to longer progression-free and overall survival in patients with advanced breast cancer.8 Therefore, it would be interesting to know whether the prospective data set reported by Li et al1 shows that the development of treatment-specific adverse effects, such as hypertension, hand-foot syndrome, and proteinuria, is related to treatment outcome.

e and overall survival in patients with advanced breast cancer.8 Therefore, it would be interesting to know whether the prospective data set reported by Li et al1 shows that the development of treatment-specific adverse effects, such as hypertension, hand-foot syndrome, and proteinuria, is related to treatment outcome. The investigators could help to address this issue by analyzing survival data according to the emergence of treatment-related adverse events. Such data could help clinicians make better treatment decisions and may shed light on the future development of VEGFR signaling-targeted therapy for gastric and gastroesophageal junction carcinomas. AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST Is Treatment-Emergent Toxicity a Biomarker of Efficacy of Apatinib in Gastric Cancer? The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or jco.ascopubs.org/site/ifc. Hyo Jin Lee No relationship to disclose Ji Young Moon No relationship to disclose Seung Woo Baek No relationship to disclose

Interpretation of oncology clinical trial data are not always straightforward or consistent. Similar trial results with disparate interventions may be interpreted differently by the oncology community. One of the main reasons for this discrepancy is the debate regarding what is the appropriate end point for demonstration of efficacy of cancer drugs. There is no doubt that overall survival (OS) is the best parameter to judge the utility of any intervention, and it is free from bias in ascertainment and measurement1; but for conditions with few treatment options and dire outcomes, the need for new agents is high and the oncology community sometimes settles on a surrogate end point that, in many cases, is progression-free survival (PFS).2 It is easy to understand why PFS is favored among the researchers: It occurs early and is not influenced by postprogression therapy. At the same time, it would make little sense to have an agent that reduces chances of dying of cancer but increases off-target deaths; hence, the need for verification of OS. Phase III trials that report on significant PFS benefits without OS prolongation become the apples of discord in the oncology community. In this commentary, we present three examples from lung, ovarian, and breast cancers and demonstrate how the oncology community interprets similar data differently. Finally, we take our best guess as to why this phenomenon happens.

benefits without OS prolongation become the apples of discord in the oncology community. In this commentary, we present three examples from lung, ovarian, and breast cancers and demonstrate how the oncology community interprets similar data differently. Finally, we take our best guess as to why this phenomenon happens. Lung Cancer: Bevacizumab and Cetuximab Bevacizumab and cetuximab have both been tested in phase III trials for use in advanced/metastatic non-small cell lung cancer (NSCLC) in combination with chemotherapy. The Eastern Cooperative Oncology Group (ECOG) 4599 trial demonstrated a significant OS prolongation with the addition of bevacizumab compared with chemotherapy alone (12.3 months v 10.3 months; hazard ratio [HR], 0.79; P = .03) but with significant toxicities, including 15 treatment-related deaths among 434 patients randomly assigned to the bevacizumab arm.3 The AVAIL (Avastin in Lung) study on the other hand found a marginal benefit in PFS, with no benefit in OS, by adding bevacizumab to chemotherapy (13.6 months v 13.1 months; HR, 0.93; P = not significant [NS]).4 A Japanese study also failed to show an OS benefit with addition of bevacizumab to chemotherapy (22.8 months v 23.4 months; HR, 0.99; P = .95).5 However, bevacizumab received approval by the US Food and Drug Administration (FDA) for use in this setting and is commonly used in practice as evidenced by its inclusion in the National Comprehensive Cancer Network (NCCN) guidelines as a category 2A recommendation for patients with EGFR, ALK negative, or unknown nonsquamous non-small cell lung cancer.6

val by the US Food and Drug Administration (FDA) for use in this setting and is commonly used in practice as evidenced by its inclusion in the National Comprehensive Cancer Network (NCCN) guidelines as a category 2A recommendation for patients with EGFR, ALK negative, or unknown nonsquamous non-small cell lung cancer.6 FLEX (First-Line Erbitux in Lung Cancer) was a randomized phase III trial comparing chemotherapy plus cetuximab with chemotherapy alone in patients with advanced NSCLC and demonstrated a significant OS benefit (11.3 months v 10.1 months; HR, 0.87; P = .044).7 However, another phase III trial, BMS099, failed to show similar benefit in OS (9.6 months v 8.3 months; HR, 0.89; P = .169).8 It is important to note here that OS was the primary end point in FLEX, whereas PFS was the primary end point in the BMS099 study. Later, a meta-analysis showed significant benefit for OS, PFS, and response rates with the addition of cetuximab to chemotherapy.9 However, cetuximab is not approved by the FDA and is widely considered a failed drug in NSCLC by the oncology community, as evidenced by its removal from the NCCN guidelines.6 Ovarian Cancer: Angiogenesis Inhibitors and Dose-Dense Chemotherapy Several attempts have been made to build on the success of the platinum-taxane combination for treating advanced or metastatic ovarian cancer, but none have been met with irrefutable success. Of those various strategies, two are the most common and the most debated: dose-dense treatment schedule and addition of an angiogenesis inhibitor to the combination.

ld on the success of the platinum-taxane combination for treating advanced or metastatic ovarian cancer, but none have been met with irrefutable success. Of those various strategies, two are the most common and the most debated: dose-dense treatment schedule and addition of an angiogenesis inhibitor to the combination. The feasibility and efficacy of a dose-dense schedule (weekly paclitaxel v every-3-week paclitaxel) was demonstrated in the Japanese Gynecologic Oncology Group (JGOG) 3016 trial, a study among 637 Japanese patients.10 This trial showed that weekly paclitaxel improved both PFS and OS. The OS advantage was not trivial; it was a sizable 38-month extension (100.5 months v 62.2 months; HR, 0.79; P = .039). However, the global oncology community adopted the addition of bevacizumab but has largely ignored the dose-dense paclitaxel schedule. Perhaps, the large benefit with weekly paclitaxel prompted clinicians to disbelief and wanting further confirmation; yet, it is hard to imagine clinicians believed a larger benefit would altogether vanish, rather than merely be attenuated. In 2014, an Italian trial failed to replicate these results, but had used a different dose schedule.11 Whether this lack of replication was due to this difference in dose of paclitaxel used or due to ethnic differences between the populations remains to be known, but the results of the Gynecologic Oncology Group (GOG-0262) trial have shown benefit with weekly paclitaxel in the US population.12

different dose schedule.11 Whether this lack of replication was due to this difference in dose of paclitaxel used or due to ethnic differences between the populations remains to be known, but the results of the Gynecologic Oncology Group (GOG-0262) trial have shown benefit with weekly paclitaxel in the US population.12 In the past few months, three important clinical trials have been published and add to the evidence (and confusion) of these two strategies: the updated results of the International Collaborative Ovarian Neoplasm 7 (ICON7) trial,13 the AGO-OVAR 12 (Standard first-line chemotherapy with or without nintedanib for advanced ovarian cancer) trial,14 and the GOG-0262 trial.12 The results of these trials and the conclusions the authors derived are of interest and importance. The ICON-7 trial showed a PFS benefit but failed to show an OS benefit with the addition of bevacizumab to the chemotherapy backbone.13 However, a subgroup analysis was performed and revealed that for high-risk patients, addition of bevacizumab did have an OS benefit. Instead of highlighting the overall negative OS data, the authors chose to emphasize the OS advantage among high-risk patients. Further, this trial was not placebo controlled and has been criticized.15

group analysis was performed and revealed that for high-risk patients, addition of bevacizumab did have an OS benefit. Instead of highlighting the overall negative OS data, the authors chose to emphasize the OS advantage among high-risk patients. Further, this trial was not placebo controlled and has been criticized.15 The AGO-OVAR 12 trial randomized a large number of patients (N = 1,366) to nintedanib, another angiogenesis inhibitor, or placebo in combination with chemotherapy.14 OS data are not available but PFS was significantly better with nintedanib versus placebo (HR, 0.84; P = .024). However, the actual gain in PFS was a mere 0.6 months. But the authors concluded, “Nintedanib in combination with carboplatin and paclitaxel is an active first-line treatment that significantly increases progression-free survival for women with advanced ovarian cancer.”14

nintedanib versus placebo (HR, 0.84; P = .024). However, the actual gain in PFS was a mere 0.6 months. But the authors concluded, “Nintedanib in combination with carboplatin and paclitaxel is an active first-line treatment that significantly increases progression-free survival for women with advanced ovarian cancer.”14 GOG 0262, the third study, compared weekly paclitaxel with every-3-week paclitaxel among patients with ovarian cancer.12 This trial also allowed patients to receive bevacizumab and prospectively stratified them according to bevacizumab status. Although every-3-week paclitaxel did not improve the PFS in the entire population (14.7 months v 14.0 months; HR, 0.89, P = .18), the PFS for those patients who did not take bevacizumab was significantly improved by 3.9 months (14.2 months v 10.3 months; HR, 0.62; P = .03).12 The OS data are not yet available. Considering that 84% of patients in this trial took bevacizumab and this could negate the overall benefit of dose-dense treatment, we assumed the trial would be interpreted as meaning weekly paclitaxel was superior to every-3-week dosing except for those patients who received additional bevacizumab. However, the results of this trial have mostly been interpreted as negative.

izumab and this could negate the overall benefit of dose-dense treatment, we assumed the trial would be interpreted as meaning weekly paclitaxel was superior to every-3-week dosing except for those patients who received additional bevacizumab. However, the results of this trial have mostly been interpreted as negative. Important information can be gleaned from summarizing these trials. There is no trial that shows OS benefit with any angiogenesis inhibitor in ovarian cancer (Table 1), whereas there is one trial that shows OS benefit with the dose-dense schedule. Unless we have data comparing chemotherapy plus bevacizumab versus dose-dense chemotherapy alone, the current evidence equally (if not more) supports the use of dose-dense chemotherapy alone compared with bevacizumab addition. Table 1. Published Phase III Trials of Angiogenesis Inhibitors in Advanced/Relapsed Ovarian Cancer Yet, the authors of such pivotal studies as ICON7 attempted to emphasize the benefit of bevacizumab, whereas those of GOG-0262 did not highlight the benefits derived from dose-dense paclitaxel. It is noteworthy that practice patterns occur despite the fact that the NCCN guidelines categorize bevacizumab addition to first-line chemotherapy as a category 3 recommendation and dose-dense paclitaxel as a category 1.21

mab, whereas those of GOG-0262 did not highlight the benefits derived from dose-dense paclitaxel. It is noteworthy that practice patterns occur despite the fact that the NCCN guidelines categorize bevacizumab addition to first-line chemotherapy as a category 3 recommendation and dose-dense paclitaxel as a category 1.21 Breast Cancer: Everolimus and Bevacizumab The use of PFS as a surrogate for OS may be valid for certain tumor types, certain classes of agents, and certain lines of therapy, but an umbrella analysis of surrogate correlation studies showed that it is unreliable in the setting of metastatic breast cancer.1 This uncertainty took on importance after the results of the E2100 trial, which showed a large improvement in PFS from the use of bevacizumab when added to taxane therapy versus taxane therapy alone.22 This finding led to accelerated approval of the drug. Yet, just a few years later, multiple randomized trials not only failed to confirm survival benefit in this setting but also failed to replicate similar magnitude of benefit in PFS. And, bevacizumab clearly increased toxicity. After a contentious fight, the drug was revoked.

ing led to accelerated approval of the drug. Yet, just a few years later, multiple randomized trials not only failed to confirm survival benefit in this setting but also failed to replicate similar magnitude of benefit in PFS. And, bevacizumab clearly increased toxicity. After a contentious fight, the drug was revoked. Now, just 4 years later, we have seen two new drug approvals for metastatic breast cancer that mirror the history of bevacizumab. Everolimus23 and palbociclib,24 both in combination with hormonal therapy, have had markedly similar results to the case of bevacizumab. Both drugs improved PFS in randomized trials, both drugs add toxicity, and neither drug has shown OS benefits. It is interesting to note that the absolute gain in PFS in the pivotal trials of these drugs is similar to that of bevacizumab seen in the E2100 trial: 4.1 months with everolimus,23 5.4 months with palbociclib,24 and 5.9 months with bevacizumab.22 However, a meta-analysis conducted later, with the addition of subsequent trials, showed that the pooled benefit in PFS with bevacizumab was only 2.5 months.25 Whether the PFS benefits with everolimus and palbociclib also are similarly reduced remains to be seen with the acquisition of more data. At least in the case of everolimus, the drug received traditional or full approval, meaning that revoking the approval on the basis of further efficacy data is unlikely, and postmarketing studies to assess the drug’s benefit on OS are not required. Still, the NCCN guidelines include both bevacizumab and everolimus as a category 2A recommendation, whereas palbociclib gets a category 1 recommendation—without having any OS data yet!26

val on the basis of further efficacy data is unlikely, and postmarketing studies to assess the drug’s benefit on OS are not required. Still, the NCCN guidelines include both bevacizumab and everolimus as a category 2A recommendation, whereas palbociclib gets a category 1 recommendation—without having any OS data yet!26 Same Data; Different Interpretations It is difficult to provide a unifying theme that explains why we treat similar data differently in oncology. Potential explanations include reimbursement incentives, historical accident, pharmaceutical marketing, perceived toxicity, clinical anecdotes, social norms, or objective and articulable differences that we have not considered. In the case of bevacizumab and cetuximab in NSCLC, the unique regulatory history and pathway for drug approval likely explain the success and validation of the former and the failure of the latter. In the case of discrepancy among the oncologists in the acceptance of dose-dense chemotherapy versus angiogenesis inhibitors in ovarian cancer, it is difficult to not consider the issue of financial reimbursement (higher with bevacizumab) and convenience to practitioners.

on of the former and the failure of the latter. In the case of discrepancy among the oncologists in the acceptance of dose-dense chemotherapy versus angiogenesis inhibitors in ovarian cancer, it is difficult to not consider the issue of financial reimbursement (higher with bevacizumab) and convenience to practitioners. A more optimistic outlook of the medical community toward targeted therapies compared with cytotoxic agents may be another potential reason. In the final case of everolimus and bevacizumab, it is possible regulators were not eager to relive the painful events leading to removal of bevacizumab’s indication, and, for that reason, gave an unwarranted traditional (full) approval to everolimus (on the basis of comparable data). This would eliminate the need for postmarketing studies and preclude a contentious withdrawal from market, as was seen for bevacizumab. Ultimately, however, our interpretation of these discrepancies must be acknowledged as speculative and other potential factors in play for these discrepancies must be explored. Given that we now have umbrella meta-analyses of the strength of surrogate correlations in oncology1,27 that show the validity of correlations between surrogates and survival in specific cancer settings (e.g., does disease-free survival predict OS among cytotoxic drugs in the adjuvant treatment of colorectal cancer?), it may now be possible for the field to move toward greater evidence-based consistency in our interpretation and regulatory use of trial data.

between surrogates and survival in specific cancer settings (e.g., does disease-free survival predict OS among cytotoxic drugs in the adjuvant treatment of colorectal cancer?), it may now be possible for the field to move toward greater evidence-based consistency in our interpretation and regulatory use of trial data. We cannot also ignore the deep issues beyond clinical data that result in discrepancies in cancer care, such as politics, emotional overlay, lobbying, and advocacy of support groups. Although we explore three instances of discrepancies in the treatment of three similar cancer settings in this paper, many discrepancies exist in cancer care. When bevacizumab was revoked for breast cancer, support groups and patient advocates protested against the decision, but when 131I-tositumomab was withdrawn from marketing, it died silently. Thus, our attitudes toward cancer care are multifactorial. As oncologists, however, we should push for uniformity in the interpretation of clinical trial results and try to achieve as much consistency in our practice as possible. Consistency would be a virtue for cancer care. Authors’ disclosures of potential conflicts of interest are found in the article online at www.jco.org. Author contributions are found at the end of this article. AUTHOR CONTRIBUTIONS Administrative support: All authors Provision of study materials or patients: All authors Manuscript writing: All authors Final approval of manuscript: All authors

Authors’ disclosures of potential conflicts of interest are found in the article online at www.jco.org. Author contributions are found at the end of this article. AUTHOR CONTRIBUTIONS Administrative support: All authors Provision of study materials or patients: All authors Manuscript writing: All authors Final approval of manuscript: All authors AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST Same Data; Different Interpretations The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or jco.ascopubs.org/site/ifc. Bishal Gyawali No relationship to disclose Vinay Prasad No relationship to disclose

Extensive-stage small-cell lung cancer (ES-SCLC) is an aggressive disease characterized by high initial response rates to first-line platinum-based chemotherapy followed inevitably by relapse, poor response to subsequent systemic treatment, and ultimately death. Long-term survival prospects for ES-SCLC are dismal, with an estimated 2-year overall survival (OS) rate of less than 5%. Recent advances in the development and regulatory approval of several new active agents against advanced non–small-cell lung cancer (NSCLC) contrast sharply with the lack of progress in the systemic treatment of ES-SCLC, where survival outcomes have changed minimally over a quarter century.1,2 In fact, the last new drug approval for ES-SCLC (ie, topotecan) occurred almost 20 years ago; meanwhile, 16 new therapies for NSCLC were approved over the same time period (eight targeted therapies, four chemotherapies, two antiangiogenic agents, and two programmed death-1 [PD-1] immune checkpoint inhibitors).

tury.1,2 In fact, the last new drug approval for ES-SCLC (ie, topotecan) occurred almost 20 years ago; meanwhile, 16 new therapies for NSCLC were approved over the same time period (eight targeted therapies, four chemotherapies, two antiangiogenic agents, and two programmed death-1 [PD-1] immune checkpoint inhibitors). In theory, if any class of drugs were to alter the natural history of ES-SCLC and improve survival, it would be immune checkpoint inhibitors (anti–cytotoxic T-cell lymphocyte-4 [anti–CTLA-4] and anti–PD-1 or anti–programmed death ligand 1 [PD-L1] antibodies). Immune checkpoint blockade is reportedly more active in cancers with hypermutated phenotypes, such as malignant melanoma, NSCLC, bladder cancer, and microsatellite instability–high tumors. The postulated mechanism is that higher neoantigen burden and mutational load render these tumors more immunogenic, with reawakened pre-existent antitumor CD8+ cytotoxic T-cell responses, when exposed to immune checkpoint blockade.3 It is thought that high tumoral mutational burden (and thus sensitivity to immunotherapy) corresponds in part to the degree or nature of prior carcinogen exposure. Indeed, smoking-associated NSCLC seems to derive more benefit from checkpoint-targeted immunotherapies than lung cancers in never-smoking patients.4 Because lung cancer with small-cell histology has the strongest association with tobacco carcinogenesis and harbors a high frequency of somatic mutations, one would posit that SCLC would preferentially benefit from immune checkpoint blockade.5,6 Furthermore, it has been hypothesized that cytotoxic chemotherapy could enhance the expression of tumoral neoantigens, thus priming the tumor for response to checkpoint inhibitor therapy. In fact, in the initial phase II trials of ipilimumab plus chemotherapy in either SCLC or NSCLC, modest improvements in immune-related progression-free survival—based on criteria that accounted for tumor shrinkage in the face of new lesions—were seen when ipilimumab was administered concurrently with chemotherapy in later cycles rather than immediately in the first cycle.7,8

s chemotherapy in either SCLC or NSCLC, modest improvements in immune-related progression-free survival—based on criteria that accounted for tumor shrinkage in the face of new lesions—were seen when ipilimumab was administered concurrently with chemotherapy in later cycles rather than immediately in the first cycle.7,8 Against this background, Reck et al9 conducted a large placebo-controlled clinical trial in ES-SCLC in which 1,132 patients were randomly assigned to receive either etoposide and platinum (cisplatin or carboplatin) for four cycles alone or together with the anti–CTLA-4 antibody ipilimumab. Disappointingly, the trial was negative; the primary end point of OS in patients who received at least one dose of ipilimumab was not improved (hazard ratio, 0.94; 95% CI, 0.81 to 1.09). The phased strategy of delivering two initial cycles of etoposide and platinum without ipilimumab is reasonable given the theoretic considerations we have described for increasing expression of immunogenic neoantigens. Besides, from a practical standpoint, the need for cytoreduction in patients often experiencing symptoms of rapidly growing SCLC is paramount; the high anticipated response rates to initial etoposide and platinum would provide an opportunity to palliate symptoms and enrich the patient population for those more likely to benefit from and tolerate subsequent ipilimumab.7,8

toreduction in patients often experiencing symptoms of rapidly growing SCLC is paramount; the high anticipated response rates to initial etoposide and platinum would provide an opportunity to palliate symptoms and enrich the patient population for those more likely to benefit from and tolerate subsequent ipilimumab.7,8 Why was this large and well-conducted trial negative? Considerations intrinsic to ES-SCLC likely contributed to the failure of ipilimumab combined with etoposide and platinum to improve outcomes. In this disease, rapid tumor growth with corresponding symptomatic disease and performance status decline can lead to patient drop off as a result of poor drug tolerability or disease progression. In fact, the primary end point in this study was altered from OS in the intent-to-treat population to OS among patients who received at least one dose of study drug commencing at cycle three. As reported by Reck et al,9 approximately 15% of randomly assigned patients did not receive the study drug. Only approximately 13% of those randomly assigned to receive ipilimumab lived long enough without progression or toxicity to receive it as maintenance. In other ES-SCLC studies, even when biomarker-driven approaches for immune checkpoint blockade have been used, excessive patient dropout has limited generalizability of clinical outcomes. For example, in KEYNOTE 028, only 24 (16%) of 147 patients with SCLC screened for PD-L1 expression actually received pembrolizumab, although 29% (42 of 147) were PD-L1 positive. Nevertheless, this therapy produced a response rate of 29%, impressive for previously treated ES-SCLC.10

eneralizability of clinical outcomes. For example, in KEYNOTE 028, only 24 (16%) of 147 patients with SCLC screened for PD-L1 expression actually received pembrolizumab, although 29% (42 of 147) were PD-L1 positive. Nevertheless, this therapy produced a response rate of 29%, impressive for previously treated ES-SCLC.10 Additional potential explanations can be derived from the experience in metastatic melanoma, where it has been reported that cytotoxic exposure before CTLA-4 blockade induces mostly subclonal mutations rather than clonal mutations.3 Such subclonal mutations may be insufficient to drive an immune response robust enough to improve survival end points. Perhaps priming doses of chemotherapy in ES-SCLC are unable to generate the appropriate level of neoantigen expression, or perhaps the so-called correct neoantigens are not sufficiently expressed to drive functional immunogenicity. Moreover, as an anti–CTLA-4 targeted agent, ipilimumab may not be the best immunotherapeutic agent to use after chemotherapy, because mechanistically its effect on cytotoxic T cells should occur during the priming phase. Anti–PD-1 or anti–PD-L1 antibodies that act locally in the tumor microenvironment during the effector phase may be more clinically relevant in this context than anti–CTLA-4 antibodies that act peripherally at the time of initial response to antigen.11 Indeed, promising overall response rates in trials combining platinum-based chemotherapy with PD-1 antibodies in NSCLC have been reported, although increased toxicity is a major concern; for example, a grade 3 and 4 adverse event rate of 45% and pneumonitis rate of 7%, resulting in discontinuation of study treatment in 21% of patients, were recently reported in a phase I study combining platinum-based chemotherapy and nivolumab.12 Maintenance trials with PD-1 antibodies in SCLC after initial cytoreduction with etoposide and platinum are under way and may represent a more tolerable strategy in the population of patients with SCLC, which often has compromised performance status resulting from medical comorbidities and tumor burden.

nd nivolumab.12 Maintenance trials with PD-1 antibodies in SCLC after initial cytoreduction with etoposide and platinum are under way and may represent a more tolerable strategy in the population of patients with SCLC, which often has compromised performance status resulting from medical comorbidities and tumor burden. Rather than priming with cytotoxic chemotherapy, combined CTLA-4 and PD-1 or PD-L1 blockade in SCLC may represent an encouraging alternative combination strategy, although increased toxicity, including risk of paraneoplastic syndromes, which are already more frequent with small-cell histology, remains a major concern. The nonoverlapping mechanisms of action of CTLA-4 and PD-1 blockade are best demonstrated by recent clinical trials reporting the combined effects of agents targeting these two pathways. In the recently published phase I/II CheckMate 032 study, durable responses to nivolumab and ipilimumab were observed, prompting a randomized phase III trial.13

ms of action of CTLA-4 and PD-1 blockade are best demonstrated by recent clinical trials reporting the combined effects of agents targeting these two pathways. In the recently published phase I/II CheckMate 032 study, durable responses to nivolumab and ipilimumab were observed, prompting a randomized phase III trial.13 Finally, the trial by Reck et al9 failed to improve outcomes in part because it did not attempt to enrich for patients who may have preferentially benefited from such a therapeutic strategy. On the basis of early results with immune checkpoint blockade in SCLC, it is likely that only a small subset of patients benefits from these drugs. Thus, continued companion biomarker development and validation to identify those patients likely to respond to immunotherapy are critical. However, tumor samples in ES-SCLC are often scant and inadequate; obtaining adequate tissue in a timely fashion to appropriately assess the tumor and immune microenvironment can be challenging. With a fast-growing cancer like SCLC, there is also a need to identify and exclude patients whose disease will progress too rapidly for potential benefit from immune checkpoint blockade.

uate; obtaining adequate tissue in a timely fashion to appropriately assess the tumor and immune microenvironment can be challenging. With a fast-growing cancer like SCLC, there is also a need to identify and exclude patients whose disease will progress too rapidly for potential benefit from immune checkpoint blockade. How do we put the study by Reck et al9 into perspective with other checkpoint immunotherapy trials in lung cancer? A similarly designed phase III trial using first-line carboplatin and paclitaxel with ipilimumab in squamous histology lung cancer is ongoing. Even if positive, it will need to be interpreted within the context of the current widespread use of approved anti–PD-1 agents in squamous cell lung cancer. Understanding the influence of sequencing of prior ipilimumab on clinical outcomes of subsequent PD-1 blockade related to changes in the tumor and immune microenvironment will be important if a meaningful improvement in survival is achieved. These results will also need to be interpreted within the context of the OS benefit recently announced for first-line pembrolizumab in patients with stage IV NSCLC harboring high PD-L1 expression (KEYNOTE 024).

the tumor and immune microenvironment will be important if a meaningful improvement in survival is achieved. These results will also need to be interpreted within the context of the OS benefit recently announced for first-line pembrolizumab in patients with stage IV NSCLC harboring high PD-L1 expression (KEYNOTE 024). In summary, Reck et al9 are to be congratulated for completing, to our knowledge, the largest SCLC trial to date and the first phase III randomized trial with immune checkpoint blockade in SCLC. Although overall survival was not improved by adding ipilimumab to chemotherapy in this trial, recent data suggest that immune checkpoint blockade with dual CTLA-4 and PD-1 inhibition may be a more effective strategy in SCLC.13 Assuming toxicity issues are adequately addressed, combined immune checkpoint blockade strategies may be more likely to break the quarter-century drought of new therapies in ES-SCLC. See accompanying article on page 3740 ACKNOWLEDGMENT Supported by an Addario and Van Auken Foundation Young Innovators Team Award (J.W.R.); by the Stand Up to Cancer Lung Cancer Dream Team (J.W.R., D.R.G.); by Merck, AstraZeneca, and Novartis research support (paid directly to institution; J.W.R.); by a Paul Calabresi Career Development Award in Clinical Oncology (Grant No. K12CA138464; J.W.R. [K12 scholar], P.N.L. [grant principal investigator]). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. We thank Arta Monjazeb for a critical review of this editorial.

d in Clinical Oncology (Grant No. K12CA138464; J.W.R. [K12 scholar], P.N.L. [grant principal investigator]). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. We thank Arta Monjazeb for a critical review of this editorial. AUTHOR CONTRIBUTIONS Manuscript writing: All authors Final approval of manuscript: All authors AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST Theory Meets Practice for Immune Checkpoint Blockade in Small-Cell Lung Cancer The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO’s conflict of interest policy, please refer to www.asco.org/rwc or jco.ascopubs.org/site/ifc. Jonathan W. Riess Honoraria: Genentech Consulting or Advisory Role: Celgene, ARIAD Pharmaceuticals, Clovis Oncology Research Funding: Onconova Therapeutics (Inst), AstraZeneca (Inst), Millennium Pharmaceuticals (Inst), Novartis (Inst), Merck (Inst) Travel, Accommodations, Expenses: Genentech, Celgene Primo N. Lara Jr Honoraria: Pfizer Consulting or Advisory Role: Clovis Oncology, Exelixis, Pfizer, TEVA Pharmaceuticals Industries, Halozyme, Novartis, Sanofi, Lpath, Eli Lilly, AstraZeneca, Physicians’ Education Resource, Bayer HealthCare Pharmaceuticals, Genentech

Research Funding: Onconova Therapeutics (Inst), AstraZeneca (Inst), Millennium Pharmaceuticals (Inst), Novartis (Inst), Merck (Inst) Travel, Accommodations, Expenses: Genentech, Celgene Primo N. Lara Jr Honoraria: Pfizer Consulting or Advisory Role: Clovis Oncology, Exelixis, Pfizer, TEVA Pharmaceuticals Industries, Halozyme, Novartis, Sanofi, Lpath, Eli Lilly, AstraZeneca, Physicians’ Education Resource, Bayer HealthCare Pharmaceuticals, Genentech Research Funding: Millennium Pharmaceuticals (Inst), Polaris (Inst), Oncogenex (Inst), GlaxoSmithKline (Inst), Genentech (Inst), Aragon Pharmaceuticals (Inst), Janssen Biotech (Inst), Heat Biologics (Inst), TRACON Pharma (Inst), Merck (Inst) David R. Gandara Consulting or Advisory Role: Merck Research Funding: Bristol-Myers Squibb (Inst), Merck (Inst)

We thank Daly et al1 for providing valuable comments regarding our recent paper on the prognostic role of muscle loss during anticancer treatment in patients with metastatic colorectal cancer.2 We have found that muscle mass decreased significantly during chemotherapy and a decrease in muscle mass was independently associated with poor survival in patients with metastatic colorectal cancer. Daly et al1 correctly note that we observed no associations between a low skeletal muscle index at baseline and reduced survival, in contrast to some but not all previous studies. In our article, we provide some explanations for this discrepancy, for example, the heterogeneity regarding treatment regimens and follow-up time.2 Daly et al add a possible explanation as there may be a possible difference in body composition reference values between a North American (Canadian) and European population. Daly et al suggest that extrapolating cutoff points from a Canadian population to a cohort of Dutch patients may have been a suboptimal approach to identify the true prevalence of low SMI and the relationship between low SMI and survival within this cohort. We acknowledge the importance of differences in body composition between countries. For example, the Dutch population, on average, is taller and the prevalence of overweight and obesity is lower compared with the Canadian population.3,4 Although a large percentage of the Canadian population consists of (European) immigrants,5 we agree that it would be better to compare our study data with normative values derived from a European, or even a Dutch, population. Although we did find a new publication with cutoff values for an Asian population,6 normative data for a European population are not available yet.

n consists of (European) immigrants,5 we agree that it would be better to compare our study data with normative values derived from a European, or even a Dutch, population. Although we did find a new publication with cutoff values for an Asian population,6 normative data for a European population are not available yet. There are several options to consider to overcome the question of ethnic variation in body composition in the near future. Data on body composition measured with computed tomography scans from recent European studies7-10 could be pooled to build a database with reference values for the European population. Another approach is to derive normative values from a healthy population, which is what our group is working on at the moment. It would then be interesting to repeat the statistical analyses of our study and to investigate whether our population truly displayed a low skeletal muscle index compared with European reference values. Only then we will be able to understand why skeletal muscle index was not associated with survival in our cohort and whether this may have been caused by choosing the wrong reference group. In the meantime, while we await reference values for different countries and/or ethnic groups, we recommend that future studies on body composition display patient characteristics with regard to ethnicity, especially when cutoff values or reference values are being used. This does not apply to Europe alone, but also to other regions across the world.

reference values for different countries and/or ethnic groups, we recommend that future studies on body composition display patient characteristics with regard to ethnicity, especially when cutoff values or reference values are being used. This does not apply to Europe alone, but also to other regions across the world. AUTHORS’ DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST Reply to L.E. Daly et al The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or jco.ascopubs.org/site/ifc. Susanne Blauwhoff-Buskermolen No relationship to disclose Marian A.E. de van der Schueren Speakers' Bureau: Nutricia Jacqueline A.E. Langius Travel, Accommodations, Expenses: Nutricia Henk M.W. Verheul Consulting or Advisory Role: Boehringer Ingelheim (Inst) Research Funding: Amgen (Inst)

To the Editor: Leyland-Jones et al1 conducted an open-label, noninferiority study to evaluate the impact of epoetin alfa (EPO) on tumor outcomes when used to treat anemia in patients who received chemotherapy for metastatic breast cancer. The primary end point was progression-free survival (PFS) on the basis of the investigators’ assessments. The study was designed on the basis of the difference between groups as measured by the hazard ratio (HR; EPO v best standard care [BSC]), with a noninferiority margin of 1.15. A total of 1,650 PFS events would provide > 80% power with a one-sided type I error of 0.025 to rule out a 15% HR increase. The study was conducted from March 2006 to July 2014, and at the end of study, 1,659 events had been observed. Estimated HR was 1.089, with a 95% CI of 0.988 to 1.200. The observed upper bound exceeded the prespecified noninferiority margin of 1.15. The authors concluded that “Overall, this study did not achieve the noninferiority objective in ruling out a 15% increased risk in PD or death.”1

dy, 1,659 events had been observed. Estimated HR was 1.089, with a 95% CI of 0.988 to 1.200. The observed upper bound exceeded the prespecified noninferiority margin of 1.15. The authors concluded that “Overall, this study did not achieve the noninferiority objective in ruling out a 15% increased risk in PD or death.”1 HR is a ratio of two hazard functions over time. Hazard, which is not a probability measure, is commonly misinterpreted as a risk of an event of interest. The observed upper bound, 1.20, of the above 95% CI does not mean that EPO has a 20% risk of increase versus BSC. In fact, it is difficult to interpret the HR in clinically meaningful terms without a hazard function estimate available from BSC. The hazard function, by itself, is difficult to estimate well without a model and difficult to interpret clinically. This issue has been extensively discussed in the clinical and statistical literature, especially for evaluating the safety of a drug or device.2-4 The summary measure using HR for this rather lengthy study does not help us to assess the value of EPO under a risk–benefit perspective.

d difficult to interpret clinically. This issue has been extensively discussed in the clinical and statistical literature, especially for evaluating the safety of a drug or device.2-4 The summary measure using HR for this rather lengthy study does not help us to assess the value of EPO under a risk–benefit perspective. An alternative is to use the restricted mean survival time (RMST) as the summary measure to quantify the group difference.2-4 For the present case, survival means PFS. Although the patient-level observations from the study by Leyland-Jones et al1 are not publicly available, we used a well-established computer algorithm to scan the Kaplan-Meier (KM) curves presented in their Figure 2A and reconstructed the observed individual times to progression and/or death.5 The resulting KM curves and HR estimates with these reconstructed observations are closely matched with the original counterparts reported in the article. With these data, an estimated RMST for PFS ≤ 48 months for EPO is the area under the KM curve in Figure 2A by 48 months, which is 9.9 months. That is, future patients who receive EPO with 48 months of follow-up would achieve a PFS of an average of 9.9 months. For BSC, the RMST estimate is 11.4 months. The difference (BSC − EPO) is 1.5 months (95% CI, 0.5 to 2.6; P < .004) in favor of BSC. This difference, coupled with an RMST of 11.4 months for BSC, provides a clinically meaningful interpretation. In any event, when quantifying a group difference with a summary measure, it is informative to have a reference value from the control arm for decision making to assess the benefit and safety profile of a treatment strategy.

, coupled with an RMST of 11.4 months for BSC, provides a clinically meaningful interpretation. In any event, when quantifying a group difference with a summary measure, it is informative to have a reference value from the control arm for decision making to assess the benefit and safety profile of a treatment strategy. There is an ongoing randomized phase III study of darbepoetin versus BSC (NT00858364), for anemia secondary to platinum-based treatment of stage IV non–small-cell lung cancer. We hope that the investigators of the study would consider a sensitivity analysis using the RMST summary measure to further inform the benefit–risk profile of erythropoietin-stimulating agents in the oncology setting. For future patients’ treatment, we may need more information beyond presenting an overall summary measure for the treatment difference. For such a relatively large study as the present one by Leyland-Jones et al,1 it would be important to use information from the patient’s baseline variables to identify a subgroup, if any, of patients who would not have safety concerns, but would benefit from EPO.6

overall summary measure for the treatment difference. For such a relatively large study as the present one by Leyland-Jones et al,1 it would be important to use information from the patient’s baseline variables to identify a subgroup, if any, of patients who would not have safety concerns, but would benefit from EPO.6 AUTHORS’ DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST How to Summarize the Safety Profile of Epoetin Alfa Versus Best Standard of Care in Anemic Patients With Metastatic Breast Cancer Receiving Standard Chemotherapy? The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or jco.ascopubs.org/site/ifc. Takahiro Hasegawa Employment: Shionogi & Co, Ltd Stock or Other Ownership: Shionogi & Co, Ltd Hajime Uno Consulting or Advisory Role: IONIS Lee-Jen Wei Other Relationship: Served as a member of the independent data monitoring committee of this trial

To the Editor: In the study by Leyland-Jones et al,1 we question whether the protocol resulted in adequate iron supplementation in both the epoetin alfa (EPO) group and the best standard of care (BSC) group, but especially in the group that was randomly assigned to receive EPO 40,000 IU once per week. The study design recognized that the availability of iron is a limiting factor in the response to EPO and suggested that, “Patients with transferrin saturation value of less than 20% were to receive iron supplementation”.1 There did not seem to be a requirement for iron supplementation after enrollment. We feel the evaluation for functional iron deficiency was inadequate. For patients treated with iron because of baseline transferrin < 20%, iron parameters were remeasured in 3 weeks, which delayed initialization of intravenous iron therapy unnecessarily in patients for whom enteric iron was prescribed. For patients with transferrin > 20%, iron parameters were not remeasured for 6 weeks after random assignment. The authors do not provide the results of iron studies at baseline or during and at the end of study; therefore, we are unable to judge the degree of iron deficiency and functional iron deficiency in the two groups.

atients with transferrin > 20%, iron parameters were not remeasured for 6 weeks after random assignment. The authors do not provide the results of iron studies at baseline or during and at the end of study; therefore, we are unable to judge the degree of iron deficiency and functional iron deficiency in the two groups. That iron and especially intravenous iron was underused in the EPO group is clear from the rate of both enteric and intravenous iron therapy. Oral iron therapy, rather than intravenous iron, was used most frequently (49% of the EPO group and 56% of the BSC group). Intravenous iron use was nearly identical in the two groups (8% in the EPO group and 9% in the BSC group), yet the increased frequency of functional iron deficiency with use of EPO is well known.2 The benefit of iron administration in patients with metastatic breast cancer, even when done primarily via oral administration, may explain why, despite chemotherapy, the hemoglobin in both groups rose during the duration of the study from approximately 10.2 g/dL to a median achieved hemoglobin of 11.6 g/dL in the EPO group and 10.9 g/dL in the BSC group. One should also acknowledge that transfusion in the BSC group provides intravenous iron, as donor cells are broken down and heme iron is recycled and used.

ups rose during the duration of the study from approximately 10.2 g/dL to a median achieved hemoglobin of 11.6 g/dL in the EPO group and 10.9 g/dL in the BSC group. One should also acknowledge that transfusion in the BSC group provides intravenous iron, as donor cells are broken down and heme iron is recycled and used. Underuse of intravenous iron may have contributed to the higher rate of thrombotic vascular events (TVEs) in the EPO group.3,4 EPO can cause functional iron deficiency. Failure to use intravenous iron when administering EPO results in an increased rate of nonresponders and a requirement for higher doses of EPO.5,6 Iron deficiency as well as functional iron deficiency is associated with thrombocytosis and increased rates of arterial and venous thrombosis.7 Higher target hemoglobin levels are also associated with an increased risk of TVE.8 Red cell transfusion is associated with an increased risk of thrombosis.9 All patients in the EPO group were treated with EPO to a target hemoglobin of 12 g/dL. Had the BSC group been treated with transfusion to the same hemoglobin target, a higher percentage of patients would have been exposed to the thrombotic risk of red cell transfusion. Risk of TVE in the BSC group is underestimated because a lower target hemoglobin was used. In this study, we think the target hemoglobin in the EPO and BSC groups should have been the same to compare TVE as an adverse outcome in the two groups.

nts would have been exposed to the thrombotic risk of red cell transfusion. Risk of TVE in the BSC group is underestimated because a lower target hemoglobin was used. In this study, we think the target hemoglobin in the EPO and BSC groups should have been the same to compare TVE as an adverse outcome in the two groups. We disagree with the authors’ conclusions that RBC transfusion should be the preferred approach for the management of anemia during first- or second-line chemotherapy for metastatic breast cancer. This recommendation does not take into account the risks of allogeneic red cell transfusion that may be missed in a small study.10 The independent review committee–determined primary outcome of progression-free survival met the study criteria of noninferiority. Finally, the difference in TVE as a secondary outcome is likely explained by the higher target hemoglobin in the EPO group, combined with suboptimal iron replacement and EPO-induced functional iron deficiency. AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST Critical Role of Iron in Epoetin Alfa Treatment of Chemotherapy-Associated Anemia The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or jco.ascopubs.org/site/ifc. Irwin Gross Employment: Accumen

AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST Critical Role of Iron in Epoetin Alfa Treatment of Chemotherapy-Associated Anemia The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or jco.ascopubs.org/site/ifc. Irwin Gross Employment: Accumen Honoraria: AMAG Pharmaceuticals Shannon Farmer Honoraria: Thieme, Elsevier Travel, Accomodations, Expenses: National Blood Authority (Australia) Other Relationship: Investigator in a Government grant-funded clinical trial investigating the use of intravenous iron in the critically ill Axel Hofmann Honoraria: Vifor Pharma Speakers' Bureau: Vifor Pharma, TEM Research Funding: CSL Behring Travel, Accommodations, Expenses: Vifor Pharma, TEM, CSL Behring Sherri Ozawa Consulting or Advisory Role: Specialty Care Travel, Accommodations, Expenses: Specialty Care Aryeh Shander Honoraria: Vifor Pharma Consulting or Advisory Role: Vifor Pharma Travel, Accommodations, Expenses: Vifor Pharma Matti Aapro Honoraria: Amgen Consulting or Advisory Role: Helsinn Healthcare, Teva Pharmaceuticals, Hospira, Merck, Sandoz, Pierre Fabre Medicament, Vifor Pharma, Tesaro, Pfizer

Travel, Accommodations, Expenses: Specialty Care Aryeh Shander Honoraria: Vifor Pharma Consulting or Advisory Role: Vifor Pharma Travel, Accommodations, Expenses: Vifor Pharma Matti Aapro Honoraria: Amgen Consulting or Advisory Role: Helsinn Healthcare, Teva Pharmaceuticals, Hospira, Merck, Sandoz, Pierre Fabre Medicament, Vifor Pharma, Tesaro, Pfizer Speakers' Bureau: Amgen, Helsinn Healthcare, Teva Pharmaceuticals, Novartis, Roche, Johnson & Johnson, Hospira, Pfizer, Sandoz, Pierre Fabre Medicament, Vifor Pharma, Tesaro, Kyowa Kirin, Taiho Pharmaceutical, Ono Pharmaceutical Research Funding: Helsinn Healthcare (Inst), Sandoz, Hospira, Novartis (Inst), Pierre Fabre Medicament (Inst), Novartis Expert Testimony: Amgen

Zhang,1 Fornaro et al,2 and Lee et al3 have posed several questions and suggestions regarding our study4 on apatinib in chemotherapy-refractory gastric cancer. According to the concept offered by the International Conference on Harmonisation’s E9 guidelines5 and the China Food and Drug Administration,6 it is acceptable to exclude from the full analysis set (FAS) any patient who did not receive at least one dose of trial medication after random assignment. Therefore, we eliminated from the FAS six patients who did not take any trial medication. This is the same approach used in a phase III trial7 in which 162 patients were randomly assigned and three were eliminated from the FAS as a result of not taking any trial medication. Of the patients in our study with gastric or gastroesophageal junction adenocarcinoma, 70% had experienced gastrectomy and 75% were male. Apatinib dosage in our phase III trial was based mainly on the dosage of the previous phase II trial of apatinib in gastric cancer.8 The phase II trial of apatinib in breast cancer, however, was an exploratory study, including dose exploring, and all patients were female.9 Use of a medicine may vary in different studies and indications, such as in the two trials of bevacizumab for breast cancer and colorectal cancer.10,11 In addition, the tolerance dose may differ between men and women patients, although this requires further observation.

ng dose exploring, and all patients were female.9 Use of a medicine may vary in different studies and indications, such as in the two trials of bevacizumab for breast cancer and colorectal cancer.10,11 In addition, the tolerance dose may differ between men and women patients, although this requires further observation. In our trial, total dosages in cycles 1, 2, and 3 of the apatinib group were 21,117.05 mg, 20,050.00 mg, and 20,371.38 mg, respectively, and the average doses were 754.2 mg, 716.0 mg, and 730.1 mg, respectively. No treatment-related death was observed throughout the trial. We appreciate the opportunity to respond to the quality-of-life (QoL) comments mentioned by Zhang.1 In our trial report, at the end of the third cycle, rates of compliance for responding to the QoL questionnaire were 34.7% in the apatinib group and 7.7% in the placebo group. It was suggested that treatment with apatinib may have an effect against the deterioration of patient QoL. Although there were some Eastern Cooperative Oncology Group performance status differences in two groups at baseline, it was not statistically significant for a randomized, double-blind trial. Thus, this would not influence the overall survival significantly.

We appreciate the opportunity to respond to the quality-of-life (QoL) comments mentioned by Zhang.1 In our trial report, at the end of the third cycle, rates of compliance for responding to the QoL questionnaire were 34.7% in the apatinib group and 7.7% in the placebo group. It was suggested that treatment with apatinib may have an effect against the deterioration of patient QoL. Although there were some Eastern Cooperative Oncology Group performance status differences in two groups at baseline, it was not statistically significant for a randomized, double-blind trial. Thus, this would not influence the overall survival significantly. The results of our study reported that grade 3 to 4 proteinuria and hypertension occurred in 2.3% and 4.5% of patients, respectively, in the apatinib group. In the REGARD study mentioned by Fornaro et al,2 grade 3 to 4 proteinuria and hypertension developed in 4% and 8% of patients, respectively, in the ramucirumab group.12 Generally, proteinuria and hypertension are recognized as the main characteristic adverse event (AE) of antiangiogenesis agents and have high clinical risk. The RAINBOW13 trial focused on European and American populations; however, gastric carcinoma in China tends toward younger patients, that is, people age 45 to 64 years have the highest incidence.14 Hence, it is well founded that we enrolled patients age < 70 years in our trial.

The results of our study reported that grade 3 to 4 proteinuria and hypertension occurred in 2.3% and 4.5% of patients, respectively, in the apatinib group. In the REGARD study mentioned by Fornaro et al,2 grade 3 to 4 proteinuria and hypertension developed in 4% and 8% of patients, respectively, in the ramucirumab group.12 Generally, proteinuria and hypertension are recognized as the main characteristic adverse event (AE) of antiangiogenesis agents and have high clinical risk. The RAINBOW13 trial focused on European and American populations; however, gastric carcinoma in China tends toward younger patients, that is, people age 45 to 64 years have the highest incidence.14 Hence, it is well founded that we enrolled patients age < 70 years in our trial. Several experimental studies and clinical trials about the efficacy and safety of apatinib combined with chemotherapy are ongoing. Preliminary data have showed synergistic effects of combination therapy and unchanged adverse drug reaction profiles. Cardiotoxicity-related AEs were observed, reported, and analyzed exactly in this trial. Cardiac toxicity of apatinib was atypical and most AEs were mild or moderate. There was no statistically significant difference in cardiac toxicity between the apatinib and placebo groups.

Several experimental studies and clinical trials about the efficacy and safety of apatinib combined with chemotherapy are ongoing. Preliminary data have showed synergistic effects of combination therapy and unchanged adverse drug reaction profiles. Cardiotoxicity-related AEs were observed, reported, and analyzed exactly in this trial. Cardiac toxicity of apatinib was atypical and most AEs were mild or moderate. There was no statistically significant difference in cardiac toxicity between the apatinib and placebo groups. The relationship between the specific AEs and the efficiency of apatinib has been noted, as Lee et al3 write. In our trial, overall survival for patients who had hypertension, proteinuria, and hand-foot skin reaction was greater than that of patients who did not experience those AEs. These specific AEs could be considered surrogate clinical biomarkers of drug activity. Relevant data analysis will be published soon. Apatinib is a small molecular and multiple-target tyrosine kinase inhibitor, whereas ramucirumab is a large molecular and humanized IgG1 monoclonal antibody. They both mainly target vascular endothelial growth factor receptor, but with different mechanisms. Ramucirumab has not yet been approved for use in China. We look forward to proceeding with head-to-head studies of the two drugs in the future.

eas ramucirumab is a large molecular and humanized IgG1 monoclonal antibody. They both mainly target vascular endothelial growth factor receptor, but with different mechanisms. Ramucirumab has not yet been approved for use in China. We look forward to proceeding with head-to-head studies of the two drugs in the future. AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST Reply to S. Zhang, L. Fornaro et al, and H.J. Lee et al The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or jco.ascopubs.org/site/ifc. Shukui Qin No relationship to disclose Jin Li Research Funding: Merck Serono (Inst), Amgen (Inst)

To the Editor: The article by Blauwhoff-Buskermolen et al1 highlights changes in body composition that occur during chemotherapy in patients with metastatic colorectal cancer. Literature on the impact of body composition, as defined by specific computed tomography criteria, in cancer management is evolving and, in our view, has the potential to become a valuable clinical biomarker across an array of cancers, with the ability to predict toxicity from systemic therapy as well as overall outcome. Altered body composition is common in many cancers. Studies that have evaluated body composition changes during anticancer treatment by using single-slice computed tomography images of the abdominal region (L3) are becoming more frequent and have focused primarily on the prognostic significance of changes in muscle mass.2,3 The current study by Blauwhoff-Buskermolen et al1 focuses on the prognostic role of loss of muscle mass during chemotherapy in patients with metastatic colorectal cancer. The authors report that patients who experienced a loss of muscle mass > 9% (lowest tertile) had significantly lower survival rates than did those who experienced a loss of < 9%, which remained significant after controlling for important prognostic covariates (hazard ratio, 4.47; 95% CI, 2.21 to 9.05; P < .001).1

ncer. The authors report that patients who experienced a loss of muscle mass > 9% (lowest tertile) had significantly lower survival rates than did those who experienced a loss of < 9%, which remained significant after controlling for important prognostic covariates (hazard ratio, 4.47; 95% CI, 2.21 to 9.05; P < .001).1 In the current study, low skeletal muscle index (SMI; skeletal muscle area at L3/height [m2]) at baseline was not associated with reduced survival, which contrasts with some,4 but not all, research findings.2,5,6 In the literature today, the most commonly used cut points for the definition of low SMI (or sarcopenia) using body composition from computed tomography scanning are those published by both Martin et al,4 who defined low SMI of < 43 cm2/m2 for men with body mass index < 25 cm2/m2, < 53 cm2/m2 for men with body mass index ≥ 25 cm2/m2, and < 41 cm2/m2 for women to be prognostic of reduced survival in a large cohort of 1,473 patients with lung and GI cancer; and by Prado et al,7 who defined low SMI of < 52.4 cm2/m2 for men and < 38.5 cm2/m2 for women as a predictor of poor overall survival in a cohort of 250 obese patients with lung and GI cancer. Both sets of cut points have been derived by using optimal stratification analysis in a North American population, and their use in other ethnicities has been questioned.8 This is largely in response to significant differences in skeletal muscle mass and in the rates of muscle loss with age that have been observed among different ethnic groups—African Americans, Whites, Hispanics, and Asians.8 Ethnic variation in cancer survival is not new, for example, East meets West in gastric cancer,9 and toxicity from chemotherapy can also vary depending on the population studied, for example, regional differences for the tolerability of fluoropyrimidines.10

hnic groups—African Americans, Whites, Hispanics, and Asians.8 Ethnic variation in cancer survival is not new, for example, East meets West in gastric cancer,9 and toxicity from chemotherapy can also vary depending on the population studied, for example, regional differences for the tolerability of fluoropyrimidines.10 Within their study, Blauwhoff-Buskermolen et al1 defined low SMI by using the cut points established by Martin et al.4 Extrapolating such cut points to a cohort of Dutch patients may have been a suboptimal approach to identify the true prevalence of low SMI and the relationship between low SMI and survival within this cohort. This may identify why changes in skeletal muscle area in this study was predictive of reduced survival, whereas low SMI at a specific time point was not. Although these North American–derived cut points have been widely applied in studies that have examined the clinical implications of low SMI, the validity of these cut points in a large European cohort has not been examined. Several of the published studies, which have reported nonsignificant relationships between low baseline SMI and reduced survival have been from European studies2,5 that have used the North American–derived cut points.

cations of low SMI, the validity of these cut points in a large European cohort has not been examined. Several of the published studies, which have reported nonsignificant relationships between low baseline SMI and reduced survival have been from European studies2,5 that have used the North American–derived cut points. Further large-scale investigations are warranted in European populations, where cut points for low SMI are devised by using optimal stratification to determine its prognostic value within this population. This would provide cut points that have been validated to predict survival in large cohorts of European patients with cancer, and would not rely on cut points previously established in populations that are not representative of those being studied. In our view, the utility of body composition analysis could be even greater if ethnic variation is accounted for. AUTHORS’ DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST Response to “Loss of Muscle Mass During Chemotherapy Is Predictive for Poor Survival of Patients With Metastatic Colorectal Cancer” The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or jco.ascopubs.org/site/ifc. Louise E. Daly No relationship to disclose Aoife M. Ryan No relationship to disclose

AUTHORS’ DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST Response to “Loss of Muscle Mass During Chemotherapy Is Predictive for Poor Survival of Patients With Metastatic Colorectal Cancer” The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or jco.ascopubs.org/site/ifc. Louise E. Daly No relationship to disclose Aoife M. Ryan No relationship to disclose Derek G. Power No relationship to disclose

To the Editor: I have concerns about the reported results in the recent apatinib trial by Li et al1 in Chinese patients with gastric cancer. In superiority trials, intention-to-treat (ITT) analyses that include all patients according to the allocated treatment are standard. This can lead to a conservative estimate of effect size and avoid bias associated with nonrandom loss of participants; however, excluding patients who did not start treatment or who stopped treatment during the trial can lead to biased estimates. This ITT analysis has been required by CONSORT statement, and the pitfalls of not performing ITT analysis have been documented previously.2 In the current study, five patients in the experimental arm and one patient in the placebo arm were excluded from the ITT analyses, which may exaggerate the modest survival benefit observed.1

alysis has been required by CONSORT statement, and the pitfalls of not performing ITT analysis have been documented previously.2 In the current study, five patients in the experimental arm and one patient in the placebo arm were excluded from the ITT analyses, which may exaggerate the modest survival benefit observed.1 Toxicity is one crucial aspect of clinical trial reporting. In a previous phase II trial that used apatinib for treatment of 25 patients with breast cancer, the dose of 750 mg once per day resulted in substantial toxicities: a dose delay of at least one cycle with dose reduction occurred in 84% of patients. Almost all patients experienced grade ≥ 3 toxicity, and treatment-related death occurred in two patients.3 Because of the heavy toxicity of apatinib, the dose was reduced to 500 mg once per day in further studies. In the current study, Li et al1 concluded that a dose of 850 mg once per day is tolerable and acceptable. Reasons for this discrepancy are not clear; however, even in the current trial, of the 40 patients that discontinued apatinib treatment, 22 patients (55%) stopped treatment as a result of toxicity, and dose reduction occurred in 21% patients who finished apatinib treatment.1 What is the total dose for patients who receive treatment with apatinib? What is the mean dose for these patients? Is there any treatment-related death? Because apatinib has been approved by the China Food and Drug Administration on the basis of the results of this trial and apatinib will now become commercially available in China, detailed clarification and explanation of the toxicity data remain essential for patient safety and evaluation of benefit versus risk, although the toxicity profile cannot be directly compared between different trials.

the basis of the results of this trial and apatinib will now become commercially available in China, detailed clarification and explanation of the toxicity data remain essential for patient safety and evaluation of benefit versus risk, although the toxicity profile cannot be directly compared between different trials. One good way to evaluate the efficacy and safety burden is quality-of-life data; however, the rate of compliance for responding to quality-of-life questionnaires in the apatinib group was surprising low (7.7%).1 This may also undermine the validity of the corresponding results. Therefore, I question the conclusion of the phase III study of apatinib in patients with gastric cancer by Li et al.1 AUTHOR’S DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST Problematic Analysis and Inadequate Toxicity Data in Phase III Apatinib Trial in Gastric Cancer The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or jco.ascopubs.org/site/ifc. Sheng Zhang No relationship to disclose

To the Editor: Li et al1 have recently reported the results of a phase III trial that evaluated apatinib treatment in patients with advanced gastric cancer who experienced disease progression after two or more lines of systemic therapy. The authors compared apatinib—a tyrosine kinase inhibitor that targets the vascular endothelial growth factor receptor-2 (VEGFR-2)—with placebo among 273 randomly assigned patients and reported an overall survival advantage in favor of the experimental treatment. This study represents an important step forward in gastric cancer research, as it confirms the value of targeting VEGFR-2 beyond the already available evidence that supports the use of ramucirumab, an anti-VEGFR-2 antibody, in the second-line setting.2,3 However, there are several potential pitfalls that may limit the value of these results. With regard to patient characteristics, an imbalance in performance status (PS) distribution is evident, with an excess of patients with a PS of 0 in the experimental arm compared with placebo (27.3% v 16.5%). The authors conclude that this difference does not formally reach statistical significance, but, in our opinion, this could have had an unpredictable impact on overall survival. Indeed, PS is one of the main determinants of survival in advanced gastric cancer from first-line to third-line setting,4-6 and even a seemingly mild deterioration in the general condition, that is, a PS of 1 according to the Eastern Cooperative Oncology Group scale, might impair outcomes in such a fragile patient scenario.

d, PS is one of the main determinants of survival in advanced gastric cancer from first-line to third-line setting,4-6 and even a seemingly mild deterioration in the general condition, that is, a PS of 1 according to the Eastern Cooperative Oncology Group scale, might impair outcomes in such a fragile patient scenario. In terms of safety, apatinib resulted in a not negligible rate of grade 3 to 4 hand-foot syndrome (8.5%), with approximately one of two patients experiencing proteinuria (generally grade 1 to 2) and 5.7% of patients experiencing grade 3 to 4 neutropenia. Is it really reasonable to conclude that apatinib “has a favorable safety profile in comparison with other antiangiogenic agents?”1(p1452) In the larger REGARD trial, single-agent ramucirumab resulted in a much lower rate of proteinuria (all grade, 4%), and grade 3 to 4 neutropenia and hand-foot syndrome were not reported.2 Whereas these adverse events are manageable when apatinib is administered as monotherapy, they may not suit well when apatinib is combined with cytotoxic agents used in gastric cancer, such as fluoropyrimidines and platinum compounds. With regard to cardiac safety, initial data with apatinib seems reassuring; however, comparisons with sunitinib and bevacizumab—as reported in the Discussion of the article by Li et al1 and for which we have greater knowledge about potentially severe cardiovascular events—seem at least premature. This is particularly true if we keep in mind that elderly patients (age > 70 years) were excluded from the trial, the number of patients between age 65 and 70 years was limited (37 patients), and the median age in the two arms (age 58 years) is lower than that observed in routine practice.1 In the RAINBOW trial with ramucirumab, despite a similar median age in the enrolled patient cohorts, the upper age limit was 84 years and the number of patients age ≥ 65 years was greater (249 patients). Neither in the REGARD trial7 nor in the RAINBOW trial8 did the age subgroup analyses report increased toxicity among older patients, with the exception of a relatively higher rate of grade ≥ 3 neutropenia in the RAINBOW trial.

age limit was 84 years and the number of patients age ≥ 65 years was greater (249 patients). Neither in the REGARD trial7 nor in the RAINBOW trial8 did the age subgroup analyses report increased toxicity among older patients, with the exception of a relatively higher rate of grade ≥ 3 neutropenia in the RAINBOW trial. Finally, no significant improvement in quality of life was noted for apatinib.1 Quality-of-life data for ramucirumab monotherapy have been recently published; ramucirumab proved to delay the worsening of symptoms and the deterioration of PS.9 It should be recognized that the population enrolled in the apatinib trial had a far more pretreated disease, but patient characteristics of PS, age, and number of metastatic sites seem comparable between trials.1,2 With an arguably increasing number of patients receiving ramucirumab in the second-line setting, it is difficult to anticipate the potential impact of apatinib in routine clinical practice. As these agents share the same target along the pathways that regulate tumor angiogenesis, the efficacy of apatinib in overcoming resistance to ramucirumab is unclear. Moreover, results with apatinib have been reported among Chinese patients. As a result of differences in the biologic background and treatment patterns between Eastern and Western patients with gastric cancer,10 other studies should be awaited to confirm these data outside Asia.

ng resistance to ramucirumab is unclear. Moreover, results with apatinib have been reported among Chinese patients. As a result of differences in the biologic background and treatment patterns between Eastern and Western patients with gastric cancer,10 other studies should be awaited to confirm these data outside Asia. To conclude, apatinib is another arrow in the bow against gastric cancer. Future trials and, we hope, the identification of predictive biomarkers will ultimately tell us how far we can hurl our hope in this challenging struggle. AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST Apatinib in Advanced Gastric Cancer: A Doubtful Step Forward The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or jco.ascopubs.org/site/ifc. Lorenzo Fornaro No relationship to disclose Enrico Vasile No relationship to disclose Alfredo Falcone No relationship to disclose

We have several responses to the correspondence by Gross et al1 on the critical role of iron in our study.2 In contrast to what the authors stated, iron parameters at baseline and iron use during the study were provided in the Data Supplement,2 and use of iron during the study is adequately described in the protocol (NT00338286 and EudraCT Number 005-001817-17). Patients in both treatment arms with transferrin < 20% were to be considered to have functional iron deficiency and were to receive iron therapy, preferably by intravenous administration. The recommended iron therapy was in agreement with the European guidelines for treatment of renal anemia at the start of the study3 and aligned with the European Organisation for Research and Treatment of Cancer guidelines on iron supplementation in patients with cancer who received chemotherapy and erythropoiesis-stimulating agents (ESAs).4 The recommendation of iron use in this study is the same as in another metastatic breast cancer study with ESAs, the Breast Cancer-Anemia and the Value of Erythropoietin study.5 This study reported concomitant iron supplementation of 85% in the epoetin beta group and 80% in the control group. Of note, in the Breast Cancer Erythropoietin Survival Trial (BEST) study,6 a daily dose of elemental iron (oral) 200 mg was recommended when transferrin was < 20%; only 1.8% of the total population received iron supplementation.

omitant iron supplementation of 85% in the epoetin beta group and 80% in the control group. Of note, in the Breast Cancer Erythropoietin Survival Trial (BEST) study,6 a daily dose of elemental iron (oral) 200 mg was recommended when transferrin was < 20%; only 1.8% of the total population received iron supplementation. Iron therapy in patients with cancer has often been contentious because of its potential contribution to tumor initiation and tumor growth, its role in the tumor microenvironment and metastasis, and the changes in the uptake and management of iron in patients with cancer.7 Particularly in breast cancer, almost 50% of all genes that are involved in the regulation or maintenance of iron metabolism were significantly associated with clinical outcome.8 We believe these findings may contribute to the cautious use of iron supplementation in patients with cancer. The comment by Gross et al1 regarding thrombotic vascular events and target hemoglobin (Hb) is interesting. As RBC transfusion guidelines with respect to initiation and/or maintenance Hb are quite different from the guidelines for anemia management with ESAs in patients with cancer, the Hb reached in the best standard care (BSC) group is obviously lower compared with the epoetin alfa (EPO) group. Whether this may potentially bias the reporting of thrombotic vascular events in both groups remains hypothetical.

te different from the guidelines for anemia management with ESAs in patients with cancer, the Hb reached in the best standard care (BSC) group is obviously lower compared with the epoetin alfa (EPO) group. Whether this may potentially bias the reporting of thrombotic vascular events in both groups remains hypothetical. Hasegawa et al9 provide an alternative approach to analyzing and interpreting the data presented in our study.2 The study was designed to show noninferiority of EPO to BSC in terms of progression-free survival (PFS) in patients with metastatic breast cancer. The prespecified noninferiority margin, as hazard ratio (HR; EPO v BSC), was 1.15. The final analysis of the study was based on a total of 2,098 patients and 1,659 PFS events. Median PFS was 7.4 months in both treatment groups and HR was 1.089 with a 95% CI of 0.988 to 1.200. By using restricted mean survival time (RMST) approach, Hasegawa et al9 showed that, with up to 48 months of follow-up, the EPO group had an average PFS of 9.9 months compared with 11.4 months in the BSC group. The difference of the two averages was 1.5 months, with a 95% CI of 0.5 to 2.6. This analysis was based on individual-level data that were reconstructed from the published results.

showed that, with up to 48 months of follow-up, the EPO group had an average PFS of 9.9 months compared with 11.4 months in the BSC group. The difference of the two averages was 1.5 months, with a 95% CI of 0.5 to 2.6. This analysis was based on individual-level data that were reconstructed from the published results. We have conducted a similar post hoc analysis on the basis of original patient-level data up to a follow-up time of 75 months. Our results showed that the EPO group had an average PFS of 10.1 months compared with 11.6 months in the BSC group. The difference of the two averages was 1.6 months, with a 95% CI of 0.45 to 2.66. Our analysis and that of Hasegawa et al9 have closely matched results, although conducted independently with different data sources and different follow-up times.

d an average PFS of 10.1 months compared with 11.6 months in the BSC group. The difference of the two averages was 1.6 months, with a 95% CI of 0.45 to 2.66. Our analysis and that of Hasegawa et al9 have closely matched results, although conducted independently with different data sources and different follow-up times. On the basis of HR, there was an observed 9% increase in risk of progression or death for the EPO group compared with BSC, and the study did not achieve the noninferiority objective per the prespecified noninferiority margin of 1.15. As Hasegawa et al9 rightly pointed out, these are not probability assessments and they lack intuitive clinical interpretations. Results based on the RMST analysis by Hasegawa et al9 showed an observed absolute 1.5 months, or 13% (1.5 divided by 11.4), relative decrease in average time of PFS for EPO compared with BSC. The study noninferiority objective cannot be evaluated on the basis of this result as there was no prespecified noninferiority margin for analysis with RMST, but the outcome has statistically ruled out a decrease in average time of PFS > 2.6 months or < 0.5 months. HR, under specific assumptions in the underlying distribution, can be considered the inverse of the median time ratio of the two pertinent treatment groups. However, these assumptions are more often than not unsatisfied, and the median as a single cross-sectional measure by no means reflects the entire time-to-event trajectory.

specific assumptions in the underlying distribution, can be considered the inverse of the median time ratio of the two pertinent treatment groups. However, these assumptions are more often than not unsatisfied, and the median as a single cross-sectional measure by no means reflects the entire time-to-event trajectory. The most notable advantages of the RSMT approach are that the validity of the method is independent of any assumptions regarding the underlying distribution and that the clinical interpretation of the results is easy and intuitive. Determining the follow-up cutoff requires careful planning at the study design stage as it could introduce bias in favor of or against one treatment arm versus the other. For a noninferiority study, determination and justification for the noninferiority margin in terms of RMST may not be an easier matter even though the method has a more intuitive appeal.

ires careful planning at the study design stage as it could introduce bias in favor of or against one treatment arm versus the other. For a noninferiority study, determination and justification for the noninferiority margin in terms of RMST may not be an easier matter even though the method has a more intuitive appeal. With regard to identification of a subgroup of patients that would not have the safety concerns,9 our study2 included analyses for eight prespecified subgroups defined by demographic and baseline characteristics (age, body mass index, Eastern Cooperative Oncology Group, line of chemotherapy, human epidermal growth factor receptor 2 [HER2]/neu status, time from initial diagnosis to metastatic disease, type of prior adjuvant therapy, and Hb level) and three additional subgroups based on hormone receptors and HER2 status (HER2 positive, hormone receptor positive or HER2 negative, and triple negative). Results from these subgroup analyses were consistent, in general, with that of the overall population. AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST Reply to T. Hasegawa et al and I. Gross et al The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or jco.ascopubs.org/site/ifc.

s are considered compensated. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or jco.ascopubs.org/site/ifc. Brian Leyland-Jones Honoraria: Janssen Research & Development Consulting or Advisory Role: Genentech Speakers' Bureau: Genentech Travel, Accommodations, Expenses: Johnson & Johnson Els Vercammen Employment: Janssen Research & Development Stock or Other Ownership: Johnson & Johnson Liang Xiu Employment: Janssen Research & Development Stock or Other Ownership: Johnson & Johnson

The current oncology drug development landscape is dominated by efforts to create therapies that are mechanistically designed to improve outcomes for patients with cancers that harbor specific molecular aberrations, which often occur across a variety of tumor types. In the evaluation of targeted therapies, basket trials have emerged as an approach to test the hypothesis that targeted therapies may be effective independent of tumor histology, as long as the molecular target is present.1 However, the term basket has been applied broadly, and there is little uniformity in the design or goals of these trials. Furthermore, the scientific goals frequently are not specified with the precision conventionally used for clinical trials, leading to some difficulties in design and interpretation. For instance, many investigative teams use the popular Simon two-stage design, independently in each basket, thus effectively treating the trial overall as a series of independent phase II clinical trials. However, the actual goals are typically more complex than those of simple phase II clinical trials of new agents. In this commentary, we present an overview of the various trials described as basket trials, clarify the distinctive goals that basket trials seek to address, discuss the inherent hidden complexities, and offer general recommendations regarding their design.

than those of simple phase II clinical trials of new agents. In this commentary, we present an overview of the various trials described as basket trials, clarify the distinctive goals that basket trials seek to address, discuss the inherent hidden complexities, and offer general recommendations regarding their design. Several approaches to evaluating targeted therapies in multiple tumor types have been described as basket trials (Fig 1). The first prototype in Figure 1 is the basket trial of vemurafenib.2 Vemurafenib is an oral inhibitor of BRAF that has greater selectivity for the BRAFV600 mutant form of the kinase than for wild-type BRAF, which had been previously approved for patients with BRAFV600E mutation–positive metastatic melanoma. Vemurafenib was targeted at a single variant in a variety of cancers with different primary disease sites and histologies, thereby defining disease-specific baskets. The second prototype in Figure 1 is the CREATE trial,3 which evaluated the use of the anaplastic lymphoma kinase and/or mesenchymal-epithelial transition factor inhibitor crizotinib. Here, although again there was a single agent under investigation, the drug inhibits multiple oncokinases including c-Met and anaplastic lymphoma kinase. Thus the baskets reflect a combination of diseases and targets. The last prototype in Figure 1 is the CUSTOM trial.4 In this trial, investigators planned to enroll patients with one of three diseases and allocate them to one of five targeted therapies, resulting in 15 disease-drug-mutation–specific baskets. Thus, this study tests the efficacy of a variety of drugs in a variety of targets and disease sites. We note that more complex trials than those presented in Figure 1, such as the NCI-MATCH,5 Genentech MyPathway,6 Novartis Signature,7 and American Society of Clinical Oncology’s Targeted Agent and Profiling Utilization Registry Study (TAPUR)8 basket trials, fall into this general framework. These ongoing studies define drug-mutation–specific baskets because their aim is to determine the efficacy of drugs that target certain pathways, typically with postmarketed drugs in nonindicated solid tumor types. The NCI-MATCH5 trial is even more complex in that genomic screening is incorporated into the therapeutic study itself and treatment assignment is determined by a matching algorithm that uses predefined levels of evidence of the gene variants.

work, the similarity of the results between baskets can be factored into a statistical model.14,15 This strategy allows the information from responses in different baskets to be shared, improving the efficiency of the statistical design and, by implication, permitting the study to reach conclusions with fewer patients. The use of these concepts has the potential to greatly improve the design of basket trials, but additional complexities must be recognized. The concept of statistical power becomes more complex in the basket setting. Power is the probability that the drug will be shown to be effective if it is truly effective, and in phase II trials, the calculation involves specifying the hypothesized true effectiveness of the drug. In basket trials, we must consider different configurations of effectiveness. For example, the drug may truly work only in one basket. Alternatively, it may actually work in two of the baskets, or in three or more. Each of these configurations can lead to markedly different statistical properties, and thus, the ideal study design is different, depending on which of these scenarios is true.

athways, typically with postmarketed drugs in nonindicated solid tumor types. The NCI-MATCH5 trial is even more complex in that genomic screening is incorporated into the therapeutic study itself and treatment assignment is determined by a matching algorithm that uses predefined levels of evidence of the gene variants. The trial aims to assess the activity of multiple drugs used in mutation-specific baskets (mutations, amplifications, or translocations), regardless of tumor origin, using a single stage design for each biomarker-defined subgroup (ie, mutation-specific basket).5 As the clinical setting becomes more complex for a study, the terms basket trial and umbrella trial begin to overlap. For example, the NCI-MATCH trial has been referred to as both an umbrella trial as a result of the multiple drugs under evaluation5 and as a basket trial because of the multiple disease populations for screening.1,9 Moreover, many basket studies evaluate multiple genomic variants in a given gene, which further complicates the clinical setting, and these variants may individually influence the likelihood of response to therapy. Fig 1. Three published basket trials. (A) Disease-specific baskets.2 (B) Disease-mutation–specific baskets (CREATE).3 (C) Disease-drug-mutation–specific baskets (CUSTOM).4 *Mutations or amplifications. ALK, anaplastic lymphoma kinase; EGFR, epidermal growth factor receptor; MET, mesenchymal-epithelial transition factor; NSCLC, non–small-cell lung cancer; SCLC, small-cell lung cancer; TM, thymic malignancy.

–specific baskets (CREATE).3 (C) Disease-drug-mutation–specific baskets (CUSTOM).4 *Mutations or amplifications. ALK, anaplastic lymphoma kinase; EGFR, epidermal growth factor receptor; MET, mesenchymal-epithelial transition factor; NSCLC, non–small-cell lung cancer; SCLC, small-cell lung cancer; TM, thymic malignancy. All the trials in Figure 1 were constructed as a series of independent phase II trials, using a conventional two-stage design, such as the Simon design,10 within each basket, individually controlling type I error at a nominal level. However, the design aspects and performance characteristics of these trials are not well understood, and the nature of the scientific goals is more complex than those of traditional disease-specific studies.1 Most clinical trials are constructed to address a single primary objective. Although secondary objectives may be articulated in the protocol, the chosen study design may not be ideal for addressing all of them, although in conventional clinical trials, the overall design is frequently suitable for addressing typical secondary goals. In the context of a basket trial, the ideal design options are not necessarily well aligned for the numerous questions being asked. Thus, careful consideration is needed to identify which question is paramount and to design the study accordingly. To focus our discussion, we limit attention to the seemingly most straightforward setting in which there is one target mutation and one drug targeting that mutation, evaluated in several disease types. In this setting, the questions the investigators seek to address may be one or more of the following: Does the drug have any efficacy at all? Does efficacy differ by disease site? If so, in which disease sites does the drug work?

target mutation and one drug targeting that mutation, evaluated in several disease types. In this setting, the questions the investigators seek to address may be one or more of the following: Does the drug have any efficacy at all? Does efficacy differ by disease site? If so, in which disease sites does the drug work? Consider the first question: Does the drug have any efficacy at all? This is the essential question for determining whether to pursue additional studies that would lead to regulatory approval for marketing the drug. A basket study that implements multiple independent two-stage designs (one per basket) has a much higher false-positive rate than a typical phase II study (ie, there is a much higher chance that the drug will be declared effective in at least one basket when in fact the drug is truly ineffective). For example, if there are five baskets and each has a 5% false-positive rate, the chance that an ineffective drug will be declared effective in one or more baskets is approximately 23%; increasing the number of baskets exacerbates this false-positive rate (for 10 baskets, approximately 40%). In short, the common practice of using a basket trial design that treats each cohort as an independent trial inevitably leads to a higher overall false-positive rate. Investigators can control the overall false-positive rate by adjusting the sample size or decision rules so that there is a lower false-positive rate within each basket. At a minimum, it is important to report the overall false-positive rate in a basket trial.

l inevitably leads to a higher overall false-positive rate. Investigators can control the overall false-positive rate by adjusting the sample size or decision rules so that there is a lower false-positive rate within each basket. At a minimum, it is important to report the overall false-positive rate in a basket trial. If the primary goal of the study is to determine the effect of the drug separately in each basket, a design structured to evaluate drug efficacy in each cancer site independently, such as a series of independent Simon designs, is an appropriate candidate. However, this strategy fails to recognize the potential inherent connectedness of the efficacy results in the different baskets and our expectations regarding this. As positive results emerge in a new cancer site, they increase our expectation of positive results in other sites, as has been observed recently in studies targeting kinase fusions.11 At the outset of the trial, we expect the efficacy results to be correlated. Evidence of such correlation can be harnessed productively in such trials, and there are two distinct ways of taking advantage of this. The first is the concept of aggregation.12,13 If, as the trial progresses, we observe good evidence of similar efficacy in a subset of the baskets, then aggregation of these baskets on the basis of an interim analysis can allow us to reach conclusions for this subset of baskets more quickly (ie, with fewer patients). The second concept is the use of statistical modeling. In this framework, the similarity of the results between baskets can be factored into a statistical model.14,15 This strategy allows the information from responses in different baskets to be shared, improving the efficiency of the statistical design and, by implication, permitting the study to reach conclusions with fewer patients.

INTRODUCTION Over the last 30 years, there has been substantial research efforts in the area of the health of sexual minority populations; however, this research has focused on sexually transmitted diseases, particularly HIV,1,2 and little is known about how cancer risk varies among sexual minorities compared with heterosexual populations.2-4 This contrasts with the increasing disease burden that is associated with cancer, which is currently the leading cause of death in high-income countries,5 and, after mental health services and circulatory diseases, cancer services make up the third largest category of spending in contemporary health care systems.6

ness. For example, the drug may truly work only in one basket. Alternatively, it may actually work in two of the baskets, or in three or more. Each of these configurations can lead to markedly different statistical properties, and thus, the ideal study design is different, depending on which of these scenarios is true. Recent research on a design and analysis strategy that permits aggregation of baskets with similar efficacy results, as determined by an interim assessment of heterogeneity, has shown large potential gains in efficiency in determining if the drug works overall.12 Efficiency is represented by the need for fewer patients, and it can be shown that this is improved, especially when the drug truly works across most of the baskets under investigation, with a modest cost with respect to power if the drug is effective in only one of the baskets. We display some results from this research that show considerable reductions in both the expected (ie, average) sample sizes and trial durations in settings in which the drug is truly effective in the preponderance of the baskets (Appendix Fig A1, online only). However, this strategy has limitations if the primary goal is to classify the effectiveness of the drug in each individual basket, because this goal requires adequate accrual to each basket. These results demonstrate that it is inevitable that the choice of design must involve compromise, balancing the ideal properties when the drug truly works in different numbers of baskets. In this context, interim analyses can be highly informative and can be constructed to make interim decisions that use the accumulating evidence more efficiently than simply following the decision rules embedded in independent Simon two-stage designs.

properties when the drug truly works in different numbers of baskets. In this context, interim analyses can be highly informative and can be constructed to make interim decisions that use the accumulating evidence more efficiently than simply following the decision rules embedded in independent Simon two-stage designs. In summary, the advent of targeted therapy has led to the introduction of a new class of clinical trials, and the term basket trial has come to be used to cover many different trial designs in this class. These different types of clinical trials share the characteristics that one or more targeted therapies are being tested and that the drugs are being investigated in distinct disease subtypes or sites. The scientific goals of these trials are typically more complex and frequently not specified with the precision conventionally used for clinical trials. Most investigators view a basket trial as a series of independent phase II clinical trials. In fact, the simplest type of basket trial, the evaluation of a single drug targeting a single mutation in multiple disease sites, presents a much more complex framework than a conventional evaluation of a single drug in a single disease. We believe that creative investigation into design options offers the potential to meet the study goals faster, with fewer patients. Such investigation must recognize the fact that most basket trials typically aim to answer multiple questions simultaneously. Most importantly, in this period of transition to precision medicine, our clinical research tools must maintain the scientific rigor embedded in the traditional clinical trials paradigm, in which hypotheses are specified precisely and the clinical trial is designed to address these hypotheses.

questions simultaneously. Most importantly, in this period of transition to precision medicine, our clinical research tools must maintain the scientific rigor embedded in the traditional clinical trials paradigm, in which hypotheses are specified precisely and the clinical trial is designed to address these hypotheses. ACKNOWLEDGMENT Supported by National Cancer Institute Grants No. CA008748 and CA163251. AUTHOR CONTRIBUTIONS Manuscript writing: All authors Final approval of manuscript: All authors AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST Basket Trials in Oncology: A Trade-Off Between Complexity and Efficiency The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/jco/site/ifc. Kristen M. Cunanan No relationship to disclose Mithat Gonen No relationship to disclose Ronglai Shen No relationship to disclose David M. Hyman Consulting or Advisory Role: Atara Biotherapeutics, CytomX Therapeutics Research Funding: Puma Biotechnology, Loxo, AstraZeneca Travel, Accommodations, Expenses: Puma Biotechnology Gregory J. Riely Consulting or Advisory Role: Novartis, Roche, Genentech Research Funding: Novartis (Inst), Roche/Genentech (Inst), Millennium (Inst), GlaxoSmithKline (Inst), Pfizer (Inst), Infinity Pharmaceuticals (Inst), ARIAD (Inst)

Research Funding: Puma Biotechnology, Loxo, AstraZeneca Travel, Accommodations, Expenses: Puma Biotechnology Gregory J. Riely Consulting or Advisory Role: Novartis, Roche, Genentech Research Funding: Novartis (Inst), Roche/Genentech (Inst), Millennium (Inst), GlaxoSmithKline (Inst), Pfizer (Inst), Infinity Pharmaceuticals (Inst), ARIAD (Inst) Colin B. Begg No relationship to disclose Alexia Iasonos No relationship to disclose Appendix Clinical Setting Investigate five disease-specific baskets, assuming a 15% null response rate and a 45% desirable response rate. Accrual assumed to be two patients per month. Independent Design Implements independent, parallel, two-stage Simon designs with 1% type 1 error rate and 20% type 2 error rate (ie, 80% power) within each basket. With these specifications, the family-wise error rate is controlled at 5%. Aggregation Design Implements an interim heterogeneity assessment to potentially aggregate baskets after first stage. This design is calibrated to control the family-wise error rate to 5% and to achieve at least 80% marginal power when the drug truly works in two or more (of the five) baskets, where marginal power is defined as the probability of correctly identifying that the drug works in an individual basket.

ets after first stage. This design is calibrated to control the family-wise error rate to 5% and to achieve at least 80% marginal power when the drug truly works in two or more (of the five) baskets, where marginal power is defined as the probability of correctly identifying that the drug works in an individual basket. Interpretation of Results In Figure A1, the blue shaded areas highlight the improved efficiency of the aggregation design in most configurations, in terms of expected (ie, average) sample size (left panel) and expected trial duration (right panel). The aggregation design is modestly inefficient with respect to sample size when the drug truly works in only one basket (gold shading). Fig A1. Gains and losses in sample size and trial duration (months) for two designs. The blue highlighted area shows gains when using an aggregation design; the gold highlighted area shows losses when using independent parallel phase II studies.

INTRODUCTION Advances in next-generation sequencing technology have led to the development of multiplex panel testing for the molecular diagnosis of inherited cancer susceptibility. Commercially available panels differ in their exact composition but usually include moderate-penetrance and high-penetrance genes (with mutations reported in the literature to be associated with a relative risk [RR] of cancer between 2 and 5, or greater than 5, respectively). Genes in which mutations are associated with susceptibility to inherited cancer have been rapidly incorporated into these panels, often before robust evidence of the magnitude of the association is known. For some genes, the relatively low prevalence of mutations makes it difficult to obtain reliable estimates of the associated cancer risk.1

s are associated with susceptibility to inherited cancer have been rapidly incorporated into these panels, often before robust evidence of the magnitude of the association is known. For some genes, the relatively low prevalence of mutations makes it difficult to obtain reliable estimates of the associated cancer risk.1 Variants of uncertain significance are frequent in panel testing and can be challenging to resolve. Despite the availability of public databases for sharing genetic variants, the development of prediction models based on protein structure and function, and the potential for laboratory-based functional analyses to determine the pathogenicity of some variants, discordant interpretation of the clinical pathogenicity of variants remains a frequent problem.2-4 Different standardized classification systems for interpretation of sequence-based results have been developed.5,6 The ClinVar database7,8 is a publicly available database that has allowed clinical laboratories to submit their identified variants and share their interpretation. All accessions of the same genetic variant provided by different submitters are maintained in ClinVar, which allows tracking of the changes and sharing of the evidence used for interpretation. However, submission to ClinVar is voluntary and not all laboratories choose to submit data. Moreover, resolution of the diverse submissions is a voluntary activity for experts and a Herculean task. From a clinical perspective, it is relevant to quantitate the frequencies and describe the patterns of variants with conflicting interpretations, particularly those that may impact medical management. Unrelated individuals or even members from the same family with the same genetic variant tested by different clinical laboratories may be given a different clinical assessment of variant pathogenicity. The ordering provider may not be aware that a different laboratory testing a patient’s relative has provided a different interpretation. It is crucial to identify variants with conflicting interpretations among laboratories, the frequencies and types of discrepancies, and the underlying reasons for these discrepancies to enable specific guidance for variant curation and to increase the consistency of variant interpretation among the laboratories.

ation. It is crucial to identify variants with conflicting interpretations among laboratories, the frequencies and types of discrepancies, and the underlying reasons for these discrepancies to enable specific guidance for variant curation and to increase the consistency of variant interpretation among the laboratories. Here, we describe the frequencies and types of genetic findings with conflicting interpretations in non-BRCA genes tested as part of panels assembled by Clinical Laboratory Improvement Amendments–approved commercial laboratories. Data were collected from individuals who underwent clinical testing and enrolled in a prospective registry. This registry includes patients with results from some commercial laboratories that currently do not deposit data in ClinVar. PATIENTS AND METHODS Overall, 1,191 individuals tested for cancer susceptibility genes as part of commercial multiplex panels self-enrolled in the ongoing Prospective Registry of Multiplex Testing (PROMPT) between September 1, 2014, and September 30, 2015, the period for which data were ascertained for this analysis.

Here, we describe the frequencies and types of genetic findings with conflicting interpretations in non-BRCA genes tested as part of panels assembled by Clinical Laboratory Improvement Amendments–approved commercial laboratories. Data were collected from individuals who underwent clinical testing and enrolled in a prospective registry. This registry includes patients with results from some commercial laboratories that currently do not deposit data in ClinVar. PATIENTS AND METHODS Overall, 1,191 individuals tested for cancer susceptibility genes as part of commercial multiplex panels self-enrolled in the ongoing Prospective Registry of Multiplex Testing (PROMPT) between September 1, 2014, and September 30, 2015, the period for which data were ascertained for this analysis. The PROMPT registry partnered with several clinical laboratories, including Ambry Genetics, Color Genomics, GeneDx, Invitae, Myriad Genetics, Pathway Genomics, and Quest Diagnostics, to recruit individuals who had undergone genetic panel testing and had at least one variant in their report. Participating laboratories advertised PROMPT within the packet of test report forms sent to individuals. Health-care providers also were educated about PROMPT at academic and industry meetings, and through e-mails. An informational Web site and video were created for participants and providers. An informational article was communicated via the Dr Susan Love Research Foundation e-newsletter to individuals enrolled in the Army of Women. Some participants found the study directly through Internet.

stry meetings, and through e-mails. An informational Web site and video were created for participants and providers. An informational article was communicated via the Dr Susan Love Research Foundation e-newsletter to individuals enrolled in the Army of Women. Some participants found the study directly through Internet. An enrollment site for PROMPT was built on a platform maintained by a partner organization, PatientCrossroads. To enroll, participants created an account with PatientCrossroads and consented to participate in an online genetic registry in either an identifiable (contact information available to research team) or de-identified (contact information not available to research team) manner. All participants completed the baseline questionnaire with personal and family cancer history of cancer, genetic testing, and demographics. Participants were given the opportunity to upload their genetic testing report to the PatientCrossroads portal or to send it directly to PROMPT registry staff.

search team) manner. All participants completed the baseline questionnaire with personal and family cancer history of cancer, genetic testing, and demographics. Participants were given the opportunity to upload their genetic testing report to the PatientCrossroads portal or to send it directly to PROMPT registry staff. From the initial 1,191 participants assessed for eligibility, individuals considered for this analysis were required to have verifiable genetic data by PROMPT staff (ie, a copy of a test report submitted, or a detailed self-reported test result in the registry that was found in ClinVar). Participants who had undergone tumor testing, who did not have any genetic findings, or whose unique finding was a BRCA result, were excluded (n = 112), as were individuals with self-reported results for which a ClinVar submission or test report was not available for confirmation (n = 410). From the 669 remaining participants, 518 (43%) had results interpreted by more than one laboratory (including at least one in ClinVar) or findings with multiple submissions reported in ClinVar, and these participants were used as the final cohort for the current analysis (Fig 1). Because commercial laboratories are not required to submit to ClinVar, if there were known differences in classification of a genetic test result among laboratories observed through reporting in the PROMPT registry, then this result was classified as a conflicting interpretation of pathogenicity. That is, a patient may have provided a report from a non-ClinVar submitting laboratory and the same finding may have been entered into ClinVar by another laboratory.

lt among laboratories observed through reporting in the PROMPT registry, then this result was classified as a conflicting interpretation of pathogenicity. That is, a patient may have provided a report from a non-ClinVar submitting laboratory and the same finding may have been entered into ClinVar by another laboratory. Fig 1. CONSORT diagram showing the flow of participants and genetic variants per participant from the PROMPT registry until inclusion for current analysis. PROMPT, Prospective Registry of Multiplex Testing. All genetic test results were checked in the ClinVar public archive (http://www.ncbi.nlm.nih.gov/clinvar/) and their clinical significance was assigned according to the submissions by different laboratories through clinical testing or submissions from research and literature curation. In ClinVar, if differences in interpretation among submitters are observed, the genetic test results are classified as conflicting interpretation of pathogenicity.7 A search for any update in reclassification was performed at the time of data analysis lock, with no change. Descriptive statistics were used to describe the study population.

rences in interpretation among submitters are observed, the genetic test results are classified as conflicting interpretation of pathogenicity.7 A search for any update in reclassification was performed at the time of data analysis lock, with no change. Descriptive statistics were used to describe the study population. RESULTS Overall, 518 participants enrolled into PROMPT were considered eligible for this analysis. Their median age was 52 years (range, 44-61 years) and 95% were female. Overall, 427 (82%) were white, and 350 (68%) had a cancer diagnosis, mostly breast cancer (n = 188; 36%). Thirty-one percent had multiple primary tumors. A total of 419 participants (81%) reported being invited into PROMPT by their health-care provider or by the laboratory where the testing had been performed (Table 1). Table 1. Study Population Enrolled in the Prospective Registry of Multiplex Testing Registry These 518 participants reported 603 genetic variants with multiple interpretations by several commercial laboratories and/or submissions to ClinVar. Of the 518 participants included in this analysis, 165 provided information from testing done in a laboratory that does not submit to ClinVar. The most frequent gene with sequence alterations reported through PROMPT was CHEK2 (n = 117), followed by ATM (n = 105) (Fig 2). Fig 2. Distribution of genetic variants with multiple submissions in ClinVar by gene (N = 603).

These 518 participants reported 603 genetic variants with multiple interpretations by several commercial laboratories and/or submissions to ClinVar. Of the 518 participants included in this analysis, 165 provided information from testing done in a laboratory that does not submit to ClinVar. The most frequent gene with sequence alterations reported through PROMPT was CHEK2 (n = 117), followed by ATM (n = 105) (Fig 2). Fig 2. Distribution of genetic variants with multiple submissions in ClinVar by gene (N = 603). Regarding the type of results according to their clinical interpretation in ClinVar, 220 (36%) were consistently classified as VUS, 191 (32%) as pathogenic/likely pathogenic, and 34 (6%) as benign/likely benign, while 155 (26%) were classified as conflicting interpretation. Among these 155, 26% of them were in CHEK2; 20% in ATM; 8% in RAD51C; 7% in PALB2; 5% in BARD1; 4% in NBN and APC; 3% in RAD50, PMS2, TP53, and MUTYH; 2% in BRIP1 and FANCC; and the remainder were distributed among the other genes at approximately 1% each (Fig 3). Of the 155 discordant findings, 56 (36%) were reported as pathogenic/likely pathogenic by at least one laboratory but not by all laboratories (ie, clinically significant; Table 2). Fig 3. Distribution of genetic variants according to ClinVar interpretation (N = 603), and the absolute number of variants with conflicting interpretation by gene (n = 155). Table 2. Genetic Variants With Conflicting Interpretation

Regarding the type of results according to their clinical interpretation in ClinVar, 220 (36%) were consistently classified as VUS, 191 (32%) as pathogenic/likely pathogenic, and 34 (6%) as benign/likely benign, while 155 (26%) were classified as conflicting interpretation. Among these 155, 26% of them were in CHEK2; 20% in ATM; 8% in RAD51C; 7% in PALB2; 5% in BARD1; 4% in NBN and APC; 3% in RAD50, PMS2, TP53, and MUTYH; 2% in BRIP1 and FANCC; and the remainder were distributed among the other genes at approximately 1% each (Fig 3). Of the 155 discordant findings, 56 (36%) were reported as pathogenic/likely pathogenic by at least one laboratory but not by all laboratories (ie, clinically significant; Table 2). Fig 3. Distribution of genetic variants according to ClinVar interpretation (N = 603), and the absolute number of variants with conflicting interpretation by gene (n = 155). Table 2. Genetic Variants With Conflicting Interpretation Of 117 findings with multiple interpretations for CHEK2, 41 (35%) were conflicting, and the majority (n = 36; 88%) would be characterized as clinically significant. Eighteen were c.470T>C (p.Ile157T) variant and 12 were c.1283C>T (p.Ser428Phe), both classified as either pathogenic/likely pathogenic or as VUS (Table 3). Other variants in this gene with a conflicting interpretation between pathogenic/likely pathogenic or VUS are described in Table 2. In addition, 32 of 105 variants (30%) in the ATM gene with multiple submissions were classified as conflicting. However, all of these ATM variants ranged from benign/likely benign or VUS (Appendix Fig A1, online only; Table 2) and, therefore, should not alter medical management. In PALB2, nine of 49 findings (18%) reported with multiple submissions were conflicting; one, c.3113G>A (p.Trp1038Ter), which was reported three times, was classified as either pathogenic or VUS, and the remainder were interpreted as either benign or VUS. In BRIP1, three of 33 findings (9%) were discordant; one of them was reported twice [c.139C>G (p.Pro47Ala)] and was classified as either likely pathogenic or VUS. In RAD51C, 13 of 26 (50%) findings (50%) were conflicting. Twelve of these findings corresponded to the variant c.790G>A (p.Gly264Ser), which was submitted as either benign or VUS, and one, the c.1026+5_1026+7delGTA variant, was interpreted as both likely pathogenic and VUS. In the NBN, six of 20 findings (30%) reported were conflicting. Two of them, c.511A>G (p.Ile171Val) and c.643C>T (p.Arg215Trp), were reported twice each, and had a two-step difference in conflicting interpretation in ClinVar (pathogenic, VUS, benign/likely benign). Other genes with conflicting interpretation are listed in Table 2.

six of 20 findings (30%) reported were conflicting. Two of them, c.511A>G (p.Ile171Val) and c.643C>T (p.Arg215Trp), were reported twice each, and had a two-step difference in conflicting interpretation in ClinVar (pathogenic, VUS, benign/likely benign). Other genes with conflicting interpretation are listed in Table 2. Table 3. Criteria Used for Clinical Interpretation of CHEK2 Variants p.S428F and p.I157K

six of 20 findings (30%) reported were conflicting. Two of them, c.511A>G (p.Ile171Val) and c.643C>T (p.Arg215Trp), were reported twice each, and had a two-step difference in conflicting interpretation in ClinVar (pathogenic, VUS, benign/likely benign). Other genes with conflicting interpretation are listed in Table 2. Table 3. Criteria Used for Clinical Interpretation of CHEK2 Variants p.S428F and p.I157K DISCUSSION One quarter of the clinical genetic results from commercially available multiplex cancer panels and reported at the PROMPT registry had conflicting interpretations within ClinVar. Most of the variants with conflicting interpretations were in CHEK2, followed by ATM, RAD51C, and PALB2. Many conflicting interpretations are of low clinical significance because the discrepancy ranged between an interpretation of benign/likely benign or VUS; therefore, medical management should default to personal and family history. However, the identification of a VUS can cause a great deal of uncertainty for patients and providers alike and increase the risk for inappropriate medical management.5 For example, it is inappropriate to recommend oophorectomy based on a VUS finding alone. Of greater concern, 36% of conflicting results appeared to be clinically relevant, because they were either reported as pathogenic/likely pathogenic or as a VUS by different clinical laboratories. In this regard, CHEK2, PALB2, and BRIP1 were most frequently identified as having discordant interpretation between these two levels of pathogenicity. As these genes are being incorporated into clinical practice as part of cancer risk assessment, tailored screening and cancer prevention recommendations, or within tumor panel sequencing for potential targeted therapy, it will be critical to standardize their curation and clinical classification of variants to provide appropriate management for mutation carriers and their families.

part of cancer risk assessment, tailored screening and cancer prevention recommendations, or within tumor panel sequencing for potential targeted therapy, it will be critical to standardize their curation and clinical classification of variants to provide appropriate management for mutation carriers and their families. A strength of this study is the inclusion of individuals who underwent testing at laboratories that do not submit data to ClinVar. Our data highlight several specific variants of interest related to differential reporting. Of 117 of the CHEK2 findings reported in PROMPT to date, 41 (35%) were conflicting. The c.470T>C (p.Ile157Thr, I157K) and the c.1283C>T (p.Ser428Phe, S428F) variants were the most common with conflicting interpretation between pathogenic/likely pathogenic and VUS. The putative pathogenicity of I157T has long been studied. Several reports analyze its association with breast cancer risk or other tumors.9-13 Functional analyses of this variant have also been published14-16 and in silico predictions are available. The frequency of this variant in the population is 0.4% in the National Heart, Lung, and Blood Institute Exome Variant Server and approximately 5% are found in some Northern European populations.9 Despite these data, there was discordant interpretation of this evidence by the different clinical genetic laboratories. For instance, two laboratories found the functional analysis data compelling enough to suggest a damaging effect on protein function and to influence variant interpretation, whereas two other laboratories felt that this variant had no effect on CHEK2 protein kinase activity and the relationship between functional studies and cancer association is unclear. Only one laboratory supported the association with cancer risk as being significant, whereas the other three laboratories reported an increased prevalence in affected individuals, but also documented high frequencies observed among controls in diverse populations. All agreed that the CHEK2 variant is located in a well-established functional domain, but only one laboratory used the supporting limited evidence provided from predictions of in silico algorithms of this variant's effect. Overall, these discrepancies in interpretation of the evidence lead to a range of clinical interpretations from VUS to pathogenic (low penetrance), likely pathogenic, and pathogenic (Table 3).

one laboratory used the supporting limited evidence provided from predictions of in silico algorithms of this variant's effect. Overall, these discrepancies in interpretation of the evidence lead to a range of clinical interpretations from VUS to pathogenic (low penetrance), likely pathogenic, and pathogenic (Table 3). Differences in interpretation of the evidence for the pathogenicity of CHEK2 S428F are less pronounced but still lead to conflicting reports of pathogenic, pathogenic/low penetrance, and VUS. All laboratories agree that the S428F variant is located in a well-established functional domain16; and all but one agree that there is evidence of an increased prevalence in affected individuals v controls.9 However, three laboratories report that functional studies are supportive of the damaging effect of CHEK2 S428F, and one laboratory concludes that the impairment is variable. Finally, two laboratories report the literature evidence from cosegregation of the CHEK2 variant with the disease; only one laboratory reports the prediction of pathogenicity by in silico algorithms, and another laboratory emphasizes the contribution of this variant to tumorigenesis by loss of heterozygosity (Table 3). The examples of the discrepancies involving CHEK2 I157T and S428F also may reflect the challenges of describing a so-called low-penetrance susceptibility allele (RR < 2) in a format designed for high-penetrance alleles.17

emphasizes the contribution of this variant to tumorigenesis by loss of heterozygosity (Table 3). The examples of the discrepancies involving CHEK2 I157T and S428F also may reflect the challenges of describing a so-called low-penetrance susceptibility allele (RR < 2) in a format designed for high-penetrance alleles.17 Another interesting variant with conflicting interpretation between likely pathogenic and VUS is the c.139C>G substitution (p.Pro47Ala) in the BRIP1 gene, which was observed in two unrelated participants from PROMPT. This variant was first described in 2001,18 when the BRIP1 protein was found to interact with BRCA1 and contribute to its DNA repair function. This variant was initially identified in an individual with early-onset breast cancer and a family history of breast and ovarian cancer; segregation analysis was not available and loss of heterozygosity in the tumor was not demonstrated. Since then, it has been reported in the literature several times as a breast or ovarian cancer susceptibility gene in affected individuals, but has also been observed in control subjects.19,20 Two clinical laboratories providing a summary of their clinical interpretation in ClinVar differ in their assessment of the meaning of loss of function observed in the experimental studies in regard to cancer susceptibility; one laboratory highlights that it is present in population databases, albeit as a very rare allele (0.04%); and only one endorses the supporting evidence provided by the prediction of in silico algorithms. In phase 2 of the PROMPT project, enrollment of family members will be encouraged and cosegregation analysis of this BRIP1 variant may be undertaken, which may help provide one more piece of supporting evidence to clarify its pathogenicity.

ses the supporting evidence provided by the prediction of in silico algorithms. In phase 2 of the PROMPT project, enrollment of family members will be encouraged and cosegregation analysis of this BRIP1 variant may be undertaken, which may help provide one more piece of supporting evidence to clarify its pathogenicity. Findings of discordant interpretation of results in genetic testing are not limited to oncologic settings nor ClinVar. Similar discrepancies also have been observed in hereditary connective-tissue disorders and are felt to be due to lack of submission of data to public databases, limited use of allele frequency data, and varying consideration of protein structure and function.3 In the current study, the discrepancies between commercial laboratories in interpretation of variants in cancer-related genes seem to be mostly based on differences in the interpretation of evidence. Efforts have already been initiated among laboratories to resolve these differences.21 The lack of a gold standard test for pathogenicity, or a noncontroversial interpretation of functional analyses, suggests that discrepant interpretation of challenging variants, particularly missense and splice-site variants, will persist for some time.

been initiated among laboratories to resolve these differences.21 The lack of a gold standard test for pathogenicity, or a noncontroversial interpretation of functional analyses, suggests that discrepant interpretation of challenging variants, particularly missense and splice-site variants, will persist for some time. Multigene testing for cancer susceptibility is a complex endeavor characterized by challenges in curation and reporting of variants; unfortunately, this study demonstrates that conflicting interpretation of those variants may be relatively frequent. The rates of discrepant interpretation reported herein support the need for initiatives focused on harmonizing variant interpretation in the context of shared data. We encourage clinical laboratories to submit their findings to ClinVar and other public databases with relevant clinical interpretations of findings, as well as the evidence on which interpretations are based. To better identify the underlying reasons for their discrepant interpretation, laboratories could explicitly describe their weighting (from very strong to moderate or supporting) for each pathogenic criterion used for classification.6 This transparency in data sharing and collaboration between academic research consortia and commercial laboratories may promote strategies to standardize clinical variant curation algorithms. Ongoing consortia such as Evidence-Based Network for the Interpretation of Germline Mutant Alleles (ENIGMA), and International Society for Gastrointestinal Hereditary Tumors (InSIGHT) have demonstrated the utility of multidisciplinary collaboration to curate and reclassify submitted VUS.22,23 Platforms such as the Leiden Open Source Variation Database (LOVD)24, which allows for collection, curation, and display of phenotypes and DNA sequence variants, and the BRCA Challenge (http://brcaexchange.org/; an international effort to review and provide vetted data on BRCA1 and BRCA2 gene variants) are additional relevant resources. PROMPT uses both crowd-sourcing and direct participant enrollment; therefore, it also can provide a platform for cohort formation and prospective follow-up of individuals and their family members harboring genetic variants. However, at this time, health-care providers and patients need to be aware that there could be conflicting interpretations of variants and those variants may be reclassified.

it also can provide a platform for cohort formation and prospective follow-up of individuals and their family members harboring genetic variants. However, at this time, health-care providers and patients need to be aware that there could be conflicting interpretations of variants and those variants may be reclassified. Our analysis has limitations. Because PROMPT is an elective, patient-oriented registry, there may be significant ascertainment bias. For example, individuals with VUS or those who were self-aware of a variant with conflicting interpretation may have been more likely to report their variant to PROMPT, leading to an overestimation of discrepant findings. On the other hand, a significant number of test results reported by participants were not submitted to ClinVar by any laboratory and, therefore, were excluded from the primary analysis, potentially leading to a misestimate of conflicting variant interpretations. Finally, because some patients did not upload their clinical test report, it was not possible to carry out more in-depth analysis in many cases. Efforts to obtain reports from all participants are ongoing.

e excluded from the primary analysis, potentially leading to a misestimate of conflicting variant interpretations. Finally, because some patients did not upload their clinical test report, it was not possible to carry out more in-depth analysis in many cases. Efforts to obtain reports from all participants are ongoing. In conclusion, clinical interpretation of genetic testing for increased cancer susceptibility as assessed by multiplex panels hinges on accurate curation and interpretation of variants. Discrepant interpretation of some genetic variants appears to be common. Internet-based registries provide a powerful tool to collect data to inform efforts to standardize classification of genetic variants, and can play an important role in efforts to minimize potential medical harms due to false alarm or false reassurance following cancer genetic testing. Processed as a Rapid Communication manuscript. Authors’ disclosures of potential conflicts of interest are found in the article online at www.jco.org. Author contributions are found at the end of this article. See accompanying editorial on page 4061

In conclusion, clinical interpretation of genetic testing for increased cancer susceptibility as assessed by multiplex panels hinges on accurate curation and interpretation of variants. Discrepant interpretation of some genetic variants appears to be common. Internet-based registries provide a powerful tool to collect data to inform efforts to standardize classification of genetic variants, and can play an important role in efforts to minimize potential medical harms due to false alarm or false reassurance following cancer genetic testing. Processed as a Rapid Communication manuscript. Authors’ disclosures of potential conflicts of interest are found in the article online at www.jco.org. Author contributions are found at the end of this article. See accompanying editorial on page 4061 ACKNOWLEDGMENT J.B. is a recipient of an award by Sociedad Española de Oncología Médica. This project was supported also by the Robert and Kate Niehaus Center for Inherited Cancer Genomics at Memorial Sloan Kettering Cancer Center (M.E.R., M.F.W., K.O.), the Breast Cancer Research Foundation (S.M.D., M.E.R., K.O., F.C., J.G.), and the Susan G. Komen Foundation (S.M.D.). We thank Myriad, Ambry, Invitae, GeneDx, Color Genomics, Pathway Genomics, and Quest Diagnostics for their active collaboration with PROMPT AUTHOR CONTRIBUTIONS Conception and design: Judith Balmaña, Mark E. Robson, Susan M. Domchek Collection and assembly of data: Judith Balmaña, Laura Digiovanni, Pragna Gaddam, Judy E. Garber, Kenneth Offit, Mark E. Robson, Susan M. Domchek

ACKNOWLEDGMENT J.B. is a recipient of an award by Sociedad Española de Oncología Médica. This project was supported also by the Robert and Kate Niehaus Center for Inherited Cancer Genomics at Memorial Sloan Kettering Cancer Center (M.E.R., M.F.W., K.O.), the Breast Cancer Research Foundation (S.M.D., M.E.R., K.O., F.C., J.G.), and the Susan G. Komen Foundation (S.M.D.). We thank Myriad, Ambry, Invitae, GeneDx, Color Genomics, Pathway Genomics, and Quest Diagnostics for their active collaboration with PROMPT AUTHOR CONTRIBUTIONS Conception and design: Judith Balmaña, Mark E. Robson, Susan M. Domchek Collection and assembly of data: Judith Balmaña, Laura Digiovanni, Pragna Gaddam, Judy E. Garber, Kenneth Offit, Mark E. Robson, Susan M. Domchek Data analysis and interpretation: Judith Balmaña, Laura Digiovanni, Michael F. Walsh, Vijai Joseph, Zsofia K. Stadler, Katherine L. Nathanson, Fergus J. Couch, Kenneth Offit, Mark E. Robson, Susan M. Domchek Manuscript writing: All authors Final approval of manuscript: All authors Accountable for all aspects of the work: All authors

Data analysis and interpretation: Judith Balmaña, Laura Digiovanni, Michael F. Walsh, Vijai Joseph, Zsofia K. Stadler, Katherine L. Nathanson, Fergus J. Couch, Kenneth Offit, Mark E. Robson, Susan M. Domchek Manuscript writing: All authors Final approval of manuscript: All authors Accountable for all aspects of the work: All authors AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST Conflicting Interpretation of Genetic Variants and Cancer Risk by Commercial Laboratories as Assessed by the Prospective Registry of Multiplex Testing The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or jco.ascopubs.org/site/ifc. Judith Balmaña No relationship to disclose Laura Digiovanni No relationship to disclose Pragna Gaddam No relationship to disclose Michael F. Walsh No relationship to disclose Vijai Joseph No relationship to disclose Zsofia K. Stadler Stock or Other Ownership: Alimera Sciences (I), Avalanche Biotechnologies (I) Consulting or Advisory Role: Alimera Sciences (I), Avalanche Biotechnologies (I), Allergan (I), Bausch and Lomb (I), Roche (I), Regeneron (I), Optos (I) Katherine L. Nathanson No relationship to disclose Judy E. Garber Consulting or Advisory Role: Pfizer, Sequenom, Novartis (I), Pfizer (I), SV Life Sciences (I)

Zsofia K. Stadler Stock or Other Ownership: Alimera Sciences (I), Avalanche Biotechnologies (I) Consulting or Advisory Role: Alimera Sciences (I), Avalanche Biotechnologies (I), Allergan (I), Bausch and Lomb (I), Roche (I), Regeneron (I), Optos (I) Katherine L. Nathanson No relationship to disclose Judy E. Garber Consulting or Advisory Role: Pfizer, Sequenom, Novartis (I), Pfizer (I), SV Life Sciences (I) Research Funding: Ambry Genetics, Myriad Genetics, Novartis (I), Pfizer (I) Fergus J. Couch Travel, Accommodations, Expenses: Ambry Genetics Kenneth Offit No relationship to disclose Mark E. Robson Honoraria: AstraZeneca Consulting or Advisory Role: Bayer, Pfizer, McKesson Research Funding: AstraZeneca (Inst), AbbVie (Inst), Myriad Genetics (Inst), Biomarin (Inst) Travel, Accommodations, Expenses: AstraZeneca, Biomarin Susan M. Domchek Honoraria: EMD Serono Research Funding: AstraZeneca (Inst), Clovis Oncology (Inst), Abbvie (Inst), Pharmamar (Inst) Appendix Fig A1. Type of conflicting interpretation by genes.

INTRODUCTION Prostate cancer (PCa) is one of the most common causes of male deaths from cancer in Europe.1 Whereas watchful waiting (WW) is an accepted method of PCa management, the risk of PCa-specific mortality can be diminished by radical treatment of localized tumors with either radical prostatectomy (RP)2,3 or radical radiotherapy (RT). For non–organ-confined disease, androgen deprivation therapy (ADT) effectively palliates and prevents PCa-related complications, but without a survival advantage in the absence of combination treatment with RT for locally advanced tumors.4 Comorbidities are medical disorders that coexist with, but are distinct from, the primary diagnosis.5 As with PCa, comorbidity is also age related and can influence the decision, timing, and modality of treatment selection. Because the survival advantage of radical therapy for PCa is typically observed only 10 years after treatment,2 current European guidelines recommend radical treatment with curative intent in patients with a > 10-year life expectancy.4 Comorbidity is highly prevalent among patients with cancer6 and may adversely affect both competing-cause and cancer-specific mortality,7,8 depending on the measures of comorbidity and survival. Evidence for an effect in patients with PCa is conflicting,1,8,9 and how comorbidity influences PCa-specific mortality is unclear. Here, by using a population-based observational cohort, we test the hypothesis that PCa-specific mortality is not affected by comorbidity after accounting for patient and tumor characteristics and treatment type.

in patients with PCa is conflicting,1,8,9 and how comorbidity influences PCa-specific mortality is unclear. Here, by using a population-based observational cohort, we test the hypothesis that PCa-specific mortality is not affected by comorbidity after accounting for patient and tumor characteristics and treatment type. PATIENTS AND METHODS Study Cohort Our study cohort is based on the Prostate Cancer Database Sweden (PCBaSe), which is described elsewhere.10,11 In brief, PCBaSe is a composite population-based data set that links the National Prostate Cancer Register, the Swedish Cancer Register, the cause of death register, and six other national registers by a unique personal identity number that is assigned to every Swedish resident. PCBaSe captures more than 98% of all cases of PCa in Sweden diagnosed since 199812 with virtually complete data on year of diagnosis, age, clinical (TNM) stage,13 Gleason tumor grade,14 or WHO15 tumor grade, diagnostic serum prostate-specific antigen (PSA) levels, planned primary treatment within 6 months of diagnosis, county of residence, marital status, educational level, socioeconomic status, comorbidity, and cancer-related events during follow-up.11 Neoadjuvant ADT with RP was not recorded, but is a rare management strategy; neoadjuvant ADT with RT has been recorded since 2008 only; and information on ADT after WW was not available. We identified a total of 129,389 men who were diagnosed with PCa between January 1998 and December 2012 and who were observed for survival (or death) until December 2014, which is the last date of update for the cause of death register linkage to PCBaSe. Causes of death were ascertained by using the cause of death register, for which the accuracy of the data on PCa has been validated and reported to be 86%.16 After exclusion of those patients whose treatment was unknown (n = 5,427), those who had died before treatment (n = 482), or those who were missing one or more tumor covariates (n = 4,937) we included all patients regardless of primary treatment or otherwise (n = 118,543). Median follow-up time for the included cohort was 8.3 years (interquartile range, 5.2 to 11.5 years). This study was approved by the central research ethics committee and the regional ethical review board in Stockholm (EPN Dnr 2012/499-31/4).

uded all patients regardless of primary treatment or otherwise (n = 118,543). Median follow-up time for the included cohort was 8.3 years (interquartile range, 5.2 to 11.5 years). This study was approved by the central research ethics committee and the regional ethical review board in Stockholm (EPN Dnr 2012/499-31/4). Comorbidity Comorbidity—described in PCBaSe by the Charlson comorbidity index (CCI)17—was estimated from registrations in the Swedish National In-Patient Register and the Swedish Cancer Register that were retrieved from 10 years before, until the date of PCa diagnosis.9 Although outpatient diagnoses were not included, the validity of these registers has been demonstrated to be high for medical diagnoses18 and most cancer diagnoses.19 CCI is a weighted scoring system that estimates the burden of 17 groups of concomitant diseases (Data Supplement) for each patient,17 which results in four comorbidity levels that are scored from 0 (no comorbidity) to ≥ 3 (severe comorbidity). CCI has been previously shown to impact treatment choices for PCa and the subsequent outcomes of patients in PCBaSe.9

imates the burden of 17 groups of concomitant diseases (Data Supplement) for each patient,17 which results in four comorbidity levels that are scored from 0 (no comorbidity) to ≥ 3 (severe comorbidity). CCI has been previously shown to impact treatment choices for PCa and the subsequent outcomes of patients in PCBaSe.9 Statistical Analyses Clinicopathologic characteristics were reported as medians and interquartile ranges. Study end points were PCa-specific and other-cause survival. Survival time was defined as the interval between the date of PCa diagnosis and the date of death, emigration, or end of follow-up. When considering one cause of death, deaths that were from a competing cause were treated as censoring time points. Overall follow-up time was calculated by using the reverse Kaplan-Meier method.20 Patients were categorized by patient factors (marital status, educational level), tumor characteristics (PSA, clinical grade and stage), and treatment type, namely, RP, RT, ADT, and WW. Data were stratified by CCI (0, 1, 2, or ≥ 3) and treatment type as detailed in the tables and figure legends, respectively. The χ2 and Wilcoxon Mann-Whitney tests were used to test for differences in the distributions of patient characteristics between CCI groups. To adjust for any imbalances in the distribution of covariates among groups, we used stabilized inverse probability weighting, a propensity score–based method for which a situation is emulated in which the groups to be compared are made to have similar characteristics at baseline on the basis of preselected adjustment variables.21 For adjustments, we used patient and tumor-related clinical characteristics that were available only at the time of treatment decision, namely, age, marital status, educational level, year of diagnosis, tumor grade, clinical stage, and PSA. Adjustment weights were constructed by using multinomial regression wherein continuous covariates were modeled as restricted cubic splines with three knots. Extreme low or high weights were truncated at 0.25 and 4, respectively. PCa-specific and other-cause survival were compared between the actual and emulated groups by using unadjusted and weighted Kaplan-Meier estimates and Cox proportional hazards regression models to calculate hazard ratios.

splines with three knots. Extreme low or high weights were truncated at 0.25 and 4, respectively. PCa-specific and other-cause survival were compared between the actual and emulated groups by using unadjusted and weighted Kaplan-Meier estimates and Cox proportional hazards regression models to calculate hazard ratios. A cause-of-death–specific analysis, rather than competing-risks analysis, was specifically used to determine whether comorbidity affected PCa-specific or other-cause mortality under the hypothetical scenario that no men in the overall cohort or each treatment group (RP, RP, ADT, or WW) would die of causes other than PCa, or from PCa, respectively.22 Statistical analyses were performed with R software v.3.1.223 using the multinom function from the nnet package as well as the Survival Analysis and Regression Modeling Strategies packages. All statistical tests were two sided and performed at a 5% significance level.

causes other than PCa, or from PCa, respectively.22 Statistical analyses were performed with R software v.3.1.223 using the multinom function from the nnet package as well as the Survival Analysis and Regression Modeling Strategies packages. All statistical tests were two sided and performed at a 5% significance level. RESULTS Baseline unadjusted patient and tumor characteristics and treatment type (RP, RT, ADT, or WW) stratified by CCI (0, 1, 2, or ≥ 3) are listed in Table 1 (the data after statistical adjustments are listed in the Data Supplement). At diagnosis, patients with greater comorbidity were generally older than those with no comorbidity. Median PSA at the time of diagnosis was also higher in the comorbid group compared with the group with no comorbidity. Consistent with this, there was a higher proportion of low-grade, localized tumors that were identified in the group of patients with no comorbidity compared with the comorbid group. Consequently, the proportion of patients who were treated by radical therapies (RP or RT) was greater in the group of patients with little or no comorbidity than in the more comorbid groups in which a greater proportion of patients were treated with ADT or WW. Of a total of 87,816 patients in CCI group 0, 16,186 patients in CCI group 1, 9,114 patients in CCI group 2, and 5,427 patients in CCI group ≥ 3 (Table 1) there were 15,403, 3,591, 1,895, and 1,134 PCa-related deaths and 15,203, 5,409, 3,683, and 2,849 deaths from other causes, respectively, by the end of the study period.

total of 87,816 patients in CCI group 0, 16,186 patients in CCI group 1, 9,114 patients in CCI group 2, and 5,427 patients in CCI group ≥ 3 (Table 1) there were 15,403, 3,591, 1,895, and 1,134 PCa-related deaths and 15,203, 5,409, 3,683, and 2,849 deaths from other causes, respectively, by the end of the study period. Table 1. Baseline Clinicopathologic and Follow-Up Data for Cohort Stratified by CCI Groups

total of 87,816 patients in CCI group 0, 16,186 patients in CCI group 1, 9,114 patients in CCI group 2, and 5,427 patients in CCI group ≥ 3 (Table 1) there were 15,403, 3,591, 1,895, and 1,134 PCa-related deaths and 15,203, 5,409, 3,683, and 2,849 deaths from other causes, respectively, by the end of the study period. Table 1. Baseline Clinicopathologic and Follow-Up Data for Cohort Stratified by CCI Groups In the complete unadjusted data set, we observed an effect of increased comorbidity on PCa-specific and other-cause mortality, which rose to a 1.99 -fold hazard (95% CI, 1.87 to 2.11) of PCa-specific mortality and a 5.62-fold hazard (95% CI, 5.40 to 5.85) of other-cause mortality for patients with a CCI score of ≥ 3 compared with those with no comorbidity (CCI 0; Table 2 and Fig 1A, left and right panels, respectively). After adjustments for patient and tumor characteristics, the effect of limited comorbidity (CCI score of 1 and 2) on PCa-specific mortality was clearly attenuated and not statistically significant, but was maintained for other-cause mortality across all CCI groups (Table 2 and Fig 1B, left and right panels, respectively). After additional adjustment for treatment type, the association between comorbidity and PCa-specific mortality was again attenuated without clear trends across CCI groups, whereas the effect of increasing comorbidity on other-cause mortality was maintained (Table 2 and Fig 1C, left and right panels, respectively). Of the individual comorbidities that constituted the CCI (Data Supplement), only congestive heart failure and dementia affected PCa-specific mortality after adjusting for patient and tumor characteristics and treatment type (Data Supplement).

ity was maintained (Table 2 and Fig 1C, left and right panels, respectively). Of the individual comorbidities that constituted the CCI (Data Supplement), only congestive heart failure and dementia affected PCa-specific mortality after adjusting for patient and tumor characteristics and treatment type (Data Supplement). Table 2. Comparison of Mortality Figures and Stabilized Inverse Probability Weighting Adjusted Subdistribution HRs for Deaths From PCa and Other Causes Between CCI Groups, Stratified by Treatment Type Fig 1. Prostate cancer (PCa)–specific (left panels) and other-cause (right panels) survival for (A) a full cohort of patients without adjustments, (B) a full cohort of patients with adjustments for patient and tumor characteristics, and (C) a full cohort of patients with adjustments for patient and tumor characteristics, and treatment type. Unadjusted and adjusted Kaplan-Meier plots display cumulative survival probability and follow-up time . The number of patients at risk in each Charlson comorbidity index (CCI) group are tabulated at each time point on the x-axis with text colors corresponding to the same color for each curve on the Kaplan-Meier plot representing CCI 0 (blue), CCI 1 (gold), CCI 2 (gray), and CCI ≥ 3 (red).

probability and follow-up time . The number of patients at risk in each Charlson comorbidity index (CCI) group are tabulated at each time point on the x-axis with text colors corresponding to the same color for each curve on the Kaplan-Meier plot representing CCI 0 (blue), CCI 1 (gold), CCI 2 (gray), and CCI ≥ 3 (red). In a subset analysis in which comparisons were repeated separately within each treatment subgroup (Table 2 and Figs 2A-2D, left panels), the only treatment for which an effect of comorbidity on PCa-specific mortality was consistently observed for all CCI groups compared with no comorbidity (CCI score, 0) in unadjusted data was WW, which was lost after adjusting for patient and tumor characteristics (Table 2). For the RP, RT, and ADT subgroups, limited differences in PCa-specific mortality were observed in both unadjusted and adjusted data (Table 2). Additional analyses of a subgroup of patients who were diagnosed and treated between 2008 and 2012 for which information was available on neoadjuvant ADT before RT did not reveal any association between CCI and PCa-specific mortality for patients who were treated with RT (Data Supplement). The most striking effect of comorbidity on other-cause mortality was observed in patients who were treated with RP, where patients had a 4.42-fold hazard (95% CI, 3.23 to 6.06) in other-cause mortality with a CCI score of ≥ 3 (Table 2 and Fig 2A, right panel).

ty for patients who were treated with RT (Data Supplement). The most striking effect of comorbidity on other-cause mortality was observed in patients who were treated with RP, where patients had a 4.42-fold hazard (95% CI, 3.23 to 6.06) in other-cause mortality with a CCI score of ≥ 3 (Table 2 and Fig 2A, right panel). Fig 2. Prostate cancer (PCa)–specific (left panels) and other-cause (right panels) survival for a cohort of patients with adjustments for patient and tumor characteristics, stratified by treatment type. (A) Radical prostatectomy, (B) radical radiotherapy, (C) androgen deprivation therapy, and (D) watchful waiting. Adjusted Kaplan-Meier plots display cumulative survival probability and follow-up time. The number of patients at risk in each Charlson comorbidity index (CCI) group are tabulated at each time point on the x-axis with text colors corresponding to the same color for each curve on the Kaplan-Meier plot representing CCI 0 (blue), CCI 1 (gold), CCI 2 (gray), and CCI ≥ 3 (red).

probability and follow-up time. The number of patients at risk in each Charlson comorbidity index (CCI) group are tabulated at each time point on the x-axis with text colors corresponding to the same color for each curve on the Kaplan-Meier plot representing CCI 0 (blue), CCI 1 (gold), CCI 2 (gray), and CCI ≥ 3 (red). DISCUSSION In this large observational study of men with PCa who were treated with RP, RT, ADT, or WW, with a maximum follow-up to 16.99 years, we did not observe a statistically significant effect of comorbidity (as determined by CCI) on PCa-specific mortality after statistical adjustments for patient and tumor characteristics and treatment type; however, we did see an association with a CCI score of ≥ 3 compared with no comorbidity (CCI score, 0) on PCa-specific mortality after adjusting for patient and tumor characteristics alone, which was lost after additional adjustments for treatment type—the reasons for this effect are unclear. Overall, our findings suggest that, after adjusting for patient and tumor characteristics, comorbidity does not seem to significantly impact the risk of dying from PCa after radical treatment (RP or RT) or WW.

alone, which was lost after additional adjustments for treatment type—the reasons for this effect are unclear. Overall, our findings suggest that, after adjusting for patient and tumor characteristics, comorbidity does not seem to significantly impact the risk of dying from PCa after radical treatment (RP or RT) or WW. Population-based studies that incorporate all disease states24-28 and a recent systematic review29 suggest that RP may be more effective than RT for the treatment of localized PCa, although a common criticism of these analyses is that the differences observed, at least in part, may be a result of the distribution of comorbidity between treatment groups. The premise here is that lesser comorbidity in the RP group may result in residual bias that accounts for the lower PCa-specific mortality compared with RT; however, considering the lack of an association between comorbidity and PCa-specific mortality for both RP and RT in our present study, differences in oncologic outcomes that have been observed in population-based studies may not be a result of comorbidity-associated bias, although residual confounding is difficult to preclude without random treatment allocation.

on between comorbidity and PCa-specific mortality for both RP and RT in our present study, differences in oncologic outcomes that have been observed in population-based studies may not be a result of comorbidity-associated bias, although residual confounding is difficult to preclude without random treatment allocation. To date, the two largest population-based studies have reported similar findings for an effect of comorbidity on other-cause mortality, with discordant findings for an effect on PCa-specific mortality using competing-risk models. A PCBaSe-based study identified an association between increased comorbidity and PCa-specific and all-cause mortality.9 A similar effect of comorbidity was identified in all-cause, but not PCa-specific mortality in a study that was based on the Prostate Cancer Outcomes Study population-based cohort from the National Cancer Institute SEER program.30 Consistent with this, another large SEER-based study, which excluded all patients who underwent radical treatment, observed similar effects of increased comorbidity on overall survival by using a competing-risks model.31 Although the reasons for the differences in findings between the PCBaSe 9 and SEER30 studies are unclear, possible explanations are the patient self-reporting of comorbidity in the SEER data set30 or the misclassification of comorbidity in the PCBaSe data set.9 Other smaller studies have demonstrated an association between increased comorbidity and risk of other cause–mortality and all cause–mortality, but not PCa-specific mortality in men with localized PCa who were treated with RP, RT, ADT, or WW.32-34 These studies are limited by potential bias associated with the retrospective nature of single-center data and small numbers34 or a complete absence of patients who were treated with ADT or WW.32,33

cause–mortality, but not PCa-specific mortality in men with localized PCa who were treated with RP, RT, ADT, or WW.32-34 These studies are limited by potential bias associated with the retrospective nature of single-center data and small numbers34 or a complete absence of patients who were treated with ADT or WW.32,33 The above population-based studies9,30,31 that used competing-risk models can help to determine whether the risk of death from PCa differs between groups, allowing certain men to be protected from PCa-specific mortality because of death from other causes. The findings may be relevant in counseling men with comorbidity about radical treatment options for which treatment choices are based on shared decision-making with knowledge of the potential benefits and harms of treatment when considering an individual’s risk of death from PCa and/or other causes. For example, a patient with comorbidity and a greater risk of other-cause mortality compared with PCa-specific mortality may wish to withhold treatment to limit PCa treatment–related morbidity without compromising oncologic outcome. In comparison, our cause-of-death–specific analyses allow us to emulate how comorbidity affects PCa-specific mortality under the hypothetical scenario that no men died of causes other than PCa. We used this method as we wished to determine whether a patient with comorbidity is more or less likely to die of PCa compared with a noncomorbid patient.

death–specific analyses allow us to emulate how comorbidity affects PCa-specific mortality under the hypothetical scenario that no men died of causes other than PCa. We used this method as we wished to determine whether a patient with comorbidity is more or less likely to die of PCa compared with a noncomorbid patient. Our study has several strengths. First, it utilizes a large, unique composite population-based data set that links nine national registries with complete data collection from the time of diagnosis and during follow-up of more than 98% of men who were diagnosed with PCa in Sweden since 1998, with well-validated patient and clinicopathologic variables. The accuracy of the study end point is high, as the cause of death register in Sweden has been found to be reliable for the correct assignment of cause of death for patients with PCa.16 As a result of long-term follow-up, it is possible to identify differences in mortality > 10 years after diagnosis and treatment, thereby increasing the sensitivity of the analyses.

the cause of death register in Sweden has been found to be reliable for the correct assignment of cause of death for patients with PCa.16 As a result of long-term follow-up, it is possible to identify differences in mortality > 10 years after diagnosis and treatment, thereby increasing the sensitivity of the analyses. We used stabilized inverse probability weighting propensity score–based statistical adjustments for preselected variables to emulate a situation in which groups to be compared are made to have similar characteristics at baseline; however, unmeasured or misclassified confounders may still bias the estimates. For example, despite the apparent validity of the cause-of-death register,16 Swedish National In-Patient Register,18 and Swedish Cancer Register, which constitute the CCI score in PCBaSe, it is conceivable that the misclassification of cause of death, or insensitivity of comorbidity estimation, or ability to detect certain conditions may result in bias. In addition, our PCa- and other cause–specific survival analyses may result in bias because of informative censoring if those patients who died early from other causes had at a different risk of death from PCa than did those who survived longer. Hence, similar analyses of large population-based PCa data sets are needed to validate our findings.

other cause–specific survival analyses may result in bias because of informative censoring if those patients who died early from other causes had at a different risk of death from PCa than did those who survived longer. Hence, similar analyses of large population-based PCa data sets are needed to validate our findings. Notwithstanding these limitations, our study suggests that comorbidity affects other cause–mortality but not PCa-specific–mortality, accounting for patient and tumor characteristics and treatment type. Regardless of radical treatment type (RP or RT), increased comorbidity does not seem to significantly affect the risk of dying from PCa. Consequently, differences in oncologic outcome that were observed in population-based comparative effectiveness studies of PCa treatments do not seem to be a result of the varying distribution of comorbidity among treatment groups. Funded by a joint Royal College of Surgeons/Cancer Research UK Clinician Scientist Fellowship in Surgery (C19198/A15339; P.R.), the Barts Charity and the Orchid Charity (P.R.), the National Institute for Health Research Oxford Biomedical Research Centre based at Oxford University Hospitals NHS Trust and the University of Oxford (P.S.), grants from the Swedish Cancer Society and and Stockholm County Council (O.A.), and the Swedish Research Council (Grant No. K2013-99X-22283-01-3; N.P.W.). The Prostate Cancer Database Sweden is funded by the Swedish Research Council (825-2012-5047).

rd University Hospitals NHS Trust and the University of Oxford (P.S.), grants from the Swedish Cancer Society and and Stockholm County Council (O.A.), and the Swedish Research Council (Grant No. K2013-99X-22283-01-3; N.P.W.). The Prostate Cancer Database Sweden is funded by the Swedish Research Council (825-2012-5047). Presented at the Annual Meeting of the British Association of Urological Surgeons Section of Academic Urology, London, United Kingdom, January 12, 2017, and at the Annual Meeting of the American Urological Association, Boston, MA, May 12, 2017. ACKNOWLEDGMENT This work was made possible by the continuous work of the National Prostate Cancer Register of Sweden steering group: Pär Stattin (chairman), Anders Widmark, Camilla Thellenberg Karlsson, Ove Andrén, Ann-Sofi Fransson, Magnus Törnblom, Stefan Carlsson, Marie Hjälm-Eriksson, David Robinson, Mats Andén, Jonas Hugosson, Ingela Franck Lissbrant, Maria Nyberg, Ola Bratt, René Blom, Lars Egevad, Calle Walller, Olof Akre, Per Fransson, Eva Johansson, Fredrik Sandin, and Karin Hellström. AUTHOR CONTRIBUTIONS Conception and design: Prabhakar Rajan, Prasanna Sooriakumaran, Tommy Nyberg, Gunnar Steineck, N. Peter Wiklund Administrative support: Gunnar Steineck Provision of study materials or patients: Olof Akre Collection and assembly of data: Tommy Nyberg Data analysis and interpretation: All authors Manuscript writing: All authors Final approval of manuscript: All authors Accountable for all aspects of the work: All authors

AUTHOR CONTRIBUTIONS Conception and design: Prabhakar Rajan, Prasanna Sooriakumaran, Tommy Nyberg, Gunnar Steineck, N. Peter Wiklund Administrative support: Gunnar Steineck Provision of study materials or patients: Olof Akre Collection and assembly of data: Tommy Nyberg Data analysis and interpretation: All authors Manuscript writing: All authors Final approval of manuscript: All authors Accountable for all aspects of the work: All authors AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST Effect of Comorbidity on Prostate Cancer–Specific Mortality: A Prospective Observational Study The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/jco/site/ifc. Prabhakar Rajan No relationship to disclose Prasanna Sooriakumaran No relationship to disclose Tommy Nyberg No relationship to disclose Olof Akre No relationship to disclose Stefan Carlsson No relationship to disclose Lars Egevad No relationship to disclose Gunnar Steineck No relationship to disclose N. Peter Wiklund Research Funding: Intuitive Surgical

.2-4 This contrasts with the increasing disease burden that is associated with cancer, which is currently the leading cause of death in high-income countries,5 and, after mental health services and circulatory diseases, cancer services make up the third largest category of spending in contemporary health care systems.6 Approximately four of 10 cancers are attributable to lifestyle and environmental risk factors, including smoking, alcohol consumption, dietary factors, occupational exposures, and sexual and reproductive history.7-9 Disparities in cancer incidence among people from sexual minorities compared with heterosexual populations are likely. Lifetime exposure to risk factors among sexual minority and heterosexual populations is likely to differ—for example, smoking initiation is higher among lesbian, gay, and bisexual young people in both the United Kingdom10 and the United States.11 Hormonal factors are also likely to be important for some cancers; child bearing and the use of oral contraceptives are associated with a risk of female breast and ovarian cancer,12 and lesbian and heterosexual women are differently exposed to these two factors.13 In addition, HIV prevalence in the United Kingdom is 0.2%, but among men who have sex with men (age 15 to 44 years), it is 5%14; immune deficiency is associated with an increased risk of several cancers.15,16 Approximately 4.8% of cancer diagnoses worldwide in 2008 were attributable to human papillomavirus (HPV), with differences in exposure burden in men and women.17 Understanding how the risk of cancer varies by sexual orientation is therefore of particular importance to help understand where best to target preventive efforts.

ximately 4.8% of cancer diagnoses worldwide in 2008 were attributable to human papillomavirus (HPV), with differences in exposure burden in men and women.17 Understanding how the risk of cancer varies by sexual orientation is therefore of particular importance to help understand where best to target preventive efforts. In the United States, the Institute of Medicine in 2011 called for research investment in the health of sexual minorities, including basic epidemiologic research, which highlighted the lack of relevant evidence.18 Similarly, ASCO has called for research on sexual orientation–related disparities and increased data collection.19 In the United Kingdom. cancer charities have highlighted the continuing limitations of data on lesbian, gay, and bisexual people with cancer.20 It is usually difficult to study sexual orientation in population health and epidemiologic studies, primarily because this information is simply not known or not collected.19 Nonetheless, provided that items on sexual orientation are included, patient experience surveys provide unique opportunities for acquiring insight into the risk of cancer in sexual minorities.21

on in population health and epidemiologic studies, primarily because this information is simply not known or not collected.19 Nonetheless, provided that items on sexual orientation are included, patient experience surveys provide unique opportunities for acquiring insight into the risk of cancer in sexual minorities.21 Against this background, we used data from two English patient surveys, the General Practice Patient Survey (GPPS) and the Cancer Patient Experience Survey (CPES), to examine two research questions: Do women and men from sexual minorities report a cancer diagnosis in the previous 5 years more or less frequently than heterosexual women and men? And, among recently treated survivors of cancer, is there variation between cancer sites in the proportion of men and women who report gay, lesbian, or bisexual sexual orientation?

n and men from sexual minorities report a cancer diagnosis in the previous 5 years more or less frequently than heterosexual women and men? And, among recently treated survivors of cancer, is there variation between cancer sites in the proportion of men and women who report gay, lesbian, or bisexual sexual orientation? PATIENTS AND METHODS Data GPPS is a national survey of the patient experience of primary care and is sent by postal mail to approximately 2.7 million patients in England age ≥ 18 years who have been continuously registered with a general practice for at least 6 months, with a respondent sample of approximately 1 million (response rate approximately 37%). A stratified sample of patients was drawn from the practice list of each general practice in England, with oversampling of patients from practices that were known from prior surveys to have low response rates. Full details are published in the technical report.22 Data from 2011/2012 were used in this analysis. Data from GPPS have been previously used to describe the patient experience, health-related quality of life, and health service utilization of sexual minorities.23,24 CPES is a separate survey of recently treated survivors of cancer25 that is sent annually to all patients age ≥ 16 years who were treated for cancer in a National Health Service hospital in England during a 3-month period. Anonymous data from 2010, 2011/2012, 2013, and 2014 were obtained via the UK Data Archive, and full survey details are available.26-29

treated survivors of cancer25 that is sent annually to all patients age ≥ 16 years who were treated for cancer in a National Health Service hospital in England during a 3-month period. Anonymous data from 2010, 2011/2012, 2013, and 2014 were obtained via the UK Data Archive, and full survey details are available.26-29 This study involved the secondary analysis of previously collected anonymous data, for which formal ethical approval is not required.30 All cancer diagnoses are included in this analysis; however, cells counts of fewer than six individuals are suppressed in reporting, which is in line with best practice.31 Overall and Site-Specific Diagnosis of Cancer In the GPPS, respondents are asked “Which, if any, of the following medical conditions do you have?,” with 16 response options, including “Cancer in the last 5 years,” plus “None of these conditions” and “I would prefer not to say”. “Prefer not to say” responses and responses for which no options were ticked were coded as missing for this analysis. No additional detail about the nature of the cancer diagnosis—that is, in relation to cancer site—is available in this survey. In CPES, patients were identified for inclusion in the survey sampling frame when the main hospital record for inpatient or outpatient treatment recorded an International Classification of Diseases, Tenth Revision cancer diagnosis code for inpatient or outpatient cancer treatment. In line with previous research,32 but with the addition of Kaposi’s sarcoma, we included 38 common and rarer cancer site groups in the analysis (Table A8, online-only).

ient treatment recorded an International Classification of Diseases, Tenth Revision cancer diagnosis code for inpatient or outpatient cancer treatment. In line with previous research,32 but with the addition of Kaposi’s sarcoma, we included 38 common and rarer cancer site groups in the analysis (Table A8, online-only). Sexual Orientation Survey questions were used to identify respondents’ sexual orientation in both surveys. In GPPS, “Which of the following best describes how you think of yourself?” had the following possible responses: “Heterosexual/straight,” “Gay/Lesbian,” “Bisexual,” “Other,” or “I would prefer not to say”. In CPES, “Which of the following best describes your sexual orientation?” has the following possible responses: “Heterosexual/straight (opposite sex),” “Bisexual (both sexes),” “Gay or Lesbian (same sex),” “Other,” of “Prefer not to answer.” Demographic Information In GPPS, survey responses for age 18 to 24 years, then 10-year age groups to age ≥ 85 years, gender, and ethnicity in five groups (ONS2011) were used. In CPES, hospital record recorded age and gender (as these are more complete), and survey-reported ethnicity in six groups (ONS2001) were used. For both surveys, the Index of Multiple Deprivation—a small geographic area measure of socioeconomic deprivation, derived from respondents’ postcodes—was used and divided into five groups by using quintile-defining cut points.33

these are more complete), and survey-reported ethnicity in six groups (ONS2001) were used. For both surveys, the Index of Multiple Deprivation—a small geographic area measure of socioeconomic deprivation, derived from respondents’ postcodes—was used and divided into five groups by using quintile-defining cut points.33 Analysis In all analyses, women and men are considered separately, and all respondents—both those from sexual minorities and those who report heterosexual sexual orientation—are included. GPPS survey data are provided with weights that account for design and nonresponse22; therefore, descriptive analysis is presented for weighted data. For CPES, as all cancer cases within the sampling period were selected, design weights are not applicable, and only unweighted data are presented. For our first analysis, using data from GPPS we performed logistic regression to examine variations in report of cancer diagnosis in the last 5 years by sexual orientation, including a test of any difference between gay or lesbian and bisexual individuals.

Analysis In all analyses, women and men are considered separately, and all respondents—both those from sexual minorities and those who report heterosexual sexual orientation—are included. GPPS survey data are provided with weights that account for design and nonresponse22; therefore, descriptive analysis is presented for weighted data. For CPES, as all cancer cases within the sampling period were selected, design weights are not applicable, and only unweighted data are presented. For our first analysis, using data from GPPS we performed logistic regression to examine variations in report of cancer diagnosis in the last 5 years by sexual orientation, including a test of any difference between gay or lesbian and bisexual individuals. For the second analysis, we used data from recently treated survivors of cancer who responded to CPES with sexual orientation as a binary outcome—grouping gay or lesbian and bisexual respondents—and cancer site as covariate. This analysis aimed to reveal overall patterns of variation between cancer sites in the proportion of men and women who report gay, lesbian, or bisexual sexual orientation. For a particular cancer site, higher or lower odds reflect differences in cancer diagnosis among people from sexual minorities compared with the reference site (breast in women, prostate in men). Our choice of these reference sites was based on analytical considerations—that is, because they were the cancers with the largest sample size, which allowed the most precise comparisons. In addition, from these models, we predicted the adjusted percentage of women and men with cancer of a particular site who were expected to report lesbian or bisexual or gay or bisexual sexual orientation should they have the same age composition as all included survey responders with this diagnosis (this percentage is also known as a recycled prediction).

dicted the adjusted percentage of women and men with cancer of a particular site who were expected to report lesbian or bisexual or gay or bisexual sexual orientation should they have the same age composition as all included survey responders with this diagnosis (this percentage is also known as a recycled prediction). Adjusting for ethnicity, deprivation, survey wave (CPES), and GP practice (GPPS) or hospital of treatment (CPES; using a random effect for organization) had a minimal effect on coefficients for cancer site, and, given this, to reduce the amount of missing observations as a result of incomplete information on deprivation and ethnicity, these variables were dropped from the main analyses, which are only adjusted for age (Appendix Tables A1 to A3, online only). Supplementary Analyses We explored a series of sensitivity analyses. First, we considered each response option to the sexual orientation question separately—that is, “Gay/Lesbian,” “Bisexual,” “Other,” “I would prefer not to say,” and missing responses were compared with the response, “Heterosexual/straight,” to examine potential associations between our outcomes and patient groups other than those that endorsed heterosexual, gay/lesbian, and bisexual response options. Second, for CPES analysis, we also restricted the analysis to people who had been diagnosed with cancer in the past year to consider a population that more closely represented incident cancer cases34 and to explore the potential impact of the same respondents being included across survey waves.

sexual response options. Second, for CPES analysis, we also restricted the analysis to people who had been diagnosed with cancer in the past year to consider a population that more closely represented incident cancer cases34 and to explore the potential impact of the same respondents being included across survey waves. RESULTS Among 796,594 respondents from the population-based GPPS sampling frame, 32,437 of all respondents (3.0%) reported cancer in the past 5 years. Of the 12,177 respondents (2.1%) from sexual minorities, 361 (1.9%) reported cancer in the past 5 years. Among the 240,010 recently treated survivors of cancer who responded to CPES between 2010 and 2014, there were 2,199 respondents (0.9%) who endorsed a sexual minority orientation. Before adjustment for age, people from sexual minorities are less likely to report cancer in the past 5 years, as, on average, people who report nonheterosexual sexual orientation are younger than those who report heterosexual sexual orientation.23 After adjusting for age, GPPS data provided no evidence of a difference between heterosexual and lesbian or bisexual women (odds ratio [OR], 1.14; 95% CI, 0.94 to 1.37; P = .19); however, gay or bisexual men were more likely to report cancer in the past 5 years than heterosexual men (OR, 1.31; 95% CI, 1.15 to 1.50; P < .001), with evidence of a difference between gay (OR, 1.45; 95% CI, 1.24 to 1.69) and bisexual men (OR, 1.00; 95% CI, 0.78 to 1.30; Table 2).

.14; 95% CI, 0.94 to 1.37; P = .19); however, gay or bisexual men were more likely to report cancer in the past 5 years than heterosexual men (OR, 1.31; 95% CI, 1.15 to 1.50; P < .001), with evidence of a difference between gay (OR, 1.45; 95% CI, 1.24 to 1.69) and bisexual men (OR, 1.00; 95% CI, 0.78 to 1.30; Table 2). Although lesbian or bisexual women represented 0.7% of all female CPES responders (any cancer site), they represented 2.3% (adjusted, 2.1%) of women with oropharyngeal cancer, 2.0%(adjusted, 1.1%) with Hodgkin lymphoma, and 1.3% (adjusted, 0.7%) of women with cervical cancer. In the same way, although gay or bisexual men represented 1.1% of all male CPES responders (any cancer site), they made up 46.4%(adjusted, 35.4%) of men with Kaposi’s sarcoma, 17.3% (adjusted, 15.7%) of men with anal cancer, 2.9% (adjusted, 1.4%) of men with Hodgkin lymphoma, 2.5% (adjusted, 0.9%) of men with testicular, and 2.5% (adjusted, 1.6%) thyroid cancers. Again, because endorsing a sexual minority orientation in the two surveys was more common in younger patients,23 crude figures are confounded by age, with the crude proportions of men and women from sexual minorities higher for cancer sites where diagnosis among young people is more common—for example, cervical and testicular cancer, and Hodgkin Lymphoma (Tables 1 and 3). Table 1. Cancer Diagnosis Frequency and Unadjusted and Adjusted Percentages of Cancer Patient Experience Survey Responders Reporting Lesbian or Bisexual Sexual Orientation by Cancer Site (women) Table 2. Cancer Prevalence by Sexual Orientation Among GPPS Responders

Although lesbian or bisexual women represented 0.7% of all female CPES responders (any cancer site), they represented 2.3% (adjusted, 2.1%) of women with oropharyngeal cancer, 2.0%(adjusted, 1.1%) with Hodgkin lymphoma, and 1.3% (adjusted, 0.7%) of women with cervical cancer. In the same way, although gay or bisexual men represented 1.1% of all male CPES responders (any cancer site), they made up 46.4%(adjusted, 35.4%) of men with Kaposi’s sarcoma, 17.3% (adjusted, 15.7%) of men with anal cancer, 2.9% (adjusted, 1.4%) of men with Hodgkin lymphoma, 2.5% (adjusted, 0.9%) of men with testicular, and 2.5% (adjusted, 1.6%) thyroid cancers. Again, because endorsing a sexual minority orientation in the two surveys was more common in younger patients,23 crude figures are confounded by age, with the crude proportions of men and women from sexual minorities higher for cancer sites where diagnosis among young people is more common—for example, cervical and testicular cancer, and Hodgkin Lymphoma (Tables 1 and 3). Table 1. Cancer Diagnosis Frequency and Unadjusted and Adjusted Percentages of Cancer Patient Experience Survey Responders Reporting Lesbian or Bisexual Sexual Orientation by Cancer Site (women) Table 2. Cancer Prevalence by Sexual Orientation Among GPPS Responders Table 3. Cancer Diagnosis Frequency and Unadjusted and Adjusted Percentages of Cancer Patient Experience Survey Responders Reporting Gay or Bisexual Sexual Identity by Cancer Site (men)

Table 1. Cancer Diagnosis Frequency and Unadjusted and Adjusted Percentages of Cancer Patient Experience Survey Responders Reporting Lesbian or Bisexual Sexual Orientation by Cancer Site (women) Table 2. Cancer Prevalence by Sexual Orientation Among GPPS Responders Table 3. Cancer Diagnosis Frequency and Unadjusted and Adjusted Percentages of Cancer Patient Experience Survey Responders Reporting Gay or Bisexual Sexual Identity by Cancer Site (men) After adjusting for age, there continues to be statistical evidence that the proportion of people from sexual minorities varies between cancer sites (women, P = .0002; men, P < .0001). Whereas evidence for overall variation is statistically significant, the number of cases for many sites is small, which resulted in wide CIs (Figs 1 and 2). For most cancer sites that were examined in our adjusted analysis (30 of 33 in women, 28 of 32 in men), there was no evidence that sexual minorities were over- or under-represented compared with the most common cancers in each gender (female, breast; male, prostate); however, there were a few notable differences, including some infection-related (HIV or HPV) cancers. Lesbian/bisexual women are more frequently represented among women with oropharyngeal cancer (OR, 3.2; 95% CI, 1.7 to 6.0), and less frequently represented in anal (OR, 0.3; 95% CI, 0.0 to 2.0), and vulval/vaginal cancers (OR, 0.7; 95% CI, 0.2 to 2.1), although CIs are wide. Gay or bisexual men are relatively more frequently represented among men with Kaposi’s sarcoma (OR, 48.2; 95% CI, 22.0 to 105.6), anal (OR, 15.5; 95% CI, 11.0 to 21.9), and penile cancer (OR, 1.8; 9% CI, 0.9 to 3.7). In addition, lesbian or bisexual women are more frequently represented among women with mesothelioma, stomach, and endometrial cancers, but less among liver and esophageal cancers. Gay or bisexual men are more frequently represented among men with thyroid and oral cancers, melanoma, and Hodgkin lymphoma, and are relatively less frequently represented in liver and stomach cancers, leukemia, and mesothelioma.

sothelioma, stomach, and endometrial cancers, but less among liver and esophageal cancers. Gay or bisexual men are more frequently represented among men with thyroid and oral cancers, melanoma, and Hodgkin lymphoma, and are relatively less frequently represented in liver and stomach cancers, leukemia, and mesothelioma. Fig 1. Odds ratios of specific cancer site diagnosis by lesbian or bisexual orientation among women with cancer, adjusted for age (Cancer Patient Experience Survey). Diagnoses represented with gold circles indicate fewer than six women with this diagnosis reporting lesbian or bisexual sexual identity (CPES); these diagnoses were included in the analysis model in the same way as other cancers, however the gold circles highlight that these results are based on relatively small numbers of cases. Fig 2. Odds ratios of specific cancer site diagnosis by gay or bisexual orientation among men with cancer, adjusted for age.

Fig 1. Odds ratios of specific cancer site diagnosis by lesbian or bisexual orientation among women with cancer, adjusted for age (Cancer Patient Experience Survey). Diagnoses represented with gold circles indicate fewer than six women with this diagnosis reporting lesbian or bisexual sexual identity (CPES); these diagnoses were included in the analysis model in the same way as other cancers, however the gold circles highlight that these results are based on relatively small numbers of cases. Fig 2. Odds ratios of specific cancer site diagnosis by gay or bisexual orientation among men with cancer, adjusted for age. In supplementary analyses, self-reported cancer prevalence among GPPS responders was lower among women with a missing response regarding their sexual orientation (P = .027) and men who stated, “Prefer not to say” (P < .001), with a small but significant (men, P = .0005; women, P = .001) variation in missing responses by cancer site, and “Prefer not to say” (men, P < .001). Considering respondents who had been diagnosed only in the past year, findings were consistent with the main analyses of CPES respondents. Variation in cancer diagnosis was also consistent when considering gay or lesbian and bisexual CPES responders separately (Appendix Tables A4 to A6, online only).

” (men, P < .001). Considering respondents who had been diagnosed only in the past year, findings were consistent with the main analyses of CPES respondents. Variation in cancer diagnosis was also consistent when considering gay or lesbian and bisexual CPES responders separately (Appendix Tables A4 to A6, online only). DISCUSSION We report large-scale evidence from two nationwide patient surveys in England, including information on self-reported sexual orientation, to explore whether women and men from sexual minorities report a cancer diagnosis in the previous 5 years more or less frequently than heterosexual women and men, and, among recently treated survivors of cancer, whether there is variation between cancer sites in the proportion of men and women who report gay, lesbian, or bisexual sexual orientation. We find that gay or bisexual men are more likely to have had cancer in the past 5 years than heterosexual men of the same age, although there was no evidence of a difference in cancer prevalence between heterosexual and lesbian or bisexual women. For both men and women, sexual orientation seems to be unrelated to the diagnosis of most cancer sites, particularly the more common cancers, but sexual minorities are over- or under-represented among patients of certain rarer sites. Cancer sites associated with HPV—and HIV—infection are those with the greatest degree of variation in the proportion of men and women from sexual minorities.

diagnosis of most cancer sites, particularly the more common cancers, but sexual minorities are over- or under-represented among patients of certain rarer sites. Cancer sites associated with HPV—and HIV—infection are those with the greatest degree of variation in the proportion of men and women from sexual minorities. Regarding cancer prevalence, against limited overall evidence, our findings are consistent with a single other similar source, which also found similar cancer prevalence among heterosexual and sexual minority women, but increased odds among gay and bisexual men compared with heterosexual men.35 With regard to HPV-associated cancers, increased prevalence of anal cancer among gay men has been previously described36; however, to our knowledge, the relative over-representation of lesbian and bisexual women among women with oropharyngeal cancer has not previously been described and constitutes a novel finding. A history of performing oral sex is a known risk factor for oropharyngeal cancer.37 Previous evidence indicates that there is a higher probability of HPV transmission via vaginal oral sex compared with penile oral sex,38,39 which would result in a higher burden of oral HPV infection among lesbian and bisexual women.

g. A history of performing oral sex is a known risk factor for oropharyngeal cancer.37 Previous evidence indicates that there is a higher probability of HPV transmission via vaginal oral sex compared with penile oral sex,38,39 which would result in a higher burden of oral HPV infection among lesbian and bisexual women. Cervical cancer is associated with HPV infection, yet we do not find that lesbian or bisexual women are under-represented among women with cervical cancer. Our findings may be explained by a relatively low uptake of screening40-42 plus an earlier initiation of sexual intercourse.43 There have been inconsistent messages in the past about whether lesbian women need to attend cervical screening programs,20,40 but our data do not suggest any reduced need. There are also some unexpected differences in cancer diagnosis by sexual orientation, in particular, the excess risk of mesothelioma and stomach cancer in lesbian women. Here, our findings may add novel insights, although with the caution that numbers are relatively low. Nonetheless, differences by sexual orientation among women with regard to occupational exposures, smoking (contributing to cancer risk across several sites), or dietary factors may be important, although this is speculative. The higher risk of endometrial cancer among lesbian or bisexual women is also surprising and is inconsistent with prior evidence.44

rientation among women with regard to occupational exposures, smoking (contributing to cancer risk across several sites), or dietary factors may be important, although this is speculative. The higher risk of endometrial cancer among lesbian or bisexual women is also surprising and is inconsistent with prior evidence.44 There are limitations that are inherent in all survey research, and, in this study, one such limitation is that the CPES analysis is based on treated patients, rather than population-based incident cases, and so rates cannot be estimated; however, our sensitivity analysis using only cases diagnosed in the last year—more similar to an incident population34—provided results that were consistent with the main analyses presented. Additional strengths of our study are its large analyses samples, the examination of both common and rare cancer sites, and the use of well-characterized national survey data.25,45-48 Although response rates for CPES are high (64% to 67%), response rates for GPPS were 37%; however, response rates alone are a poor indicator of bias.49 In addition, a randomized controlled trial demonstrated no variation in GPPS response rates when a question about the sexual orientation of the participant was included or excluded; this does not exclude, but greatly mitigates, the potential for survey nonparticipation bias by sexual orientation.45

indicator of bias.49 In addition, a randomized controlled trial demonstrated no variation in GPPS response rates when a question about the sexual orientation of the participant was included or excluded; this does not exclude, but greatly mitigates, the potential for survey nonparticipation bias by sexual orientation.45 We acknowledge that some people who identify as lesbian, gay, or bisexual may be unwilling to acknowledge their identity in a survey50; increased homelessness among young lesbian, gay, and bisexual people in the United Kingdom may also lead to additional under-representation among survey responders.51 In addition, item nonresponse is another concern, with people who do not respond to the sexual orientation question at all (in GPPS) possibly also being less likely to report any long-term conditions. People who report greater concerns about privacy are less likely to respond to sensitive demographic survey questions,52 and it is likely that the same mechanism may apply in our study context. Sexual orientation and sexual behavior are different constructs; in this work, we consider cancer diagnosis associated with sexual orientation, although we acknowledge that HPV infection–associated cancer risk is primarily related to sexual behavior. The survey instruments encompassed heterosexual, gay, lesbian, and bisexual orientation, which is consistent with survey questions that were developed and validated by the UK Office of National Statistics53 that, however, do not encompass all sexual orientation and gender identity groups.

ly related to sexual behavior. The survey instruments encompassed heterosexual, gay, lesbian, and bisexual orientation, which is consistent with survey questions that were developed and validated by the UK Office of National Statistics53 that, however, do not encompass all sexual orientation and gender identity groups. HPV vaccination presents an important opportunity for cancer prevention.43 In the United Kingdom, the current vaccination program covers girls age 12 to 13 years, and a recent pilot HPV vaccination schedule for men who have sex with men was rolled out in 2016.54 Modeling work is still in progress to decide whether all boys should receive the vaccination at age 12 to 13 years alongside girls.55 The research presented here provides additional epidemiologic evidence to inform decisions about the most equitable, effective, and cost-effective HPV vaccination schedules. Evidence from the United States suggests that HPV vaccination rates among lesbian women are low.56 Our research provides additional evidence that should particularly support efforts to encourage vaccination among lesbian and bisexual women.

ut the most equitable, effective, and cost-effective HPV vaccination schedules. Evidence from the United States suggests that HPV vaccination rates among lesbian women are low.56 Our research provides additional evidence that should particularly support efforts to encourage vaccination among lesbian and bisexual women. This work presents population-based evidence about cancer prevalence among men and women from sexual minorities and about the relative frequencies of people from sexual minorities with common and rarer cancer diagnoses among recently treated survivors of cancer. Demographic data on cancer among people from sexual minorities are scarce20; these findings begin to address this evidential need. Finally, our research also highlights the importance of HPV vaccination among gay, lesbian, and bisexual women and men. G.L. is supported by a Cancer Research UK Advanced Clinician Scientist Fellowship (C18081/A18180). AUTHOR CONTRIBUTIONS Conception and design: Catherine L. Saunders Collection and assembly of data: Catherine L. Saunders, Gary A. Abel, Georgios Lyratzopoulos Data analysis and interpretation: All authors Manuscript writing: All authors Final approval of manuscript: All authors Accountable for all aspects of the work: All authors

AUTHOR CONTRIBUTIONS Conception and design: Catherine L. Saunders Collection and assembly of data: Catherine L. Saunders, Gary A. Abel, Georgios Lyratzopoulos Data analysis and interpretation: All authors Manuscript writing: All authors Final approval of manuscript: All authors Accountable for all aspects of the work: All authors AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST Associations Between Sexual Orientation and Overall and Site-Specific Diagnosis of Cancer: Evidence From Two National Patient Surveys in England The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/jco/site/ifc. Catherine L. Saunders No relationship to disclose Catherine Meads No relationship to disclose Gary A. Abel No relationship to disclose Georgios Lyratzopoulos No relationship to disclose

AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST Associations Between Sexual Orientation and Overall and Site-Specific Diagnosis of Cancer: Evidence From Two National Patient Surveys in England The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/jco/site/ifc. Catherine L. Saunders No relationship to disclose Catherine Meads No relationship to disclose Gary A. Abel No relationship to disclose Georgios Lyratzopoulos No relationship to disclose Appendix Appendix Tables A1 to A3 In the main analyses presented in this article, we adjust only for age as a categorical variable. We also consider as possible confounders race and ethnicity, an additional age term, socioeconomic inequalities, survey wave (year), and hospital/general practice of treatment. As including these variables did not change the magnitude of the effect sizes of the association between sexual orientation and cancer diagnosis—results presented in Appendix Table A1 for General Practice Patient Survey (GPPS) and Appendix Tables A2 and A3 for Cancer Patient Experience Survey (CPES)—we did not include them in the final analysis models, which were presented only adjusted for age and stratified by sex. In addition, including additional covariates in our analysis models led to issues with smaller sample sizes—deprivation and ethnicity information is missing for some respondents—and issues of perfect prediction for some rarer cancer site diagnoses. Although the final analysis models are therefore parsimonious, unmeasured confounding is not expected to be a major concern.

nalysis models led to issues with smaller sample sizes—deprivation and ethnicity information is missing for some respondents—and issues of perfect prediction for some rarer cancer site diagnoses. Although the final analysis models are therefore parsimonious, unmeasured confounding is not expected to be a major concern. The only coefficient that does change is that for Kaposi’s sarcoma in men after adjusting for hospital (reduction from 48.5 to 19.2; Appendix Table A3). Even after adjustment, this remains the cancer diagnosis with the strongest association with sexual orientation, and reflects the fact that patients with Kaposi’s sarcoma were only treated in a small number of hospitals overall, rather than more general confounding by region, and do not change the overall conclusions of this work. As we are interested in population-level estimates, rather than within-hospital estimates, in this analysis, we did not include this random effect in the final model.

y treated in a small number of hospitals overall, rather than more general confounding by region, and do not change the overall conclusions of this work. As we are interested in population-level estimates, rather than within-hospital estimates, in this analysis, we did not include this random effect in the final model. Appendix Tables A4 to A6 Survey questions were used to identify respondents’ sexual orientation in both surveys. In GPPS, “Which of the following best describes how you think of yourself?” had the following possible responses: “Heterosexual/straight,” “Gay/Lesbian,” “Bisexual,” “Other,” or “I would prefer not to say.” In CPES, “Which of the following best describes your sexual orientation?” had the following possible responses: “Heterosexual/straight (opposite sex),” “Bisexual (both sexes),” “Gay or Lesbian (same sex),” “Other,” or “Prefer not to answer.” To explore the sensitivity of our findings to different response options, we investigated the associations between each of the sexual orientation response options and the overall and site-specific diagnoses of cancer in a series of analyses. In particular, in the CPES analyses, the models for which the responses, “Bisexual” and “Gay” or “Lesbian,” are not grouped are consistent with the main analyses presented here, although CIs are wider, and variability in effect sizes are likely to reflect this imprecision.

c diagnoses of cancer in a series of analyses. In particular, in the CPES analyses, the models for which the responses, “Bisexual” and “Gay” or “Lesbian,” are not grouped are consistent with the main analyses presented here, although CIs are wider, and variability in effect sizes are likely to reflect this imprecision. Appendix Table A5 Results from the main analysis model (presented in Fig 1) are presented in column 1 in this table. Additional analyses explore whether there is any variation between cancer sites in the proportion of women who report “Other” or “Prefer not to say” sexual orientation, or who do not respond to this question at all, and can be compared with the findings from the main model. Although there is evidence that the proportion of women who did not respond to the sexual orientation question varies between cancer sites, the magnitude of these differences are small (all odds ratios are close to 1), particularly compared with the variation in the main model

red with the findings from the main model. Although there is evidence that the proportion of women who did not respond to the sexual orientation question varies between cancer sites, the magnitude of these differences are small (all odds ratios are close to 1), particularly compared with the variation in the main model Appendix Table A6 Although from these analyses there is evidence that the proportion of men who responded with the “Prefer not to say” option or who did not respond to the sexual orientation quesiont varies between cancer sites, the magnitude of these differences are small (all odds ratios are close to 1), particularly compared with the variation in the main model. Men with anal cancer have 2.4 times greater odds of giving a “Prefer not to say” response to the survey question about sexual identity than men with prostate cancer, and men with Kaposi’s sarcoma have 11.0 times greater odds of giving a “Prefer not to say” response. Clearly, there is some increased use of the “Prefer not to say” option among gay or bisexual men with these diagnoses, but it is not possible to determine the extent to which this option is used more frequently by all gay or bisexual men, or whether this is particularly related to these diagnoses that perhaps have greater stigma attached. Appendix Tables A7 and A8 These tables present details of the sample derivation and clinical coding of cancer sites used in the main analysis.

Appendix Table A6 Although from these analyses there is evidence that the proportion of men who responded with the “Prefer not to say” option or who did not respond to the sexual orientation quesiont varies between cancer sites, the magnitude of these differences are small (all odds ratios are close to 1), particularly compared with the variation in the main model. Men with anal cancer have 2.4 times greater odds of giving a “Prefer not to say” response to the survey question about sexual identity than men with prostate cancer, and men with Kaposi’s sarcoma have 11.0 times greater odds of giving a “Prefer not to say” response. Clearly, there is some increased use of the “Prefer not to say” option among gay or bisexual men with these diagnoses, but it is not possible to determine the extent to which this option is used more frequently by all gay or bisexual men, or whether this is particularly related to these diagnoses that perhaps have greater stigma attached. Appendix Tables A7 and A8 These tables present details of the sample derivation and clinical coding of cancer sites used in the main analysis. Notes to Support the Comparison of Results From Tables 1 to 3 In Table 2, this analysis uses data from the GPPS to answer the research question, “Do women and men from sexual minorities report a cancer diagnosis in the previous 5 years more or less frequently than heterosexual women and men?”

Appendix Tables A7 and A8 These tables present details of the sample derivation and clinical coding of cancer sites used in the main analysis. Notes to Support the Comparison of Results From Tables 1 to 3 In Table 2, this analysis uses data from the GPPS to answer the research question, “Do women and men from sexual minorities report a cancer diagnosis in the previous 5 years more or less frequently than heterosexual women and men?” In Tables 1 and 3, the CPES is used to explore the question, “Among recently treated survivors of cancer, is there variation between cancer sites in the proportion of men and women who report gay, lesbian, or bisexual sexual orientation?” GPPS data are a population-based sample of the population of England, and we can therefore estimate cancer prevalence among men and women from sexual minorities—with and without cancer in the past 5 years—and compare this with people who report a heterosexual sexual orientation. This is a unique data source, as the large sample size—nearly 1 million people—and the availability of sampling and design weights allow us to estimate cancer prevalence in the last 5 years in England, stratified by sexual orientation, then adjusted for age. These results are presented in Table 2.

l sexual orientation. This is a unique data source, as the large sample size—nearly 1 million people—and the availability of sampling and design weights allow us to estimate cancer prevalence in the last 5 years in England, stratified by sexual orientation, then adjusted for age. These results are presented in Table 2. CPES data are also a unique survey resource of recently treated survivors of cancer, sent annually to all patients age ≥ 16 years. The strengths of this data resource are the large number of people with cancer—249,010 patients, including 920 lesbian or bisexual women and 1,279 gay or bisexual men (these numbers of people from sexual minorities with cancer are much higher compared with the population-based GPPS sample, which includes both people with and without cancer—and the availability of specific International Classification of Diseases, Tenth Revision, diagnosis codes, which are not available for the GPPS data; however, the two limitations in the CPES data are, first, the sample is hospital based, rather than population based (recently treated survivors of cancer), and, second, that we do not have a comparison population of people without cancer. This means that we cannot estimate cancer prevalence by using the CPES data. Instead, it is possible to estimate whether there is variation between cancer sites in the proportion of men and women who report gay, lesbian, or bisexual sexual orientation by using a case-only analysis.

population of people without cancer. This means that we cannot estimate cancer prevalence by using the CPES data. Instead, it is possible to estimate whether there is variation between cancer sites in the proportion of men and women who report gay, lesbian, or bisexual sexual orientation by using a case-only analysis. These proportions allow us to explore whether people from sexual minorities are relatively over- or under-represented among people with a particular cancer diagnosis compared with baseline diagnosis—that is, are there more people from sexual minorities with a particular cancer site diagnosis than would be expected were there no variation between sites? This over- or under-representation will reflect higher or lower incidence, or higher or lower survival, among people from sexual minorities, but incidence and prevalence cannot be measured directly, and the comparison is therefore between cancer sites, rather than with people without cancer. In practical terms, this means for the analysis that results in Tables 1 to 3 are complementary but not directly comparable, as the populations and the analyses are different. In Table 2, results tell us about cancer prevalence in the whole population of England (age ≥ 18 years)—that prevalence is higher among gay or bisexual men, but not among lesbian or bisexual women. In Tables 1 and 3, results tell us about variation between cancer sites, but not about comparisons with the whole population.

Table 2, results tell us about cancer prevalence in the whole population of England (age ≥ 18 years)—that prevalence is higher among gay or bisexual men, but not among lesbian or bisexual women. In Tables 1 and 3, results tell us about variation between cancer sites, but not about comparisons with the whole population. Table A1. Exploration of Potential Confounders in the Relationship Between Cancer Prevalence and Sexual Orientation (General Practice Patient Survey) Table A2. Exploration of Potential Confounders on the Odds Ratios of Specific Cancer Site Diagnosis by Lesbian/Bisexual Orientation Among Women With Cancer (Cancer Patient Experience Survey) Table A3. Exploration of Potential Confounders on the Odds Ratios of Specific Cancer Site Diagnosis by Gay/Bisexual Orientation Among Men With Cancer (Cancer Patient Experience Survey) Table A4. Cancer Prevalence and Alternative Response Options to the Sexual Orientation Question in Women and Men (General Practice Patient Survey) Table A5. Odds Ratios of Specific Cancer Site Diagnosis by Lesbian/Bisexual Orientation, and Alternative Sexual Orientation Question Responses in Women (Cancer Patient Experience Survey) Table A6. Odds Ratios of Specific Cancer Site Diagnosis by Gay/Bisexual Orientation, and Alternative Sexual Orientation Question Responses in Men (Cancer Patient Experience Survey) Table A7. GPPS and CPES Survey Responses Flowchart Table A8. Cancer Diagnosis by ICD-10 Code

INTRODUCTION Half of patients with cancer suffer from pain.1-3 Despite the availability of pain management guidelines, poor outcomes are common.4-10 Common shortcomings in pain management include unstructured assessment, use of treatment guidelines that lack explicit algorithms and do not address clinicians’ concerns about prescribing opioids, and lack of systematic monitoring of outcomes, including adverse effects.11,12 Admission to a cancer center provides an important opportunity to improve pain outcomes. We developed a simple clinician-administered bedside tool—the Edinburgh Pain Assessment and management Tool (EPAT)—that builds on the concept of pain as the fifth vital sign.13-16 EPAT aims to change routine practice by directing a systematic assessment of cancer-related pain, guiding treatment using linked algorithms, and prompting the regular reassessment of pain to determine both efficacy and adverse effects of treatment. We tested EPAT in a single cancer center and found preliminary evidence of its feasibility and efficacy.16 The current study aimed to determine the effectiveness of implementing EPAT in multiple cancer center inpatient units. Specifically, we sought to determine whether the percentage of inpatients with improved pain increased in cancer centers that implemented EPAT compared with cancer centers that continued to deliver usual care (UC). We also aimed to find out whether implementing EPAT improved prescribing practice and whether it increased opioid-related adverse effects.

ine whether the percentage of inpatients with improved pain increased in cancer centers that implemented EPAT compared with cancer centers that continued to deliver usual care (UC). We also aimed to find out whether implementing EPAT improved prescribing practice and whether it increased opioid-related adverse effects. PATIENTS AND METHODS Study Design and Patients We used a two-arm, parallel group, cluster randomized controlled trial to evaluate the effect on pain outcomes of introducing EPAT into cancer centers (Appendix, online only, list of study group members). The trial clusters were the inpatient units of regional cancer centers in the United Kingdom. In each center, the trial was conducted in two phases. In the first, pre–random assignment phase, 50 patients with cancer-related pain were enrolled, and their pain outcomes were measured after management in accordance with UC. Centers were then randomly assigned to either implement EPAT (EPAT centers) or to continue providing UC (UC centers). In the second, post–random assignment phase, an additional 50 patients were enrolled and their pain outcomes were measured.

and their pain outcomes were measured after management in accordance with UC. Centers were then randomly assigned to either implement EPAT (EPAT centers) or to continue providing UC (UC centers). In the second, post–random assignment phase, an additional 50 patients were enrolled and their pain outcomes were measured. Cancer centers were eligible to participate if they did not have an existing bedside pain management system, could recruit 100 patients in the required time frame, and did not anticipate organizational changes that might affect pain management policies. Within each center, patients were eligible to participate if they were adults ( ≥ 18 years of age) with active cancer and cancer-related pain, with a worst pain in the past 24 hours score (assessed within 24 hours of admission) of ≥ 4 on a scale of 0 to 10. The protocol (Data Supplement) lists complete inclusion and exclusion criteria. The trial was approved by the Scotland A Research Ethics Committee and was overseen by a trial steering committee. Conduct of the trial in each center followed study guidance procedures (Data Supplement). All centers and participants provided written informed consent.

Cancer centers were eligible to participate if they did not have an existing bedside pain management system, could recruit 100 patients in the required time frame, and did not anticipate organizational changes that might affect pain management policies. Within each center, patients were eligible to participate if they were adults ( ≥ 18 years of age) with active cancer and cancer-related pain, with a worst pain in the past 24 hours score (assessed within 24 hours of admission) of ≥ 4 on a scale of 0 to 10. The protocol (Data Supplement) lists complete inclusion and exclusion criteria. The trial was approved by the Scotland A Research Ethics Committee and was overseen by a trial steering committee. Conduct of the trial in each center followed study guidance procedures (Data Supplement). All centers and participants provided written informed consent. Random Assignment and Masking A database software algorithm, implemented by the Edinburgh Clinical Trials Unit, randomly allocated cancer centers to implement EPAT or continue providing UC in a 1:1 ratio using variable-size permuted blocks. Because of the cluster design, we were unable to mask cancer center clinicians, patients, or data collectors to intervention. However, the clinicians in the UC centers did not know the content of EPAT, and patients self-rated their pain and knew only that their cancer center was taking part in a pain study.

ted blocks. Because of the cluster design, we were unable to mask cancer center clinicians, patients, or data collectors to intervention. However, the clinicians in the UC centers did not know the content of EPAT, and patients self-rated their pain and knew only that their cancer center was taking part in a pain study. Procedures Patient enrollment. All patients admitted to the participating cancer centers were given information about the study. A research nurse obtained patients’ verbal consent to eligibility screening. The nurse provided eligible patients with a further explanation of the study and obtained their written consent for participation. Enrollment was done independently of the clinical team. Delivery of pain management. The clinicians who delivered pain management (EPAT or UC) were oncology nurses, nursing care assistants, oncology trainee doctors, and senior oncologists. In UC centers, the clinical team managed patients’ pain according to their clinical judgment and existing local guidelines. In EPAT centers, the clinical team was provided with the EPAT (Data Supplement) and given brief (maximum, 1 hour) training in its use. EPAT was designed to address the aforementioned key barriers to effective pain management by prompting clinicians to systematically assess pain using simple questions that should not be shortened or paraphrased; to follow linked treatment algorithms, rather than broad guidelines, including instructions on opioid prescribing; and to regularly reassess pain and opioid-related adverse effects.

pain management by prompting clinicians to systematically assess pain using simple questions that should not be shortened or paraphrased; to follow linked treatment algorithms, rather than broad guidelines, including instructions on opioid prescribing; and to regularly reassess pain and opioid-related adverse effects. EPAT integrates pain assessment into routine care by its inclusion in the patient’s bedside chart. Clinicians are prompted to assess pain using a two-step procedure every time the patient’s vital signs are recorded. In step 1 (Fig 1), the patient is asked to rate his or her worst pain (since last being assessed) on a scale from 0 (no pain) to 10 (worst pain imaginable), and this score is recorded. The chart categorizes pain scores using the following color system: 0 to 2 is classified as gray, 3 to 4 as yellow, and 5 to 10 as blue. For patients with a yellow or blue score, the clinician is prompted to proceed to step 2 (Data Supplement), which explores the location and nature of the pain, exacerbating and relieving factors, and symptoms that may be caused by opioids. Flags linked to recorded responses then prompt the clinician to use the appropriate algorithm to guide prescribing. The chart also prompts reassessment of pain and opioid adverse effects 1 hour after administration of opioid medication. Fig 1. Edinburgh Pain Assessment and management Tool (EPAT) step 1 as it appears on the vital signs chart.

EPAT integrates pain assessment into routine care by its inclusion in the patient’s bedside chart. Clinicians are prompted to assess pain using a two-step procedure every time the patient’s vital signs are recorded. In step 1 (Fig 1), the patient is asked to rate his or her worst pain (since last being assessed) on a scale from 0 (no pain) to 10 (worst pain imaginable), and this score is recorded. The chart categorizes pain scores using the following color system: 0 to 2 is classified as gray, 3 to 4 as yellow, and 5 to 10 as blue. For patients with a yellow or blue score, the clinician is prompted to proceed to step 2 (Data Supplement), which explores the location and nature of the pain, exacerbating and relieving factors, and symptoms that may be caused by opioids. Flags linked to recorded responses then prompt the clinician to use the appropriate algorithm to guide prescribing. The chart also prompts reassessment of pain and opioid adverse effects 1 hour after administration of opioid medication. Fig 1. Edinburgh Pain Assessment and management Tool (EPAT) step 1 as it appears on the vital signs chart. Outcomes The primary trial outcome (measured at the cancer center level) was the change in the percentage of participants with a clinically significant improvement in pain. We defined a clinically significant improvement in pain as a reduction of ≥ 2 points in the severity of worst pain reported over the previous 24 hours measured between admission and reassessment (3 to 5 days after admission). Worst pain is an item on the Brief Pain Inventory Short Form (BPI-SF), which has 11 questions, each rated on a scale of 0 to 10, and is validated for use in the evaluation of cancer pain.17 Worst pain is the recommended outcome measure for trials, and a 2-point change on this item has been found to be meaningful to patients.18-20

is an item on the Brief Pain Inventory Short Form (BPI-SF), which has 11 questions, each rated on a scale of 0 to 10, and is validated for use in the evaluation of cancer pain.17 Worst pain is the recommended outcome measure for trials, and a 2-point change on this item has been found to be meaningful to patients.18-20 The secondary trial outcomes (also measured as cancer center–level changes) were the percentage of participants with controlled pain (worst pain BPI-SF score < 4); the mean changes in BPI-SF scores (worst pain, pain subscale, pain interference subscale, and total score); the mean change in global distress in the past 24 hours using the National Comprehensive Cancer Network thermometer (0 to 10 thermometer, where 0 is no distress and 10 is extreme distress)21; the mean change in opioid adverse effects (using the mean of 0 to 10 scales for drowsiness, confusion, disorientation, shadows, vivid dreams, hallucinations, and muscle twitching; a modification of a previously used scale)22; the percentage of participants receiving good practice prescribing (rated based on the appropriateness of the medication[s] prescribed, along with how they were prescribed, compared with guidelines; Data Supplement); the percentage of participants readmitted to the cancer center with uncontrolled pain within 14 days of discharge; and in EPAT centers, satisfaction with the attention given to pain and the ease of use of EPAT rated on a 0 to 10 scale by patients and nurses, respectively.

ed, compared with guidelines; Data Supplement); the percentage of participants readmitted to the cancer center with uncontrolled pain within 14 days of discharge; and in EPAT centers, satisfaction with the attention given to pain and the ease of use of EPAT rated on a 0 to 10 scale by patients and nurses, respectively. Outcome data were collected from participants, medical records, and relevant staff, and as appropriate, by the research nurses. If participants had been discharged from the cancer center before the time of the follow-up assessment, the research nurse collected the outcome data by telephone. Statistical Analysis We estimated that two treatment groups of nine cancer centers each, with 100 participants per cancer center (50 recruited in the first phase before random assignment of the cancer center, and 50 recruited in the second phase after random assignment) would give 80% power at the 5% significance level to detect a difference of at least 15% between the trial arms in the primary outcome. The main analysis was conducted (using SAS version 9.2 software; SAS Institute, Cary, NC) at the end of the trial, using an intent-to-treat principle that included all participants who provided usable outcome data.

significance level to detect a difference of at least 15% between the trial arms in the primary outcome. The main analysis was conducted (using SAS version 9.2 software; SAS Institute, Cary, NC) at the end of the trial, using an intent-to-treat principle that included all participants who provided usable outcome data. Because there were, as planned, relatively few clusters (cancer centers) and little variability in the cluster size, we performed the analyses using summary measures.23 For each analysis, we calculated a single summary measure for each cancer center and compared the UC centers with the EPAT centers using a two-sample t test. Hence, the summary statistic for the primary outcome was the difference between the percentage of participants who achieved a clinically significant reduction in pain before random assignment (phase 1) and the percentage who achieved a clinically significant reduction in pain after random assignment (phase 2). For continuous outcome measures, the summary statistic was the difference in the mean score before and after random assignment.

eved a clinically significant reduction in pain before random assignment (phase 1) and the percentage who achieved a clinically significant reduction in pain after random assignment (phase 2). For continuous outcome measures, the summary statistic was the difference in the mean score before and after random assignment. In a prespecified sensitivity analysis, hierarchical patient-level analyses were performed using random-effects models. The results for the primary end point (achieving a reduction of ≥ 2 points in the severity of worst pain reported over the previous 24 hours) and a secondary end point (the magnitude of the reduction in the severity of worst pain reported over the previous 24 hours) are reported. In addition to the main analysis, we performed a post hoc sensitivity analysis of the primary outcome and of good practice prescribing excluding cancer centers that had been randomly assigned to EPAT but were unable to implement it. RESULTS The United Kingdom has 40 cancer centers. We selected 20 centers from those eligible to participate, aiming for geographical spread and a range of center sizes. Each center took part in the trial for approximately 1 year (including both pre– and post–random assignment phases). Center participation was staggered and occurred between December 2007 and January 2013. One center dropped out before random assignment. Ten cancer centers were randomly assigned to implement EPAT, and nine were assigned to continue providing UC (Fig 2). Fig 2. CONSORT diagram. EPAT, Edinburgh Pain Assessment and management Tool.

RESULTS The United Kingdom has 40 cancer centers. We selected 20 centers from those eligible to participate, aiming for geographical spread and a range of center sizes. Each center took part in the trial for approximately 1 year (including both pre– and post–random assignment phases). Center participation was staggered and occurred between December 2007 and January 2013. One center dropped out before random assignment. Ten cancer centers were randomly assigned to implement EPAT, and nine were assigned to continue providing UC (Fig 2). Fig 2. CONSORT diagram. EPAT, Edinburgh Pain Assessment and management Tool. Of the 42,000 patients admitted to the cancer centers during the trial, 8,400 were assessed as having moderate or severe pain (worst pain in the past 24 hours of ≥ 4 on a 0 to 10 scale), and 1,921 patients were enrolled (985 before center random assignment and 936 after assignment; Data Supplement). The most common reason for excluding patients was that their admission was for planned chemotherapy and too brief to allow completion of the protocol. Only 2.7% of eligible patients (53 of 1,974 patients) declined to participate. Participants had a mean age of 60 years (range, 20 to 90 years), and 49% were female (Table 1). Patients had a variety of cancers, the most common of which were genitourinary (272 of 1,921 patients; 14.2%), GI (260 of 1,921 patients; 13.5%), breast (234 of 1,921 patients; 12%), and lung (220 of 1,921 patients; 11.5%). These characteristics were well balanced between EPAT and UC centers. We obtained outcome data from 93% of participants (1,795 of 1,921 patients). These data included 150 telephone follow-up assessments for patients who had been discharged from hospital (the number of telephone assessments was similar in both trial arms). The mean number of days from admission to the primary outcome assessment was 4 and was similar between trial arms.

s (1,795 of 1,921 patients). These data included 150 telephone follow-up assessments for patients who had been discharged from hospital (the number of telephone assessments was similar in both trial arms). The mean number of days from admission to the primary outcome assessment was 4 and was similar between trial arms. Table 1. Participant Baseline Characteristics Introducing EPAT into cancer centers had an effect on pain outcomes. The percentage of participants with a clinically significant improvement in pain increased after random assignment in eight of the 10 EPAT centers and in three of the nine UC centers (Fig 3). In EPAT centers, the mean percentage of participants with a clinically significant improvement in pain increased from 47.7% (before random assignment) to 54.1% (after random assignment), an absolute increase of 6.4%. In UC centers, the mean percentage of participants with a clinically significant improvement in pain decreased from 50.6% (before random assignment) to 46.4% (after random assignment), an absolute decrease of 4.2%. Thus, the absolute difference between trial arms was 10.7% (95% CI, 0.2% to 21.1%; P = .046). There was no difference in outcomes between participants assessed by telephone after discharge and those who were assessed as inpatients. Fig 3. Plot of primary outcome by center. EPAT, Edinburgh Pain Assessment and management Tool.

Introducing EPAT into cancer centers had an effect on pain outcomes. The percentage of participants with a clinically significant improvement in pain increased after random assignment in eight of the 10 EPAT centers and in three of the nine UC centers (Fig 3). In EPAT centers, the mean percentage of participants with a clinically significant improvement in pain increased from 47.7% (before random assignment) to 54.1% (after random assignment), an absolute increase of 6.4%. In UC centers, the mean percentage of participants with a clinically significant improvement in pain decreased from 50.6% (before random assignment) to 46.4% (after random assignment), an absolute decrease of 4.2%. Thus, the absolute difference between trial arms was 10.7% (95% CI, 0.2% to 21.1%; P = .046). There was no difference in outcomes between participants assessed by telephone after discharge and those who were assessed as inpatients. Fig 3. Plot of primary outcome by center. EPAT, Edinburgh Pain Assessment and management Tool. Two of the centers randomly assigned to implement EPAT were unable to do so because of unanticipated organizational changes. In these centers, the percentage of participants with a clinically significant improvement in pain actually decreased after random assignment (Fig 3). A post hoc sensitivity analysis that excluded these two centers found a larger difference between EPAT and UC (absolute difference, 15.4%; 95% CI, 5.8% to 25.0%; P = .004). The remaining eight centers that implemented EPAT used it in at least 90% of patient assessments as indicated by entries on the participants’ charts.

post hoc sensitivity analysis that excluded these two centers found a larger difference between EPAT and UC (absolute difference, 15.4%; 95% CI, 5.8% to 25.0%; P = .004). The remaining eight centers that implemented EPAT used it in at least 90% of patient assessments as indicated by entries on the participants’ charts. Regarding the trial’s secondary outcomes, EPAT centers had greater improvements in good practice prescribing (a difference that was larger when the two centers unable to implement EPAT were excluded) and greater changes in the mean worst pain item and in mean pain subscale scores. However, there were no statistically significant differences between EPAT and UC centers in the percentage of participants with controlled pain, the mean pain interference score, the mean total pain score, or the mean severity of global distress. There was also no difference between EPAT and UC centers in the percentage of patients who had received strong opioids (80%), in the mean total 24-hour oral morphine equivalent (10 mg), or importantly, in opioid-related adverse effects (Table 2). We were unable to analyze readmissions to hospital because the available data were inadequate. Participants who received EPAT reported (on a 0 to 10 scale) high satisfaction with the attention given to their pain (mean score, 8.6; standard deviation, 1.8), and nurses reported (on a 0 to 10 scale) moderate satisfaction with the ease of using EPAT (mean score, 6.4; standard deviation, 2.5). Table 2. Trial Outcomes

Regarding the trial’s secondary outcomes, EPAT centers had greater improvements in good practice prescribing (a difference that was larger when the two centers unable to implement EPAT were excluded) and greater changes in the mean worst pain item and in mean pain subscale scores. However, there were no statistically significant differences between EPAT and UC centers in the percentage of participants with controlled pain, the mean pain interference score, the mean total pain score, or the mean severity of global distress. There was also no difference between EPAT and UC centers in the percentage of patients who had received strong opioids (80%), in the mean total 24-hour oral morphine equivalent (10 mg), or importantly, in opioid-related adverse effects (Table 2). We were unable to analyze readmissions to hospital because the available data were inadequate. Participants who received EPAT reported (on a 0 to 10 scale) high satisfaction with the attention given to their pain (mean score, 8.6; standard deviation, 1.8), and nurses reported (on a 0 to 10 scale) moderate satisfaction with the ease of using EPAT (mean score, 6.4; standard deviation, 2.5). Table 2. Trial Outcomes From the prespecified patient-level sensitivity analysis of the primary end point, the estimated intracluster correlation coefficient was 0.004, and the estimated treatment effect was 11.6% (95% CI, 2.4% to 20.9%; P = .014). For the corresponding analysis of the magnitude of the reduction in the patients’ worst pain scores, the estimated intracluster correlation coefficient was 0. The estimated treatment effect was 0.82 (95% CI, 0.31 to 1.32; P = .002).

0.004, and the estimated treatment effect was 11.6% (95% CI, 2.4% to 20.9%; P = .014). For the corresponding analysis of the magnitude of the reduction in the patients’ worst pain scores, the estimated intracluster correlation coefficient was 0. The estimated treatment effect was 0.82 (95% CI, 0.31 to 1.32; P = .002). DISCUSSION The findings of this multicenter, cluster randomized trial indicate that a policy of integrating systematic pain assessment and management into routine cancer center care using a simple tool (EPAT) improves pain outcomes for patients with moderate or severe cancer-related pain. The difference between EPAT and UC centers in the percentage of patients with a clinically significant reduction in their worst pain was 10.7%. This difference increased to 15.4% when the two centers that had been unable to implement EPAT were excluded. It should be noted, however, that the 95% CI for the 10.7% difference is wide (0.2% to 21.1%).

and UC centers in the percentage of patients with a clinically significant reduction in their worst pain was 10.7%. This difference increased to 15.4% when the two centers that had been unable to implement EPAT were excluded. It should be noted, however, that the 95% CI for the 10.7% difference is wide (0.2% to 21.1%). Inspection of the results by cancer center revealed that the difference in the primary outcome between centers delivering EPAT and centers continuing UC reflected not only an improvement in pain management with EPAT (except in the two centers that failed to implement EPAT), but also a deterioration in pain management in most of the centers that continued UC. We did not find any differences in patient or cancer characteristics in the samples studied in each phase that could account for this. Therefore, we suggest that a likely explanation for the deterioration in outcome in some, but not all, of the UC centers in the post–random assignment period reflects a return to pretrial standards of pain management after improvement during the pre–random assignment phase of the trial. This short-term effect on clinicians’ behavior might be expected as a result of awareness that their center was participating in a study that monitored their patients’ pain scores.24 This effect would also be expected to decline over the many months of study in centers continuing with UC. The effect was not seen in centers implementing EPAT, which overall had better outcomes.

pected as a result of awareness that their center was participating in a study that monitored their patients’ pain scores.24 This effect would also be expected to decline over the many months of study in centers continuing with UC. The effect was not seen in centers implementing EPAT, which overall had better outcomes. EPAT centers prescribed analgesics more appropriately (as defined in the Data Supplement) and not in higher doses. Although the concept of pain as the fifth vital sign has been criticized as a potential cause of increased opioid-related adverse effects, the use of EPAT did not increase opioid adverse effects.25,26 This absence of increased adverse effects despite better pain management may be because EPAT alerts clinicians to monitor adverse effects of pain treatment, as well as efficacy. EPAT did not improve all of the secondary trial outcomes. The percentage of participants with controlled pain, the severity of general distress, and the degree to which pain interfered with activities did not differ between trial arms. The short duration of this study and inpatient setting arguably made it difficult to change these outcomes.

l of the secondary trial outcomes. The percentage of participants with controlled pain, the severity of general distress, and the degree to which pain interfered with activities did not differ between trial arms. The short duration of this study and inpatient setting arguably made it difficult to change these outcomes. An additional and notable finding is that implementation of EPAT was hampered by organizational and leadership changes in two trial centers, which also subsequently had worse patient outcomes. This observation highlights the importance of organizational factors and leadership, as well as clinician education, in achieving positive changes in patient care.27 As expected, the patient-level sensitivity analyses yielded parameter estimates and CIs that were broadly similar to those obtained in the primary cluster-level analyses and with levels of statistical significance that were more extreme. Previously published smaller studies have evaluated the effects of systematic management for cancer pain. Two randomized trials compared algorithm-based pain management by pain specialists, rather than oncology teams, with variable results.28,29 Two preliminary studies evaluated the integration of systematic pain management into the usual clinical care delivered by oncology teams with more promising findings.30,31 Finally, it is notable that published studies of integrated and systematic approaches to the management of other symptoms in patients with cancer, in particular depression, have also found these to be effective.32-34

management into the usual clinical care delivered by oncology teams with more promising findings.30,31 Finally, it is notable that published studies of integrated and systematic approaches to the management of other symptoms in patients with cancer, in particular depression, have also found these to be effective.32-34 The strengths of this trial include the participation of many cancer centers in the United Kingdom and negligible missing outcome data. However, the trial has limitations. First, the trial was carried out within one particular health care system (the United Kingdom National Health Service), although the issues associated with cancer pain assessment and management are similar in most developed countries. Second, we do not have information on participants’ longer term outcomes. Third, oncology teams and those collecting the outcome data from patients could not be masked to the treatment allocation. In conclusion, this study is, to our knowledge, the first large randomized evaluation of the integration of systematic pain assessment and management into routine care of patients with cancer. The implementation of EPAT improved both prescribing practice and pain outcomes. Furthermore, it did not increase opioid-related adverse effects. This latter finding is important given concerns that the measurement of pain as a vital sign and linked opioid prescribing can be harmful.14,15 The findings of this trial add to the accumulating evidence for the efficacy of more integrated and systematic approaches to symptom management in patients with cancer.

ffects. This latter finding is important given concerns that the measurement of pain as a vital sign and linked opioid prescribing can be harmful.14,15 The findings of this trial add to the accumulating evidence for the efficacy of more integrated and systematic approaches to symptom management in patients with cancer. Supported by Cancer Research UK Grant No. C17958/A6823. J.W. is supported by Sir Michael Sobell House Hospice, Oxford, and the National Institute for Health Research (NIHR) Collaboration for Leadership in Applied Health Research and Care Oxford at Oxford Health NHS Foundation Trust. M.S. is supported by the NIHR Senior Investigator program. The views expressed are those of the author(s) and not necessarily those of the NHS, the NIHR or the Department of Health. Clinical trial information: NCT00595777. See accompanying Editorial on page 1272 ACKNOWLEDGMENT We thank the trial participants, their families and caregivers, and the staff of the participating cancer centers. AUTHOR CONTRIBUTIONS Conception and design: Marie Fallon, Lesley Colvin, Gordon Murray, Michael Sharpe Collection and assembly of data: Marie Fallon, Gordon Murray Data analysis and interpretation: All authors Manuscript writing: All authors Final approval of manuscript: All authors Accountable for all aspects of the work: All authors

AUTHOR CONTRIBUTIONS Conception and design: Marie Fallon, Lesley Colvin, Gordon Murray, Michael Sharpe Collection and assembly of data: Marie Fallon, Gordon Murray Data analysis and interpretation: All authors Manuscript writing: All authors Final approval of manuscript: All authors Accountable for all aspects of the work: All authors AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST Pain Management in Cancer Center Inpatients: A Cluster Randomized Trial to Evaluate a Systematic Integrated Approach—The Edinburgh Pain Assessment and Management Tool The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/jco/site/ifc. Marie Fallon Research Funding: Pfizer (Inst) Jane Walker No relationship to disclose Lesley Colvin No relationship to disclose Aryelly Rodriguez No relationship to disclose Gordon Murray No relationship to disclose Michael Sharpe No relationship to disclose

AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST Pain Management in Cancer Center Inpatients: A Cluster Randomized Trial to Evaluate a Systematic Integrated Approach—The Edinburgh Pain Assessment and Management Tool The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/jco/site/ifc. Marie Fallon Research Funding: Pfizer (Inst) Jane Walker No relationship to disclose Lesley Colvin No relationship to disclose Aryelly Rodriguez No relationship to disclose Gordon Murray No relationship to disclose Michael Sharpe No relationship to disclose Appendix The Edinburgh Pain Assessment and Management Tool Study Group Debra Gordon, Moira Ross, Lucy Norris, Eleanor Clausen, Elaine Boland, Sara Booth, Alison Coakley, Mary Comiskey, Andrew Davies, Carol Davis, Jane Edgecombe, Karen Forbes, Fiona Hicks,† Jamal Humaira, Jane Maher, Wendy Makin, Mary Nugent, Nikki Pease, Allan Price, Julia Riley, Joy Ross, Kirsten Saharia, John Speakman, John Walley, Bee Wee, Andrew Wilcock, Pauline Wilkinson, Teresa Young, and Andrew Walker. †Deceased September 2015.

INTRODUCTION Women with estrogen receptor (ER) –positive primary breast cancer are generally offered adjuvant endocrine therapy for 5 years. More than 50% of recurrences occur after that time, and several studies have indicated that extending treatment beyond 5 years can improve disease outcome.1-5 However, this improvement is relatively modest, and extended therapy carries a risk of adverse effects. Few tools have been developed for selecting patients as candidates for extended endocrine therapy or alternatively identifying those who might be spared such therapy. One approach is to identify patients whose risk after 5 years is so low that any benefit would be outweighed by potential adverse effects.

verse effects. Few tools have been developed for selecting patients as candidates for extended endocrine therapy or alternatively identifying those who might be spared such therapy. One approach is to identify patients whose risk after 5 years is so low that any benefit would be outweighed by potential adverse effects. Clinicopathologic parameters such as tumor size, nodal status, and histopathologic grade are routinely used to estimate risk of breast cancer recurrence at diagnosis; we previously reported a clinical treatment score that integrates these factors to estimate prognosis.6 Some of these factors have been reported to be associated with risk after 5 years; for example, we found nodal status was a powerful prognostic marker for late recurrence,7,8 whereas tumor size and particularly grade were less prognostic after 5 years. Recently, an overview analysis of > 60,000 women with ER-positive disease, who were scheduled to receive 5 years of endocrine therapy and remained disease free at 5 years, reported the subsequent risk of distant recurrence.9 Even in patients with T1N0 disease, the estimated risk of distant recurrence between years 5 and 20 was 10% for those with low, 13% for those with intermediate, and 17% for those with high histologic grades, respectively. Although these data unequivocally demonstrate the importance of these clinicopathologic factors, they include studies from 40 years ago, possibly limiting their relevance for contemporary patients with breast cancer. The data were presented largely as categories (eg, T1, T2), limiting precise estimates of risk for individual patients. Lastly, the largely tamoxifen-treated population did not allow assessment of possible differences between tamoxifen and aromatase inhibitors (AIs) with regard to long-term risk.

ts with breast cancer. The data were presented largely as categories (eg, T1, T2), limiting precise estimates of risk for individual patients. Lastly, the largely tamoxifen-treated population did not allow assessment of possible differences between tamoxifen and aromatase inhibitors (AIs) with regard to long-term risk. We aimed to develop and test the validity of a simple prognostic tool to estimate risk of late distant recurrence (Clinical Treatment Score post–5 years [CTS5]) on the basis of clinicopathologic parameters measured in virtually all patients with breast cancer at diagnosis. We used data from the ATAC (Arimidex, Tamoxifen, Alone or in Combination) trial10 as the training set and from the BIG (Breast International Group) 1-98 trial as the testing set.11

Score post–5 years [CTS5]) on the basis of clinicopathologic parameters measured in virtually all patients with breast cancer at diagnosis. We used data from the ATAC (Arimidex, Tamoxifen, Alone or in Combination) trial10 as the training set and from the BIG (Breast International Group) 1-98 trial as the testing set.11 PATIENTS AND METHODS Study Populations CTS5 (ATAC) was trained using data from the ATAC trial (International Standard Randomized Controlled Trial identifier ISRCTN18233230), in which postmenopausal women with ER-positive or ER-unknown early breast cancer were randomly assigned to receive anastrozole 1 mg per day, tamoxifen 20 mg per day, or a combination for 5 years.10 The combination arm was discontinued after the first report of trial results.12 We included data from women with ER-positive breast cancer randomly assigned to receive anastrozole alone or tamoxifen alone, who were distant recurrence free after 5 years of follow-up and for whom all clinicopathologic data were available (N = 4,735; Appendix Fig A1, online only). Median follow-up was 9.8 years. Data from BIG 1-98 (ClinicalTrials.gov identifier NCT00004205) was used to validate CTS5 (ATAC). BIG 1-98 initially (1998 to 2000) randomly assigned postmenopausal women with hormone receptor–positive early-stage breast cancer to receive 5 years of letrozole 2.5 mg per day or tamoxifen 20 mg per day. Later (1999 to 2003), sequential therapy was also randomly assigned (2 years of letrozole followed by 3 years of tamoxifen or opposite sequence).11,13 Median follow-up was 8.1 years. For this analysis, all women were included who were distant recurrence free at 5 years and for whom all clinicopathologic data were available (N = 6,711; Appendix Fig A1). For both trials, women were included in the analysis regardless of whether they received chemotherapy.

quence).11,13 Median follow-up was 8.1 years. For this analysis, all women were included who were distant recurrence free at 5 years and for whom all clinicopathologic data were available (N = 6,711; Appendix Fig A1). For both trials, women were included in the analysis regardless of whether they received chemotherapy. Prognostic value of the following variables for post–5-year (late) distant recurrence was determined by univariable Cox regression analyses: nodes, tumor size (in millimeters), grade (1, 2, or 3), age at start of endocrine therapy (years), and type of assigned endocrine treatment. Type of endocrine treatment was not significant for late distant recurrence in univariable analyses and not included in the final model. The log hazard was almost linear for five nodal status groups (negative, one positive, two to three positive, four to nine positive, and > nine positive) but not for continuous tumor size alone. Therefore, a negative quadratic term was introduced, and tumor size was capped at 30 mm, where risk plateaued. The final CTS5 (ATAC) model included age (continuous), tumor size (continuous), quadratic tumor size, nodal status (five groups: 0, negative; 1, one positive; 2, two to three positive; 3, four to nine positive; and 4, > nine positive), and grade (three groups: 1, low; 2, intermediate; and 3, high) and is given by: CTS5 (ATAC) = 0.471 × nodes + 0.980 × (0.164 × size − 0.003 × size2 + 0.312 × grade + 0.03 × age)

Prognostic value of the following variables for post–5-year (late) distant recurrence was determined by univariable Cox regression analyses: nodes, tumor size (in millimeters), grade (1, 2, or 3), age at start of endocrine therapy (years), and type of assigned endocrine treatment. Type of endocrine treatment was not significant for late distant recurrence in univariable analyses and not included in the final model. The log hazard was almost linear for five nodal status groups (negative, one positive, two to three positive, four to nine positive, and > nine positive) but not for continuous tumor size alone. Therefore, a negative quadratic term was introduced, and tumor size was capped at 30 mm, where risk plateaued. The final CTS5 (ATAC) model included age (continuous), tumor size (continuous), quadratic tumor size, nodal status (five groups: 0, negative; 1, one positive; 2, two to three positive; 3, four to nine positive; and 4, > nine positive), and grade (three groups: 1, low; 2, intermediate; and 3, high) and is given by: CTS5 (ATAC) = 0.471 × nodes + 0.980 × (0.164 × size − 0.003 × size2 + 0.312 × grade + 0.03 × age) A shrinkage factor of 0.980 for the nonnodal part of the score was calculated using a nested Cox model14 and applied to allow for the small amount of overfitting. Separate models developed for patients receiving chemotherapy or not did not perform significantly better for either group than a single model including all patients (data not shown).

0 for the nonnodal part of the score was calculated using a nested Cox model14 and applied to allow for the small amount of overfitting. Separate models developed for patients receiving chemotherapy or not did not perform significantly better for either group than a single model including all patients (data not shown). Statistical Analyses Analyses were performed according to a prespecified analysis plan, approved by both trial groups, and are summarized here. Full details are provided in the Appendix (online only). The primary end point was time to distant recurrence, defined as metastatic disease, excluding contralateral disease, and locoregional and ipsilateral recurrences. The end point was censored at last follow-up visit or death before distant recurrence such that risk is a pure risk calculation ignoring deaths.

line only). The primary end point was time to distant recurrence, defined as metastatic disease, excluding contralateral disease, and locoregional and ipsilateral recurrences. The end point was censored at last follow-up visit or death before distant recurrence such that risk is a pure risk calculation ignoring deaths. Cox proportional hazards models were used to create the model in ATAC, and the CTS5 (ATAC) score was tested in BIG 1-98. Likelihood ratio statistics (LR-χ2) and Kaplan-Meier survival estimates with corresponding 95% CIs were used to determine the prognostic performance of CTS5 (ATAC) in BIG 1-98. The 5- to 10-year distant recurrence risk groups were determined in ATAC and defined as: low risk, < 5%; intermediate risk, 5% to 10%; and high risk, > 10%. To compare the prognostic performance of CTS5 (ATAC) between ATAC and BIG 1-98 trials, CTS5 (ATAC) was normalized to have unit variance, and hazard ratios (HRs) and associated 95% CIs were estimated from Cox models. All statistical analyses were two sided, and P < .05 was regarded as statistically significant. We compared the newly developed CTS5 (ATAC) with the published CTS (termed CTS0 here) developed for estimating prognosis from the time of disease presentation.6 All analyses were performed with STATA software (version 13.1; College Station, TX).

ses were two sided, and P < .05 was regarded as statistically significant. We compared the newly developed CTS5 (ATAC) with the published CTS (termed CTS0 here) developed for estimating prognosis from the time of disease presentation.6 All analyses were performed with STATA software (version 13.1; College Station, TX). RESULTS The ATAC training set and the BIG 1-98 test set consisted of 4,735 and 6,711 postmenopausal patients, respectively, assigned to receive 5 years of endocrine therapy (Table 1). Women in the ATAC cohort were significantly older by an average of approximately 3 years and had more node-negative disease (68% v 61%) and more grade 3 tumors (25% v 20%), and fewer women received adjuvant chemotherapy compared with women in the BIG 1-98 set (19.5% v 24.2%). Tumor size was similar between the two trials. In the training set, 330 (7.0%) late distant recurrences were recorded, with an annual hazard rate of 0.79% (95% CI, 0.71% to 0.88%). In BIG 1-98, a total of 370 (5.5%) late distant recurrences occurred, with an annual hazard rate of 0.66% (95% CI, 0.60% to 0.73%), which was significantly lower than in ATAC (P = .014; Table 1). Table 1. Demographic and Clinical Characteristics According to Trial of Patients Distant Recurrence Free at 5 Years After Random Assignment

RESULTS The ATAC training set and the BIG 1-98 test set consisted of 4,735 and 6,711 postmenopausal patients, respectively, assigned to receive 5 years of endocrine therapy (Table 1). Women in the ATAC cohort were significantly older by an average of approximately 3 years and had more node-negative disease (68% v 61%) and more grade 3 tumors (25% v 20%), and fewer women received adjuvant chemotherapy compared with women in the BIG 1-98 set (19.5% v 24.2%). Tumor size was similar between the two trials. In the training set, 330 (7.0%) late distant recurrences were recorded, with an annual hazard rate of 0.79% (95% CI, 0.71% to 0.88%). In BIG 1-98, a total of 370 (5.5%) late distant recurrences occurred, with an annual hazard rate of 0.66% (95% CI, 0.60% to 0.73%), which was significantly lower than in ATAC (P = .014; Table 1). Table 1. Demographic and Clinical Characteristics According to Trial of Patients Distant Recurrence Free at 5 Years After Random Assignment Training Set (ATAC) Appendix Table A1 (online only) shows the comparisons of the published CTS06 with CTS5 (ATAC) for prediction of late distant recurrence between years 5 and 10. CTS5 (ATAC) provided significantly more prognostic information compared with CTS0 (CTS5 [ATAC]: LR-χ2 = 308.6 [5 df]; CTS0: LR-χ2 = 285.0 [9 df]), and larger effect sizes were observed (HR, 2.47 v 2.04, respectively). CTS5 (ATAC) was slightly more prognostic in chemotherapy-free women compared with those who received chemotherapy (HR, 2.50; 95% CI, 2.22 to 2.81 v 2.39; 95% CI, 1.94 to 2.95), but the interaction with chemotherapy use was not significant (P = .76).

and larger effect sizes were observed (HR, 2.47 v 2.04, respectively). CTS5 (ATAC) was slightly more prognostic in chemotherapy-free women compared with those who received chemotherapy (HR, 2.50; 95% CI, 2.22 to 2.81 v 2.39; 95% CI, 1.94 to 2.95), but the interaction with chemotherapy use was not significant (P = .76). The prognostic value of CTS5 (ATAC) for risk of distant recurrence (± 95% CI) between years 5 and 10 is shown in Figure 1A for the whole population and in Figure 1B for node-positive and node-negative populations separately. Cutoffs in the ATAC population to separate low-, intermediate-, and high-risk populations were 4.35 and 5.02, respectively (Fig 1A). As expected, most but not all low-risk patients were node negative, and conversely, most high-risk patients were node positive (Fig 1B). Fig 1. Predicted distant recurrence (DR) risk in years 5 to 10 since random assignment (start of adjuvant endocrine therapy) for ATAC (Arimidex, Tamoxifen, Alone or Combination) trial (A) overall population and (B) node-negative and node-positive patients. Solid vertical lines indicate cutoff points for risk groups. CTS5, Clinical Treatment Score post–5 years.

isk in years 5 to 10 since random assignment (start of adjuvant endocrine therapy) for ATAC (Arimidex, Tamoxifen, Alone or Combination) trial (A) overall population and (B) node-negative and node-positive patients. Solid vertical lines indicate cutoff points for risk groups. CTS5, Clinical Treatment Score post–5 years. Overall, 42.0% were categorized as low risk, 31.3% as intermediate risk, and 26.7% as high risk for late distant recurrence (Table 2). Those categorized as low risk had a mean 5- to 10-year distant recurrence risk of 2.5% (95% CI, 1.8% to 3.4%), as compared with 7.7% (95% CI, 6.3% to 9.5%) for intermediate-risk and 20.3% (95% CI, 17.2% to 24.0%) for high-risk groups (Fig 2). Those at intermediate or high risk had a 3.42-fold (95% CI, 2.37- to 4.95-fold) or 9.43-fold (95% CI, 6.71- to 13.25-fold), respectively, higher risk of late distant recurrence than the low-risk group. Notably only two of 133 patients with one to three positive nodes and categorized as low risk had a distant recurrence between years 5 and 10 (Table 2). Virtually all patients with ≥ four positive nodes were categorized as high risk. Approximately one fifth of patients with two or three positive nodes had risk categorized as low or intermediate, whereas 42.9% with one positive node were categorize as high risk. Only 57.7% of node-negative patients were categorized as low risk. Table 2. Distribution of Risk Categories in the ATAC and BIG 1-98 Cohorts According to Tumor Size, Grade, and Nodal Involvement

Overall, 42.0% were categorized as low risk, 31.3% as intermediate risk, and 26.7% as high risk for late distant recurrence (Table 2). Those categorized as low risk had a mean 5- to 10-year distant recurrence risk of 2.5% (95% CI, 1.8% to 3.4%), as compared with 7.7% (95% CI, 6.3% to 9.5%) for intermediate-risk and 20.3% (95% CI, 17.2% to 24.0%) for high-risk groups (Fig 2). Those at intermediate or high risk had a 3.42-fold (95% CI, 2.37- to 4.95-fold) or 9.43-fold (95% CI, 6.71- to 13.25-fold), respectively, higher risk of late distant recurrence than the low-risk group. Notably only two of 133 patients with one to three positive nodes and categorized as low risk had a distant recurrence between years 5 and 10 (Table 2). Virtually all patients with ≥ four positive nodes were categorized as high risk. Approximately one fifth of patients with two or three positive nodes had risk categorized as low or intermediate, whereas 42.9% with one positive node were categorize as high risk. Only 57.7% of node-negative patients were categorized as low risk. Table 2. Distribution of Risk Categories in the ATAC and BIG 1-98 Cohorts According to Tumor Size, Grade, and Nodal Involvement Fig 2. Kaplan-Meier curves and 5- to 10-year distant recurrence (DR) rates since random assignment for the overall population according to trial (solid lines, ATAC [Arimidex, Tamoxifen, Alone or Combination]; dashed lines, BIG [Breast International Group] 1-98).

Table 2. Distribution of Risk Categories in the ATAC and BIG 1-98 Cohorts According to Tumor Size, Grade, and Nodal Involvement Fig 2. Kaplan-Meier curves and 5- to 10-year distant recurrence (DR) rates since random assignment for the overall population according to trial (solid lines, ATAC [Arimidex, Tamoxifen, Alone or Combination]; dashed lines, BIG [Breast International Group] 1-98). A total of 77 patients experienced local recurrence but no distant recurrence in years 0 to 5, with CTS5 (ATAC) ranking most as intermediate or high risk. Among these 77, CTS5 (ATAC) predicted 24.3 distant recurrences, and 25 were observed. Exclusion of these 77 patients marginally increased the HR for one standard deviation change, from 2.47 (95% CI, 2.24 to 2.73) to 2.53 (95% CI, 2.26 to 2.82). Validation Set (BIG 1-98) CTS5 (ATAC) performed non-significantly better in the validation BIG 1-98 cohort than CTS0 (CTS5 [ATAC]: HR 2.07; 95% CI, 1.88 to 2.28; LR-χ2 = 212.1 [1 df] v CTS0: HR 1.84; 95% CI, 1.70 to 1.98; LR-χ2 = 214.9 [1 df]). CTS5 (ATAC) was significantly prognostic in women who did not receive chemotherapy (HR, 2.20; 95% CI, 1.96 to 2.47; P < .001; LR-χ2 = 168.7 [1 df]) and more so when compared with those who did (HR, 1.76; 95% CI, 1.46 to 2.13; P < .001; LR-χ2 = 34.7 [1 df]; Appendix Table A1), but the interaction with chemotherapy was not statistically significant (P = .06).

c in women who did not receive chemotherapy (HR, 2.20; 95% CI, 1.96 to 2.47; P < .001; LR-χ2 = 168.7 [1 df]) and more so when compared with those who did (HR, 1.76; 95% CI, 1.46 to 2.13; P < .001; LR-χ2 = 34.7 [1 df]; Appendix Table A1), but the interaction with chemotherapy was not statistically significant (P = .06). The number of observed distant recurrences was compared with those expected by CTS5 (ATAC) in deciles of risk for node-negative and node-positive patients, separately (Figs 3A and 3B). In each case, there were no significant differences between the observed and expected numbers for any of the deciles. The correlation (r) between the observed versus expected numbers across the deciles was 0.89 for node-negative and 0.95 for node-positive groups. Using CTS0, a number of deciles showed significant χ2 values (Appendix Fig A2, online only), and the r values were also lower, at 0.78 and 0.87, respectively. Concordance between the estimated and actual distant recurrence rates was also shown to be better with CTS5 using the Goran-Heller C-index (CTS5 [ATAC], 0.678; CTS0, 0.656). Fig 3. Observed versus expected number of events and χ2 values in the BIG (Breast International Group) 1-98 trial according to deciles of Clinical Treatment Score post–5 years (ATAC [Arimidex, Tamoxifen, Alone or Combination]) for (A) node-negative and (B) node-positive patients. None of the χ2 were statistically significant.

Observed versus expected number of events and χ2 values in the BIG (Breast International Group) 1-98 trial according to deciles of Clinical Treatment Score post–5 years (ATAC [Arimidex, Tamoxifen, Alone or Combination]) for (A) node-negative and (B) node-positive patients. None of the χ2 were statistically significant. We used predefined cutoff points of 4.35 and 5.02 from ATAC to determine risk groups for late distant recurrence in BIG 1-98 (Figs 1A and 1B). These cut points intersected the risk curves for BIG 1-98 at 5.4% and 9.9% for node-negative patients and 5.5% and 9.5% for node-positive patients, respectively, and therefore were strongly validated by this test set. The distribution of patients in low-, intermediate-, and high-risk groups was also similar in the BIG 1-98 data set to that observed in the training set (Table 2). The mean 5- to 10-year distant recurrence risk of patients in BIG 1-98 in those three categories was 3.6% (95% CI, 2.7% to 4.9%), 6.9% (95% CI, 5.6% to 8.5%), and 17.3% (95% CI, 14.8% to 20.1%), respectively (Table 2; Fig 2). Thus, for each category, the actual mean risk for each category fitted well with that of the predicted risk. The curves for node-negative and node-positive women were almost identical in the CTS5 (ATAC) regions of overlap in BIG 1-98.

CI, 5.6% to 8.5%), and 17.3% (95% CI, 14.8% to 20.1%), respectively (Table 2; Fig 2). Thus, for each category, the actual mean risk for each category fitted well with that of the predicted risk. The curves for node-negative and node-positive women were almost identical in the CTS5 (ATAC) regions of overlap in BIG 1-98. Significant separation between low- versus intermediate-risk groups (HR, 2.19; 95% CI, 1.61 to 2.98) and low- versus high-risk groups (HR, 5.33; 95% CI, 4.02 to 7.07) was observed (Fig 2). Notably, only four of 304 patients with one to three positive nodes and categorized as low risk had a recurrence between years 5 and 10. As with the ATAC data set, in BIG 1-98, virtually all patients with ≥ four positive nodes were categorized as high risk (Table 2). The distribution of patients in the risk categories across histologic grades and nodal categories was similar between ATAC and BIG 1-98. Again, approximately one fifth of patients with two or three positive nodes had risk categorized as low or intermediate, but a somewhat smaller proportion of patients with one positive node were categorize as high risk (29.7% v 42.9%). In BIG 1-98, 62.5% of node-negative patients were categorized as low risk, compared with 57.7% in ATAC. Combined ATAC and BIG 1-98 Sets To increase the precision of the risk estimates, we combined the ATAC and BIG 1-98 data sets such that new coefficients were fitted using the same variables as in the training or validation cohort. The final CTS5 is represented by the following model:

Significant separation between low- versus intermediate-risk groups (HR, 2.19; 95% CI, 1.61 to 2.98) and low- versus high-risk groups (HR, 5.33; 95% CI, 4.02 to 7.07) was observed (Fig 2). Notably, only four of 304 patients with one to three positive nodes and categorized as low risk had a recurrence between years 5 and 10. As with the ATAC data set, in BIG 1-98, virtually all patients with ≥ four positive nodes were categorized as high risk (Table 2). The distribution of patients in the risk categories across histologic grades and nodal categories was similar between ATAC and BIG 1-98. Again, approximately one fifth of patients with two or three positive nodes had risk categorized as low or intermediate, but a somewhat smaller proportion of patients with one positive node were categorize as high risk (29.7% v 42.9%). In BIG 1-98, 62.5% of node-negative patients were categorized as low risk, compared with 57.7% in ATAC. Combined ATAC and BIG 1-98 Sets To increase the precision of the risk estimates, we combined the ATAC and BIG 1-98 data sets such that new coefficients were fitted using the same variables as in the training or validation cohort. The final CTS5 is represented by the following model: CTS5 = 0.438 × nodes + 0.988 × (0.093 × size − 0.001 × size2 + 0.375 × grade + 0.017 × age)

Combined ATAC and BIG 1-98 Sets To increase the precision of the risk estimates, we combined the ATAC and BIG 1-98 data sets such that new coefficients were fitted using the same variables as in the training or validation cohort. The final CTS5 is represented by the following model: CTS5 = 0.438 × nodes + 0.988 × (0.093 × size − 0.001 × size2 + 0.375 × grade + 0.017 × age) The relationship between the final CTS5 and risk of distant recurrence is shown in Figure 4, with a table of CTS5 values that relate to one-unit intervals of distant recurrence risk. New cutoff points for low- (CTS5 < 3.13), intermediate- (3.13 to 3.86), and high-risk (> 3.86) groups were derived from this final model. An example of the calculation of CTS5 and the associated risk estimate is given in Figure 4. Fig 4. Predicted 5- to 10-year distant recurrence (DR) risk since random assignment and Clinical Treatment Score post–5 years (CTS5) values for the combined data set. Solid vertical lines indicate cutoff points for risk groups. Arrows indicate the CTS5 and equivalent 5- to 10-year risks of a patient age 54 years with a 12-mm, node-negative, grade 2 tumor. Using the formula CTS5 = 0.438 × nodes + 0.988 × (0.093 × size − 0.001 × size2 + 0.375 × grade + 0.017 × age), her CTS5 score is 2.61 and her 5- to 10-year risk of DR is 3%.

for risk groups. Arrows indicate the CTS5 and equivalent 5- to 10-year risks of a patient age 54 years with a 12-mm, node-negative, grade 2 tumor. Using the formula CTS5 = 0.438 × nodes + 0.988 × (0.093 × size − 0.001 × size2 + 0.375 × grade + 0.017 × age), her CTS5 score is 2.61 and her 5- to 10-year risk of DR is 3%. DISCUSSION Over the last three decades, there have been major increases in invasive breast cancer incidence in Western countries; in the United States, it was estimated that > 250,000 women would be diagnosed with invasive breast cancer in 2017,15 with a large majority of cases localized to the breast. Approximately 80% of patients are now diagnosed as ER positive, and almost all of these are prescribed 5 years of adjuvant endocrine therapy. Although such treatment markedly reduces mortality (eg, by approximately 30% with 5 years of tamoxifen and approximately 40% with an AI in postmenopausal women), recurrences continue to occur after the 5-years treatment ends. The observation that these events can be decreased by continued treatment1,2,16 means that decisions about whether to continue with therapy at 5 years are at the forefront of patient management at that time. We expect that the CTS5 tool reported and validated here will prove helpful to oncologists and patients in making a decision about continued treatment. The integration of clinical pathologic features that are measured in all patients at diagnosis should mean that risk is calculable at little expense globally; the table in Figure 4 will allow a direct readout, and an online tool will be provided to facilitate estimates of continuous risk.

ion about continued treatment. The integration of clinical pathologic features that are measured in all patients at diagnosis should mean that risk is calculable at little expense globally; the table in Figure 4 will allow a direct readout, and an online tool will be provided to facilitate estimates of continuous risk. Strengths of the study include its use of two large sets of registration-standard randomized clinical trial data with detailed clinical follow-up for 10 years. The ATAC training set included the AI anastrozole as well as tamoxifen as adjuvant treatment, and although the specific endocrine adjuvant therapy did not feature in the algorithm, this allowed us to infer that the score would be valid for both tamoxifen- and AI-treated patients. This is consistent with the overview analysis of AIs versus tamoxifen.17 Median 5-year follow-up for the two trials combined occurred approximately 12 years ago. Therefore, it is possible that our risk estimates may not accurately reflect those of current patients reaching 5 years. However, the only major change to the management of primary ER-positive breast cancer since the completion of recruitment to ATAC and BIG 1-98 has been the introduction of trastuzumab for patients with human epidermal growth factor receptor 2–positive disease. CTS5 should be applied with caution in such patients until validated specifically for that population. All patients in the two cohorts were postmenopausal at diagnosis. Although risk of distant recurrence post–5 years has been reported to be similar across age groups, other than for the small group of patients diagnosed at age < 35 years,9 the present algorithm cannot be extended to premenopausal patients without further validation.

in the two cohorts were postmenopausal at diagnosis. Although risk of distant recurrence post–5 years has been reported to be similar across age groups, other than for the small group of patients diagnosed at age < 35 years,9 the present algorithm cannot be extended to premenopausal patients without further validation. Neither trial collected complete information on the use of extended adjuvant endocrine therapy. However, the first significant data supporting the use of an AI after tamoxifen1 emerged close to the end of the treatment period for the trials, and we estimate that < 1% of tamoxifen-treated patients in ATAC and < 5% in BIG 1-98 received such extended therapy. This would be expected to have minimal impact on our estimates of risk when extended therapy is not used. Also similar to the report by the Early Breast Cancer Trialists’ Collaborative Group, we found that whether patients had received chemotherapy at presentation had no significant impact on residual risk of recurrence when taking the other factors into account. This may relate in part to the observation that the bulk of the benefit from adjuvant chemotherapy is shown over the first 5 years of follow-up.18

we found that whether patients had received chemotherapy at presentation had no significant impact on residual risk of recurrence when taking the other factors into account. This may relate in part to the observation that the bulk of the benefit from adjuvant chemotherapy is shown over the first 5 years of follow-up.18 The categories of low, intermediate, and high risk were chosen to closely parallel those defined by several molecular profiling tools for managing patients with ER-positive breast cancer.19-21 However, those tools are applied immediately after surgery, largely in making the decision of whether to administer adjuvant chemotherapy; what is considered low or high risk in that setting may not be the same when considering the appropriateness of extended adjuvant therapy. In discussions with individual patients whose preferences for continuing or ceasing endocrine therapy at 5 years are likely to vary markedly, the use of a continuous risk estimate from CTS5 is likely to be more informative than the categorical estimates (ie, low, intermediate, and high) used here for illustrative and comparative purposes. The agreement between the ATAC and BIG1-98 data was almost complete within the low- and intermediate-risk categories but somewhat less beyond the intermediate/high cutoff. Thus, the instrument may be used with greatest confidence in defining 5- to 10-year distant recurrence risk when < 10% and will be of greatest use in assessing the potential value of extended therapy on the basis of risk estimates below that level.

-risk categories but somewhat less beyond the intermediate/high cutoff. Thus, the instrument may be used with greatest confidence in defining 5- to 10-year distant recurrence risk when < 10% and will be of greatest use in assessing the potential value of extended therapy on the basis of risk estimates below that level. Our report deals only with clinicopathologic profiles. Multigene expression profiles have significantly increased the ability to predict distant recurrence over 10 years after diagnosis in ER-positive breast cancer.22 Several of these signatures, such as the Oncotype Dx recurrence score,23 PAM50-based Prosigna risk of recurrence score,19,24 Breast Cancer Index,25,26 EndoPredict test,20,27,28 and Netherlands Cancer Institute 70-gene signature,29 are commercially available and endorsed by several guidelines.30-33 Although a number of them estimate risk of late as well as early recurrence, these tests were developed to manage patients with breast cancer at diagnosis and have not been calibrated for application 5 years after diagnosis. Over the first 10 years of follow-up, clinicopathologic and molecular factors have nearly completely independent prognostic value, and their optimal use for prognosis requires their integration.34 It is near certain that the same is true for the 5- to 10-year period. CTS5 provides a straightforward starting point for combining with molecular scores.

ow-up, clinicopathologic and molecular factors have nearly completely independent prognostic value, and their optimal use for prognosis requires their integration.34 It is near certain that the same is true for the 5- to 10-year period. CTS5 provides a straightforward starting point for combining with molecular scores. Supported by the Royal Marsden National Institute of Health Biomedical Research Centre, Breast Cancer Now Grant No. CTR-Q4-Y1, Cancer Research UK Grant No. C569/A16891, and Susan G. Komen for the Cure Promise Grant No. KG080081 (M.M.R.,G.V.). The BIG 1-98 trial was supported by Novartis. The International Breast Cancer Study Group, which coordinated the BIG 1-98 trial, was also supported by the Swedish Cancer Society, Swedish Research Council, Cancer Council Australia, Australia and New Zealand Breast Cancer Trials Group, Frontier Science and Technology Research Foundation, Swiss Group for Clinical Cancer Research, Cancer Research Switzerland/Oncosuisse, and Foundation for Clinical Cancer Research of Eastern Switzerland. Presented orally at the 2017 San Antonio Breast Cancer Symposium, San Antonio, TX, December 5-9, 2017. Clinical trial information: ISRCTN18233230, NCT00004205. AUTHOR CONTRIBUTIONS Conception and design: Mitch Dowsett, Ivana Sestak, Meredith M. Regan, Beat Thürlimann, Marco Colleoni, Jack Cuzick Provision of study materials or patients: Beat Thürlimann Collection and assembly of data: Ivana Sestak, Meredith M. Regan, Giuseppe Viale

Clinical trial information: ISRCTN18233230, NCT00004205. AUTHOR CONTRIBUTIONS Conception and design: Mitch Dowsett, Ivana Sestak, Meredith M. Regan, Beat Thürlimann, Marco Colleoni, Jack Cuzick Provision of study materials or patients: Beat Thürlimann Collection and assembly of data: Ivana Sestak, Meredith M. Regan, Giuseppe Viale Data analysis and interpretation: Mitch Dowsett, Ivana Sestak, Meredith M. Regan, Andrew Dodson, Beat Thürlimann, Marco Colleoni, Jack Cuzick Manuscript writing: All authors Final approval of manuscript: All authors Accountable for all aspects of the work: All authors AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST Integration of Clinical Variables for the Prediction of Late Distant Recurrence in Patients With Estrogen Receptor–Positive Breast Cancer Treated With 5 Years of Endocrine Therapy: CTS5 The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/jco/site/ifc. Mitch Dowsett Honoraria: Pfizer, Myriad Genetics, Roche Consulting or Advisory Role: Roche/Genentech, GTx, Radius Health Research Funding: AstraZeneca (Inst), Novartis (Inst), Pfizer (Inst), Radius Health (Inst) Other Relationship: Institute of Cancer Research Rewards for Inventors Ivana Sestak Honoraria: Myriad Genetics, NanoString Technologies

Mitch Dowsett Honoraria: Pfizer, Myriad Genetics, Roche Consulting or Advisory Role: Roche/Genentech, GTx, Radius Health Research Funding: AstraZeneca (Inst), Novartis (Inst), Pfizer (Inst), Radius Health (Inst) Other Relationship: Institute of Cancer Research Rewards for Inventors Ivana Sestak Honoraria: Myriad Genetics, NanoString Technologies Consulting or Advisory Role: Myriad Genetics Meredith M. Regan Consulting or Advisory Role: Merck, Ipsen (Inst) Research Funding: Veridex (Inst), OncoGenex (Inst), Pfizer (Inst), Ipsen (Inst), Novartis (Inst), Merck (Inst), Ferring Pharmaceuticals (Inst), Celgene (Inst), AstraZeneca (Inst), Pierre Fabre (Inst), Bristol-Myers Squibb (Inst) Andrew Dodson No relationship to disclose Giuseppe Viale Honoraria: Merck Sharp & Dohme Oncology Consulting or Advisory Role: DAKO, Roche/Genentech, AstraZeneca, Bristol-Myers Squibb, Astellas Pharma Research Funding: Roche/Genentech, Ventana Medical Systems (Inst) Travel, Accommodations, Expenses: Roche, Celgene Beat Thürlimann Stock or Other Ownership: Roche, Novartis Honoraria: Roche Consulting or Advisory Role: Roche, Eli Lilly, Amgen, Pfizer, AstraZeneca Expert Testimony: AstraZeneca Travel, Accommodations, Expenses: Roche Marco Colleoni Honoraria: Novartis Consulting or Advisory Role: Pierre Fabre, Pfizer, OBI Pharma, Puma Biotechnology, Celldex, AstraZeneca Jack Cuzick Research Funding: AstraZeneca (Inst)