Browse the corpus

Introduction Heart failure is common, important, and costly. More than 15 million people are thought to have heart failure in Europe.1 It is estimated that in the USA alone the total direct costs of heart failure care is more than $20 billion per annum.2 Despite improvements in diagnosis and the development of effective therapies for patients with heart failure, the case-fatality rate at 5-years is 50%.3–6 One of the major causes of heart failure is acute myocardial infarction with symptoms developing in those patients who have sustained significant myocardial injury and ventricular impairment.7,8 However, with the development of high-sensitivity cardiac troponin assays we increasingly identify patients with minor myocardial injury.9–12 Whether cardiac troponin concentration remains a useful predictor in this group of patients is uncertain. Furthermore, high sensitivity cardiac troponin assays are increasingly used to risk-stratify patients to identify those likely to benefit from hospitalization and/or further investigation.13 We therefore examined the association and predictive utility of cardiac troponin concentration for subsequent admission to hospital with heart failure in consecutive patients with suspected acute coronary syndrome.

tratify patients to identify those likely to benefit from hospitalization and/or further investigation.13 We therefore examined the association and predictive utility of cardiac troponin concentration for subsequent admission to hospital with heart failure in consecutive patients with suspected acute coronary syndrome. Methods We included 4748 consecutive patients who presented to the Emergency Department with suspected acute coronary syndrome to three secondary and tertiary hospitals in Edinburgh, Scotland between 1 June 2013 and 31 January 2014 and who survived to discharge. These patients were enrolled in the standard care arm of a stepped-wedge cluster randomized controlled trial (ClinicalTrials.gov registration NCT01852123). All patients who had cardiac troponin requested by the attending clinician for suspected acute coronary syndrome were included. Patients were excluded if they had been admitted previously during the study period or did not reside in Scotland.13 The study was performed with the approval of the National Research Ethics Committee, and in accordance with the Declaration of Helsinki.

ing clinician for suspected acute coronary syndrome were included. Patients were excluded if they had been admitted previously during the study period or did not reside in Scotland.13 The study was performed with the approval of the National Research Ethics Committee, and in accordance with the Declaration of Helsinki. High-sensitivity cardiac troponin I assay The ARCHITECTSTAT high-sensitive troponin I assay (Abbott Laboratories, Abbott Park, IL, USA) was used to measure cardiac troponin I concentration in all patients. The limit of detection (LoD) of this assay is 1.2 ng/L and the upper reference limit (URL) or 99% centile in women is 16 and 34 ng/L in women and men, respectively.13 The lowest cardiac troponin I concentration with an inter-assay coefficient of variation (CV) of less than 10% was 4.7 ng/L according to the manufacturer and was 6 ng/L in our laboratory.14,15 In patients with more than one measure of cardiac troponin, the highest concentration was used. Values below the LoD were assigned a value of 1.2 ng/L (n = 506, 10.7%).

on with an inter-assay coefficient of variation (CV) of less than 10% was 4.7 ng/L according to the manufacturer and was 6 ng/L in our laboratory.14,15 In patients with more than one measure of cardiac troponin, the highest concentration was used. Values below the LoD were assigned a value of 1.2 ng/L (n = 506, 10.7%). Classification of myocardial infarction The electronic patient record system TrakCare (InterSystems Corporation, Cambridge, MA, USA) was used to acquire baseline characteristics for each patient, including cardiovascular risk factors and past medical events.13 Hyperlipidaemia and hypertension were defined as a documented history of the condition, or by the respective use of lipid-lowering or anti-hypertensive medication. Smoking was defined as current or ex-smoking at admission. Killip class was determined by the attending clinician. Myocardial infarction type was classified by two investigators (A.S., A.A.) independently using the Third Universal Definition of Myocardial Infarction,16 with discrepancies being adjudicated by third investigator (N.M.). Patients were classified using all available clinical information at the time of the index admission (including peak troponin and serial change in troponin when these were available). Type 1 myocardial infarction was identified when myocardial necrosis occurred in the context of a presentation with suspected acute coronary syndrome with chest pain or evidence of myocardial ischaemia on the electrocardiogram. Patients with symptoms and signs of myocardial ischaemia on the electrocardiogram that were thought to be due to increased oxygen demand or decreased supply (e.g. tachyarrhythmia, hypotension, or anaemia) and myocardial necrosis were classified as having a type 2 myocardial infarction. Myocardial injury was defined as evidence of myocardial necrosis in the absence of any clinical features of myocardial ischaemia. A detailed description of the classification criteria was published elsewhere.13

a, hypotension, or anaemia) and myocardial necrosis were classified as having a type 2 myocardial infarction. Myocardial injury was defined as evidence of myocardial necrosis in the absence of any clinical features of myocardial ischaemia. A detailed description of the classification criteria was published elsewhere.13 Outcomes Outcomes were censored on 31 March 2014 (median follow-up 156 days, interquartile range 98–218 days). Hospitalization and death during the follow-up period were obtained via linkage to the national hospital database (‘the Scottish Morbidity Record’, SMR01). Heart failure was defined as any hospitalization assigned the code I50. All heart failure diagnoses were adjudicated by a physician and cardiologist independently (D.S., A.S.). Criteria for the adjudication of heart failure were based on review of inpatient clinical records and based on symptoms of congestive cardiac failure supported by imaging evidence of left ventricular dysfunction. Agreement was 100% between the two investigators. Cardiac death was defined as deaths due to myocardial infarction, arrhythmia, or heart failure (ICD-10 codes I21/22 and I46-50).

Outcomes Outcomes were censored on 31 March 2014 (median follow-up 156 days, interquartile range 98–218 days). Hospitalization and death during the follow-up period were obtained via linkage to the national hospital database (‘the Scottish Morbidity Record’, SMR01). Heart failure was defined as any hospitalization assigned the code I50. All heart failure diagnoses were adjudicated by a physician and cardiologist independently (D.S., A.S.). Criteria for the adjudication of heart failure were based on review of inpatient clinical records and based on symptoms of congestive cardiac failure supported by imaging evidence of left ventricular dysfunction. Agreement was 100% between the two investigators. Cardiac death was defined as deaths due to myocardial infarction, arrhythmia, or heart failure (ICD-10 codes I21/22 and I46-50). Statistical analysis For all analyses, patients who died during the index admission were excluded. Analyses were performed for cardiac troponin I concentration as categorical and continuous variable. Patients were categorized into five groups according to the highest measured cardiac troponin I concentration. Group 5 included all patients who had cardiac troponin concentrations above the sex-specific 99th centile URL (>34 ng/L for men and >16 ng/L for women).13,17 The remaining patients were split into quartiles (Groups 1–4) by sex: 1.2–2.0, 2.1–4.0, 4.1–8.0, and 8.1–34.0 ng/L for men; 1.2–1.9, 2.0–2.9, 3.0–6.0, and 6.1–16.0 ng/L for women. These categories were assigned prior to analysing associations between groups and any of the study outcomes. For the analysis as a continuous variable, cardiac troponin concentrations were log-transformed as a linearizing transformation. Summary statistics were obtained for baseline characteristics by category. Hazard ratios (HRs) for first subsequent hospitalization with heart failure and for first subsequent hospitalization with heart failure or cardiac death were estimated according to group in Cox regression models adjusting for age, sex, cardiovascular risk factors (diabetes mellitus, hypertension, ischaemic heart disease, and previous myocardial infarction), and clinical features (systolic blood pressure and creatinine concentration). As there were neither heart failure hospitalizations nor cardiac deaths in the two groups with the lowest cardiac troponin concentrations, these two groups were collapsed. Similar models were used to estimate associations for troponin (log-transformed) as a continuous variable. Polynomial terms and penalized splines were used to present the associations, which were non-linear, in tables and figures, respectively.

he lowest cardiac troponin concentrations, these two groups were collapsed. Similar models were used to estimate associations for troponin (log-transformed) as a continuous variable. Polynomial terms and penalized splines were used to present the associations, which were non-linear, in tables and figures, respectively. Sensitivity analyses were performed having excluded patients with heart failure at the index presentation. Subgroup analyses of log-troponin concentration were conducted for patients with cardiac troponin concentrations below and above the 99th centile URL. Patients with elevated cardiac troponin concentrations were additionally analysed according to the diagnosis at the index presentation: type 1 myocardial infarction, type 2 myocardial infarction, and myocardial injury. The discrimination of troponin (with and without additional variables) was estimated using area under the receiver operating characteristic curves (C-statistics). Confidence intervals (CIs) for the C-statistics were obtained via bootstrapping.18 Analyses were performed in IBM SPSS Statistics Version 22.0.0 (Armonk, NY: IBM, USA, 2014) and R Version 3.0.1 (R project for statistical computing, Vienna, Austria).

rea under the receiver operating characteristic curves (C-statistics). Confidence intervals (CIs) for the C-statistics were obtained via bootstrapping.18 Analyses were performed in IBM SPSS Statistics Version 22.0.0 (Armonk, NY: IBM, USA, 2014) and R Version 3.0.1 (R project for statistical computing, Vienna, Austria). Results There were 4870 patients with suspected acute coronary syndrome (mean age 61 ± 16 years, 57% men) enrolled between 1 June 2013 and 31 January 2014. One hundred and twenty-two patients died during the index presentation and these patients were excluded from this analysis. The median cardiac troponin concentration was 5 ng/L (interquartile range 2–22 ng/L). There were 1151 (24%) patients with cardiac troponin concentrations above the URL (Group 5); 723 (15.2%) patients were classified as having type 1 myocardial infarction, 158 (3.3%) type 2 myocardial infarction, and 270 (5.7%) with myocardial injury. Patients with higher cardiac troponin concentrations were older, had a higher Killip class, and more cardiovascular risk factors. They were also more likely to have been treated with angiotensin converting enzyme inhibitors, angiotensin receptor blockers or beta blockers, and had higher creatinine concentrations (Table 1). Table 1 Baseline characteristics of patients with suspected acute coronary syndrome stratified by cardiac troponin concentration

y were also more likely to have been treated with angiotensin converting enzyme inhibitors, angiotensin receptor blockers or beta blockers, and had higher creatinine concentrations (Table 1). Table 1 Baseline characteristics of patients with suspected acute coronary syndrome stratified by cardiac troponin concentration Patients stratified by peak troponin concentration Group 1 Group 2 Group 3 Group 4 Group 5a n = 900 (19%) n = 899 (19%) n = 899 (19%) n = 899 (19%) n = 1151 (24%) Troponin, ng/L (median, range) Men 1.9 (1.2–2.0) 3.0 (2.1–4.0) 5.9 (4.1–8.0) 15.0 (8.1–34.0) 484.5 (34.1–50 000) Women 1.2 (1.2–1.9) 2.0 (2.0–2.9) 3.0 (3.0–6.0) 9.0 (6.1–16.0) 100.0 (16.1–50 000) Age, years (mean, SD) 49 (13) 58 (14) 66 (14) 71 (14) 71 (15) Females 378 (42%) 378 (42%) 378 (42%) 378 (42%) 545 (47%) Diabetes mellitus 68 (9%) 115 (15%) 130 (16%) 162 (21%) 191 (18%) Hypertension 120 (16%) 224 (29%) 290 (36%) 316 (40%) 444 (41%) Hyperlipidaemia 119 (16%) 209 (27%) 223 (28%) 238 (30%) 336 (31%) Ischaemic heart disease 93 (12%) 201 (26%) 304 (38%) 372 (47%) 409 (38%) Previous myocardial infarction 58 (6%) 120 (13%) 159 (18%) 209 (23%) 244 (21%) Previous stroke 27 (3%) 48 (5%) 54 (6%) 90 (10%) 107 (9%) Heart failure at index presentation 0 (0%) 7 (1%) 21 (2%) 83 (9%) 243 (21%) Current or ex-smoker 315 (57%) 309 (58%) 287 (56%) 252 (59%) 401 (63%) Admission ACE inhibitor/ARB 88 (16%) 160 (28%) 209 (36%) 221 (42%) 270 (38%) Admission beta blocker 65 (12%) 108 (19%) 166 (29%) 189 (36%) 236 (33%) Killip class 1 784 (99%) 784 (97%) 786 (96%) 699 (88%) 923 (85%) 2 8 (1%) 25 (3%) 30 (4%) 84 (11%) 108 (10%) 3 0 0 0 (0%) 14 (2%) 52 (5%) 4 0 0 0 1 (0%) 3 (0%) Creatinine, mg/dL (mean, SD) 0.84 (0.14) 0.86 (0.21) 0.91 (0.26) 1.07 (0.63) 1.13 (0.75) Heart rate, b.p.m. (mean, SD) 79 (18) 78 (20) 78 (20) 83 (24) 86 (29) Systolic blood pressure, mmHg (mean, SD) 136 (21) 139 (25) 138 (23) 140 (28) 139 (30) Values are numbers (proportion), except where indicated.

0 0 1 (0%) 3 (0%) Creatinine, mg/dL (mean, SD) 0.84 (0.14) 0.86 (0.21) 0.91 (0.26) 1.07 (0.63) 1.13 (0.75) Heart rate, b.p.m. (mean, SD) 79 (18) 78 (20) 78 (20) 83 (24) 86 (29) Systolic blood pressure, mmHg (mean, SD) 136 (21) 139 (25) 138 (23) 140 (28) 139 (30) Values are numbers (proportion), except where indicated. SD, standard deviation; ACE, angiotensin converting enzyme; ARB, angiotensin receptor blocker. aGroup 5—all patients with cardiac troponin concentrations >upper reference limit. Heart failure hospitalization Eighty-three patients were hospitalized with heart failure (40/1000 person years) during a total of 2071 person years follow-up. Patients with cardiac troponin concentrations above the URL (Group 5) were more likely to be hospitalized with heart failure than patients with lower troponin concentrations (118/1000 person years vs. 2/1000 in Groups 1 and 2 combined, HR: 47.2, Table 2). Similar associations were evident after adjustment for age and sex, and after further adjustment for diabetes mellitus, hypertension, ischaemic heart disease, previous myocardial infarction, systolic blood pressure, and creatinine concentration (Table 2). Table 2 Heart failure hospitalization or cardiac death in suspected acute coronary syndrome stratified by troponin concentration

ex, and after further adjustment for diabetes mellitus, hypertension, ischaemic heart disease, previous myocardial infarction, systolic blood pressure, and creatinine concentration (Table 2). Table 2 Heart failure hospitalization or cardiac death in suspected acute coronary syndrome stratified by troponin concentration Group 1 Group 2 Group 3 Group 4 Group 5 All patients n = 900 (19%) n = 899 (19%) n = 899 (19%) n = 899 (19%) n = 1151 (24%) n = 4748 (100%) Troponin (ng/L): Men 1.2–2.0 2.1–4.0 4.1–8.0 8.1–34.0 34.1–50 000 1.2–50 000 Troponin (ng/L): Women 1.2–1.9 2.0–2.9 3.0–6.0 6.1–16.0 16.1–50 000 1.2–50 000 Heart failure hospitalization Events, n 0 2 4 21 56 83 Person years 407 410 390 390 473 2071 Incidence (per 1000 person years) 0 5 10 54 118 40 HR, unadjusted 1 4.1 21.8 47.2 HR, model 1 1 2.4 9.9 21.4 HR, model 2 1 3.0 11.7 28.9 HR, model 1, continuousa 2.80 (1.81–4.31) 1.03 (0.96–1.12) Heart failure hospitalization or cardiac death Events, n 0 2 5 29 84 120 Person years 407 410 390 390 473 2071 Incidence (per 1000 person years) 0 5 13 74 178 58 HR, unadjusted 1 5.2 30.2 71.3 HR, model 1 1 2.7 11.6 27.6 HR, model 2 1 2.7 14.4 34.1 HR, model 1, continuousa 3.03 (2.05–4.48) 1.03 (0.97–1.10) Hazard ratio (95% CI). Model 1 adjusts for age and sex; model 2 additionally adjusts for diabetes mellitus, hypertension, and ischaemic heart disease, previous myocardial infarction, systolic blood pressure at the index presentation, creatinine at the index presentation and an interaction term between ischaemic heart disease and previous myocardial infarction. HR: hazard ratio.

Group 1 Group 2 Group 3 Group 4 Group 5 All patients n = 900 (19%) n = 899 (19%) n = 899 (19%) n = 899 (19%) n = 1151 (24%) n = 4748 (100%) Troponin (ng/L): Men 1.2–2.0 2.1–4.0 4.1–8.0 8.1–34.0 34.1–50 000 1.2–50 000 Troponin (ng/L): Women 1.2–1.9 2.0–2.9 3.0–6.0 6.1–16.0 16.1–50 000 1.2–50 000 Heart failure hospitalization Events, n 0 2 4 21 56 83 Person years 407 410 390 390 473 2071 Incidence (per 1000 person years) 0 5 10 54 118 40 HR, unadjusted 1 4.1 21.8 47.2 HR, model 1 1 2.4 9.9 21.4 HR, model 2 1 3.0 11.7 28.9 HR, model 1, continuousa 2.80 (1.81–4.31) 1.03 (0.96–1.12) Heart failure hospitalization or cardiac death Events, n 0 2 5 29 84 120 Person years 407 410 390 390 473 2071 Incidence (per 1000 person years) 0 5 13 74 178 58 HR, unadjusted 1 5.2 30.2 71.3 HR, model 1 1 2.7 11.6 27.6 HR, model 2 1 2.7 14.4 34.1 HR, model 1, continuousa 3.03 (2.05–4.48) 1.03 (0.97–1.10) Hazard ratio (95% CI). Model 1 adjusts for age and sex; model 2 additionally adjusts for diabetes mellitus, hypertension, and ischaemic heart disease, previous myocardial infarction, systolic blood pressure at the index presentation, creatinine at the index presentation and an interaction term between ischaemic heart disease and previous myocardial infarction. HR: hazard ratio. aAnalysis of troponin as a continuous variable among patients with troponin levels below the upper reference limit (Groups 1–4) and above the upper reference limit (Group 5).

Group 1 Group 2 Group 3 Group 4 Group 5 All patients n = 900 (19%) n = 899 (19%) n = 899 (19%) n = 899 (19%) n = 1151 (24%) n = 4748 (100%) Troponin (ng/L): Men 1.2–2.0 2.1–4.0 4.1–8.0 8.1–34.0 34.1–50 000 1.2–50 000 Troponin (ng/L): Women 1.2–1.9 2.0–2.9 3.0–6.0 6.1–16.0 16.1–50 000 1.2–50 000 Heart failure hospitalization Events, n 0 2 4 21 56 83 Person years 407 410 390 390 473 2071 Incidence (per 1000 person years) 0 5 10 54 118 40 HR, unadjusted 1 4.1 21.8 47.2 HR, model 1 1 2.4 9.9 21.4 HR, model 2 1 3.0 11.7 28.9 HR, model 1, continuousa 2.80 (1.81–4.31) 1.03 (0.96–1.12) Heart failure hospitalization or cardiac death Events, n 0 2 5 29 84 120 Person years 407 410 390 390 473 2071 Incidence (per 1000 person years) 0 5 13 74 178 58 HR, unadjusted 1 5.2 30.2 71.3 HR, model 1 1 2.7 11.6 27.6 HR, model 2 1 2.7 14.4 34.1 HR, model 1, continuousa 3.03 (2.05–4.48) 1.03 (0.97–1.10) Hazard ratio (95% CI). Model 1 adjusts for age and sex; model 2 additionally adjusts for diabetes mellitus, hypertension, and ischaemic heart disease, previous myocardial infarction, systolic blood pressure at the index presentation, creatinine at the index presentation and an interaction term between ischaemic heart disease and previous myocardial infarction. HR: hazard ratio. aAnalysis of troponin as a continuous variable among patients with troponin levels below the upper reference limit (Groups 1–4) and above the upper reference limit (Group 5). In all patients the association between cardiac troponin concentration and heart failure hospitalization was non-linear (P for non-linearity <0.001), with a plateau evident at around 30 ng/L as shown via penalized spline smoothing functions (Figure 1, Supplementary material online, Table S2 for model coefficients).

aAnalysis of troponin as a continuous variable among patients with troponin levels below the upper reference limit (Groups 1–4) and above the upper reference limit (Group 5). In all patients the association between cardiac troponin concentration and heart failure hospitalization was non-linear (P for non-linearity <0.001), with a plateau evident at around 30 ng/L as shown via penalized spline smoothing functions (Figure 1, Supplementary material online, Table S2 for model coefficients). Figure 1 Association between peak cardiac troponin concentration and time to first event for heart failure hospitalization, and heart failure hospitalization or cardiac death. Departures from linearity were explored using penalized spline smoothing functions (grey band). The association was also analysed after stratifying patients into those with a peak troponin concentration less than the upper reference limit (URL) (red line), and those with troponin concentrations above the URL, with a specific index diagnosis (type 1 myocardial infarction, green line; type 2 myocardial infarction, blue; myocardial injury, purple).

after stratifying patients into those with a peak troponin concentration less than the upper reference limit (URL) (red line), and those with troponin concentrations above the URL, with a specific index diagnosis (type 1 myocardial infarction, green line; type 2 myocardial infarction, blue; myocardial injury, purple). Among patients with cardiac troponin concentrations below the URL there was a nearly three-fold increase in risk of heart failure hospitalization per doubling of troponin concentration (HR: 2.80, 95% CI 1.81–4.31), whereas among all patients with cardiac troponin concentration above this threshold there was no evidence of an association between increasing cardiac troponin concentrations and heart failure hospitalization (HR: 1.03, 95% CI 0.96–1.12, Table 2). On stratifying by the adjudicated diagnosis, patients with type 1 and type 2 myocardial infarction were consistent with negative or weakly positive associations between peak cardiac troponin concentration and hospitalization with heart failure (1.06, 95% CI 0.96–1.17 and 0.81, 95% CI 0.57–1.14, respectively). For patients with myocardial injury there was a positive association (HR: 1.21, 95% CI 1.00–1.47), although this was much weaker than for patients with cardiac troponin concentrations below the URL and the CI included the null (Supplementary material online, Table S1). Similar associations were found in analyses using presentation instead of maximal cardiac troponin concentrations (Supplementary material online, Table S1).

h this was much weaker than for patients with cardiac troponin concentrations below the URL and the CI included the null (Supplementary material online, Table S1). Similar associations were found in analyses using presentation instead of maximal cardiac troponin concentrations (Supplementary material online, Table S1). Similar associations were found in women and men (Supplementary material online, Figure S1). The HR in patients with cardiac troponin concentrations below the URL was 3.01 (95% CI 1.50–6.03) for women and 2.68 (95% CI 1.54–4.69) for men. Among patients with troponin concentration above the URL the HR was 1.11 (95% CI 1.01–1.21) for women and 0.88 (95% CI 0.76–1.03) for men. Heart failure hospitalization or cardiac death Similar associations were evident for the combined outcome of hospitalization with heart failure or cardiac death in unadjusted analyses and after adjusting for age and sex, and after further adjustment for diabetes mellitus, hypertension, ischaemic heart disease, previous myocardial infarction, systolic blood pressure, and creatinine concentration (Table 2 and Figure 1).

come of hospitalization with heart failure or cardiac death in unadjusted analyses and after adjusting for age and sex, and after further adjustment for diabetes mellitus, hypertension, ischaemic heart disease, previous myocardial infarction, systolic blood pressure, and creatinine concentration (Table 2 and Figure 1). Sensitivity analyses Sensitivity analyses after the exclusion of 354 patients with heart failure [111 patients with troponin <99% centile URL (3.1%), 132 patients with type 1 MI (18.3%), 31 patients with type 2 MI (19.6%), and 80 patients with myocardial injury (29.6%)] identified during the index presentation identified similar associations for cardiac troponin and subsequent heart failure hospitalization, and the composite endpoint of heart failure hospitalization or cardiac death (Supplementary material online, Tables S2 and S4).

6%), and 80 patients with myocardial injury (29.6%)] identified during the index presentation identified similar associations for cardiac troponin and subsequent heart failure hospitalization, and the composite endpoint of heart failure hospitalization or cardiac death (Supplementary material online, Tables S2 and S4). Discrimination For heart failure hospitalization, cardiac troponin concentration alone achieved similar discrimination to a model incorporating clinical features (age, sex, diabetes mellitus, hypertension and ischaemic heart disease, previous myocardial infarction, systolic blood pressure, and creatinine concentration) (C-statistic 0.80, 95% CI 0.76–0.83 and 0.80, 95% CI 0.75–0.85, respectively, Table 3). Moreover, when troponin concentration was added to the risk factors model, the prediction improved to 0.86 (95% CI 0.82–0.89). Similar discrimination was evident for the prediction of heart failure hospitalization or cardiac death (C-statistic 0.87, 95% CI 0.84–0.90). Discrimination was similar in men and women (0.86, 95% CI 0.81–0.91 and 0.87, 95% CI 0.82–0.92, respectively). Table 3 Discrimination of cardiac troponin and a model based on clinical features for heart failure hospitalization or cardiac death

ilure hospitalization or cardiac death (C-statistic 0.87, 95% CI 0.84–0.90). Discrimination was similar in men and women (0.86, 95% CI 0.81–0.91 and 0.87, 95% CI 0.82–0.92, respectively). Table 3 Discrimination of cardiac troponin and a model based on clinical features for heart failure hospitalization or cardiac death Area under the curvea Troponinb Modelc Troponinb + modelc Heart failure hospitalization 0.80 (0.76–0.83) 0.80 (0.75–0.85) 0.86 (0.82–0.89) Men 0.80 (0.75–0.85) 0.81 (0.74–0.88) 0.86 (0.81–0.91) Women 0.81 (0.77–0.86) 0.80 (0.73–0.87) 0.87 (0.82–0.92) Heart failure hospitalization or cardiac death 0.81 (0.79–0.84) 0.82 (0.78–0.86) 0.87 (0.84–0.90) Men 0.82 (0.79–0.86) 0.84 (0.79–0.89) 0.88 (0.84–0.92) Women 0.82 (0.79–0.86) 0.81 (0.75–0.87) 0.88 (0.84–0.92) a95% CI in brackets. bIncluding troponin squared. cModel of clinical features age, sex, diabetes mellitus, hypertension, ischaemic heart disease, previous myocardial infarction, systolic blood pressure, and creatinine concentration.

Area under the curvea Troponinb Modelc Troponinb + modelc Heart failure hospitalization 0.80 (0.76–0.83) 0.80 (0.75–0.85) 0.86 (0.82–0.89) Men 0.80 (0.75–0.85) 0.81 (0.74–0.88) 0.86 (0.81–0.91) Women 0.81 (0.77–0.86) 0.80 (0.73–0.87) 0.87 (0.82–0.92) Heart failure hospitalization or cardiac death 0.81 (0.79–0.84) 0.82 (0.78–0.86) 0.87 (0.84–0.90) Men 0.82 (0.79–0.86) 0.84 (0.79–0.89) 0.88 (0.84–0.92) Women 0.82 (0.79–0.86) 0.81 (0.75–0.87) 0.88 (0.84–0.92) a95% CI in brackets. bIncluding troponin squared. cModel of clinical features age, sex, diabetes mellitus, hypertension, ischaemic heart disease, previous myocardial infarction, systolic blood pressure, and creatinine concentration. Discussion In consecutive patients with suspected acute coronary syndrome high-sensitivity cardiac troponin I concentrations predict an increased risk of subsequent hospitalization with heart failure or cardiac death. Interestingly the relationship between cardiac troponin concentration and heart failure was strongest for patients without myocardial infarction. In these patients, the risk of subsequent hospitalization increased three-fold for every doubling in cardiac troponin concentration and the addition of troponin to a model with clinical features and cardiovascular risk factors markedly improved discrimination.

e was strongest for patients without myocardial infarction. In these patients, the risk of subsequent hospitalization increased three-fold for every doubling in cardiac troponin concentration and the addition of troponin to a model with clinical features and cardiovascular risk factors markedly improved discrimination. Heart failure may occur following myocardial infarction in patients with significant myocardial injury and left ventricular systolic impairment. Our observations from a large cohort of consecutive patients with suspected acute coronary syndrome demonstrate that any increase in cardiac troponin concentration <99th centile is associated with an increase in the risk of developing heart failure. Interestingly, in patients with cardiac troponin concentrations >99th centile, further increases in cardiac troponin did not identify those at higher risk. This was true even among patients with type 1 myocardial infarction in whom one might expect more extensive myocardial injury to be associated with a greater risk of heart failure.19

patients with cardiac troponin concentrations >99th centile, further increases in cardiac troponin did not identify those at higher risk. This was true even among patients with type 1 myocardial infarction in whom one might expect more extensive myocardial injury to be associated with a greater risk of heart failure.19 The observation that the magnitude of cardiac troponin concentration is not associated with the risk of heart failure in patients with myocardial infarction should perhaps be interpreted with caution, as it has not previously been reported. It is possible that differences in the timing of cardiac troponin measurement could obscure a relationship between troponin concentration and heart failure. In our cohort the majority of patients with troponin concentrations >99th centile had measures at presentation and 12 h (71%) after the onset of symptoms (approximately 6 h following presentation), as recommended by international guidelines for the diagnosis of myocardial infarction.16 However, it is now recognized that cardiac troponin concentrations measured during the plateau phase (24–72 h after presentation) are more closely related to the structural sequelae of myocardial infarction as identified on cardiac magnetic resonance imaging, such as scarring and left ventricular impairment, than are troponin concentrations at 6–12 h which are obtained for diagnosis.20 An alternative explanation is that other clinical factors are more important than peak troponin concentration in determining whether patients with myocardial infarction develop heart failure, such as the timing and completeness of revascularization, the extent and severity of coronary heart disease, and the presence of comorbid conditions. Competing risks is an unlikely explanation, as similar associations were evident for the composite outcome of heart failure hospitalization or cardiac death as were observed for heart failure hospitalization alone.

cularization, the extent and severity of coronary heart disease, and the presence of comorbid conditions. Competing risks is an unlikely explanation, as similar associations were evident for the composite outcome of heart failure hospitalization or cardiac death as were observed for heart failure hospitalization alone. Interestingly, in those patients without myocardial infarction where cardiac troponin concentrations were within the normal reference range, troponin concentration was a powerful independent predictor of heart failure hospitalization. Those patients with cardiac troponin concentrations in the lowest two quartiles did not go on to have heart failure or cardiac death. This provides further support to our previous observations that patients with suspected acute coronary syndrome and low cardiac troponin concentrations (<5 ng/L) are at very low risk of cardiovascular events.13 It may be that some patients with cardiac troponin concentrations between 5 ng/L and the URL have asymptomatic structural heart disease and are at risk of subsequent decompensation. Echocardiography is recommended in patients with myocardial infarction, but is not routinely performed in those in whom the diagnosis is excluded. Further studies are now needed to determine whether those patients with cardiac troponin concentrations within the normal reference range, but greater than 5 ng/L, would benefit from additional investigations, and treatments to prevent heart failure events.

outinely performed in those in whom the diagnosis is excluded. Further studies are now needed to determine whether those patients with cardiac troponin concentrations within the normal reference range, but greater than 5 ng/L, would benefit from additional investigations, and treatments to prevent heart failure events. Indeed, the discrimination of cardiac troponin for predicting subsequent heart failure and cardiac death is excellent.21 When added to clinical features and cardiovascular risk factors the area under the receiver operator characteristic score was 0.87, a level indicating that a test is potentially useful for event prediction in individual patients.22 Importantly, discrimination was similar in men and women. Previous authors have reported associations between high-sensitivity troponin I/T and heart failure events in the general adult population and in patients with stable coronary artery disease.23–26 We now demonstrate that cardiac troponin predicts heart failure admission in unselected patients presenting with suspected acute coronary syndrome, in whom troponin measurement is performed routinely. Our discrimination estimate needs to be validated in an external cohort, but it is unlikely that we have overestimated the C-statistic through over fitting, as only a single continuous variable (troponin) was added to the baseline model.

cted acute coronary syndrome, in whom troponin measurement is performed routinely. Our discrimination estimate needs to be validated in an external cohort, but it is unlikely that we have overestimated the C-statistic through over fitting, as only a single continuous variable (troponin) was added to the baseline model. A strength of this study was that it included all consecutive patients presenting to either secondary or tertiary care hospitals, following referral or self-presentation, in whom acute coronary syndrome was suspected, making this a truly representative sample. However, this approach did mean that the timing of troponin sampling was at the discretion of the attending physician and will have varied by service-related or patient factors. Nonetheless, our findings are generalizable as they reflect usual clinical practice and because they were similar for maximal and presentation cardiac troponin concentrations. A second limitation of this study is that patients were censored at 10 months, and associations between cardiac troponin and heart failure hospitalization or cardiac death at later times could not be determined. However, the first 6-month period is arguably of greatest clinical interest to physicians in preventing subsequent hospitalization. Finally, we do not have N-terminal pro-BNP or BNP concentrations available in this cohort, which would likely further improved the predictive power of our model for heart failure.23,27–29

However, the first 6-month period is arguably of greatest clinical interest to physicians in preventing subsequent hospitalization. Finally, we do not have N-terminal pro-BNP or BNP concentrations available in this cohort, which would likely further improved the predictive power of our model for heart failure.23,27–29 Conclusion High-sensitivity cardiac troponin I is an excellent predictor of heart failure hospitalizations and cardiac death in patients with suspected acute coronary syndrome. Troponin concentrations may, in particular, be used to identify patients without myocardial infarction who are at risk of heart failure. It may be that this group of patients will benefit from N-terminal pro-BNP testing and/or echocardiography. Intervention studies, ideally randomized clinical trials, are needed to determine whether the costs of such a strategy are justified by benefits such as reducing or delaying heart failure admissions. Supplementary material Supplementary material is available at European Heart Journal – Quality of Care and Clinical Outcomes online. Funding British Heart Foundation (SP/12/10/29922 and PG/15/51/31596); NHS Scotland Health Informatics Challenge Grant (HICG/1/40) from the Chief Scientists Office; Intermediate Clinical Fellowship by the Wellcome Trust (201492/Z/16/Z to D.M.); Butler Senior Clinical Research Fellowship from the British Heart Foundation (FS/16/14/32023 to N.L.M.); Research Fellowship from Chest Heart and Stroke Scotland (RES/Fell/A163 to A.A.). Abbott Laboratories provided the troponin I assay reagents, calibrators, and controls without charge.

Trust (201492/Z/16/Z to D.M.); Butler Senior Clinical Research Fellowship from the British Heart Foundation (FS/16/14/32023 to N.L.M.); Research Fellowship from Chest Heart and Stroke Scotland (RES/Fell/A163 to A.A.). Abbott Laboratories provided the troponin I assay reagents, calibrators, and controls without charge. Conflict of interest: A.S.V.S. has acted as a consultant for Abbott Laboratories. N.L.M. has acted as a consultant for Abbott Laboratories, Beckman-Coulter, Roche, and Singulex. A.A. and A.R.C. report personal fees from Abbott Diagnostics, outside the submitted work. The other authors declare no competing interests. Supplementary Material Supplementary Data Click here for additional data file.

Introduction Health records from different health systems might provide insights into the care of patients with chronic diseases and the long-term outcomes of these conditions,1,2 but there have been few comparisons across countries. National hospital data are collected and coded in health systems in many countries and such data (compared with voluntary registries or consented studies) may provide samples that are larger, more nationally representative, and not limited to the study of any one disease, or any one stage of its development.3 However, there are important concerns about the quality and validity of such data. In coronary disease, most studies of outcomes following myocardial infarction (MI) have focused on the acute phase post-MI, typically up to 1 year. However, given marked improvements over the past decade in short-term and long-term mortality following MI,4–6 there is a growing need to characterize the outcomes experienced by patients in whom follow-up begins after the acute phase. By the time of the first anniversary following admission for an acute MI, dual antiplatelet therapy,7–10 cardiac rehabilitation, and cardiologist follow-up11 have commonly ended, and uptake of secondary prevention medication may be declining.12 Recent clinical guidelines7–10 do not directly address the care of patients in this chronic phase of disease, whereas a recent trial found that prolonged dual antiplatelet therapy beyond the first year after an acute MI lowers the risk of cardiovascular death, MI, and stroke.13

prevention medication may be declining.12 Recent clinical guidelines7–10 do not directly address the care of patients in this chronic phase of disease, whereas a recent trial found that prolonged dual antiplatelet therapy beyond the first year after an acute MI lowers the risk of cardiovascular death, MI, and stroke.13 To deliver better long-term care for patients surviving MI, two central questions need addressing. First, what is the risk of major clinical outcomes following the high-risk acute post-MI phase? Nearly all previous studies14,15 of MI outcomes start in the acute hospital setting rather than in the community, and it is well known that early events predominate in estimates of long-term risk. Most of the information on long-term outcomes available so far is derived from trials and voluntary registries, whose risks may not extrapolate to the wider patient population.16 Secondly, how do long-term clinical outcomes vary in different health systems? While international comparisons of cancer outcomes17 have influenced policy and quality-improvement initiatives, in coronary disease comparisons have been limited to the acute hospital care setting.5,18,19

to the wider patient population.16 Secondly, how do long-term clinical outcomes vary in different health systems? While international comparisons of cancer outcomes17 have influenced policy and quality-improvement initiatives, in coronary disease comparisons have been limited to the acute hospital care setting.5,18,19 To answer these questions, we sought national, unselected, ongoing sources of data provided by the health systems in four countries. While these data sources have been used for acute MI outcomes research within countries,20 their use in evaluations of the chronic phase of disease has been much less common, and the present study is the first to use such data to compare outcomes between the USA and European countries (Sweden, England, and France). Our objective was to test the validity of using such hospital record data to estimate and compare across countries the risk of three prognostic outcomes among MI survivors: all-cause death; composite of MI, stroke, or all-cause death; and hospitalized bleeding.

USA and European countries (Sweden, England, and France). Our objective was to test the validity of using such hospital record data to estimate and compare across countries the risk of three prognostic outcomes among MI survivors: all-cause death; composite of MI, stroke, or all-cause death; and hospitalized bleeding. Methods Health record data sources and study population We analysed anonymized patient data from national ongoing hospital sources that use the International Classification of Diseases (ICD) coding system. In Sweden, we used nationwide (100% population coverage) administrative linked data (not directly used for reimbursement) obtained from mandatory Swedish national registries: the National Inpatient Register, the Swedish Prescribed Drug Register, and the Cause of Death Register. In the USA, we used an administrative claims database (Medicare) obtained from the Centers for Medicare & Medicaid Service's standard analytic files that are publicly available; these contain a nationally representative 5% random sample of all Medicare beneficiaries, based on selecting records with 05, 20, 45, 70, or 95 in positions 8 and 9 of the Social Security Number (SSN) (Centers for Medicaid and Medicare, Standard Analytical Files. https://www.cms.gov/research-statistics-data-and-systems/files-for-order/limiteddatasets/standardanalyticalfiles.html, accessed 17 December 2015). Patients are linked across the enrolment and eligibility file and service claims files using a unique encrypted SSN. Deaths are determined by linkage to the National Death file. In England, a single primary care electronic health record (EHR) covers >95% of the population and we used a 4% sample available for research. We used the CALIBER research platform of primary care EHRs (Clinical Practice Research Datalink), linked via the unique identifier of the National Health Service number with other record sources [the Myocardial Ischaemia National Audit Project (MINAP), the Hospital Episodes Statistics database, and the nationwide cause-specific mortality database]. The CALIBER data resource has been shown to be representative of the general population,21–23 and valid for cardiovascular research.24–28 In France, the source data came from the administrative claims insurance database, which covers 95% of the French population. The sample [Echantillon Généraliste des Bénéficiaires (EGB)] available for researchers was built by randomly selecting patients from their national id check number (97 random possibilities).

–28 In France, the source data came from the administrative claims insurance database, which covers 95% of the French population. The sample [Echantillon Généraliste des Bénéficiaires (EGB)] available for researchers was built by randomly selecting patients from their national id check number (97 random possibilities). This permanent 1/97 sample has been shown to be representative in terms of age, sex, social status, and overall medical expenses.29–33 The EGB health insurance claims data are linked to hospital discharge summaries and death registry through the unique healthcare identifier number.

–28 In France, the source data came from the administrative claims insurance database, which covers 95% of the French population. The sample [Echantillon Généraliste des Bénéficiaires (EGB)] available for researchers was built by randomly selecting patients from their national id check number (97 random possibilities). This permanent 1/97 sample has been shown to be representative in terms of age, sex, social status, and overall medical expenses.29–33 The EGB health insurance claims data are linked to hospital discharge summaries and death registry through the unique healthcare identifier number. Our study population was defined by the presence of three characteristics. First, we identified an index acute MI as the patient being admitted to hospital with a primary diagnosis of MI [ICD, Tenth Revision (ICD-10): I21 (Sweden, England, France), I22 (England, France); ICD, Ninth Revision, Clinical Modification (ICD-9-CM): 410.x (excluding 410.x2) (USA)] between 2005 and 2009 (England), 2005 and 2010 (France), 2006 and 2011 (Sweden), and 2002 and 2009 (USA). Where data permitted (England and the USA), the index MI was classified as ST-elevation MI (STEMI) or non-STEMI (NSTEMI) based on MI registry diagnoses (England) or by ICD-9-CM codes (STEMI, 410.0–410.6, 410.8; NSTEMI, 410.7) (USA).34 Patients had to have continuous registration in the respective data sets for at least 12 months before the index MI (the first MI admission during the study period). Second, we identified those patients who at 12 months after their index acute MI were alive, with no further MI. We defined the study entry date as 12 months after the date of admission for the index MI. Third, we restricted the population to patients aged 65 years and older at study entry with no upper age bound, because Medicare predominantly covers this age group (the USA has no national unselected sources of data in younger patients).

d the study entry date as 12 months after the date of admission for the index MI. Third, we restricted the population to patients aged 65 years and older at study entry with no upper age bound, because Medicare predominantly covers this age group (the USA has no national unselected sources of data in younger patients). The study was approved by the Independent Scientific Advisory Committee of the Medicines and Healthcare products Regulatory Agency (protocol number 13_163) in England, regional ethics committee in Linköping, Sweden (reference number 2013/294-31), and Centers for Medicare & Medicaid Services Data Use Agreement in the USA. No ethical approval is required in France for the use of anonymized data. Baseline risk factors and co-morbidities We included demographics (age, sex) and cardiovascular and non-cardiovascular co-morbidities (ICD-9 and ICD-10 codes in Supplementary material online, Table S1) appearing as primary or secondary diagnoses in hospital admissions before the study entry date. We considered patients as currently receiving a medication (codes in Supplementary material online, Table S2) if their last active prescription or dispensation ended <60 days before study entry. No prescription data were available in the Medicare data. We included percutaneous coronary intervention (PCI) and coronary artery bypass graft (CABG) procedures performed on the day of the index MI up to the following 12 months.

e S2) if their last active prescription or dispensation ended <60 days before study entry. No prescription data were available in the Medicare data. We included percutaneous coronary intervention (PCI) and coronary artery bypass graft (CABG) procedures performed on the day of the index MI up to the following 12 months. Endpoints We studied three outcomes of interest: all-cause death; a composite of death, hospital admission for MI, or hospital admission for stroke; and hospitalized bleeding. The ICD-9/ICD-10 codes used to define these outcomes are shown in Supplementary material online, Table S3. Stroke types included ischaemic, haemorrhagic, and unclassified. Hospitalized bleeding was defined as hospital admission with a bleeding cause as a primary diagnosis. Patients were censored at the earliest of experiencing the event of interest (with censoring specific to that event type), deregistration from the primary care practice (England), or end of study period.

unclassified. Hospitalized bleeding was defined as hospital admission with a bleeding cause as a primary diagnosis. Patients were censored at the earliest of experiencing the event of interest (with censoring specific to that event type), deregistration from the primary care practice (England), or end of study period. Statistics Data from each of the four countries were analysed independently following a common protocol. We estimated the direct age- and sex-standardized prevalence of co-morbidities in each country using as reference the 2012 World Health Organization world population truncated to ages 65 years and older. For each country and endpoint, we estimated observed (Kaplan–Meier) and predicted risks, adjusted to the average characteristics of the Swedish patients (aged 78 years, with covariate values shown in Supplementary material online, Table S4). We chose Sweden as the reference population because it had the largest sample size. Predicted risks were based on incrementally adjusted Cox models (fitted separately per country): Model 1 included age, sex, and year of index MI; Model 2 included Model 1 covariates plus co-morbidities [history of more than one MI, diabetes, renal disease, heart failure, peripheral arterial disease (PAD), atrial fibrillation, stroke, hospitalized bleeding, chronic obstructive pulmonary disease, and cancer]; Model 3 included Model 2 covariates plus revascularization procedures (CABG or PCI) received in the 12 months following the index MI. Annual risks were estimated as the average annual risks over the first 3 years.

), atrial fibrillation, stroke, hospitalized bleeding, chronic obstructive pulmonary disease, and cancer]; Model 3 included Model 2 covariates plus revascularization procedures (CABG or PCI) received in the 12 months following the index MI. Annual risks were estimated as the average annual risks over the first 3 years. We estimated the relative risks (RRs) for each endpoint in each country and the 95% confidence intervals (CIs) for 3 years of follow-up using as reference the corresponding risks estimated for Sweden. For a time point t the RR for country A vs. country B is RR_t=(risk(t)_A)/(risk(t)_B). The overall RR reported is the mean of RR_t{t=0,0.5,…3years}. We verified the proportional hazards assumption of the Cox model within countries by plotting the Schoenfeld residuals and confirmed that RRs did not change with time by plotting time-specific RRs estimated for every half year between 0 and 3 years of follow-up (Supplementary material online, Figure S5). We compared the associations of age, sex, co-morbidities, and revascularization treatments with the outcomes across the different countries based on the adjusted hazard ratios (HRs) in Model 3. The overall mean HR for a risk factor was estimated by combining country-specific HRs via random-effects meta-analysis. For France, risk of hospitalized bleeding was adjusted only for Model 1, owing to the small number of events (n = 23). Analyses were performed in R version 15 and SAS version 9.3.

os (HRs) in Model 3. The overall mean HR for a risk factor was estimated by combining country-specific HRs via random-effects meta-analysis. For France, risk of hospitalized bleeding was adjusted only for Model 1, owing to the small number of events (n = 23). Analyses were performed in R version 15 and SAS version 9.3. Results Patients Of the 220 738 patients hospitalized for MI during the study period, 114 364 (54 841 in Sweden, 53 909 in the USA, 4653 in England, and 961 in France) were eligible for inclusion in the analysis (alive, aged 65 years and older, and without subsequent MI at 12-month follow-up; Supplementary material online, Figure S1). Median follow-up ranged from 1.5 years (England) to 3.2 years (USA), during which a total of 37 626 deaths, 45 072 events of MI/stroke/death, and 4697 bleeding hospitalizations were observed in the four countries. Baseline characteristics Baseline characteristics of the post-MI survivors from each country are shown in Table 1. Mean age ranged from 77.5 years in England to 78.6 years in the USA. After standardization for age and sex, we found that compared with patients from Sweden, England, and France, US patients had a higher prevalence of diabetes, heart failure, PAD, renal disease, and chronic obstructive pulmonary disease, and were more likely to have undergone CABG (Figure 1). Table 1 Baseline characteristics for 114 364 myocardial infarction survivors aged 65 years and older in four countries

Baseline characteristics Baseline characteristics of the post-MI survivors from each country are shown in Table 1. Mean age ranged from 77.5 years in England to 78.6 years in the USA. After standardization for age and sex, we found that compared with patients from Sweden, England, and France, US patients had a higher prevalence of diabetes, heart failure, PAD, renal disease, and chronic obstructive pulmonary disease, and were more likely to have undergone CABG (Figure 1). Table 1 Baseline characteristics for 114 364 myocardial infarction survivors aged 65 years and older in four countries Sweden USA England France Index MI, n 80 327 99 343 6653 1308 MI survivor study population, n (%) 54 841 (68.3) 53 909 (54.3) 4653 (70.0) 961 (73.5) Follow-up, years, median (IQR) 2.4 (1.2–3.8) 3.2 (1.6–5.3) 1.5 (0.7–2.5) 3.0 (1.7–3.0) Demographics Women, n (%) 23 280 (42.4) 26 524 (49.2) 1933 (41.5) 422 (43.9) Mean age, years (SD) 78.0 (8.0) 78.6 (7.5) 77.5 (7.7) 77.6 (7.3) White ethnicity, n (%) Not recorded 48 044 (89.1) 3679 (94.6) Not recorded NSTEMI (index MI), n (%) Not recorded 34 576 (64.1) 2393 (51.4) Not recorded Mean BMI, kg/m2 (SD) 27.5 (4.8)a Not recorded 27.3 (5.0) Not recorded Current smoking, n (%) Not recorded Not recorded 444 (10.3) Not recorded Co-morbidities and medical history, n (%) Diabetesa 13 351 (24.3) 18 907 (35.1) 1087 (23.4) 256 (26.6) >1 MI 8786 (16.0) 6465 (12.0) 651 (14.0) 129 (13.4) Heart failure 18 170 (33.1) 24 283 (45.0) 1245 (26.8) 319 (33.2) Cancer 7892 (14.4) 4508 (8.4) 499 (6.9) 167 (17.4) Atrial fibrillation 13 931 (25.4) 15 215 (28.2) 1152 (24.8) 200 (20.8) Hypertension 34 689 (63.3) 42 981 (79.7) 3246 (69.8) 663 (69.0) Stroke 7156 (13.0) 3695 (6.9) 436 (9.4) 45 (4.7) PAD 2230 (4.1) 5460 (10.1) 353 (7.6) 4 (0.4) COPD 5478 (10.0) 14 859 (27.6) 556 (11.9) 116 (12.1) Renal disease 3343 (6.1) 1809 (3.4) 452 (9.7) 99 (10.3) Dementia 2291 (4.2) 1156 (2.1) 110 (2.4) 49 (5.1) Previous hospitalized bleeding 5528 (10.1) 9159 (17.0) 398 (8.6) 41 (4.3) Medication use,bn (%) Aspirin 44 645 (81.4) Not recorded 3606 (77.5) 723 (75.2) ADP-receptor blocker 12 741 (23.2) Not recorded 2357 (50.7) 597 (62.1) Dual antiplatelet 10 932 (19.9) Not recorded 1832 (39.4) 469 (48.8) Statin 38 144 (69.6) Not recorded 3942 (84.7) 729 (75.9) β-blocker 43 913 (80.1) Not recorded 3078 (66.2) 687 (71.5) ACEIs/ARBs 37 317 (68.0) Not recorded 3594 (77.2) 667 (69.4) Calcium channel blocker 12 032 (21.9) Not recorded 1017 (21.9) 198 (20.6) Warfarin 5081 (9.3) Not recorded 408 (8.8) 107 (11.1) Revascularization (1-year post-index MI), n (%) CABG 6970 (12.7) 9134 (16.9) 474 (10.2) 59 (6.1) PCI 26 656 (48.6) 23 099 (42.9) 1519 (32.6) 562 (58.5) ACEI, angiotensin-converting enzyme inhibitor; ADP, adenosine diphosphate; ARB, angiotensin receptor blocker; BMI, bo

farin 5081 (9.3) Not recorded 408 (8.8) 107 (11.1) Revascularization (1-year post-index MI), n (%) CABG 6970 (12.7) 9134 (16.9) 474 (10.2) 59 (6.1) PCI 26 656 (48.6) 23 099 (42.9) 1519 (32.6) 562 (58.5) ACEI, angiotensin-converting enzyme inhibitor; ADP, adenosine diphosphate; ARB, angiotensin receptor blocker; BMI, bo dy mass index; CABG, coronary artery bypass graft; COPD, chronic obstructive pulmonary disease; IQR, interquartile range; MI, myocardial infarction; NSTEMI, non-ST-segment-elevation myocardial infarction; PAD, peripheral arterial disease; PCI, percutaneous coronary intervention; SD, standard deviation. aBased on medications (UK, France, Sweden) or diagnosis in primary (UK) or secondary care (UK, Sweden, USA). bRecorded prescription/dispensing or most recent prescription ending <60 days before study entry. Figure 1 Age- and sex-standardized prevalence of co-morbidities and secondary prevention treatments in post- myocardial infarction survivors aged 65 years and older. Estimates correspond to the direct age- and sex-standardized prevalence of co-morbidities in each country using as reference the 2012 World Health Organization world population truncated to age 65 years and older. ACEI, angiotensin-converting enzyme inhibitor; ADP, adenosine diphosphate; ARB, angiotensin receptor blocker; CABG, coronary artery bypass graft; COPD, chronic obstructive pulmonary disease; CVD, cardiovascular disease; MI, myocardial infarction; PAD, peripheral arterial disease; PCI, percutaneous coronary intervention.

CEI, angiotensin-converting enzyme inhibitor; ADP, adenosine diphosphate; ARB, angiotensin receptor blocker; CABG, coronary artery bypass graft; COPD, chronic obstructive pulmonary disease; CVD, cardiovascular disease; MI, myocardial infarction; PAD, peripheral arterial disease; PCI, percutaneous coronary intervention. All-cause death There were large differences in the unadjusted (Kaplan–Meier) risk of all-cause death across the four countries (Figure 2). Event rates remained high throughout follow-up, with fairly constant risks per year. The 3-year cumulative risk of death was lowest in England [19.6% (95% CI, 18.0–21.2)] and France [22.1% (19.3–24.9)], higher in Sweden [26.9% (26.5–27.4)], and highest in the USA [30.2% (29.8–30.7)]. These differences were progressively attenuated to not statistically significant (95% CI for the RR vs. Sweden crossing 1) after sequential adjustments for age, sex, year of index MI, co-morbidities, and revascularization treatments, except for the USA where the RR of death compared with Sweden was slightly higher [RR USA vs. Sweden, 1.14 (95% CI, 1.04–1.26)]. Based on the mean covariates in the Swedish sample as per Table 1, the fully adjusted 3-year cumulative risks ranged from 12.8% (England) to 19.5% (USA).

es, and revascularization treatments, except for the USA where the RR of death compared with Sweden was slightly higher [RR USA vs. Sweden, 1.14 (95% CI, 1.04–1.26)]. Based on the mean covariates in the Swedish sample as per Table 1, the fully adjusted 3-year cumulative risks ranged from 12.8% (England) to 19.5% (USA). Figure 2 Risks of all-cause death in post-myocardial infarction survivors aged 65 years and older followed from 1 year after the index myocardial infarction. Observed (Kaplan–Meier) risks (top left), adjusted risks (top right), and relative risks vs. Sweden (bottom) in post-myocardial infarction survivors from Sweden (n = 54 841), USA (n = 53 909), England (n = 4653), and France (n = 961). CABG, coronary artery bypass graft; CI, confidence interval; KM, Kaplan–Meier; PCI, percutaneous coronary intervention; RR, relative risk.

sks (top right), and relative risks vs. Sweden (bottom) in post-myocardial infarction survivors from Sweden (n = 54 841), USA (n = 53 909), England (n = 4653), and France (n = 961). CABG, coronary artery bypass graft; CI, confidence interval; KM, Kaplan–Meier; PCI, percutaneous coronary intervention; RR, relative risk. Myocardial infarction, stroke, and all-cause death There were large differences in the unadjusted (Kaplan–Meier) risk of the composite endpoint MI, stroke, or death across the four countries (Figure 3). Event rates remained high throughout follow-up, with fairly constant risks per year. The lowest risk was observed in France [26.0% (95% CI, 23.0–29.0)] and the highest in the USA [36.2% (95% CI, 35.7–36.6)]; risks were similar in Sweden [34.3% (95% CI, 33.8–34.7)] and England [32.5% (95% CI, 30.6–34.4)]. After adjustments, the risk of MI/stroke/death was similar across all four countries (RRs vs. Sweden were not statistically significant). Based on the mean covariates in the Swedish sample as per Table 1, the fully adjusted 3-year cumulative risk of MI, stroke, or death ranged from 24.4% (France) to 28.9% (England).

]. After adjustments, the risk of MI/stroke/death was similar across all four countries (RRs vs. Sweden were not statistically significant). Based on the mean covariates in the Swedish sample as per Table 1, the fully adjusted 3-year cumulative risk of MI, stroke, or death ranged from 24.4% (France) to 28.9% (England). Figure 3 Risks of the composite of myocardial infarction, stroke, and all-cause death in post-myocardial infarction survivors aged 65 years and older followed from 1 year after the index myocardial infarction. Observed (Kaplan–Meier) risks (top left), adjusted risks (top right), and relative risks vs. Sweden (bottom) in post-myocardial infarction survivors from Sweden (n = 54 841), USA (n = 53 909), England (n = 4653), and France (n = 961). CABG, coronary artery bypass graft; CI, confidence interval; KM, Kaplan–Meier; PCI, percutaneous coronary intervention; RR, relative risk. The proportion of deaths attributed to cardiovascular disease (CVD) was 57.9% (8309/14 341) in Sweden and 46.7% (280/599) in England. English patients had lower observed risks for MI, stroke, or CVD death [23.0% (95% CI, 21.3–24.8)] than Swedish patients [26.1% (95% CI, 25.7–26.5)], but a similar risk after adjustment for age, co-morbidities, and revascularization treatments [RR 0.94 (95% CI, 0.77–1.13)] (Supplementary material online, Figure S2).

d. English patients had lower observed risks for MI, stroke, or CVD death [23.0% (95% CI, 21.3–24.8)] than Swedish patients [26.1% (95% CI, 25.7–26.5)], but a similar risk after adjustment for age, co-morbidities, and revascularization treatments [RR 0.94 (95% CI, 0.77–1.13)] (Supplementary material online, Figure S2). Hospitalized bleeding The observed 3-year cumulative risk of hospitalized bleeding was lowest in France (3.1%) and Sweden (3.2%), higher in England (4.6%), and highest in the USA (5.3%) (Figure 4). The adjusted 3-year risk of hospitalized bleeding ranged from 2.7% (Sweden) to 4.0% (USA and England). Compared with Sweden, the fully adjusted RR of bleeding for French and English patients was close to 1.0 (not statistically significant), but was >50% higher for US patients [RR 1.54 (95% CI, 1.21–1.96)]. Figure 4 Risks of hospitalized bleeding events in post-myocardial infarction survivors aged 65 years and older followed from 1 year after the index myocardial infarction. Observed (Kaplan–Meier) risks (top left), adjusted risks (top right), and relative risks (bottom) for hospitalized bleeding events among post-myocardial infarction survivors from Sweden (n = 54 841), USA (n = 53 909), England (n = 4653), and France (n = 961). CABG, coronary artery bypass graft; CI, confidence interval; KM, Kaplan–Meier; PCI, percutaneous coronary intervention; RR, relative risk.

tive risks (bottom) for hospitalized bleeding events among post-myocardial infarction survivors from Sweden (n = 54 841), USA (n = 53 909), England (n = 4653), and France (n = 961). CABG, coronary artery bypass graft; CI, confidence interval; KM, Kaplan–Meier; PCI, percutaneous coronary intervention; RR, relative risk. Outcome predictors Each of the three outcomes showed consistent and strong (majority of HRs >1.5) age- and sex-adjusted associations across the four countries for 12 baseline variables assessed, including risk factors and cardiovascular and non-cardiovascular co-morbidities. The strongest associations (approximately two-fold increase in risk) with the composite of MI, stroke, or death (Figure 5) or with all-cause death alone (Supplementary material online, Figure S3) were observed for history of renal disease, heart failure, chronic obstructive pulmonary disease, and cancer. For hospitalized bleeding, the strongest associations were observed with history of previous hospitalized bleeding, renal disease, heart disease, PAD, and atrial fibrillation (Supplementary material online, Figure S4).

erved for history of renal disease, heart failure, chronic obstructive pulmonary disease, and cancer. For hospitalized bleeding, the strongest associations were observed with history of previous hospitalized bleeding, renal disease, heart disease, PAD, and atrial fibrillation (Supplementary material online, Figure S4). Figure 5 Age- and sex-adjusted hazard ratios (95% confidence interval) for the association of age, sex, and medical history with the composite of myocardial infarction, stroke, and all-cause death among post-myocardial infarction survivors from Sweden (n = 54 841), USA (n = 53 909), England (n = 4653), and Francea (n = 961). aIncidence of PAD in the French study was <0.5%; hence, it was not possible to obtain estimates of association with outcomes. CI, confidence interval; COPD, chronic obstructive pulmonary disease; HR, hazard ratio; MI, myocardial infarction; PAD, peripheral arterial disease. Discussion In one of the first US–European uses of hospital record data to evaluate long-term fatal and non-fatal clinical outcomes in CVD, we present two findings that suggest that such data have useful validity and are informative in CVD outcomes research.

Figure 5 Age- and sex-adjusted hazard ratios (95% confidence interval) for the association of age, sex, and medical history with the composite of myocardial infarction, stroke, and all-cause death among post-myocardial infarction survivors from Sweden (n = 54 841), USA (n = 53 909), England (n = 4653), and Francea (n = 961). aIncidence of PAD in the French study was <0.5%; hence, it was not possible to obtain estimates of association with outcomes. CI, confidence interval; COPD, chronic obstructive pulmonary disease; HR, hazard ratio; MI, myocardial infarction; PAD, peripheral arterial disease. Discussion In one of the first US–European uses of hospital record data to evaluate long-term fatal and non-fatal clinical outcomes in CVD, we present two findings that suggest that such data have useful validity and are informative in CVD outcomes research. First, there was a consistency across all four countries in the high level of risk of further MI, stroke, or death. This occurred in about a third of the patients aged 65 years and above over the next 3 years. This suggests that the high risk is an international phenomenon, rather than a problem with one healthcare system or resulting from the different natures of the underlying record systems. This high risk was considerably higher than that reported in the few smaller previous studies conducted in selected populations,16 highlighting the value of examining less-selected patient samples.

rather than a problem with one healthcare system or resulting from the different natures of the underlying record systems. This high risk was considerably higher than that reported in the few smaller previous studies conducted in selected populations,16 highlighting the value of examining less-selected patient samples. Second, there was a consistency across all four countries in the magnitudes of association between 12 baseline risk factors and each of the three disease outcomes. These associations were highly consistent with published findings from smaller, consented studies, supporting the validity of our risk adjustment and comparison of outcomes. Thus, as in previous studies in post-MI survivors,35–37 we found strong associations between MI, stroke, or death (with heart failure, stroke, PAD, diabetes, renal disease, and chronic obstructive pulmonary disease) and for hospitalized bleeding (with renal disease, history of hospitalized bleeding, and atrial fibrillation). This provides some evidence of the prognostic validity of the hospital record data coded in different healthcare systems, despite the diversity of data collection systems.

hronic obstructive pulmonary disease) and for hospitalized bleeding (with renal disease, history of hospitalized bleeding, and atrial fibrillation). This provides some evidence of the prognostic validity of the hospital record data coded in different healthcare systems, despite the diversity of data collection systems. Our approach was to use hospital healthcare records that have features of ‘big data’: being characterized by large sample sizes (‘volume’), diverse data sources, collected for different purposes, and using different coding systems (‘variety’) and lack of researcher control over the meaning of the data (‘veracity’). This approach has been widely advocated in understanding and improving the outcomes of disease,1 but seldom applied in international contexts.38 The strengths of this approach (compared with voluntary registries or consented studies) lie in direct health system relevance, less bias (larger samples, unselected population-based samples, long-term follow-up with minimal losses), and potential scalability to a wide range of clinical start points and endpoints.3 Such record data are also more widely accessible to the research community than those from consented studies.

lth system relevance, less bias (larger samples, unselected population-based samples, long-term follow-up with minimal losses), and potential scalability to a wide range of clinical start points and endpoints.3 Such record data are also more widely accessible to the research community than those from consented studies. Our study has important limitations, which are largely inherent in these diverse data sources. First, in only one country (Sweden) were nationwide data accessed; the sample of national data available for research in France was particularly small, but it is, nonetheless, representative of the French population. Second, such health record data will inevitably lack relevant data items. For example, MI subtype (STEMI or NSTEMI) was not recorded across all four countries and could not therefore be included in the model adjustments. However, there is strong evidence that, at 1 year following the index MI, STEMI and NSTEMI shared similar mortality, suggesting that MI subclass is unlikely to have influenced our comparisons.39 Information on younger patients, socioeconomic position, ethnicity, drug use, primary care, and cause-specific death was not simultaneously available in all four countries. It is a challenge to these health systems to improve the coverage, depth, and quality of data as part of efforts to expand international comparisons.

formation on younger patients, socioeconomic position, ethnicity, drug use, primary care, and cause-specific death was not simultaneously available in all four countries. It is a challenge to these health systems to improve the coverage, depth, and quality of data as part of efforts to expand international comparisons. We observed an annual risk of death ranging from 6.5% (England) to 10.0% (USA), more than double those in the general population [ranging from 2.9% (France) to 3.7% (UK and USA) in age group 75–79 years] (Supplementary material online, Table S5). Since 57.9% of deaths are due to CVD (based on Swedish data), our study population is in the high-risk category based on the 2012 American College of Cardiology/American Heart Association guidelines (where high risk is defined as >3% annual risk of cardiovascular death)40 or the 2013 European Society of Cardiology guidelines (where high risk is defined as >3% annual risk of all-cause death).41 However, these guidelines are described in the context of the wider population of patients with stable coronary artery disease (many of whom have no history of MI). Also, most of the information comes from meta-analyses of clinical trial data, in which survival is generally higher owing to enrolment of lower-risk populations and better adherence to therapy.

in the context of the wider population of patients with stable coronary artery disease (many of whom have no history of MI). Also, most of the information comes from meta-analyses of clinical trial data, in which survival is generally higher owing to enrolment of lower-risk populations and better adherence to therapy. Our finding of higher adjusted death rates and hospitalized bleeding rates in the USA than in Sweden could be artefactual but warrants further investigation. The higher death rates are consistent with the lower life expectancy at age 65 years in the USA compared with Europe (Supplementary material online, Table S5).42 It is possible that the case mix of patients differs in ways that were not included in our adjustments (e.g. related to the substantially higher prevalence of obesity in the general US population).42 We did find that US patients had higher age- and sex-standardized prevalences of diabetes, heart failure, PAD, renal disease, and chronic obstructive pulmonary disease—but each of these factors was included in the risk adjustment models. The USA might also have a higher proportion of ethnic minorities, which could confound between-country comparisons. It is also possible that care differs. Studies in the USA indicate that previously uninsured populations may delay seeking care before becoming eligible for Medicare,43,44 and mortality may remain elevated for up to 10 years, compared with those with private insurance.45 In contrast, European Union study populations would have had continuous access to healthcare before the age of 65 years.46 It is possible that in the USA compared with Europe secondary prevention medications including dual antiplatelet therapy (aspirin and clopidogrel) are used more or at higher doses;47 however, evidence of this in unselected populations of MI survivors is lacking. Reported use of other CVD medications in Medicare populations indicates that treatment rates are similar to those observed in the EU study population for β-blockers and calcium channel blockers, but somewhat lower for angiotensin-converting enzyme inhibitors and lipid-lowering therapies.48–53

f MI survivors is lacking. Reported use of other CVD medications in Medicare populations indicates that treatment rates are similar to those observed in the EU study population for β-blockers and calcium channel blockers, but somewhat lower for angiotensin-converting enzyme inhibitors and lipid-lowering therapies.48–53 Our findings have clinical implications. First, our results provide evidence for clinicians and regulators when considering new interventions, and when assessing the generalizability of results from clinical trials.13,43 The recently reported PEGASUS-TIMI-54 trial results in 1-year MI survivors are the first to demonstrate a role for long-term (i.e. beyond 1 year) dual antiplatelet use.13 We applied the trial inclusion and exclusion criteria to our real-world patients (Supplementary material online, Figure S1) and demonstrated that the ‘trial-like’ population represents a large proportion (e.g. 66% in Sweden) of the overall MI survivor population, and identified a population at high risk (Supplementary material online, Figure S6). Second, our findings suggest the value of considering MI in a chronic-disease management framework, e.g. with a 1-year health check after acute MI optimizing behavioural, secondary preventive, and wider health interventions. We found that a substantial proportion of deaths are from non-cardiovascular causes (53% in England and 42% in Sweden), suggesting the importance of a multidisciplinary team approach in primary care. Guidelines need to be developed for this population that recognize the multitude of cardiovascular co-morbidities (atrial fibrillation, heart failure, diabetes, and PAD) and non-cardiovascular co-morbidities (renal disease, chronic obstructive pulmonary disease) that are highly prevalent among long-term survivors of MI.

uidelines need to be developed for this population that recognize the multitude of cardiovascular co-morbidities (atrial fibrillation, heart failure, diabetes, and PAD) and non-cardiovascular co-morbidities (renal disease, chronic obstructive pulmonary disease) that are highly prevalent among long-term survivors of MI. In conclusion, analysing hospital record data in the USA and three European countries reveals a consistently high adjusted risk of death, further MI, and stroke in the chronic phase after MI. Inherently, diverse data produced by different health systems may provide insights that are useful in evaluating and comparing the care of patients with chronic diseases and the long-term outcomes of these conditions. Supplementary material Supplementary material is available at European Heart Journal – Quality of Care and Clinical Outcomes online.

In conclusion, analysing hospital record data in the USA and three European countries reveals a consistently high adjusted risk of death, further MI, and stroke in the chronic phase after MI. Inherently, diverse data produced by different health systems may provide insights that are useful in evaluating and comparing the care of patients with chronic diseases and the long-term outcomes of these conditions. Supplementary material Supplementary material is available at European Heart Journal – Quality of Care and Clinical Outcomes online. Authors' contributions E.R., M.T., P.B., N.D., M.S., C.E., P. Hasvold, S.J., D.J.C., T.J., N.M., M.J., and H.H.: conceived and designed the research. E.R., E.Y., M.S., P. Hasvold, T.J., and M.J.: acquired the data. E.R., M.T., E.Y., P.B., D.S., and P. Hasvold: performed statistical analysis. E.R., M.S., P. Hasvold, E.J., N.M., and H.H.: handled funding and supervision. M.P.-R.: handled supervision. E.R.: drafted the manuscript. F.T.-D.: was involved in the study design, data interpretation, and review of manuscript. E.R., M.T., E.Y., P.B., P. Hunt, S.-C.C., D.S., M.P.-R., A.T., S.C.D., N.D., C.E., P. Hasvold, E.J., S.J., D.J.C., T.J., N.M., M.J., and H.H.: made critical revision of the manuscript for key intellectual content. E.R., M.T., E.Y., P.B., P. Hunt, S.-C.C., D.S., M.P.-R., A.T., S.C.D., N.D., M.S., F.T.-D., C.E., P. Hasvold, E.J., S.J., D.J.C., T.J., N.M., M.J., and H.H.: given final approval of the submitted manuscript.

S.J., D.J.C., T.J., N.M., M.J., and H.H.: made critical revision of the manuscript for key intellectual content. E.R., M.T., E.Y., P.B., P. Hunt, S.-C.C., D.S., M.P.-R., A.T., S.C.D., N.D., M.S., F.T.-D., C.E., P. Hasvold, E.J., S.J., D.J.C., T.J., N.M., M.J., and H.H.: given final approval of the submitted manuscript. Funding This study was funded by AstraZeneca, the Medical Research Council Prognosis Research Strategy (PROGRESS) Partnership (H.H., grant G0902393/99558), and by awards to establish the Farr Institute of Health Informatics Research, London from the Medical Research Council, Arthritis Research UK, British Heart Foundation, Cancer Research UK, Chief Scientist Office, Economic and Social Research Council, Engineering and Physical Sciences Research Council, National Institute for Health Research (UK), National Institute for Social Care and Health Research (UK), and Wellcome Trust (E.R., D.S., M.P.-R., and S.D.). S.-C.C. was supported by the Medical Research Population Health Scientist Fellowship (grant MR/M015084/1). T.J. was supported by the Swedish Heart and Lung Foundation. The views expressed in this paper do not necessarily represent the views of the funding bodies. Funding to pay the Open Access publication charges for this article was provided by the Wellcome Trust.

opulation Health Scientist Fellowship (grant MR/M015084/1). T.J. was supported by the Swedish Heart and Lung Foundation. The views expressed in this paper do not necessarily represent the views of the funding bodies. Funding to pay the Open Access publication charges for this article was provided by the Wellcome Trust. Conflict of interest: E.R.: nothing to disclose. M.T.: personal fees from AstraZeneca AB during the conduct of the study; personal fees from AstraZeneca AB outside the submitted work. E.Y.: other from AstraZeneca Pharmaceuticals during the conduct of the study; other from other pharmaceutical consulting clients outside the submitted work: Amgen, Celgene, Takeda, Janssen, Pfizer, and Piramal. P.B.: grants from AstraZeneca during the conduct of the study. P. Hunt: employee of AstraZeneca; other from AstraZeneca Pharmaceuticals during the conduct of the study; other from other pharmaceutical consulting clients outside the submitted work: Roche, Sanofi, Medtronic, Boehringer Ingelheim, and Pfizer. S.-C.C.: nothing to disclose. D.S.: nothing to disclose. M.P.-R.: grants from AstraZeneca during the conduct of the study. A.T.: nothing to disclose. S.C.D.: nothing to disclose. N.D.: grants, personal fees, and non-financial support from Amgen, AstraZeneca, Eli Lilly, and Sanofi; grants and personal fees from Bayer, Daiichi Sankyo, and MSD; personal fees and non-financial support from Servier; personal fees from GSK, Novartis, Novo-Nordisk, Pfizer, Roche, and Boehringer Ingelheim; during the conduct of the study. M.S.: other from AstraZeneca during the conduct of the study. F.T.-D.: employee of AstraZeneca. C.E.: employee of AstraZeneca. P. Hasvold: other from AstraZeneca during the conduct of the study; other from AstraZeneca outside the submitted work; personal fees from AstraZeneca during the conduct of the study; personal fees from AstraZeneca outside the submitted work. E.J.: employee of AstraZeneca. S.J.: employee of AstraZeneca. D.J.C.: grants and personal fees from AstraZeneca and Eli Lilly; grants from Daiichi Sankyo; outside the submitted work. T.J.: the APOLLO-project is financed by AstraZeneca. N.M.: grants and personal fees from AstraZeneca during the conduct of the study; grants from most pharma companies outside the submitted work. M.J.: personal fees from SanofiAventis and AstraZeneca outside the submitted work. H.H.: grants from AstraZeneca during the conduct of the study.

ect is financed by AstraZeneca. N.M.: grants and personal fees from AstraZeneca during the conduct of the study; grants from most pharma companies outside the submitted work. M.J.: personal fees from SanofiAventis and AstraZeneca outside the submitted work. H.H.: grants from AstraZeneca during the conduct of the study. Supplementary Material Supplementary Tables Click here for additional data file. Acknowledgements Editorial support funded by AstraZeneca was provided by Oxford PharmaGenesis, Oxford, UK.

Background Recent decades have seen substantial reductions in the incidence of acute myocardial infarction (AMI) and its mortality.1,2 Much of the decrease in incidence has been attributable to a decrease in ST-elevation myocardial infarction (STEMI). Rates of non-ST elevation myocardial infarction (NSTEMI) have not decreased and may be increasing perhaps due to population ageing or clinical awareness.3 Patients with NSTEMI tend to be older and have more co-morbidities than patients with STEMI, increasing their risk of death in the longer term.4 Much of the decrease in AMI mortality has been attributed to improvements in care, but it is not clear if this has been optimized for all patient groups.5 Some high-risk groups have received particular attention and in diabetes, for example, ischaemic presentations may be atypical and thresholds for investigation and treatment are set at a lower level compared with non-diabetic patients.6 However, other groups have received less attention and chronic obstructive pulmonary disease (COPD), in particular, has been under-studied in patients with AMI despite it being common, affecting ∼1.5% of the European population, although the true prevalence may be as high as 10% as many patients remain undiagnosed.7 In the developed world, the most important risk factor for COPD is tobacco smoking; other risk factors include increased age, indoor and outdoor pollution, poor nutrition, and low socio-economic status.8

of the European population, although the true prevalence may be as high as 10% as many patients remain undiagnosed.7 In the developed world, the most important risk factor for COPD is tobacco smoking; other risk factors include increased age, indoor and outdoor pollution, poor nutrition, and low socio-economic status.8 Chronic obstructive pulmonary disease is associated with an increased risk of many other diseases, which are thought to be due, in part, to ‘spill over’ of inflammation in the lung to the systemic circulation9 (Figure 1). Cardiovascular disease is perhaps the most common co-morbidity and people with COPD, particularly in younger age groups, are at increased risk of AMI, independent of smoking status.10,11 Inflammation, endothelial dysfunction, and increased arterial stiffness, in addition to shared risk factors, are all thought to contribute to cardiovascular risk in COPD.12 Most people with COPD do not die from respiratory diseases,13 with cardiovascular disease being a major cause, accounting for ∼30% of all deaths.14 Figure 1 Diagram representing how inflammation in chronic obstructive pulmonary disease may ‘spill over’ into the systemic circulation and increase the risk of several diseases including cardiovascular disease. Original image from Barnes.9 This article will review contemporary literature on how COPD affects the presentation, management, and outcomes of AMI and how these may be interrelated.

Figure 1 Diagram representing how inflammation in chronic obstructive pulmonary disease may ‘spill over’ into the systemic circulation and increase the risk of several diseases including cardiovascular disease. Original image from Barnes.9 This article will review contemporary literature on how COPD affects the presentation, management, and outcomes of AMI and how these may be interrelated. Presentation of acute myocardial infarction The prevalence of previously diagnosed COPD among patients presenting to hospital with AMI has been estimated as 10–17%.15–18 The true prevalence, including patients with undiagnosed COPD, may be significantly higher. Several studies have reported that patients with COPD are less likely to present with typical chest pain than patients without COPD, and are more likely to present with breathlessness, atypical chest pain, and palpitations.15–17 They are also more likely than patients without COPD to present with NSTEMI than STEMI and to have lower diagnostic biomarker levels, including troponin and creatine kinase.18–21 In one study, COPD has been associated with late presentation >12 h after onset of symptoms.21

s, atypical chest pain, and palpitations.15–17 They are also more likely than patients without COPD to present with NSTEMI than STEMI and to have lower diagnostic biomarker levels, including troponin and creatine kinase.18–21 In one study, COPD has been associated with late presentation >12 h after onset of symptoms.21 Recognition and management of acute myocardial infarction One possible consequence of the differences in presentation of AMI in COPD is that its recognition is delayed or missed altogether. Around 8% of patients admitted to hospital with an acute exacerbation of COPD meet the Universal Definition for Myocardial Infarction,22 but it is unclear whether this represents type 2 AMI triggered by the exacerbation or type 2 AMI misdiagnosed as the exacerbation. At least 33% of patients admitted with COPD in whom there is evidence of prior AMI have no recorded cardiac diagnosis and the proportion is even higher among women with COPD.23 The erroneous attribution of symptoms to COPD rather than AMI may delay diagnosis and the delivery of reperfusion therapy with adverse consequences for infarct size and prognosis. In an analysis of over 300 000 first AMIs in the UK, Rothnie et al.18 found that COPD patients presenting with STEMI were more likely to have an initial incorrect diagnosis and a longer median time to reperfusion compared with patients without COPD [153 min (IQR, 74–706 min) vs. 109 min (IQR, 50–260 min)]. The difference persisted after adjustment for age, sex, and co-morbidities and was only apparent in those COPD patients in whom diagnosis was delayed.

ave an initial incorrect diagnosis and a longer median time to reperfusion compared with patients without COPD [153 min (IQR, 74–706 min) vs. 109 min (IQR, 50–260 min)]. The difference persisted after adjustment for age, sex, and co-morbidities and was only apparent in those COPD patients in whom diagnosis was delayed. Recent studies conducted in Sweden and the UK have shown that patients with COPD are less likely than patients without COPD to receive primary percutaneous intervention (pPCI) or other reperfusion strategies after a STEMI,16,18 confirming earlier US studies.17,21 A more recent US study, however, found no difference in rates of pPCI between patients with and without COPD, suggesting a change in practice and emphasizing the importance of observational data for identifying inequalities in patient management.20

sion strategies after a STEMI,16,18 confirming earlier US studies.17,21 A more recent US study, however, found no difference in rates of pPCI between patients with and without COPD, suggesting a change in practice and emphasizing the importance of observational data for identifying inequalities in patient management.20 NSTEMI guidelines recommend in-hospital cardiac catheterization within 72 h for patients with a ≥3% predicted risk of death at 6 months.24,25 Percutaneous intervention, as indicated, improves outcomes, patients at highest risk having most to gain from the intervention.26,27 Several studies have shown that patients with COPD who present with NSTEMI are less likely to receive in-hospital angiography compared with patients without COPD, despite being at higher risk.16–18,20,21 A potential explanation for this difference is that COPD patients are older and more likely to be deemed sicker or frailer than non-COPD patients, and not appropriate for more aggressive intervention. However, when comparisons are made after exclusion of patients inappropriate for angiography due, for example, to advanced cancer or dementia, findings of under-treatment are unchanged and rates of angiography remain lower in patients with than without COPD.18

ents, and not appropriate for more aggressive intervention. However, when comparisons are made after exclusion of patients inappropriate for angiography due, for example, to advanced cancer or dementia, findings of under-treatment are unchanged and rates of angiography remain lower in patients with than without COPD.18 Under-treatment of patients with COPD presenting with AMI extends beyond the acute phase. Contemporary guideline recommendations, based on randomized clinical trials, are for secondary prevention treatment with a β-blocker, an ACE inhibitor or angiotensin receptor blocker, a statin, and dual antiplatelet therapy (aspirin indefinitely and P2Y12 receptor antagonist for 1 year) for all patients with AMI unless there are clear contraindications.24,25 It has been the widely held belief that COPD contraindicates treatment with β-blockers because of the potential risk of bronchospasm caused by unopposed activation of α1 adrenergic receptors that result in smooth muscle constriction. However, many studies have shown that cardioselective β-blockers that are primarily active at cardiac β1 receptors, not bronchial β2 receptors, are not associated with a change in FEV1 or an increase in exacerbations of COPD.28 Despite this, β-blockers continue to be underused in patients with COPD who are less likely than patients without COPD to receive a prescription after AMI.16–18,21 The under-treatment of patients with COPD extends to other secondary prevention drugs, all of which, with the exception of P2Y12 receptor antagonists, tend to be prescribed less commonly in patients with COPD, although the differences are less marked compared with β-blockers.16–21 Findings from studies that have investigated differences in treatment between COPD and non-COPD patients after AMI are summarized in Table 1. Interestingly, differences in management between COPD and non-COPD patients are not apparent in all settings and appear to have changed over time. As previously mentioned, differences in rates of pPCI between patients with and without COPD appear to have narrowed over time in the USA,20 where prescription of β-blockers to patients with COPD has increased unlike in Europe.16–18 These differences between countries suggest two things: that differences in treatment between COPD and non-COPD patients do represent under-treatment, and that change is possible.

appear to have narrowed over time in the USA,20 where prescription of β-blockers to patients with COPD has increased unlike in Europe.16–18 These differences between countries suggest two things: that differences in treatment between COPD and non-COPD patients do represent under-treatment, and that change is possible. Table 1 Summary of studies that investigated differences in treatment after myocardial infarction between chronic obstructive pulmonary disease and non-chronic obstructive pulmonary disease patients Study Design and setting Population Differences in management Andell et al. 201416 Cohort study within the Swedish SWEDEHEART registry between 2005 and 2010 Consecutive patients admitted to Swedish coronary care units. COPD diagnosis ascertained through linkage to the Swedish National Patient Registry In-hospital management Percutaneous coronary intervention COPD: 37.7% Non-COPD: 55.7% P < 0.001 Coronary angiography COPD: 72.5% Non-COPD: 55.4% P < 0.001 Discharge medicines ACE inhibitors COPD: 50.6% Non-COPD: 55.5% P < 0.001 Angiotensin receptor blockers COPD: 12.6% Non-COPD: 11.1% P = 0.001 Aspirin COPD: 85.5% Non-COPD: 90.1% P < 0.001 β-Blockers COPD: 77.7% Non-COPD: 86.1% P < 0.001 Statin COPD: 68.4% Non-COPD: 79.2% P < 0.001 P2Y12 inhibitor COPD: 62.5% Non-COPD: 72.2% P < 0.001 Bursi et al. 201021 Cohort study in Olmsted County, MN from 1979 to 2007 3438 local residents in Olmsted County. ICD-10 codes used to ascertain COPD In-hospital management Reperfusion COPD: 41% Non-COPD: 52% P < 0.01 Angiography in-hospital COPD: 51% Non-COPD: 59% P < 0.01

Discharge medicines ACE inhibitors COPD: 50.6% Non-COPD: 55.5% P < 0.001 Angiotensin receptor blockers COPD: 12.6% Non-COPD: 11.1% P = 0.001 Aspirin COPD: 85.5% Non-COPD: 90.1% P < 0.001 β-Blockers COPD: 77.7% Non-COPD: 86.1% P < 0.001 Statin COPD: 68.4% Non-COPD: 79.2% P < 0.001 P2Y12 inhibitor COPD: 62.5% Non-COPD: 72.2% P < 0.001 Bursi et al. 201021 Cohort study in Olmsted County, MN from 1979 to 2007 3438 local residents in Olmsted County. ICD-10 codes used to ascertain COPD In-hospital management Reperfusion COPD: 41% Non-COPD: 52% P < 0.01 Angiography in-hospital COPD: 51% Non-COPD: 59% P < 0.01 Discharge medicines ACE inhibitor COPD: 37% Non-COPD: 29% P < 0.01 β-Blocker COPD: 47% Non-COPD: 61% P < 0.01 Diuretic COPD: 34% Non-COPD: 23% P < 0.01 Statin COPD: 29% Non-COPD: 30% P = 0.61 Enriquez et al. 201320 Cross-sectional study of National Cardiovascular Data Registry in the USA between January 2008 and December 2010 158 890 patients with an acute MI. Chronic obstructive pulmonary disease was ascertained from history of COPD or were using long-term inhaled or oral β-agonists, inhaled anti-inflammatory agents, leukotriene receptor antagonists, or inhaled steroids STEMIs In-hospital management Primary percutaneous coronary intervention COPD: 83.1% Non-COPD: 85.4% P < 0.001 Overall reperfusion COPD:92.8% Non-COPD: 94.3% P < 0.001

tory of COPD or were using long-term inhaled or oral β-agonists, inhaled anti-inflammatory agents, leukotriene receptor antagonists, or inhaled steroids STEMIs In-hospital management Primary percutaneous coronary intervention COPD: 83.1% Non-COPD: 85.4% P < 0.001 Overall reperfusion COPD:92.8% Non-COPD: 94.3% P < 0.001 Discharge medicines Aspirin COPD: 97.8% Non-COPD: 98.7% P < 0.001 β-Blocker COPD: 89.4% Non-COPD: 93.1% P < 0.001 ACE inhibitor or angiotensin receptor blocker COPD: 78.0% Non-COPD: 78.4% P = ‘not statistically significant’ Statin COPD: 92.9% Non-COPD: 94.7% P < 0.001 P2Y12 inhibitor COPD: 79.6% Non-COPD: 86.6% P < 0.001 NSTEMIs In-hospital management Cardiac catheterization COPD: 69.9% Non-COPD:81.2% P < 0.001 Percutaneous coronary intervention within 48 hours COPD: 37.2% Non-COPD 48.9% P < 0.001 Discharge medicines Aspirin COPD: 95.9% Non-COPD: 97.3 P < 0.001 β-Blocker COPD: 85.5% Non-COPD: 90.5% P < 0.001 ACE inhibitor or angiotensin receptor blocker COPD: 69.6% Non-COPD: 69.6% P = ‘not statistically significant’ Statin COPD: 85.9% Non-COPD: 89.5% P < 0.001 P2Y12 inhibitor COPD: 65.5% Non-COPD: 71.6% P < 0.001 Rothnie et al. 201518 All UK patients admitted to hospital in the MINAP registry between 2003 and 2013 300 161 patients with a first MI STEMI In-hospital management Primary PCI OR 0.87 (95% CI 0.83–0.92)a Discharge medicines Aspirin OR 0.90 (95% CI 0.85–0.94)a β-Blocker OR 0.26 (95% CI 0.25–0.27)a ACE inhibitor or angiotensin receptor blocker OR 0.89 (95% CI 0.85–0.93)a Statin OR 0.91 (95% CI 0.86–0.95)a P2Y12 inhibitor OR 0.98 (95% CI 0.94–1.03)a

Discharge medicines Aspirin COPD: 95.9% Non-COPD: 97.3 P < 0.001 β-Blocker COPD: 85.5% Non-COPD: 90.5% P < 0.001 ACE inhibitor or angiotensin receptor blocker COPD: 69.6% Non-COPD: 69.6% P = ‘not statistically significant’ Statin COPD: 85.9% Non-COPD: 89.5% P < 0.001 P2Y12 inhibitor COPD: 65.5% Non-COPD: 71.6% P < 0.001 Rothnie et al. 201518 All UK patients admitted to hospital in the MINAP registry between 2003 and 2013 300 161 patients with a first MI STEMI In-hospital management Primary PCI OR 0.87 (95% CI 0.83–0.92)a Discharge medicines Aspirin OR 0.90 (95% CI 0.85–0.94)a β-Blocker OR 0.26 (95% CI 0.25–0.27)a ACE inhibitor or angiotensin receptor blocker OR 0.89 (95% CI 0.85–0.93)a Statin OR 0.91 (95% CI 0.86–0.95)a P2Y12 inhibitor OR 0.98 (95% CI 0.94–1.03)a NSTEMI In-hospital management Angiography in-hospital OR 0.69 (95% CI 0.66–0.71)a Discharge medicines Aspirin OR 0.91 (95% CI 0.88–0.94)a β-Blocker OR 0.25 (95% CI 0.24–0.25)a ACE inhibitor or angiotensin receptor blocker OR 0.94 (95% CI 0.91–0.97)a Statin OR 0.93 (95% CI 0.90–0.96)a P2Y12 inhibitor OR 0.97 (95% CI 0.94–1.01)a Salisbury et al. 200719 Cohort study in 19 centres in the USA between 2003 and 2004 2481 MI patients in PREMIER study restricted to patients discharged alive after MI In-hospital management Cardiac catheterization COPD: 45.7% Non-COPD: 41.2% P = 0.094 Percutaneous coronary intervention COPD: 50.9% Non-COPD: 62.9% P < 0.001

NSTEMI In-hospital management Angiography in-hospital OR 0.69 (95% CI 0.66–0.71)a Discharge medicines Aspirin OR 0.91 (95% CI 0.88–0.94)a β-Blocker OR 0.25 (95% CI 0.24–0.25)a ACE inhibitor or angiotensin receptor blocker OR 0.94 (95% CI 0.91–0.97)a Statin OR 0.93 (95% CI 0.90–0.96)a P2Y12 inhibitor OR 0.97 (95% CI 0.94–1.01)a Salisbury et al. 200719 Cohort study in 19 centres in the USA between 2003 and 2004 2481 MI patients in PREMIER study restricted to patients discharged alive after MI In-hospital management Cardiac catheterization COPD: 45.7% Non-COPD: 41.2% P = 0.094 Percutaneous coronary intervention COPD: 50.9% Non-COPD: 62.9% P < 0.001 Discharge medicines Aspirin COPD: 87.8% Non-COPD: 94.5% P < 0.001 β-Blocker COPD: 86.2% Non-COPD: 92.6% P < 0.001 Stefan et al. 201217 Cohort study of patients hospitalized with acute MI at greater Worcester, MA between 1997 and 2007 6290 patients hospitalized with acute MI in greater Worcester, MA medical centres In-hospital management Cardiac catheterization OR 0.56 (95% CI 0.48–0.65)b Percutaneous coronary intervention OR 0.64 (95% CI 0.54–0.77)b Discharge medicines β-Blocker OR 0.44 (95% CI 0.35–0.50)b Anticoagulant OR 0.81 (95% CI 0.69–0.95)b Statin OR 0.70 (95% CI 0.60–0.82)b Calcium channel blocker OR 1.31 (95% CI 1.13–1.52)b aAll ORs compared COPD with non-COPD patients and are adjusted for age, sex, smoking status, and co-morbidities. bORs compare COPD with non-COPD patients and are adjusted for age, sex, year, cardiovascular disease history, renal failure, length of stay, and type of MI (STEMI or NSTEMI).

Discharge medicines β-Blocker OR 0.44 (95% CI 0.35–0.50)b Anticoagulant OR 0.81 (95% CI 0.69–0.95)b Statin OR 0.70 (95% CI 0.60–0.82)b Calcium channel blocker OR 1.31 (95% CI 1.13–1.52)b aAll ORs compared COPD with non-COPD patients and are adjusted for age, sex, smoking status, and co-morbidities. bORs compare COPD with non-COPD patients and are adjusted for age, sex, year, cardiovascular disease history, renal failure, length of stay, and type of MI (STEMI or NSTEMI). Outcomes after myocardial infarction in people with chronic obstructive pulmonary disease All-cause mortality Studies in a variety of settings have demonstrated an increased risk of death during follow-up after AMI for patients with COPD, but whether this applies to in-hospital mortality is less certain, some studies reporting increased mortality16,17,20,21,29–31 and others finding no difference15,32 compared with patients without COPD. The evidence has now been appraised in a systematic review and meta-analysis,11 which concluded that after pooling maximally adjusted estimates from several studies, there is only weak evidence for a difference in in-hospital mortality for patients with COPD (OR 1.13, 95% CI 0.97–1.31) but strong evidence for an increased risk of death during follow-up (HR 1.26, 95% CI 1.13–1.40) (Figure 2). However, effects were heterogeneous between studies perhaps because of the international differences in treatment of AMI between patients with and without COPD. If some of the increased risk of death associated with COPD is due to these treatment differences, this is likely to have contributed to the heterogeneous outcomes identified in the systematic review.

en studies perhaps because of the international differences in treatment of AMI between patients with and without COPD. If some of the increased risk of death associated with COPD is due to these treatment differences, this is likely to have contributed to the heterogeneous outcomes identified in the systematic review. Figure 2 Long-term risk of death following MI comparing chronic obstructive pulmonary disease with non-chronic obstructive pulmonary disease patients. Original image from Rothnie et al.11

en studies perhaps because of the international differences in treatment of AMI between patients with and without COPD. If some of the increased risk of death associated with COPD is due to these treatment differences, this is likely to have contributed to the heterogeneous outcomes identified in the systematic review. Figure 2 Long-term risk of death following MI comparing chronic obstructive pulmonary disease with non-chronic obstructive pulmonary disease patients. Original image from Rothnie et al.11 The effect of COPD on risk of death following AMI is modified by mode of presentation, a recent UK study reporting that the adjusted odds of in-hospital and 6-month mortality were higher for NSTEMI [(OR 1.40, 95% CI 1.30–1.52) and (OR 1.63, 95% CI 1.56–1.70)] compared with STEMI [(OR 1.27, 95% CI 1.16–1.39) and (OR 1.43, 95% CI 1.29–1.58)].18 Similar findings have been reported in a US study.20 The effect of COPD on risk after AMI appears to be greater in younger than in older patients (Figure 3), suggesting that the ‘excess’ risk of death, attributable to COPD, is clustered in younger patients.18 Dziewierz et al.30 made a similar observation, reporting that COPD was associated with an increased mortality risk after AMI only in patients aged <75. The increased AMI mortality estimates in studies that have compared patients with and without COPD are likely to be underestimates based on the atypical presentations that characterize these patients, a proportion of whom, no doubt, escape diagnosis altogether. Further contribution to the underestimation of risk in these patients is the absence of data on pre-hospital mortality, all existing studies being confined to patients admitted to hospital.

based on the atypical presentations that characterize these patients, a proportion of whom, no doubt, escape diagnosis altogether. Further contribution to the underestimation of risk in these patients is the absence of data on pre-hospital mortality, all existing studies being confined to patients admitted to hospital. Figure 3 Effect of chronic obstructive pulmonary disease on risk of death 6 months after myocardial infarction split by age group. Adapted from data presented in Rothnie et al.18

based on the atypical presentations that characterize these patients, a proportion of whom, no doubt, escape diagnosis altogether. Further contribution to the underestimation of risk in these patients is the absence of data on pre-hospital mortality, all existing studies being confined to patients admitted to hospital. Figure 3 Effect of chronic obstructive pulmonary disease on risk of death 6 months after myocardial infarction split by age group. Adapted from data presented in Rothnie et al.18 Other outcomes Outcome analyses after AMI in patients with COPD show that the risk of other endpoints, apart from mortality, may also be increased. This applies particularly to heart failure both in the acute phase and after discharge from hospital. Thus, Stefan et al.17 found that after adjusting for confounders, people with COPD were more likely to experience acute heart failure (OR 1.59, 95% CI 1.37–1.83), compared with patients without COPD, but not atrial fibrillation, cardiogenic shock, or stroke. Similar findings have been reported in unadjusted analyses.15,20 Studies of longer-term complications of AMI in patients with COPD confirm that the increased risk of heart failure compared with patients without COPD extends to the chronic phase after discharge from hospital, Andell et al. reporting a hazard ratio of 1.35 (95% CI 1.24–1.47) during the first year.16 Findings were similar in another study that included patients with heart failure or left ventricular systolic dysfunction, and reported a hazard ratio for admission with heart failure of 1.19 (95% CI 1.05–1.34) among patients with COPD during the first 3 years after AMI.33 In the same study, the hazard of sudden death was also higher in patients with COPD (HR 1.26, 95% CI 1.03–1.53), although whether this applies in less selected populations is unclear. Certainly, there is no convincing evidence that patients with COPD are at higher risk of recurrent AMI, stroke, angina, or major bleeds compared with non-COPD patients.16,19,33

death was also higher in patients with COPD (HR 1.26, 95% CI 1.03–1.53), although whether this applies in less selected populations is unclear. Certainly, there is no convincing evidence that patients with COPD are at higher risk of recurrent AMI, stroke, angina, or major bleeds compared with non-COPD patients.16,19,33 Are differences in recognition and management associated with differences in outcomes? A key question that arises from the tendency of patients with COPD to present atypically and receive under-treatment of AMI is the extent to which this might explain their adverse outcomes, particularly their heightened risk of death and heart failure compared with patients without COPD.

d with differences in outcomes? A key question that arises from the tendency of patients with COPD to present atypically and receive under-treatment of AMI is the extent to which this might explain their adverse outcomes, particularly their heightened risk of death and heart failure compared with patients without COPD. The association of atypical presentation of AMI with adverse outcomes has been previously reported.34,35 Patients who present atypically are less likely to receive guideline-recommended reperfusion therapy or invasive management and are less likely to receive β-blockers, statins, or antiplatelet therapy on discharge from hospital.35 The tendency of patients with COPD to present with atypical symptoms is, therefore, important because delayed diagnosis of AMI and its under-treatment with reperfusion therapy and secondary prevention drugs has now been shown to explain some of the excess mortality for patients with COPD.18 Similar findings have been reported by Andell et al.16 who found that hazard ratios for mortality in COPD patients fell from 1.32 (95% CI 1.24–1.40) to 1.14 (95% CI 1.07–1.21) following adjustment for in-hospital and discharge treatment. These findings point strongly to delayed diagnosis of AMI and its under-treatment as being important mediators of the adverse outcomes for patients with COPD. They also suggest that differences in treatment between countries may be a plausible reason for heterogeneity in the effects of COPD on risk of death.

treatment. These findings point strongly to delayed diagnosis of AMI and its under-treatment as being important mediators of the adverse outcomes for patients with COPD. They also suggest that differences in treatment between countries may be a plausible reason for heterogeneity in the effects of COPD on risk of death. In considering delayed diagnosis of AMI and its under-treatment as causes of excess mortality, potential direct effects of COPD should not be overlooked (Figure 4). Chronic obstructive pulmonary disease severity, defined by degree of airflow obstruction, appears to be a risk factor for AMI,36 but lung function data are unavailable in national AMI registries and it is unclear if it is also a risk factor for outcomes after AMI. Exacerbations of COPD, however, and the associated systemic inflammation, are important drivers of mortality,11,37–39 but whether ‘frequent exacerbators’ are at heightened risk of death after AMI is uncertain. There is greater certainty about the risks associated with smoking that is often responsible for COPD and is also a major risk factor for death and recurrent coronary events after AMI.40 Indeed, quitting smoking after AMI is one of the most effective preventive strategies, but in heavily dependent COPD patients may be hard to achieve. Smoking cessation pharmacotherapy is underused, and although it may be less effective after AMI,41,42 the excess mortality in patients with COPD identifies them as a group that needs targeting.

g after AMI is one of the most effective preventive strategies, but in heavily dependent COPD patients may be hard to achieve. Smoking cessation pharmacotherapy is underused, and although it may be less effective after AMI,41,42 the excess mortality in patients with COPD identifies them as a group that needs targeting. Figure 4 Schematic diagram of the possible mechanisms underlying the relationship between chronic obstructive pulmonary disease and risk of death after acute myocardial infarction.

g after AMI is one of the most effective preventive strategies, but in heavily dependent COPD patients may be hard to achieve. Smoking cessation pharmacotherapy is underused, and although it may be less effective after AMI,41,42 the excess mortality in patients with COPD identifies them as a group that needs targeting. Figure 4 Schematic diagram of the possible mechanisms underlying the relationship between chronic obstructive pulmonary disease and risk of death after acute myocardial infarction. The under-treatment of AMI in patients with COPD is important because it is potentially modifiable, and provides a means of narrowing the mortality gap between patients without COPD. Although under-treatment can be identified across the management spectrum, it is β-blockers that stand out as the drugs that clinicians often avoid for fear of exacerbating airways obstruction, and this despite there being clear evidence that cardio-selective agents are safe for COPD patients with AMI and also effective for secondary prevention. Thus, Quint et al.43 conducted a propensity matched cohort study in COPD patients with AMI and showed that patients started on a β-blocker during hospital admission had significantly better survival than patients not prescribed a β-blocker (HR 0.50, 95% CI 0.36–0.69). Similar results have been reported for a heart failure population with AMI33 in which COPD did not appear to modify the effect of β-blockers on mortality. The continuing reluctance of clinicians to prescribe β-blockers to COPD patients needs addressing because it may drive much of the increased risk of heart failure and death in the months and years following AMI.

failure population with AMI33 in which COPD did not appear to modify the effect of β-blockers on mortality. The continuing reluctance of clinicians to prescribe β-blockers to COPD patients needs addressing because it may drive much of the increased risk of heart failure and death in the months and years following AMI. Other potential contributors to the risk management paradox44 that characterizes AMI patients with COPD include the poor performance of risk algorithms and therapeutic nihilism. Thus, the GRACE score appears to perform less well in patients with COPD, but whether this makes a significant contribution to under-treatment seems unlikely because COPD patients with the same GRACE score as non-COPD patients remain less likely to receive guideline-recommended investigation and treatment.45 Potentially more important is therapeutic nihilism whereby COPD patients are seen as too old and frail to make interventional management and secondary prevention worthwhile, even though cardiovascular disease is a leading cause of death in patients with COPD and many of the excess deaths are in younger patients.

ent.45 Potentially more important is therapeutic nihilism whereby COPD patients are seen as too old and frail to make interventional management and secondary prevention worthwhile, even though cardiovascular disease is a leading cause of death in patients with COPD and many of the excess deaths are in younger patients. Conclusions Chronic obstructive pulmonary disease increases the risk of heart failure and death after AMI, particularly in the months after discharge from hospital. Effects are greater in younger patients and those with NSTEMI. Although direct effects of COPD likely contribute to the increased risk, delays in diagnosis and under-treatment are also important. It is the under-treatment of these patients, particularly with β-blockers, that provides the most modifiable target for reducing mortality. Further research is needed to investigate the extent and impact of missed AMI diagnosis in patients with COPD. Researchers should also focus on investigating how AMI risk scores function in COPD and how they should be used to guide treatment. Funding This work was supported by a Medical Research Council Industry Collaboration Agreement (G0902135). Funding to pay the Open Access publication charges for this article was provided by Imperial College London. Conflict of interest: J.K.Q. reports grants from Medical Research Council and GSK during the conduct of the study; grants from Medical Research Council, BLF, Wellcome Trust, and the Chartered Society of Physiotherapy, and personal fees from AZ and GSK, outside the submitted work.

Introduction When introducing a diagnostic test, the first step is to establish its accuracy in comparison with the gold-standard referent investigation. Historically, this has often been the main prerequisite for its adoption into clinical practice. However, increasingly diagnostic nvestigations are required to demonstrate not only diagnostic accuracy but also the clinical and cost effectiveness of the findings on subsequent patient diagnosis, risk stratification, investigations, treatments, and finally clinical outcomes. It is these serial downstream effects on clinical management and outcomes that ultimately define the clinical utility of a diagnostic test. The rapid technological advances of computed tomography coronary angiography (CTCA) have raised promise that this imaging modality may fulfil the role of a gold-standard non-invasive investigation of chest pain. Recent research has focused on investigating the merit of integrating CTCA into patient management by aiming to determine its effect on treatment and clinical outcomes.

nary angiography (CTCA) have raised promise that this imaging modality may fulfil the role of a gold-standard non-invasive investigation of chest pain. Recent research has focused on investigating the merit of integrating CTCA into patient management by aiming to determine its effect on treatment and clinical outcomes. Initial assessment and management Chest pain is a common and often concerning symptom that frequently precipitates attendance to a primary care physician with onward referral to specialist cardiology services. The aim of referral is to ascertain the cause of symptoms and to identify those patients with angina pectoris secondary to coronary heart disease (CHD). This would in turn lead to risk stratification and the initiation of evidence-based treatments, with the ultimate goal of improving symptoms and reducing the risk of future adverse cardiovascular events.

the cause of symptoms and to identify those patients with angina pectoris secondary to coronary heart disease (CHD). This would in turn lead to risk stratification and the initiation of evidence-based treatments, with the ultimate goal of improving symptoms and reducing the risk of future adverse cardiovascular events. Whilst cardiology clinics are effective at identifying high-risk patients with angina, a significant number of patients can be misdiagnosed. Specifically, one-third of CHD events occur in patients who have been initially diagnosed as having ‘non-cardiac’ chest pain.1 These patients are younger and less likely to have typical symptoms. Furthermore, a report investigating outcomes of patients attending cardiology clinics with new onset chest pain found that, when applying national guideline recommendations, two-thirds of patients were excluded from further cardiac investigation due to the perception of low risk. However, 10% of patients not offered investigation were subsequently diagnosed as having significant CHD.2 This highlights the fact that misclassification can lead to adverse outcomes and reflects the need for a clearer diagnosis in low-risk populations. Indeed, this represents the majority of patients attending cardiology clinics with recent onset chest pain and a test that could reliably exclude CHD in this group of patients may not only provide reassurance but also reduce adverse outcomes.1,2

mes and reflects the need for a clearer diagnosis in low-risk populations. Indeed, this represents the majority of patients attending cardiology clinics with recent onset chest pain and a test that could reliably exclude CHD in this group of patients may not only provide reassurance but also reduce adverse outcomes.1,2 An initial assessment of the patient with chest pain often involves the estimation of cardiovascular risk using traditional risk estimation models, providing the clinician with a guide on which to base the choice of diagnostic pathway. The Diamond and Forrester prediction rule was first published in 1979 and continues to form the basis of current international guidelines.3,4 However, such traditional risk factor models overestimate the probability of CHD in the general population and especially in women.3,5,6 This can lead to the over-investigation of patients or the initiation and maintenance of unnecessary medical treatments. Following meticulous history taking and an estimation of probability, many clinicians will seek the support of a diagnostic test in order to confirm or to exclude the diagnosis of angina pectoris secondary to CHD. This approach is supported by international guidelines.7,8

An initial assessment of the patient with chest pain often involves the estimation of cardiovascular risk using traditional risk estimation models, providing the clinician with a guide on which to base the choice of diagnostic pathway. The Diamond and Forrester prediction rule was first published in 1979 and continues to form the basis of current international guidelines.3,4 However, such traditional risk factor models overestimate the probability of CHD in the general population and especially in women.3,5,6 This can lead to the over-investigation of patients or the initiation and maintenance of unnecessary medical treatments. Following meticulous history taking and an estimation of probability, many clinicians will seek the support of a diagnostic test in order to confirm or to exclude the diagnosis of angina pectoris secondary to CHD. This approach is supported by international guidelines.7,8 Current guidelines There exists an abundance of non-invasive testing strategies that serve to further improve risk stratification and refine the probability of myocardial ischaemia secondary to CHD. At present, there is no widely adopted strategy of a single gold-standard non-invasive investigation. Indeed, the performance of an individual investigation in the clinical setting is closely dependent on the pretest probability of CHD. In selecting a test, a clinician must use this information in order to select the most appropriate investigation to maximize diagnostic certainty and to minimize the risk of false-positive or false-negative results.

individual investigation in the clinical setting is closely dependent on the pretest probability of CHD. In selecting a test, a clinician must use this information in order to select the most appropriate investigation to maximize diagnostic certainty and to minimize the risk of false-positive or false-negative results. Evidence has demonstrated that selective referral for angiography based on the results of non-invasive testing is both safe and cost effective.9,10 The National Institute for Clinical Excellence (United Kingdom) guidelines recommend invasive angiography for diagnostic purposes in patients with a pretest likelihood of CHD of 61–90%.11 The European and American guidelines reserve invasive angiography for those patients with severe symptoms despite medical therapy, left ventricular dysfunction, or suspected high-risk disease.7,8 Whilst guidelines across the UK, Europe, and the USA differ in their recommended diagnostic pathway, a common recommendation is the utilization of a functional testing strategy. However, the guidelines are inconsistent and recommend different approaches (Table 1). Moreover, prevalent practices are at odds with these guidelines. For example, the American College of Cardiology/American Heart Association guidelines primarily recommend exercise electrocardiography, whilst the majority of North American clinicians will undertake nuclear perfusion scans as the non-invasive stress test of choice.12 Conversely, the European Society of Cardiology guidelines suggest a ‘preference’ for stress imaging tests above exercise electrocardiography where expertise and resources are available. These recommendations have been made based on empirical clinical practice and studies assessing comparative diagnostic accuracy and patient risk stratification but not on clinical outcomes.

es suggest a ‘preference’ for stress imaging tests above exercise electrocardiography where expertise and resources are available. These recommendations have been made based on empirical clinical practice and studies assessing comparative diagnostic accuracy and patient risk stratification but not on clinical outcomes. Table 1 Current guideline recommendations

es suggest a ‘preference’ for stress imaging tests above exercise electrocardiography where expertise and resources are available. These recommendations have been made based on empirical clinical practice and studies assessing comparative diagnostic accuracy and patient risk stratification but not on clinical outcomes. Table 1 Current guideline recommendations Guideline Risk prediction model Estimated likelihood Recommendation for further investigation Recommendation for CTCA European Society of Cardiology7 Diamond–Forrester Model (updated and extended to include 70 years and older) <15% Can be managed without further testing Alternative to stress imaging for ruling out CHD in patients in whom good image quality can be expected 15–65% Exercise ECG if feasible. Stress imaging preferable 66–85% Non-invasive functional test >85% OMT and risk stratification National Institute for Clinical Excellence (NICE) United Kingdom11 Diamond–Forrester Model Duke Database <10% Consider other causes 10–29% CT calcium scoring If calcium score 1–400 30–60% Functional Imaging 61–90% Invasive angiography >90% Manage as angina American Heart Association/American College of Cardiology8 Diamond–Forrester Model Coronary Artery Surgery Study Duke Database Recommendation based on ability to exercise, resting ECG, and history of previous revascularization Low to intermediate If resting ECG interpretable and able to exercise—exercise ECG. if unable to exercise—Pharm stress ECHO Incapable of moderate physical activity or have disabling comorbidity Intermediate Exercise ECG. If unable to exercise—Pharm stress MPI/ECHO or Pharm CMR or CCTA May be reasonable for patients who have at least moderate physical functioning/no disabling comorbidity Intermediate to high If able to exercise—MPI or ECHO with exercise or pharm CMR. If unable to exercise—Pharm stress MPI/ECHO or Pharm CMR or CCTA If stress testing contra-indicated or unable to exercise OMT, optimal medical therapy; MPI, myocardial perfusion imaging; CMR, cardiac magnetic resonance.

sabling comorbidity Intermediate to high If able to exercise—MPI or ECHO with exercise or pharm CMR. If unable to exercise—Pharm stress MPI/ECHO or Pharm CMR or CCTA If stress testing contra-indicated or unable to exercise OMT, optimal medical therapy; MPI, myocardial perfusion imaging; CMR, cardiac magnetic resonance. Functional testing Diagnostic accuracy In a registry of over 600 000 patients undergoing angiography, results of non-invasive testing had only a weak correlation with likelihood of obstructive disease, and patients with a positive result of a non-invasive test were only moderately more likely to have obstructive CHD compared with those who did not undergo any testing.13 In this patient population, the most utilized non-invasive test was single photon emission computed tomography (SPECT) myocardial perfusion imaging (performed in 78.1%), whereas CTCA was performed in a minority (2.1%). Younger patients, women, and those with atypical symptoms were more likely to have non-obstructive coronary artery disease.13

on, the most utilized non-invasive test was single photon emission computed tomography (SPECT) myocardial perfusion imaging (performed in 78.1%), whereas CTCA was performed in a minority (2.1%). Younger patients, women, and those with atypical symptoms were more likely to have non-obstructive coronary artery disease.13 A recent meta-analysis assessed the diagnostic accuracy of myocardial perfusion imaging compared with invasive angiography plus fractional flow reserve (FFR) and found that the sensitivity and specificity of myocardial perfusion imaging with SPECT in detecting obstructive disease were 74 and 79%, respectively, whereas stress echocardiography (ECHO) yielded a sensitivity and specificity of 69 and 84%, respectively.14 In this study, stress myocardial perfusion with magnetic resonance imaging, CT, and positron emission tomography (PET) performed better, with substantially higher diagnostic accuracy (Table 2). Indeed, in head-to-head comparisons, magnetic resonance has outperformed SPECT with a negative predictive value of 91% compared with 79%, respectively.18 Table 2 Diagnostic accuracy of functional tests

A recent meta-analysis assessed the diagnostic accuracy of myocardial perfusion imaging compared with invasive angiography plus fractional flow reserve (FFR) and found that the sensitivity and specificity of myocardial perfusion imaging with SPECT in detecting obstructive disease were 74 and 79%, respectively, whereas stress echocardiography (ECHO) yielded a sensitivity and specificity of 69 and 84%, respectively.14 In this study, stress myocardial perfusion with magnetic resonance imaging, CT, and positron emission tomography (PET) performed better, with substantially higher diagnostic accuracy (Table 2). Indeed, in head-to-head comparisons, magnetic resonance has outperformed SPECT with a negative predictive value of 91% compared with 79%, respectively.18 Table 2 Diagnostic accuracy of functional tests First author/year Study design Aims Patients (n) Main findings Mahajan et al., 201015 Meta-analysis To compare diagnostic accuracy of MPI and SE for the diagnosis of left main stem and triple vessel disease 3713 SE had higher pooled sensitivity than MPI (94 vs. 75%, P < 0.001). No difference in pooled specificity for SE and MPI (40 and 48%, P = 0.16) Chinnaiyan et al., 201216 Prospective Non-randomized registry data To assess correlation and compare the diagnostic performance of CTCA and stress testing in patients undergoing ICA 6198 Stress test results did not accurately predict CHD on ICA. Only 59% of patients with abnormal stress tests had >50% stenosis on ICAa Patel et al., 201413 Observational Registry Data To investigate relationship between clinical characteristics, NIT results, and likelihood of CHD 661 063 NIT findings had minimal incremental value beyond clinical factors for predicting obstructive disease (C-index = 0.75 for clinical factors vs. 0.74 for NIT findings) Neglia et al., 201517 Prospective multicentre, non-randomized To compare the diagnostic accuracy of functional imaging and CTCA in detecting significant CHD defined by ICA 475 MPI sensitivity and specificity 74 and 73%, respectively.a Stress ECHO/CMR sensitivity and specificity 49 and 92%, respectively Takx et al., 201514 Meta-analysis Comparison of non-invasive imaging (functional and CTCA) with ICA and FFR in detection of functionally significant CHD 2048 MRI sensitivity and specificity 89 and 87%, respectively. PET 84 and 87% CT 88 and 80% SPECT 74 and 79% ECHO 69 and 84% Greenwood et al., 201218 Prospective cohort study To investigate the diagnostic accuracy of CMR and compare CMR and SPECT 752 CMR sensitivity 87% and specificity of 83% Sensitivity of SPECT 67% and specificity 83% NIT, non-invasive tests; ICA, invasive coronary angiography; FFR, fractional flow reserve.

ECHO 69 and 84% Greenwood et al., 201218 Prospective cohort study To investigate the diagnostic accuracy of CMR and compare CMR and SPECT 752 CMR sensitivity 87% and specificity of 83% Sensitivity of SPECT 67% and specificity 83% NIT, non-invasive tests; ICA, invasive coronary angiography; FFR, fractional flow reserve. aNo information regarding location and degree of positive stress tests. Risk stratification Functional testing is a predictor of clinical outcomes. Patients with a low-risk exercise electrocardiogram (ECG) have an annual cardiovascular mortality of <1%.7 Some evidence has suggested that myocardial perfusion imaging adds incremental prognostic value over standard diagnostic tests including electrocardiography and ECHO.19 Furthermore, a normal perfusion scan is associated with an excellent prognosis, even in the presence of anatomically detected CHD.20 In a meta-analysis of the prognostic value of functional testing, the negative predictive value of myocardial perfusion imaging for myocardial infarction (MI) and cardiac death was 98.8% over 3 years of follow-up, with an annualized event rate of 0.45% for a negative test (Table 3).21 Table 3 Functional testing and risk stratification

Risk stratification Functional testing is a predictor of clinical outcomes. Patients with a low-risk exercise electrocardiogram (ECG) have an annual cardiovascular mortality of <1%.7 Some evidence has suggested that myocardial perfusion imaging adds incremental prognostic value over standard diagnostic tests including electrocardiography and ECHO.19 Furthermore, a normal perfusion scan is associated with an excellent prognosis, even in the presence of anatomically detected CHD.20 In a meta-analysis of the prognostic value of functional testing, the negative predictive value of myocardial perfusion imaging for myocardial infarction (MI) and cardiac death was 98.8% over 3 years of follow-up, with an annualized event rate of 0.45% for a negative test (Table 3).21 Table 3 Functional testing and risk stratification First author, year Study design Aims Patients (n) Main findings Metz et al., 200721 Meta-analysis To determine prognostic value of normal exercise MPI texts and SE 11 029 NPV for MI and cardiac death 98.5% for MPI and 98.4% for SE. Annualized event rates 0.45% (MPI) and 0.54% (SE) Daly et al., 20066 Euro heart Survey Prospective observational cohort study To identify key prognostic features in CHD and construct score to aid risk prediction 3031 Having no stress test associated with increased risk of death or MI (HR 3.78, 95% CI 2.04–7.00). Positive stress test associated with slightly increased risk (HR 1.43, 95% CI 0.76–2.70) Gimelli et al., 200919 Observational cohort study To investigate the prognostic value of MPI with gated SPECT 676 Perfusion abnormalities independent predictor of event free survival (SDS HR 1.15, 95% CI 1.03–1.27) Sicari et al., 200322 Multicentre prospective observational study To investigate the prognostic value of stress ECHO 7333 Patients with negative SE at low risk of death (<1%/year). Positive test associated with increased risk of cardiac mortality (RR 2.2, 95% CI 1.6–3.1) Candell-Riera et al., 201323 Prospective observational study To investigate the incremental prognostic value of MPI SPECT compared with exercise electrocardiography 5672 Adding MPI SPECT to exercise ECG improves prediction of major cardiovascular events but does not improve prediction of death Piccini et al., 201024 Prospective observational study To investigate whether SPECT MPI enables risk stratification for SCD in patients with CHD and LVEF > 35% 4865 The addition of perfusion data associated with increased discrimination for SCD events (C-index 0.728) SCD, sudden cardiac death; SE, stress echocardiography; SDS, summed difference score, indicating the extent of reversible perfusion defects; NPV, negative predictive value.

SCD in patients with CHD and LVEF > 35% 4865 The addition of perfusion data associated with increased discrimination for SCD events (C-index 0.728) SCD, sudden cardiac death; SE, stress echocardiography; SDS, summed difference score, indicating the extent of reversible perfusion defects; NPV, negative predictive value. Whilst evidence has concluded that a negative stress test correlates with a favourable prognosis, the relationship between a positive result and adverse outcome is less clear. Results from the Euro Heart Survey demonstrated that, whilst not having any functional assessment was an indicator of increased risk, a positive result from a non-invasive stress test was not associated with an adverse outcome.6 The weak correlation between a positive stress test and adverse outcome reflects the finding that stress testing strategies are less reliable in accurately diagnosing CHD.16

onal assessment was an indicator of increased risk, a positive result from a non-invasive stress test was not associated with an adverse outcome.6 The weak correlation between a positive stress test and adverse outcome reflects the finding that stress testing strategies are less reliable in accurately diagnosing CHD.16 Selection of patients for invasive coronary angiography Despite guidelines recommending the use of non-invasive tests to identify and risk stratify those patients with a high likelihood of CHD, a large proportion of diagnostic angiograms are normal. In a study of 398 978 patients throughout 663 hospitals in America, only 38% of patients undergoing elective angiography had obstructive CHD and 39% had normal coronary arteries.25 Similarly, in a multicentre international trial throughout European centres, only 42% of 2260 patients undergoing elective angiography had evidence of obstructive CHD.5 This reflects the lack of certainty regarding the diagnosis and the residual concern of missing underlying CHD. There is therefore a major need for an improved diagnostic strategy and improved patient selection for invasive angiography.

of 2260 patients undergoing elective angiography had evidence of obstructive CHD.5 This reflects the lack of certainty regarding the diagnosis and the residual concern of missing underlying CHD. There is therefore a major need for an improved diagnostic strategy and improved patient selection for invasive angiography. CTCA in the investigation of stable chest pain The diagnostic accuracy of CTCA has been demonstrated in large multicentre studies that have compared this imaging modality with invasive angiography.26,27,28 The results have demonstrated that, in the detection of CHD, CTCA has a sensitivity and specificity which is similar to invasive coronary angiography. However, its positive predictive value in detecting severe stenosis is lower, and the degree of stenosis can be overestimated especially in the presence of marked coronary calcification. Following the emergence of evidence highlighting the comparable diagnostic accuracy of CTCA when compared with invasive angiography, research has now focused on the clinical application of CTCA and its role in patient management and prognosis. Two large randomized controlled trials have recently addressed this question.

Following the emergence of evidence highlighting the comparable diagnostic accuracy of CTCA when compared with invasive angiography, research has now focused on the clinical application of CTCA and its role in patient management and prognosis. Two large randomized controlled trials have recently addressed this question. The PROMISE trial The PROspective Multicenter Imaging Study for Evaluation of chest pain (PROMISE) trial was a large multicentre study of 10 003 participants undergoing non-invasive investigation for suspected CHD who were randomized to an anatomical testing strategy with CTCA or a functional testing strategy, which included exercise electrocardiography, stress ECHO, or radionuclide perfusion imaging.12 The primary endpoint was a composite of all-cause mortality, MI, hospitalization for unstable angina, and major complications of cardiovascular procedures. Secondary endpoints included invasive catheterization showing normal coronary arteries. The study population had an intermediate risk of CHD with a mean pretest likelihood of obstructive CHD of 53%. Only 12% of the population had typical angina, whereas 11% had non-anginal chest pain. A proportion of patients (27%) had a primary symptom other than chest pain including breathlessness, fatigue, weakness, or palpitations. The choice of functional testing varied, with two-thirds of patients undergoing radionuclide perfusion imaging and 10% undergoing exercise electrocardiography.12

n-anginal chest pain. A proportion of patients (27%) had a primary symptom other than chest pain including breathlessness, fatigue, weakness, or palpitations. The choice of functional testing varied, with two-thirds of patients undergoing radionuclide perfusion imaging and 10% undergoing exercise electrocardiography.12 The PROMISE trial reported that compared with functional testing, CTCA led to an increase in invasive coronary angiography, although it was less likely to demonstrate normal coronary arteries (4.3 vs. 3.4%; P = 0.02) and more likely to lead to coronary revascularization at 90 days (6.2 vs. 3.2%; P<0.001).12 At 12 months, the risk of death or non-fatal MI was lower in the CTCA group [hazard ratio (HR), 0.66; 95% CI, 0.44–1.00; P = 0.049], although this benefit did not persist throughout study follow-up. Ultimately, the event rate in PROMISE was lower than expected (3%) for the pre-specified analysis, and there was no significant difference in outcomes between the two patient groups.12

e CTCA group [hazard ratio (HR), 0.66; 95% CI, 0.44–1.00; P = 0.049], although this benefit did not persist throughout study follow-up. Ultimately, the event rate in PROMISE was lower than expected (3%) for the pre-specified analysis, and there was no significant difference in outcomes between the two patient groups.12 The SCOT-HEART trial The Scottish COmputed Tomography of the HEART (SCOT-HEART) trial recruited patients with suspected angina pectoris due to CHD from cardiology clinics and randomized participants (1 : 1) to CTCA plus standard care or standard care alone. This served to investigate the complementary role of CTCA in addition to other clinical tools, as opposed to a direct head-to-head comparison with functional testing strategies.29 All participants underwent clinical evaluation including cardiovascular risk assessment. Clinicians were asked to document whether the patient was diagnosed with (i) CHD and (ii) angina pectoris secondary to CHD, as well as their confidence in these diagnoses at both baseline and 6 weeks of follow-up.

ing strategies.29 All participants underwent clinical evaluation including cardiovascular risk assessment. Clinicians were asked to document whether the patient was diagnosed with (i) CHD and (ii) angina pectoris secondary to CHD, as well as their confidence in these diagnoses at both baseline and 6 weeks of follow-up. The SCOT-HEART trial recruited a broad and representative population of patients referred to the cardiology clinic and included 40% of all patients referred and 47% of those eligible for trial participation. Indeed, this trial specifically included patients who had previously been excluded from diagnostic accuracy studies, such as those with high calcium scores, high body mass index, or atrial fibrillation. Importantly, the majority of patients (85%) underwent exercise electrocardiography in the clinic. This was abnormal in 15% of patients and inconclusive in a further 15%.29 At baseline, the attending clinician diagnosed 47% of patients as having CHD and 36% as having angina secondary to CHD. By 6 weeks, the diagnosis changed in 1 of the 4 patients who underwent CTCA compared with only 1% in the standard care group. Specifically, the use of CTCA increased the certainty of the diagnosis of angina secondary to CHD. Interestingly, the overall diagnostic rate of CHD increased, whilst the diagnosis of angina pectoris secondary to CHD appeared to fall with the use of CTCA.

underwent CTCA compared with only 1% in the standard care group. Specifically, the use of CTCA increased the certainty of the diagnosis of angina secondary to CHD. Interestingly, the overall diagnostic rate of CHD increased, whilst the diagnosis of angina pectoris secondary to CHD appeared to fall with the use of CTCA. Changes in diagnosis led to alterations in further investigations, with CTCA leading to the cancellation of 121 functional tests and 29 invasive angiograms. Whilst the use of CTCA was associated with an early rise in referrals for invasive angiography (n = 94), the majority of these patients had obstructive disease and over half were referred for surgical revascularization due to the presence of high-risk disease.29 Consistent with the changing patterns of diagnoses, the use of CTCA was associated with an increase in recommendations for preventive therapy in patients with documented CHD, whilst the use of unnecessary anti-anginal medication fell with the exclusion of obstructive disease. By clarifying and excluding the diagnosis of CHD, unnecessary medications were discontinued, and this may have important implications for patients' health-related quality of life.

erapy in patients with documented CHD, whilst the use of unnecessary anti-anginal medication fell with the exclusion of obstructive disease. By clarifying and excluding the diagnosis of CHD, unnecessary medications were discontinued, and this may have important implications for patients' health-related quality of life. Similar to the PROMISE trial, the overall event rate was low in the SCOT-HEART trial, with a 2% overall absolute event rate during 1.7 years of follow-up. This is reflective of the fact that the majority of patients had either normal coronary arteries or non-obstructive CHD, and only 30% were ultimately diagnosed with angina pectoris due to CHD. However, despite this, there was an apparent reduction in CHD death or non-fatal MI with CTCA [38% relative risk (RR) reduction; P = 0.0527].29 The observed low event rate reflects the generally good prognosis of patients with recent onset stable chest pain and implies a positive effect of current treatment. This also highlights that documenting a clear improvement in prognosis through the effect of a diagnostic test can be challenging. Nevertheless, the results from the SCOT-HEART trial demonstrate that CTCA plays an important role in clarifying the diagnosis of angina pectoris secondary to CHD and leads to important changes in further management, which may ultimately reduce coronary events.

through the effect of a diagnostic test can be challenging. Nevertheless, the results from the SCOT-HEART trial demonstrate that CTCA plays an important role in clarifying the diagnosis of angina pectoris secondary to CHD and leads to important changes in further management, which may ultimately reduce coronary events. CTCA in the investigation of unstable chest pain As well as determining the role of CTCA in the investigation of recent onset stable chest pain, investigators have also sought to determine its value in the Emergency Department setting. The majority of patients presenting to the Emergency Department with acute chest pain do not have an acute coronary syndrome, with pain attributable to non-cardiac causes.30,31 However, because of diagnostic uncertainty, many patients are admitted to hospital for a period of monitoring, serial ECGs and biochemical markers, and specialist review. Chest pain accounts for up to 1 in 4 acute hospital admissions,30 and population-based rates of hospitalization for suspected cardiovascular disease have been increasing over the past decade.32 With such a high negative predictive value, a clear strength of CTCA lies in the reliable exclusion of significant CHD, especially in the low-to-intermediate risk population. This has driven research to investigate the role of early CTCA in the triage and management of low-risk patients with acute chest pain.

decade.32 With such a high negative predictive value, a clear strength of CTCA lies in the reliable exclusion of significant CHD, especially in the low-to-intermediate risk population. This has driven research to investigate the role of early CTCA in the triage and management of low-risk patients with acute chest pain. The CT-STAT trial compared the use of CTCA with radionuclide myocardial perfusion imaging in the triage of low-risk patients with acute chest pain and demonstrated that the use of CTCA resulted in a 54% reduction in time to diagnosis and 38% reduction in total Emergency Department costs of care. These findings also demonstrated that the presence and severity of atherosclerotic plaque on CTCA were predictive of acute coronary syndrome.31 The Rule Out Myocardial Infarction using Computer Assisted Tomography (ROMICAT-II) trial randomly assigned 1000 patients with symptoms suggestive of acute coronary syndrome but negative initial troponin tests and non-ischaemic ECG changes to early CTCA or to standard treatment in the Emergency Department. CTCA led to a reduction in mean length of hospital stay by 7.6 h, and a greater proportion of patients were discharged directly from the Emergency Department (47 vs. 12% P < 0.001). This had no adverse effect, and there were no cases of missed diagnosis of acute coronary syndrome.33

eatment in the Emergency Department. CTCA led to a reduction in mean length of hospital stay by 7.6 h, and a greater proportion of patients were discharged directly from the Emergency Department (47 vs. 12% P < 0.001). This had no adverse effect, and there were no cases of missed diagnosis of acute coronary syndrome.33 Patients with a negative CTCA have an excellent prognosis, with low event rates and a ‘warranty period’ that can extend for a number of years.34,35,36 Therefore, in addition to facilitating safe and time-efficient discharge, its integration into the management of patients in the acute setting may provide the opportunity to reassure both patients and clinicians of the exclusion of CHD. Cost effectiveness Coronary heart disease represents a significant economical burden to the European Union and the rest of the world. In the European Union, CHD is estimated to cost the economy €60 billion each year, with 33% of this sum attributable to direct healthcare costs. Furthermore, CHD represents an important cause of disability, accounting for 8% of all disability adjusted life years.37

urden to the European Union and the rest of the world. In the European Union, CHD is estimated to cost the economy €60 billion each year, with 33% of this sum attributable to direct healthcare costs. Furthermore, CHD represents an important cause of disability, accounting for 8% of all disability adjusted life years.37 Whilst invasive coronary angiography remains the gold standard for the diagnosis of CHD, this is an expensive test associated with a small yet significant risk of major complications.38 Economic assessments have concluded that selective referral for invasive coronary angiography is cost effective.7,8 However, the proportion of normal invasive angiograms in current practice may suggest poor selection of patients for an initial invasive assessment, or reflect the poor diagnostic accuracy of currently selected non-invasive tests. CTCA, especially when used in in patients with a low-to-intermediate pretest likelihood of CHD, has the potential to improve selection of patients for invasive testing or revascularization, avoiding unnecessary risks and costs associated with invasive angiography. Furthermore, its use in the Emergency Department can allow cost- and time-efficient discharge of patients. Nonetheless, when considering the cost effectiveness of CTCA in routine clinical care, and as a gatekeeper to invasive angiography, further evidence is needed to ascertain fully the resulting healthcare costs, including the impact on downstream testing.

partment can allow cost- and time-efficient discharge of patients. Nonetheless, when considering the cost effectiveness of CTCA in routine clinical care, and as a gatekeeper to invasive angiography, further evidence is needed to ascertain fully the resulting healthcare costs, including the impact on downstream testing. Ideal diagnostic pathway From current evidence, no single non-invasive test has achieved the diagnostic accuracy and clinical utility to merit use as a single gold-standard investigation. Whilst evidence has demonstrated that revascularization according to the functional impact of atherosclerosis improves outcomes over assessing the degree of stenosis alone,39 adopting a functional test in isolation can lead to unnecessary invasive angiography as a consequence of poor sensitivity and specificity in the detection of CHD. To overcome this, an appropriate strategy would be to combine anatomical and functional tests in order to refine risk stratification and improve selection of patients for revascularization. However, close consideration needs to be given to both the costs and the risks to patients including radiation burden, when multiple testing is incurred.

propriate strategy would be to combine anatomical and functional tests in order to refine risk stratification and improve selection of patients for revascularization. However, close consideration needs to be given to both the costs and the risks to patients including radiation burden, when multiple testing is incurred. We would suggest that a safe, cost-effective, and accessible method of achieving combined functional and anatomical assessments is the use of serial exercise electrocardiography testing combined with follow-on CTCA as required. This plays to the strengths of both techniques and protects against their inherent weaknesses. For example, the strength of exercise electrocardiography is the functional assessment of the reproducibility and severity of symptoms combined with high specificity for the presence of obstructive CHD. Its main weakness relates to the poor sensitivity (∼50%) for diagnosing CHD. CTCA compensates for this with a very high sensitivity for CHD and high negative predictive value. The weakness of CTCA in overestimating or poorly defining obstructive disease due to calcification can be mitigated by considering the functional assessment afforded by the exercise ECG. Thus, combining a simple functional test with a highly sensitive anatomical test may represent an ideal strategy in the diagnosis and risk stratification of patients with suspected CHD.

ining obstructive disease due to calcification can be mitigated by considering the functional assessment afforded by the exercise ECG. Thus, combining a simple functional test with a highly sensitive anatomical test may represent an ideal strategy in the diagnosis and risk stratification of patients with suspected CHD. Future developments Combining anatomical and functional imaging Until recently, an important limitation of CTCA has been the inability to gain functional information about the impact of potential coronary stenoses. However, the development of non-invasive measurements of fractional flow reserve from CTCA (CT-FFR) raises the promise for the potential to gain both anatomical and functional information from a single non-invasive imaging modality. Evidence from recent trials has highlighted that this technique improves the specificity of CTCA and thereby may reduce the number of false-positive tests and unnecessary invasive angiograms.40,41,42,43 With the advent of dynamic volume scanners, computed tomography perfusion (CTP) is another promising technique that, when used as an adjunct to CTCA, allows determination of both the anatomical and functional significance of CHD. This technique has yielded sensitivities of 83–91% and specificities of 72–98% when compared with other functional imaging modalities.44 Furthermore, it adds incremental diagnostic accuracy to CTCA alone in patients with high calcium scores.45 Whilst this technique still has limitations, it has the potential to evolve as an effective addition to CTCA.

ivities of 83–91% and specificities of 72–98% when compared with other functional imaging modalities.44 Furthermore, it adds incremental diagnostic accuracy to CTCA alone in patients with high calcium scores.45 Whilst this technique still has limitations, it has the potential to evolve as an effective addition to CTCA. CTCA in the detection of high-risk plaque In addition to identifying anatomically and functionally significant CHD, CTCA has the advantage of being able to detect the presence of features associated with the vulnerable plaque including low attenuation, microcalcification, and positive remodelling.46 The presence of high-risk plaque features on CTCA and plaque progression through serial imaging have both been highlighted as independent risk factors for acute coronary syndrome. In a study of over 3000 patients, acute coronary syndrome frequency was 16% in patients with CTCA confirmed high-risk plaque compared with 1.6% of patients with no evidence of high-risk plaque characteristics (Figure 1).46 Figure 1 Computed tomography coronary angiography image of a plaque with high-risk characteristics including low attenuation (red arrow).

CTCA in the detection of high-risk plaque In addition to identifying anatomically and functionally significant CHD, CTCA has the advantage of being able to detect the presence of features associated with the vulnerable plaque including low attenuation, microcalcification, and positive remodelling.46 The presence of high-risk plaque features on CTCA and plaque progression through serial imaging have both been highlighted as independent risk factors for acute coronary syndrome. In a study of over 3000 patients, acute coronary syndrome frequency was 16% in patients with CTCA confirmed high-risk plaque compared with 1.6% of patients with no evidence of high-risk plaque characteristics (Figure 1).46 Figure 1 Computed tomography coronary angiography image of a plaque with high-risk characteristics including low attenuation (red arrow). The majority of acute ischaemic events arise as a consequence of rupture of a non-flow limiting vulnerable plaque.47 Therefore, adopting a functional testing strategy may not identify patients at risk of MI who would benefit from the initiation of treatment. However, despite significant advances in our understanding of the biology of the high-risk plaque,48 the clinical utility of documentation of the vulnerable plaque remains uncertain. Future research should focus on the merit of medical or interventional management in patients with non-obstructive high-risk atherosclerotic plaque morphology (Figure 2). Figure 2 Example of a high-risk proximal left main stem plaque (red arrow) with evidence of positive remodelling and calcification.

The majority of acute ischaemic events arise as a consequence of rupture of a non-flow limiting vulnerable plaque.47 Therefore, adopting a functional testing strategy may not identify patients at risk of MI who would benefit from the initiation of treatment. However, despite significant advances in our understanding of the biology of the high-risk plaque,48 the clinical utility of documentation of the vulnerable plaque remains uncertain. Future research should focus on the merit of medical or interventional management in patients with non-obstructive high-risk atherosclerotic plaque morphology (Figure 2). Figure 2 Example of a high-risk proximal left main stem plaque (red arrow) with evidence of positive remodelling and calcification. Conclusions The diagnostic accuracy of CTCA, combined with evidence of its impact on clinical decision-making and outcomes, makes this a powerful and potentially cost-effective tool when integrated into the management of patients with chest pain. In a population of patients with suspected angina secondary to CHD, its use serves to improve patient selection for invasive angiography and revascularization, as well as excluding CHD in those patients who may otherwise be subjected to unnecessary further investigation or life-long medications. Ultimately, CTCA appears to reduce the risk of fatal and non-fatal MI, something that no previous non-invasive diagnostic strategy has been able to achieve.

phy and revascularization, as well as excluding CHD in those patients who may otherwise be subjected to unnecessary further investigation or life-long medications. Ultimately, CTCA appears to reduce the risk of fatal and non-fatal MI, something that no previous non-invasive diagnostic strategy has been able to achieve. Funding D.E.N. is funded by the British Heart Foundation (CH/09/002) and is supported by a Wellcome Trust Senior Investigator Award (WT103782AIA). He was chief investigator of the SCOT-HEART trial. Conflict of interest: none declared.

Introduction Acute coronary syndrome (ACS) is a leading cause of premature illness and death, and one of the most common reasons for an emergency admission to hospital.1,2 Reducing the public health burden from ACS is a key priority for health care providers and governments. In the United Kingdom (UK) in 2013/2014, there were 491 647 inpatient episodes attributed to ischaemic heart disease (IHD),3 and the number of myocardial infarction (MI) events being recorded in England, Wales, and Northern Ireland has increased (2011/2012—79 433; 2014/2015—83 842).4,5 The Myocardial Ischaemia National Audit Project (MINAP) provides information on national standards of care for patients with heart attack in England, Wales, and Northern Ireland but not in Scotland, where joined-up systems for reporting contemporary secondary care activities and related patient outcomes are lacking. Such information has crucial importance as practice variation in ACS patients has been shown to be associated with mortality,6–8 while comparative research across countries might help to improve health systems and prevent deaths.9,10 The National Health Service (NHS) is the sole provider of secondary care services for hospitalized patients with an ACS in the UK. Given the large number of patient episodes and the complexity of this health care system, the process and outcome of individual patients who have been hospitalized with ACS is difficult to describe and uncertain when based on individual hospital records alone.

e services for hospitalized patients with an ACS in the UK. Given the large number of patient episodes and the complexity of this health care system, the process and outcome of individual patients who have been hospitalized with ACS is difficult to describe and uncertain when based on individual hospital records alone. The project was set up as a Joint Working Project between NHS health boards, including Greater Glasgow and Clyde (NHS GGC) and the Golden Jubilee National Hospital (GJNH) and AstraZeneca UK Ltd. The overall aim of the project was to develop and implement an updatable electronic registry (e-Registry), in a regional network that provides secondary care for patients hospitalized for a known or suspected ACS, with applications for audit, research, and health care improvement. In addition, the project team wished to collaborate and demonstrate that the pharmaceutical industry could work as a trusted partner to the NHS in adding value to health care beyond medicines through shared skills, expertise, and resources to further enhance patient outcomes and service performance.

h care improvement. In addition, the project team wished to collaborate and demonstrate that the pharmaceutical industry could work as a trusted partner to the NHS in adding value to health care beyond medicines through shared skills, expertise, and resources to further enhance patient outcomes and service performance. Methods Setting The e-Registry was established in the NHS in the West of Scotland. During the study period, NHS GGC provided acute secondary care services through seven hospitals serving a population of approximately 1.2 million. The GJNH is a regional cardiothoracic intervention centre that provides invasive cardiology and cardiothoracic services for this population, amongst others, but is administratively distinct from NHS GGC. These hospitals participate in a Managed Clinical Network to provide strategic health care delivery. Semi-electronic patient records were implemented across all secondary care clinical and administration systems in NHS GGC and the GJNH by June 201211,12 enabling capture of key components of ACS care. Governance The project was supported by the National Advisory Committee for Coronary Heart Disease on behalf of the Scottish Government. The Joint Working Project described within this report was approved by hospital management and the Caldicott Guardian for clinical governance in each Health Board.

Semi-electronic patient records were implemented across all secondary care clinical and administration systems in NHS GGC and the GJNH by June 201211,12 enabling capture of key components of ACS care. Governance The project was supported by the National Advisory Committee for Coronary Heart Disease on behalf of the Scottish Government. The Joint Working Project described within this report was approved by hospital management and the Caldicott Guardian for clinical governance in each Health Board. Design and methodology An executable system was developed to identify, extract, and link usual care electronic health records for patients hospitalized with a suspected or known ACS. No new clinical records or manual data entry from paper archives were required. The Community Health Index (CHI) is a 10-digit unique identifier for each person registered with a general practitioner in Scotland and is present on almost all health care encounters. This identifier was used to link the records of patients in the extracted data sets. Hospital patient episodes were used to create episodes of care using (i) the Intersystems TRAKCare Patient Admin System with data extracts based on ICD-1013 diagnosis codes, (ii) Scottish Care Information (SCI) Gateway14 electronic referrals for invasive cardiovascular procedures, (iii) a bespoke hospital-level patient and catheter laboratory record developed in the GJNH (Figure 1). Figure 1 Data sources and linkage.

Hospital patient episodes were used to create episodes of care using (i) the Intersystems TRAKCare Patient Admin System with data extracts based on ICD-1013 diagnosis codes, (ii) Scottish Care Information (SCI) Gateway14 electronic referrals for invasive cardiovascular procedures, (iii) a bespoke hospital-level patient and catheter laboratory record developed in the GJNH (Figure 1). Figure 1 Data sources and linkage. Data were extracted from these core clinical systems used to manage patients with ACS and deposited within an existing repository for electronic health data (an NHS Safe Haven).15 Patients within these data sets were then linked to National Records Scotland (NRS) death certificates where appropriate. The electronic records were pseudonymised within the Safe Haven before being securely transferred to the analysts who were employed by AstraZeneca UK Ltd under a data sharing agreement signed by all parties. Statistical analysis was supported by a co-funded PhD from the University of Glasgow.

ficates where appropriate. The electronic records were pseudonymised within the Safe Haven before being securely transferred to the analysts who were employed by AstraZeneca UK Ltd under a data sharing agreement signed by all parties. Statistical analysis was supported by a co-funded PhD from the University of Glasgow. The ACS diagnoses were based on the discharge summary recorded by the attending clinician(s) in usual care health records captured in the clinical systems. In the local hospital patient administration system (TRAKCare), the diagnoses are subsequently coded per the International Classification of Disease (ICD)-10 and in the invasive centre (IC) the discharge diagnosis is recorded in a standardized text format. An algorithm was developed to assign the most appropriate diagnosis to an episode of care. Where a patient had multiple diagnoses in a linked episode of care from different clinical settings or systems, the diagnosis on discharge from the IC was taken as the most accurate representation. Where a patient did not attend the IC as part of their care pathway, the ICD-10 diagnosis from the local hospital was used. As well as this a hierarchy of diagnoses was used to ensure that the most severe or specific final diagnosis was acknowledged, with ST-elevation myocardial infarction (STEMI) at the top of the hierarchy (see Supplementary material online).

rt of their care pathway, the ICD-10 diagnosis from the local hospital was used. As well as this a hierarchy of diagnoses was used to ensure that the most severe or specific final diagnosis was acknowledged, with ST-elevation myocardial infarction (STEMI) at the top of the hierarchy (see Supplementary material online). Data were extracted from TRAKCare for all admissions (1 October 2013–30 September 2014) with an ICD-10 diagnosis of angina (I200–I209), myocardial infarction (I210–I229), other IHD (I240–I249), or heart failure (I50) to ensure complete capture of suspected ACS events. This was linked to referrals for invasive cardiovascular procedures made through the SCI-Gateway system, cardiac interventions performed in the IC of the GJNH and all-cause mortality data from NRS. This linked data set was then analysed to look at diagnoses and patient characteristics, and to identify episodes of care, which were then categorized into distinct clinical care pathways. Those with a final diagnosis of ACS were isolated for analysis on referrals for invasive cardiovascular procedures, treatment durations, service delivery metrics and outcomes, including mortality.

atient characteristics, and to identify episodes of care, which were then categorized into distinct clinical care pathways. Those with a final diagnosis of ACS were isolated for analysis on referrals for invasive cardiovascular procedures, treatment durations, service delivery metrics and outcomes, including mortality. After the classification of patients and pathways, data on deprivation status of patients was provided by the NHS Safe Haven identified based on the postcode of the patient’s home address and measured using quintiles of the Scottish Index of Multiple Deprivation (SIMD) 2012 measure.16,17 Quintile 1 represents the highest level of deprivation with Q5 representing the least deprived. The top 20% most deprived data zones in Scotland are in the first quintile, with the distribution of Glasgow City’s data zones being 49%, 19%, 13%, 10.5%, 8.5% (Q1–Q5).16 Pre-specified health care outcomes The pre-specified primary outcomes were 30 day and 1 year all-cause mortality. The receipt of cardiac interventions, duration of hospital stay and pathways of care were the pre-specified secondary outcomes.

After the classification of patients and pathways, data on deprivation status of patients was provided by the NHS Safe Haven identified based on the postcode of the patient’s home address and measured using quintiles of the Scottish Index of Multiple Deprivation (SIMD) 2012 measure.16,17 Quintile 1 represents the highest level of deprivation with Q5 representing the least deprived. The top 20% most deprived data zones in Scotland are in the first quintile, with the distribution of Glasgow City’s data zones being 49%, 19%, 13%, 10.5%, 8.5% (Q1–Q5).16 Pre-specified health care outcomes The pre-specified primary outcomes were 30 day and 1 year all-cause mortality. The receipt of cardiac interventions, duration of hospital stay and pathways of care were the pre-specified secondary outcomes. Data validation A quality assurance procedure was conducted by the NHS in Glasgow to assess the robustness of the data extraction process and the accuracy of the outputs from the analysis programme, as compared with source clinical data assessed and verified by independent, trained observers. The review focused on (i) a quality assessment of 200 individual patient episodes, taken from a randomly selected calendar month (30 days) of the e-Registry with all consecutive episodes included, and (ii) an assessment of the causes of death of 44 patients. The observers accessed usual care health records using TRAKCare and other executable systems including Clinical Portal18 and SCI-STORE19 that are used at the point of care in hospital.

Data validation A quality assurance procedure was conducted by the NHS in Glasgow to assess the robustness of the data extraction process and the accuracy of the outputs from the analysis programme, as compared with source clinical data assessed and verified by independent, trained observers. The review focused on (i) a quality assessment of 200 individual patient episodes, taken from a randomly selected calendar month (30 days) of the e-Registry with all consecutive episodes included, and (ii) an assessment of the causes of death of 44 patients. The observers accessed usual care health records using TRAKCare and other executable systems including Clinical Portal18 and SCI-STORE19 that are used at the point of care in hospital. A team of five research nurses with a cardiovascular background employed by the NHS in Glasgow were tasked with assessing source clinical data against the data from the e-registry to confirm or query the patient episodes, including dates and the primary and secondary causes of these events. In addition, two cardiologists (K.M., C.B.) assessed any queries that were raised by the research nurses. The reviewers focused on 4 aspects of the analysis programme; care pathway assignment, diagnosis assignment, assessment of invasive procedures received and assessment of mortality.

A team of five research nurses with a cardiovascular background employed by the NHS in Glasgow were tasked with assessing source clinical data against the data from the e-registry to confirm or query the patient episodes, including dates and the primary and secondary causes of these events. In addition, two cardiologists (K.M., C.B.) assessed any queries that were raised by the research nurses. The reviewers focused on 4 aspects of the analysis programme; care pathway assignment, diagnosis assignment, assessment of invasive procedures received and assessment of mortality. Statistical analyses As this project involved exploratory health services delivery research, a power calculation was not performed. Categorical variables are expressed as number and percentage of patients. Most continuous variables followed a normal distribution and are therefore presented as means together with standard deviation. Those variables that did not follow a normal distribution are presented as medians with interquartile range. Differences in continuous variables between groups were assessed by the t-test or Mann-Whitney test as appropriate. Differences in categorical variables between groups were assessed using the Fisher’s test. Cox proportional hazards regressions were used to evaluate the effect of risk factors and intervention procedures on all-cause mortality for ST-segment elevation myocardial infarction (STEMI) and non-STEMI (NSTEMI) patients. Multivariable logistic or zero-inflated negative binomial models were performed to evaluate differences in service delivery. Survival analyses will be performed using the Kaplan-Meier method. Analyses were conducted using SAS Enterprise Guide (v5.1).

vation myocardial infarction (STEMI) and non-STEMI (NSTEMI) patients. Multivariable logistic or zero-inflated negative binomial models were performed to evaluate differences in service delivery. Survival analyses will be performed using the Kaplan-Meier method. Analyses were conducted using SAS Enterprise Guide (v5.1). Results Demographics and clinical characteristics Between 1 October 2013 and 30 September 2014 (12 months), 2327 unique patients had 2472 distinct episodes of care across 7 acute hospitals and one regional cardiothoracic centre. Of these patients, the final diagnosis was STEMI in 586 (25.2%) episodes, NSTEMI in 1068 (45.9%), unspecified MI in 146 (6.3%), and unstable angina (UA) in 527 (22.6%) for their first hospitalization within the study period. ST-elevation myocardial infarction patients were generally younger with a higher proportion of males than the other diagnoses and relatively more deprived than NSTEMI (77.6% in SIMD Q1–3 vs. 73.6%) (see Table 1). Table 1 Demographic characteristics by final diagnosis

their first hospitalization within the study period. ST-elevation myocardial infarction patients were generally younger with a higher proportion of males than the other diagnoses and relatively more deprived than NSTEMI (77.6% in SIMD Q1–3 vs. 73.6%) (see Table 1). Table 1 Demographic characteristics by final diagnosis All (n = 2327) STEMI (n = 586) NSTEMI (n = 1068) Unspecified MI (n = 146) Unstable angina (n = 527) Age at admission (years) 67.5 (13.9) 62.8 (14.2) 68.4 (13.4) 77.0 (11.6) 68.3 (13.2) Gender Female 989 (42.5%) 182 (31.1%) 458 (42.9%) 70 (47.9%) 279 (52.9%) Male 1338 (57.5%) 404 (68.9%) 610 (57.1%) 76 (52.1%) 248 (47.1%) Ethnicity White 1636 (70.4%) 404 (69.2%) 650 (60.9%) 130 (89.0%) 452 (85.8%) Other 48 (2.1%) 8 (1.4%) 23 (2.2%) 4 (2.7%) 13 (2.5%) Unknown 643 (27.6%) 174 (29.5%) 395 (37.0%) 12 (8.2%) 62 (11.8%) Missing 2 2 0 0 0 SIMD quintile Q1 (most deprived) 906 (43.6%) 214 (41.2%) 407 (43.4%) 63 (44.7%) 222 (46.3%) Q2 372 (17.9%) 95 (18.3%) 167 (17.8%) 24 (17.0%) 86 (18.0%) Q3 293 (14.1%) 94 (18.1%) 116 (12.4%) 25 (17.7%) 58 (12.1%) Q4 245 (11.8%) 50 (9.6%) 113 (12.0%) 15 (10.6%) 67 (14.0%) Q5 (least deprived) 262 (12.6%) 67 (12.9%) 135 (14.4%) 14 (9.9%) 46 (9.6%) Missing 249 66 130 5 48 Pathway Emergency direct to IC 333 (14.3%) 304 (51.9%) 29 (2.7%) 0 (0.0%) 0 (0.0%) Local A&E to IC 155 (6.7%) 148 (25.3%) 7 (0.7%) 0 (0.0%) 0 (0.0%) Acute invasive 492 (21.1%) 57 (9.7%) 426 (39.9%) 2 (1.4%) 7 (1.3%) Elective invasive 208 (8.9%) 5 (0.9%) 198 (18.5%) 0 (0.0%) 5 (0.9%) Local hospital only 1081 (46.5%) 68 (11.6%) 364 (34.1%) 144 (98.6%) 505 (95.8%) Elective direct to IC 58 (2.5%) 4 (0.7%) 44 (4.1%) 0 (0.0%) 10 (1.9%) Data are mean (SD) or number (%).

) Acute invasive 492 (21.1%) 57 (9.7%) 426 (39.9%) 2 (1.4%) 7 (1.3%) Elective invasive 208 (8.9%) 5 (0.9%) 198 (18.5%) 0 (0.0%) 5 (0.9%) Local hospital only 1081 (46.5%) 68 (11.6%) 364 (34.1%) 144 (98.6%) 505 (95.8%) Elective direct to IC 58 (2.5%) 4 (0.7%) 44 (4.1%) 0 (0.0%) 10 (1.9%) Data are mean (SD) or number (%). SIMD, Scottish Index of Multiple Deprivation; IC, Invasive Centre. For first admission in study period. Six distinct treatment pathways were identified (Figure 2). The treatment pathway was mapped for each episode of care giving 53% of patients admitted to the IC; 14% directly as emergency episodes; 3% directly on an elective basis; 7% via local A&E; 21% after an inpatient stay in a local hospital for acute invasive care; and 9% discharged home from the local hospital for elective invasive care at a later date. The other 47% stayed within the local hospital only, either managed conservatively with no following invasive treatment or dying in hospital (Table 1).

A&E; 21% after an inpatient stay in a local hospital for acute invasive care; and 9% discharged home from the local hospital for elective invasive care at a later date. The other 47% stayed within the local hospital only, either managed conservatively with no following invasive treatment or dying in hospital (Table 1). Figure 2 All acute coronary syndrome patients (1st hospitalization in timeframe) by care pathway. Note sizes are proportional to percentage of analysis population. Pathway definitions: 1—Emergency direct admission to invasive centre [Golden Jubilee National Hospital (GJNH)]; 2—Local A&E in regional hospital followed by direct transfer to invasive centre; 3—Admission to local hospital followed by inter-hospital transfer to the invasive centre (acute invasive); 4—Admission to local hospital followed by referral to the invasive centre on an urgent outpatient basis (elective invasive); 5—Local hospital only (no referral for invasive management); 6—Elective direct to the invasive centre (no local hospital referral). As expected and dictated by local protocol, STEMI patients tend to be admitted to the IC directly or by transfer from the local A&E for immediate invasive management while NSTEMI patients tend to be transferred to the IC after admission to a local hospital (Table 1).

Figure 2 All acute coronary syndrome patients (1st hospitalization in timeframe) by care pathway. Note sizes are proportional to percentage of analysis population. Pathway definitions: 1—Emergency direct admission to invasive centre [Golden Jubilee National Hospital (GJNH)]; 2—Local A&E in regional hospital followed by direct transfer to invasive centre; 3—Admission to local hospital followed by inter-hospital transfer to the invasive centre (acute invasive); 4—Admission to local hospital followed by referral to the invasive centre on an urgent outpatient basis (elective invasive); 5—Local hospital only (no referral for invasive management); 6—Elective direct to the invasive centre (no local hospital referral). As expected and dictated by local protocol, STEMI patients tend to be admitted to the IC directly or by transfer from the local A&E for immediate invasive management while NSTEMI patients tend to be transferred to the IC after admission to a local hospital (Table 1). Most records of clinical characteristics and cardiovascular risk factors came from the IC or electronic referrals for invasive cardiovascular procedures. These are provided in Tables 2 and 3 for the STEMI and NSTEMI patients undergoing invasive management, respectively. Patients not undergoing invasive management did not have these data recorded as they did not enter a care pathway with systems that provided this information. Where information was available, patients with unstable angina/NSTEMI tended to have more concomitant diseases including hypertension, hypercholesterolaemia, diabetes, a family history of CHD and histories of MI, PCI, CABG and symptomatic peripheral vascular disease than patients with STEMI. However, patients with STEMI were more likely to be current smokers. Table 2 Clinical characteristics for ST-elevation myocardial infarction patients undergoing invasive treatment by intervention

a family history of CHD and histories of MI, PCI, CABG and symptomatic peripheral vascular disease than patients with STEMI. However, patients with STEMI were more likely to be current smokers. Table 2 Clinical characteristics for ST-elevation myocardial infarction patients undergoing invasive treatment by intervention Angiogram PCI after angiogram Yes (n = 504) No (n = 36) Yes (n = 468) Risk factors Age at admission (years) 60.9 (13.1) 59.0 (16.2) 61.0 (12.8) Gender Female 142 (28.2%) 6 (16.7%) 136 (29.1%) Male 362 (71.8%) 30 (83.3%) 332 (70.9%) SIMD quintile Q1 (most deprived) 184 (36.5%) 10 (27.8%) 174 (37.2%) Q2 83 (16.5%) 3 (8.3%) 80 (17.1%) Q3 82 (16.3%) 6 (16.7%) 76 (16.2%) Q4 45 (8.9%) 4 (11.1%) 41 (8.8%) Q5 (least deprived) 56 (11.1%) 5 (13.9%) 51 (10.9%) Hypertension 190 (37.7%) 16 (44.4%) 174 (37.2%) Hypercholesterolaemia 162 (32.1%) 12 (33.3%) 150 (32.1%) Diabetes 36 (7.1%) 3 (8.3%) 33 (7.1%) Smoking status Current 250 (49.6%) 11 (30.6%) 239 (51.1%) Ex 101 (20.0%) 14 (38.9%) 87 (18.6%) Never 119 (23.6%) 6 (16.7%) 113 (24.1%) Family history of CHD 192 (38.1%) 13 (36.1%) 179 (38.2%) Medical history Previous PCI 54 (10.7%) 2 (5.6%) 52 (11.1%) Previous cardiac surgery CABG 10 (2.0%) 1 (2.8%) 9 (1.9%) CABG; valve 2 (0.4%) 0 (0.0%) 2 (0.4%) None/missing 490 (97.2%) 35 (97.2%) 455 (97.2%) Other cardiac 1 (0.2%) 0 (0.0%) 1 (0.2%) Valve 1 (0.2%) 0 (0.0%) 1 (0.2%) Previous MI 73 (14.5%) 6 (16.7%) 67 (14.3%) Data are mean (SD) or number (%) out of group total.

.1%) Previous cardiac surgery CABG 10 (2.0%) 1 (2.8%) 9 (1.9%) CABG; valve 2 (0.4%) 0 (0.0%) 2 (0.4%) None/missing 490 (97.2%) 35 (97.2%) 455 (97.2%) Other cardiac 1 (0.2%) 0 (0.0%) 1 (0.2%) Valve 1 (0.2%) 0 (0.0%) 1 (0.2%) Previous MI 73 (14.5%) 6 (16.7%) 67 (14.3%) Data are mean (SD) or number (%) out of group total. CABG, coronary artery bypass grafting surgery; CHD, coronary heart disease; PCI, percutaneous coronary intervention; SIMD, Scottish Index of Multiple Deprivation. Table 3 Clinical characteristics for non-ST-elevation myocardial infarction patients undergoing invasive treatment by intervention

.1%) Previous cardiac surgery CABG 10 (2.0%) 1 (2.8%) 9 (1.9%) CABG; valve 2 (0.4%) 0 (0.0%) 2 (0.4%) None/missing 490 (97.2%) 35 (97.2%) 455 (97.2%) Other cardiac 1 (0.2%) 0 (0.0%) 1 (0.2%) Valve 1 (0.2%) 0 (0.0%) 1 (0.2%) Previous MI 73 (14.5%) 6 (16.7%) 67 (14.3%) Data are mean (SD) or number (%) out of group total. CABG, coronary artery bypass grafting surgery; CHD, coronary heart disease; PCI, percutaneous coronary intervention; SIMD, Scottish Index of Multiple Deprivation. Table 3 Clinical characteristics for non-ST-elevation myocardial infarction patients undergoing invasive treatment by intervention Angiogram PCI after angiogram Yes (n = 678) No (n = 334) Yes (n = 344) Risk factors Age at admission (years) 63.7 (11.7) 63.8 (11.7) 63.7 (11.7) Gender Female 253 (37.3%) 133 (39.8%) 120 (34.9%) Male 425 (62.7%) 201 (60.2%) 224 (65.1%) SIMD quintile Q1 (most deprived) 263 (38.8%) 140 (41.9%) 123 (35.8%) Q2 104 (15.3%) 47 (14.1%) 57 (16.6%) Q3 76 (11.2%) 39 (11.7%) 37 (10.8%) Q4 64 (9.4%) 31 (9.3%) 33 (9.6%) Q5 (least deprived) 85 (12.5%) 33 (9.9%) 52 (15.1%) Hypertension 363 (53.5%) 179 (53.6%) 184 (53.5%) Hypercholesterolaemia 297 (43.8%) 146 (43.7%) 151 (43.9%) Diabetes 67 (9.9%) 28 (8.4%) 39 (11.3%) Smoking status Current 221 (32.6%) 98 (29.3%) 123 (35.8%) Ex 166 (24.5%) 82 (24.6%) 84 (24.4%) Never 222 (32.7%) 119 (35.6%) 103 (29.9%) Family history of CHD 299 (44.1%) 144 (43.1%) 155 (45.1%) BMI (kg/m2) 29.0 (5.7) 28.8 (5.9) 29.2 (5.6) Missing 129 59 70 GRACE Score 134.9 (37.6) 135.2 (39.2) 134.6 (36.0) Missing 89 38 51 Medical history Previous PCI 101 (14.9%) 47 (14.1%) 54 (15.7%) Previous Cardiac Surgery CABG 41 (6.0%) 18 (5.4%) 23 (6.7%) CABG; Valve 1 (0.1%) 0 (0.0%) 1 (0.3%) Congenital cardiac 1 (0.1%) 0 (0.0%) 1 (0.3%) None/missing 628 (92.6%) 313 (93.7%) 315 (91.6%) Other cardiac 5 (0.7%) 2 (0.6%) 3 (0.9%) Valve 2 (0.3%) 1 (0.3%) 1 (0.3%) Previous MI 173 (25.5%) 90 (26.9%) 83 (24.1%) Data are mean (SD) or number (%) out of group total. GRACE scores are recorded at time of referral.

0.3%) Congenital cardiac 1 (0.1%) 0 (0.0%) 1 (0.3%) None/missing 628 (92.6%) 313 (93.7%) 315 (91.6%) Other cardiac 5 (0.7%) 2 (0.6%) 3 (0.9%) Valve 2 (0.3%) 1 (0.3%) 1 (0.3%) Previous MI 173 (25.5%) 90 (26.9%) 83 (24.1%) Data are mean (SD) or number (%) out of group total. GRACE scores are recorded at time of referral. BMI, body mass index; CABG, coronary artery bypass grafting surgery; CHD, coronary heart disease; PCI, percutaneous coronary intervention; SIMD, Scottish Index of Multiple Deprivation.

0.3%) Congenital cardiac 1 (0.1%) 0 (0.0%) 1 (0.3%) None/missing 628 (92.6%) 313 (93.7%) 315 (91.6%) Other cardiac 5 (0.7%) 2 (0.6%) 3 (0.9%) Valve 2 (0.3%) 1 (0.3%) 1 (0.3%) Previous MI 173 (25.5%) 90 (26.9%) 83 (24.1%) Data are mean (SD) or number (%) out of group total. GRACE scores are recorded at time of referral. BMI, body mass index; CABG, coronary artery bypass grafting surgery; CHD, coronary heart disease; PCI, percutaneous coronary intervention; SIMD, Scottish Index of Multiple Deprivation. Service Delivery Intervention rates and the mean number of days to invasive procedure varied depending on care pathway. The mean total duration of hospital stay for all patients with a diagnosis of STEMI was lower than for all patients with a diagnosis of NSTEMI (5.5, 95% CI 4.8–6.2 days vs. 7.5, 95% CI 6.8–8.1 days; P < 0.0001). These durations include both those in invasive and non-invasive pathways. Among all patients with STEMI (including those who remained in the local hospital), 470 (80.2%) underwent reperfusion therapy by percutaneous coronary intervention (PCI), while NSTEMI (33.1%) and UA (1.9%) patients underwent revascularization by PCI significantly less frequently. For patients managed invasively, STEMI bed days after intervention were higher than for NSTEMI (1.4, 95% CI 1.3–1.5 days vs. 1.1, 95% CI 1.0–1.2 days; P < 0.0001) whereas NSTEMI patients spent longer in hospital prior to intervention (0.7, 95% CI 0.4–1.0 days vs. 4.4, 95% CI 4.0–4.8 days; P < 0.0001). This finding is related to the differences observed in the repatriation of patients to local hospitals depending on a STEMI or NSTEMI diagnosis. 85.3% of STEMI patients treated in the IC were repatriated following treatment as compared to 19.9% of NSTEMI patients, who tended to be discharged home directly from the IC. Other characteristics of service delivery by diagnosis (all pathways combined) are provided in the Supplementary material online, Table S8.

osis. 85.3% of STEMI patients treated in the IC were repatriated following treatment as compared to 19.9% of NSTEMI patients, who tended to be discharged home directly from the IC. Other characteristics of service delivery by diagnosis (all pathways combined) are provided in the Supplementary material online, Table S8. Of the episodes of care where a STEMI patient accessed the IC for intended invasive treatment, 90.7% of episodes involved the patient receiving PCI. The median door-to-balloon (time from arrival at IC to receipt of PCI) (DTB) and call-to-balloon (time from call to emergency services to receipt of PCI) (CTB) times were 22 and 95 min, respectively for STEMI patients (Table 4). This compares favourably to the median DTB and CTB times presented in the most recent MINAP Annual Report5 for England which were 41 min and 115 min, respectively. Myocardial Ischaemia National Audit Project also measures the proportions of STEMI patients who receive their PCI against different time metrics. Times to receipt of PCI for patients treated in Glasgow again compare favourably to England for all reported metrics; proportion of patients receiving PCI within (i) 90 min of arriving at IC (Glasgow 98.5% vs. England 88.9%), (ii) 150 min of call for help whether direct or transfer (85.9% vs. 82.3%), (iii) 150 min of call for help with direct admission (94.8% vs. 82.8%), (iv) 150 min of call for help admitted by transfer (60.9% vs. 50.5%), and (v) 120 min of call for help with direct admission (87.5% vs. 53.9%). Table 4 Service delivery by pathway for ST-elevation myocardial infarction and non-ST-elevation myocardial infarction for those intended for invasive treatment

4.8% vs. 82.8%), (iv) 150 min of call for help admitted by transfer (60.9% vs. 50.5%), and (v) 120 min of call for help with direct admission (87.5% vs. 53.9%). Table 4 Service delivery by pathway for ST-elevation myocardial infarction and non-ST-elevation myocardial infarction for those intended for invasive treatment All Emergency direct to IC Local A&E to IC Acute invasive Elective invasive Elective direct to IC STEMI (n) 518 304 148 57 5 4 Receipt of coronary angiography Yes 504 (97.3%) 301 (99.0%) 146 (98.6%) 49 (86.0%) 5 (100.0%) 3 (75.0%) No 14 (2.7%) 3 (1.0%) 2 (1.4%) 8 (14.0%) 0 (0.0%) 1 (25.0%) Duration from admission to angiography (days) 0 [0, 0] 0 [0, 0] 0 [0, 0] 1 [0, 3] 18 [18, 21] 0 [0, 0] Receipt of PCI Yes 470 (90.7%) 285 (93.8%) 138 (93.2%) 40 (70.2%) 3 (60.0%) 4 (100.0%) No 48 (9.3%) 19 (6.3%) 10 (6.8%) 17 (29.8%) 2 (40.0%) 0 (0.0%) Duration from admission to PCI (days) 0 [0, 0] 0 [0, 0] 0 [0, 0] 1 [0, 3] 21 [11, 23] 0 [0, 0] Call-to-balloon (min) 95 [83, 116] 91 [81, 104] 132 [99, 191] 76 [74, 108] NR NR n 325 232 87 5 NR NR Door-to-balloon (min) 22 [18, 27] 21 [18, 27] 22 [18, 29] 25 [14, 27] NR NR n 400 263 126 10 NR NR NSTEMI (n) 704 29 7 426 198 44 Receipt of coronary angiography Yes 678 (96.3%) 27 (93.1%) 7 (100.0%) 408 (95.8%) 198 (100.0%) 38 (86.4%) No 26 (3.7%) 2 (6.9%) 0 (0.0%) 18 (4.2%) 0 (0.0%) 6 (13.6%) Duration from admission to angiography (days) 5 [2, 14] 0 [0, 0] 0 [0, 0] 4 [2, 6] 23 [16, 29] 0 [0, 0] Receipt of PCI Yes 354 (50.3%) 18 (62.1%) 6 (85.7%) 233 (54.7%) 72 (36.4%) 25 (56.8%) No 350 (49.7%) 11 (37.9%) 1 (14.3%) 193 (45.3%) 126 (63.6%) 19 (43.2%) Duration from admission to PCI (days) 4 [1, 9] 0 [0, 0] 0 [0, 0] 4 [2, 6] 23 [15, 27] 0 [0, 0] The recommended call-to-balloon time is within 150 min and door-to-balloon time within 90 min for STEMI.5,20 The recommended duration from admission to angiography for NSTEMI is within 72 h (3 days).21 Data are number (%) or median [IQR]. Where duration from admission to angiography/PCI is 0 days, this means the procedure happened on the same day as admission.

thin 150 min and door-to-balloon time within 90 min for STEMI.5,20 The recommended duration from admission to angiography for NSTEMI is within 72 h (3 days).21 Data are number (%) or median [IQR]. Where duration from admission to angiography/PCI is 0 days, this means the procedure happened on the same day as admission. NR, Not relevant. For the 63.5% of NSTEMI patients who were referred to the IC for intended invasive treatment, 96.3% underwent angiography, but only 50.3% then went on to have PCI. In comparison, the MINAP report indicates that 79.0% of patients in England were referred for angiography during their admission for NSTEMI in 2014–2015. The proportion of NSTEMI patients seen by a cardiologist for Glasgow and England were also able to be compared (93.3% vs. 95.1%). Time to angiography for NSTEMI patients differed by pathway of care (Table 4). For NSTEMI patients referred to the IC via the acute invasive pathway, median time to angiography was 4 days with 25% receiving angiography within 2 days and 75% receiving angiography within 6 days. For those who accessed the IC via the elective invasive pathway median time to angiography was 23 days (IQR 16–29 days). Associations with all-cause mortality and service delivery The unadjusted all-cause mortality rate at 30 days was 9.0% in all STEMI patients and 3.0% in all NSTEMI patients rising to 11.9% in STEMI and 11.6% in NSTEMI at 1 year. Age-, gender-, and pathway-adjusted 30-day mortality in STEMI was significantly higher than that in NSTEMI (HR 4.63, 95% CI 2.7–7.92) and remains higher at 1 year (HR 1.72, 95% CI 1.16–2.53).

t 30 days was 9.0% in all STEMI patients and 3.0% in all NSTEMI patients rising to 11.9% in STEMI and 11.6% in NSTEMI at 1 year. Age-, gender-, and pathway-adjusted 30-day mortality in STEMI was significantly higher than that in NSTEMI (HR 4.63, 95% CI 2.7–7.92) and remains higher at 1 year (HR 1.72, 95% CI 1.16–2.53). Age-adjusted associations of all-cause death in STEMI and NSTEMI patients at 30 days (Table 5) and 1 year (Figure 3 and 4) were assessed. Compared to conservative non-invasive treatment, the receipt of angiography alone is not associated with increased survival at 30 days or 1 year in STEMI patients but the receipt of PCI was associated with higher survival (30 days: HR 3.07, 95% CI 1.61–5.85, 1 year: HR 3.06, 95% CI 1.75–5.34). Initial presentation in a local hospital compared to being admitted to the IC directly is associated with higher mortality for STEMI patients (30 days: HR 1.86, 95% CI 1.04–3.32; P = 0.036, 1 year: HR 1.96, 95% CI 1.19–3.24; P = 0.009) but the association disappears after adjusting for the rate of PCI performed. For NSTEMI patients, the receipt of angiography alone (30 days: HR 6.95, 95% CI 1.55–31.17, 1 year: HR 4.03, 95% CI 2.17–7.48) and angiography with follow-on PCI (30 days: HR 3.66, 95% CI 1.18–11.40, 1 year: HR 4.68, 95% CI 2.48–8.83) are both associated with higher survival, with no statistically significant difference between the two. GRACE score was predictive of all-cause mortality at 1-year for NSTEMI (P = 0.021) but not at 30 days (P = 0.402). There was no difference in 30-day (P = 0.238) or 1 year (P = 0.676) mortality between local admitting hospitals. Table 5 Associations with 30-day mortality in all ST-elevation myocardial infarction and non-ST-elevation myocardial infarction patients

ality at 1-year for NSTEMI (P = 0.021) but not at 30 days (P = 0.402). There was no difference in 30-day (P = 0.238) or 1 year (P = 0.676) mortality between local admitting hospitals. Table 5 Associations with 30-day mortality in all ST-elevation myocardial infarction and non-ST-elevation myocardial infarction patients 30-day mortality and associations STEMI NSTEMI Hazard ratio (95% CI) P-value Hazard ratio (95% CI) P-value Age (5 year) 1.31 (1.18, 1.46) <0.0001 1.44 (1.23, 1.69) <0.0001 Male vs. female 1.05 (0.59, 1.87) 0.8717 1.82 (0.88, 3.77) 0.1056 SIMD quintile 0.1148 0.7562 Q2 vs. Q1 0.61 (0.22, 1.64) 1.20 (0.45, 3.20) Q3 vs. Q1 1.60 (0.77, 3.29) 1.08 (0.35, 3.36) Q4 vs. Q1 1.04 (0.41, 2.67) 1.46 (0.55, 3.89) Q5 vs. Q1 0.27 (0.06, 1.19) 0.57 (0.16, 2.03) Admission to local hospital vs. direct admission to IC (all patients) 1.86 (1.04, 3.32) 0.0363 NA Admission to local hospital vs. direct admission to IC (patients intended for invasive management) 0.70 (0.21, 2.28) 0.5488 NA GRACE (low vs. high) NA 0.31 (0.02, 4.86) 0.4019 Invasive management 0.0012 0.0082 None vs. angiogram only 1.10 (0.40, 3.05) 6.95 (1.55, 31.17) None vs. angiogram and PCI 3.07 (1.61, 5.85) 3.66 (1.18, 11.40) Angiogram only vs. angiogram and PCI 2.79 (1.08, 7.25) 0.53 (0.10, 2.88) Local admitting hospital NA 0.2376 Hospital 2 vs. 1 0.21 (0.03, 1.58) Hospital 3 vs. 1 0.89 (0.37, 2.16) Hospital 4 vs. 1 1.04 (0.30, 3.66) Hospital 5 vs. 1 — Hospital 6 vs. 1 0.39 (0.11, 1.38) Hospital 7 vs. 1 0.20 (0.05, 0.89) GRACE scores are recorded at time of referral. GRACE score: low (≤140), high (>140).

2.88) Local admitting hospital NA 0.2376 Hospital 2 vs. 1 0.21 (0.03, 1.58) Hospital 3 vs. 1 0.89 (0.37, 2.16) Hospital 4 vs. 1 1.04 (0.30, 3.66) Hospital 5 vs. 1 — Hospital 6 vs. 1 0.39 (0.11, 1.38) Hospital 7 vs. 1 0.20 (0.05, 0.89) GRACE scores are recorded at time of referral. GRACE score: low (≤140), high (>140). IC, Intervention Centre; PC, percutaneous coronary intervention; SIMD, Scottish Index of Multiple Deprivation; Q1, most deprived; Q5, least deprived. Figure 3 Associations with all-cause mortality at 1 year for ST-elevation myocardial infarction patients. SIMD, Scottish Index of Multiple Deprivation. Q1 represents the highest level of deprivation. Figure 4 Associations with all-cause mortality at 1 year for non-ST-segment elevation myocardial infarction patients. SIMD, Scottish Index of Multiple Deprivation. Q1 represents the highest level of deprivation. Figure 5 The Kaplan–Meier plot for all-cause mortality by diagnosis and management.

Figure 3 Associations with all-cause mortality at 1 year for ST-elevation myocardial infarction patients. SIMD, Scottish Index of Multiple Deprivation. Q1 represents the highest level of deprivation. Figure 4 Associations with all-cause mortality at 1 year for non-ST-segment elevation myocardial infarction patients. SIMD, Scottish Index of Multiple Deprivation. Q1 represents the highest level of deprivation. Figure 5 The Kaplan–Meier plot for all-cause mortality by diagnosis and management. All-cause mortality by diagnosis and management is shown in Figure 5. Age-adjusted associations of service delivery are shown in Table 6. If referred for intervention, NSTEMI patients with high GRACE scores tend to be fast-tracked through an acute invasive pathway (P < 0.001) but did not differ in the rate of angiography or PCI compared to those referred with low GRACE scores. Males experiencing NSTEMI were more likely to undergo angiography (OR 1.40, 95% CI 1.06–1.87) and PCI (OR 1.41, 95% CI 1.08–1.85) than females but gender was not apparently associated with whether a patient was referred to the intervention centre (OR 1.31, 95% CI 0.97–1.77) or whether a referral was acute invasive or elective invasive (OR 1.14, 95% CI 0.80–1.64). However, crude referral rates for females and males were 55.1% and 69.0%, respectively, suggesting that there is a numerical difference between gender. In those referred, female patients tended to have higher GRACE scores with 48.2% having a GRACE score >140 vs. 40.9% in males. Male gender was not associated with increased 30-day or 1-year mortality. Males were more likely than females to be diagnosed with NSTEMI vs. unstable angina (OR 1.50, 95% CI 1.22–1.85). Younger patients were more likely to be referred for angiography and PCI, but if referred, older patients tended to be fast-tracked through an acute invasive pathway, likely to be in part explained by the effect of age on the GRACE score. There was a large proportion (46%) of NSTEMI patients referred for angiography that did not receive follow-on PCI. Referral and angiography rates for NSTEMI patients differed significantly depending on the local admitting hospital (P = 0.03 and P = 0.011, respectively). Local admitting hospital also had an association with whether a patient was referred to the IC via an acute or elective invasive pathway (P < 0.0001) as well as the time from local hospital referral to receipt of PCI (P < 0.0001). Despite the differences in angiography referrals, there was ultimately no significant association between local admitting hospital and receipt of PCI (Table 6).

was referred to the IC via an acute or elective invasive pathway (P < 0.0001) as well as the time from local hospital referral to receipt of PCI (P < 0.0001). Despite the differences in angiography referrals, there was ultimately no significant association between local admitting hospital and receipt of PCI (Table 6). Table 6 Associations with service delivery in all ST-elevation myocardial infarction and non-ST-elevation myocardial infarction patients

was referred to the IC via an acute or elective invasive pathway (P < 0.0001) as well as the time from local hospital referral to receipt of PCI (P < 0.0001). Despite the differences in angiography referrals, there was ultimately no significant association between local admitting hospital and receipt of PCI (Table 6). Table 6 Associations with service delivery in all ST-elevation myocardial infarction and non-ST-elevation myocardial infarction patients Service delivery and associations STEMI NSTEMI Odds ratio/estimate (95% CI) P-value Odds ratio/estimate (95% CI) P-value Angiography (%) 86.0 63.5 Age (5 year) 0.68 (0.61, 0.75) <0.0001 0.64 (0.60, 0.69) <0.0001 Male vs. Female 1.43 (0.85, 2.41) 0.1742 1.40 (1.06, 1.87) 0.0198 SIMD quintile 0.4777 0.6751 Q2 vs. Q1 1.23 (0.56, 2.68) 0.90 (0.59, 1.38) Q3 vs. Q1 1.48 (0.68, 3.24) 1.09 (0.67, 1.79) Q4 vs. Q1 2.62 (0.90, 7.60) 0.92 (0.57, 1.49) Q5 vs. Q1 1.19 (0.63, 2.68) 1.31 (0.83, 2.09) Admission to local hospital vs. direct admission to IC 0.01 (0.01, 0.03) <0.0001 NA GRACE (low vs. high) NA 2.36 (0.71, 7.83) 0.1590 Local admitting hospital NA 0.0111 Hospital 2 vs. 1 0.52 (0.30, 0.90) Hospital 3 vs. 1 1.00 (0.65, 1.55) Hospital 4 vs. 1 0.80 (0.41, 1.55) Hospital 5 vs. 1 — Hospital 6 vs. 1 0.45 (0.29, 0.71) Hospital 7 vs. 1 0.75 (0.48, 1.16) Percutaneous coronary intervention (%) 80.2 33.1 Age (5 year) 0.79 (0.73, 0.86) <0.0001 0.82 (0.78, 0.86) <0.0001 Male vs. female 1.13 (0.72, 1.78) 0.5949 1.41 (1.08, 1.85) 0.0132 SIMD quintile 0.8286 0.1363 Q2 vs. Q1 1.30 (0.66, 2.53) 1.29 (0.87, 1.92) Q3 vs. Q1 1.12 (0.59, 2.12) 1.17 (0.74, 1.84) Q4 vs. Q1 1.44 (0.63, 3.30) 1.19 (0.74, 1.90) Q5 vs. Q1 0.91 (0.46, 1.81) 1.75 (1.15, 2.68) Admission to local hospital vs. direct admission to IC 0.04 (0.02, 0.07) <0.0001 NA GRACE (low vs. high) NA 0.83 (0.55, 1.24) 0.3566 Local admitting hospital NA 0.1059 Hospital 2 vs. 1 0.80 (0.46, 1.38) Hospital 3 vs. 1 1.13 (0.76, 1.70) Hospital 4 vs. 1 0.93 (0.50, 1.74) Hospital 5 vs. 1 — Hospital 6 vs. 1 0.61 (0.38, 0.97) Hospital 7 vs. 1 1.29 (0.86, 1.92) Referral to intervention centre (invasive transfer vs. local hospital only) (%) 47.7 63.2 Age (5 year) 0.66 (0.56, 0.77) <0.0001 0.64 (0.59, 0.68) <0.0001 Male vs. female 1.84 (0.76, 4.48) 0.1774 1.31 (0.97, 1.77) 0.0751 SIMD quintile 0.0843 0.6356 Q2 vs. Q1 1.57 (0.42, 5.91) 0.96 (0.62, 1.50) Q3 vs. Q1 1.17 (0.30, 4.59) 1.15 (0.69, 1.91) Q4 vs. Q1 5.14 (1.23, 21.42) 0.98 (0.60, 1.61) Q5 vs. Q1 0.39 (0.09, 1.66) 1.41 (0.87, 2.29) Local admitting hospital NA 0.0301 Hospital 2 vs. 1 0.61 (0.35, 1.07) Hospital 3 vs. 1 0.98 (0.63, 1.52) Hospital 4 vs. 1 0.81 (0.41, 1.58) Hospital 5 vs.

2, 5.91) 0.96 (0.62, 1.50) Q3 vs. Q1 1.17 (0.30, 4.59) 1.15 (0.69, 1.91) Q4 vs. Q1 5.14 (1.23, 21.42) 0.98 (0.60, 1.61) Q5 vs. Q1 0.39 (0.09, 1.66) 1.41 (0.87, 2.29) Local admitting hospital NA 0.0301 Hospital 2 vs. 1 0.61 (0.35, 1.07) Hospital 3 vs. 1 0.98 (0.63, 1.52) Hospital 4 vs. 1 0.81 (0.41, 1.58) Hospital 5 vs. 1 — Hospital 6 vs. 1 0.46 (0.29, 0.73) Hospital 7 vs. 1 0.71 (0.46, 1.11) Acute vs. elective invasive transfer (%) 91.9 68.3 Age (5 year) 1.27 (0.88, 1.82) 0.1973 1.25 (1.16, 1.35) <0.0001 Male vs. female — 0.9566 1.14 (0.80, 1.64) 0.4740 SIMD quintile 0.5882 0.1113 Q2 vs. Q1 — 0.54 (0.32, 0.90) Q3 vs. Q1 0.36 (0.02, 7.72) 0.97 (0.54, 1.76) Q4 vs. Q1 0.20 (0.01, 5.81) 1.17 (0.60, 2.30) Q5 vs. Q1 0.10 (0.01, 1.51) 0.72 (0.41, 1.26) GRACE (low vs. high) NA 0.07 (0.04, 0.13) <0.0001 Local admitting hospital NA <0.0001 Hospital 2 vs. 1 4.50 (1.52, 13.32) Hospital 3 vs. 1 0.60 (0.36, 0.98) Hospital 4 vs. 1 0.70 (0.33, 1.48) Hospital 5 vs. 1 — Hospital 6 vs. 1 0.33 (0.19, 0.58) Hospital 7 vs. 1 1.00 (0.58, 1.70) Time from admission to PCI (days) Age (5 year) −0.23 (−0.52, 0.05) 0.1050 −0.12 (−0.18, −0.06) <0.0001 Weekday vs. weekend admission 1.08 (−0.36, 2.52) 0.1426 0.03 (−0.29, 0.34) 0.8610 Admission to local hospital vs. direct admission to IC 4.78 (4.01, 5.55) <0.0001 NA Local admitting hospital NA <0.0001 Hospital 2 vs. 1 −0.43 (−0.91, 0.04) Hospital 3 vs. 1 0.63 (0.29, 0.97) Hospital 4 vs. 1 0.15 (−0.36, 0.67) Hospital 5 vs. 1 — Hospital 6 vs. 1 0.42 (0.02, 0.82) Hospital 7 vs. 1 −0.09 (−0.41, 0.23) GRACE scores are recorded at time of referral. GRACE score: low (≤140) high (>140).

A Local admitting hospital NA <0.0001 Hospital 2 vs. 1 −0.43 (−0.91, 0.04) Hospital 3 vs. 1 0.63 (0.29, 0.97) Hospital 4 vs. 1 0.15 (−0.36, 0.67) Hospital 5 vs. 1 — Hospital 6 vs. 1 0.42 (0.02, 0.82) Hospital 7 vs. 1 −0.09 (−0.41, 0.23) GRACE scores are recorded at time of referral. GRACE score: low (≤140) high (>140). IC, Intervention Centre; PCI, percutaneous coronary intervention; SIMD, Scottish Index of Multiple Deprivation; Q1, Most deprived; Q5, Least deprived. A patient’s deprivation status (as measured by the SIMD) was not seen to be associated with either mortality at 30 days or 1 year, or the pathway of care for STEMI or NSTEMI patients including delivery of angiography or PCI. Outcome of data validation Following review of source clinical data, pathway assignment was confirmed to be correct in all [n = 200 (100%)] of the patient episodes. The diagnosis assignment was confirmed in 199 (99%) and one subject had missing data. The occurrence of invasive coronary angiography, PCI and the dates of these procedures, were confirmed for all subjects [n = 200 (100%)]. Forty-four individual patient records were included in the verification of mortality assignment. All deaths were confirmed. The cause of death was confirmed to be correct in 100% of the deceased patients with available records [n = 36 (82%)]. Verification of cause of death was not possible in eight subjects (18%) in whom primary care records could not be accessed. The date of death was confirmed correct in all (100%) subjects.

re confirmed. The cause of death was confirmed to be correct in 100% of the deceased patients with available records [n = 36 (82%)]. Verification of cause of death was not possible in eight subjects (18%) in whom primary care records could not be accessed. The date of death was confirmed correct in all (100%) subjects. Discussion The primary objective of this proof-of-concept project was to implement a contemporary secondary care e-Registry for patients hospitalized with suspected or established ACS, utilizing only electronic records collected as part of usual clinical care in a complex regional health care system; and to be able to produce clinically meaningful analyses from these data without the need for additional manual data collection. The pilot has demonstrated that implementation of an e-registry is possible with the main outcomes being (i) longitudinal follow-up of patient episodes for all-cause and cause-specific mortality, (ii) a new system that has potential applications for quality and health care improvement, service evaluation and electronic health record-enabled research, and (iii) a system that could enable reporting for national audit.

being (i) longitudinal follow-up of patient episodes for all-cause and cause-specific mortality, (ii) a new system that has potential applications for quality and health care improvement, service evaluation and electronic health record-enabled research, and (iii) a system that could enable reporting for national audit. The analyses performed on the data suggest possible variation in standards of care and service delivery for patients experiencing STEMI and NSTEMI, including differing service provision depending on the local admitting hospital. As has previously been demonstrated in other studies, whether a patient receives invasive management or not is clearly associated with patient outcomes including mortality. The 2015 ESC Guidelines for the management of ACS in patients presenting without persistent ST-segment elevation21 recommend that patients with at least 1 intermediate risk factor should receive angiography within 72 h (3 days) of hospital admission. At the time of data extraction for this analysis the acute invasive pathway was the most likely route through which intermediate to high-risk NSTEMI patients would access the IC. However, in this pathway less than 50% of NSTEMI patients achieved this recommendation. Based partly on the evidence provided by the e-registry a ‘Direct NSTEMI’ service has recently been implemented whereby patients with high-risk characteristics would be directly transferred by the Ambulance Service to the IC, rather than the usual admission pathway to the nearest local hospital. This new service was implemented subsequent to this analysis, hence the impact of this service change is not captured. There is the opportunity for evaluation of the impact of the new service through our e-Registry with subsequent data extracts. These data provide the opportunity to look at the characteristics of patients and factors affecting their referral for invasive management to support identification of patients who will benefit most from an invasive strategy, and potentially to identify the group of patients for whom referral for angiography is unnecessary, for example those who receive angiography but do not ultimately require PCI. It is currently unknown of this group how many go on to receive cardiac surgery in place of PCI and this will be addressed in the future development of the data set.

ntify the group of patients for whom referral for angiography is unnecessary, for example those who receive angiography but do not ultimately require PCI. It is currently unknown of this group how many go on to receive cardiac surgery in place of PCI and this will be addressed in the future development of the data set. In addition to this, the ability to record GRACE score for patients who remain in the local hospitals would give further insights into the factors associated with the decision to refer for invasive treatment, or not. The differences observed in the rates of invasive management for male and female patients experiencing NSTEMI warrants further investigation. The statistical analysis suggests that there is no difference in referral rates between men and women, however the crude referral rates suggest, at least a numerical difference. When this is considered in light of a higher proportion of females referred with a GRACE score greater than 140, it would appear that males are more likely to be referred for, and to receive, angiography despite having a larger proportion of low risk scores. This raises questions about the underlying factors influencing the decision to refer for and perform angiography and PCI. The data from the e-Registry presents an opportunity to study other factors associated with gender to better understand the differences in intervention rates and this is an aspect that will be focused on as part of the continuation of the project.

s influencing the decision to refer for and perform angiography and PCI. The data from the e-Registry presents an opportunity to study other factors associated with gender to better understand the differences in intervention rates and this is an aspect that will be focused on as part of the continuation of the project. Visibility of this type of information is crucial for local clinicians and decision makers in the NHS to be able to identify focus areas for improved services and outcomes, reduce variation in care and to be able to measure the impact of any changes made. Since the e-Registry is based on linkage using the CHI number, other databases could be linked, such as for drug prescriptions, longer term follow-up in primary care and the ambulance service. The e-Registry has the potential to permit queries at the point-of-care in near-real time following further development of the functionality and creation of an NHS-Focused reporting system.

other databases could be linked, such as for drug prescriptions, longer term follow-up in primary care and the ambulance service. The e-Registry has the potential to permit queries at the point-of-care in near-real time following further development of the functionality and creation of an NHS-Focused reporting system. The ACS e-Registry has its roots in the UK MINAP with some differences. The e-Registry includes variables with definitions that are identical to those that are collected in MINAP. However, the data in the e-Registry can be routinely updated based on executable computer programmes, and no additional manual data recording is needed. Because MINAP involves some de novo manual data reporting by individual hospitals, data completeness and accuracy may be qualified, and this is especially the case for patients with NSTEMI or unstable angina. The e-Registry makes use of usual care records without de novo data entry, and so, theoretically, patient identification is more complete as recording of diagnosis in the clinical systems is mandatory. Comparison with annual figures for myocardial infarction hospital activity published by Information Services Division (ISD) Scotland22 shows that the number of cases of MI identified in the e-Registry is broadly in line with published figures. The methodologies used for each analysis differ. Certain epidemiological methods have been applied to the data from the e-Registry to allow meaningful statistical analysis, such as only considering a patient’s first admission to hospital in the time period studied, therefore it is difficult to draw conclusions on the level of similarity. However, taking these factors into account the number of MI events observed during the study period is consistent with the expected rate for the population served. The benefit of near complete case identification from use of electronic records is contrasted with potential data quality issues. Where data fields in the clinical systems are not mandatory some clinical detail may be less well reported due to poor completion of non-mandatory fields, or poor quality and consistency of data recording in electronic systems.

ntification from use of electronic records is contrasted with potential data quality issues. Where data fields in the clinical systems are not mandatory some clinical detail may be less well reported due to poor completion of non-mandatory fields, or poor quality and consistency of data recording in electronic systems. Three percent of subjects with an ACS had an elective referral. This finding was unexpected and may be explained by how episodes are created. For example, patients with stable symptoms referred on an elective basis experience an ACS while on the waiting list for invasive management, or the referral for an ACS patient is incorrectly categorized as elective by the referring clinician (commonly a trainee doctor). In addition to this, the study period for this analysis started in October 2013. Prior to this date, we do not have any historical data for patients, therefore we may be missing information on prior interactions with the health system which might have been able to clarify what had happened to these patients. This is likely to become less of an issue as new data is added to the e-Registry as patients will have increased history and follow-up.

rical data for patients, therefore we may be missing information on prior interactions with the health system which might have been able to clarify what had happened to these patients. This is likely to become less of an issue as new data is added to the e-Registry as patients will have increased history and follow-up. Despite this limitation, the results of the Data Validation support the conclusion that data extracted from clinical systems to create this e-Registry and the analyses that have been performed on the data set accurately represent what is happening to these patients in the clinical setting. The cause of death for some subjects could not be confirmed using secondary care electronic records, especially for patients who died in the community. This gap points to the need for enhanced communication between primary and secondary care health care systems.

t is happening to these patients in the clinical setting. The cause of death for some subjects could not be confirmed using secondary care electronic records, especially for patients who died in the community. This gap points to the need for enhanced communication between primary and secondary care health care systems. The secondary objective to demonstrate the ability of the pharmaceutical industry to work as a trusted partner to the NHS and academia has been successful. Under the Joint Working Agreement, due to run until April 2019, the skills and resources of all parties have been deployed equally to deliver the ambition of the e-Registry, with the NHS providing clinical leadership, Safe Haven and IT system support, the University providing data science and statistical support and the industry partner providing project management and analytical support. The industry partner has had a unique, enabling role in the development of this e-Registry. Through use of legal and data sharing agreements between all parties, utilizing project-specific anonymized data extracts, which protected individual patient identities and limited access to only appropriate data, the industry partner was able to provide guidance and expertise to the process of managing and analysing the data. The involvement of an industry partner also provided a wider view of the utility and interpretation of the data and brought knowledge of other available data sources for comparison with the e-Registry. For the remainder of the agreement the continued collaboration between the existing partners and inclusion of other stakeholders is viewed as vital to the further development of this work. However, the participation of the industry partner was always intended to be finite; Onward funding and management of the e-Registry will then exclusively rest with the NHS in Scotland.

boration between the existing partners and inclusion of other stakeholders is viewed as vital to the further development of this work. However, the participation of the industry partner was always intended to be finite; Onward funding and management of the e-Registry will then exclusively rest with the NHS in Scotland. The future development of the e-Registry should involve other regions in NHS Scotland with the aim of developing a national e-Registry. The current project represents a proof-of-concept ACS e-Registry in one Health Board (GGC), representing approximately 25% of the population in Scotland and could be rapidly upscaled to capture additional areas as the electronic systems and pathways for the other Health Boards in the region are comparable. In addition, linkage with other health-related databases would enable an end-to-end picture of management of patients experiencing ACS. Further to that, by including a broader range of ICD-10 codes for cardiovascular conditions and procedures, the e-Registry could be expanded to include a wider range of cardiovascular patient episodes.

age with other health-related databases would enable an end-to-end picture of management of patients experiencing ACS. Further to that, by including a broader range of ICD-10 codes for cardiovascular conditions and procedures, the e-Registry could be expanded to include a wider range of cardiovascular patient episodes. Limitations Utilizing data recorded as part of usual care records without any additional data capture means that some data completeness and quality issues may exist for non-mandatory or free-text fields. As a result, many of the clinical characteristics collected within the project were only available for patients with a particular diagnosis who followed a particular pathway. The amount of missing data ranged from 0.5% for history of hypertension in STEMI patients intended for invasive management, to over 42% of GRACE scores for all NSTEMI patients (13% in NSTEMI patients referred for angiography). This has shown a need for improved recording of key data elements and this has been fed back to the clinical care teams. In addition, using an automated analysis programme to define episodes of care and categorize the data can present issues when episodes of care are complex or information is incomplete. The assumption that this methodology allows complete capture of patients experiencing ACS relies on accurate recording of diagnoses in the clinical systems. The e-Registry may not have complete clinical records for patients due to the limited data sources included, for example the current lack of data from the primary care and outpatient settings. This issue is being addressed as part of further development. This study does not permit inference on causality, and other interpretations of the data are possible therefore further development and studies are warranted.

urces included, for example the current lack of data from the primary care and outpatient settings. This issue is being addressed as part of further development. This study does not permit inference on causality, and other interpretations of the data are possible therefore further development and studies are warranted. Conclusions The project has demonstrated that implementation of an e-registry of ACS hospitalizations from existing health records in a complex health care system is possible, without the need for additional data collection or excessive use of NHS resources. It is recommended to look at the electronic systems used to collect these records and address some of the issues surrounding missing data for example making certain fields such as GRACE score mandatory. Also additional guidance for clinicians and administrators should be considered in order to support the consistency and completeness of routinely collected data as part of quality improvement activities to enable the use of these data by the NHS to inform clinical practice and patient care. The data presents further opportunities to study the ACS population in more detail to gain insights into factors affecting clinical pathways and outcomes. The Joint Working Project is an example of how the NHS and pharmaceutical industry can work together to facilitate the delivery of projects that are valuable to the NHS and patients. Supplementary material Supplementary material is available at European Heart Journal – Quality of Care and Clinical Outcomes online.

Conclusions The project has demonstrated that implementation of an e-registry of ACS hospitalizations from existing health records in a complex health care system is possible, without the need for additional data collection or excessive use of NHS resources. It is recommended to look at the electronic systems used to collect these records and address some of the issues surrounding missing data for example making certain fields such as GRACE score mandatory. Also additional guidance for clinicians and administrators should be considered in order to support the consistency and completeness of routinely collected data as part of quality improvement activities to enable the use of these data by the NHS to inform clinical practice and patient care. The data presents further opportunities to study the ACS population in more detail to gain insights into factors affecting clinical pathways and outcomes. The Joint Working Project is an example of how the NHS and pharmaceutical industry can work together to facilitate the delivery of projects that are valuable to the NHS and patients. Supplementary material Supplementary material is available at European Heart Journal – Quality of Care and Clinical Outcomes online. Supplementary Material Supplementary Data Click here for additional data file.

Conclusions The project has demonstrated that implementation of an e-registry of ACS hospitalizations from existing health records in a complex health care system is possible, without the need for additional data collection or excessive use of NHS resources. It is recommended to look at the electronic systems used to collect these records and address some of the issues surrounding missing data for example making certain fields such as GRACE score mandatory. Also additional guidance for clinicians and administrators should be considered in order to support the consistency and completeness of routinely collected data as part of quality improvement activities to enable the use of these data by the NHS to inform clinical practice and patient care. The data presents further opportunities to study the ACS population in more detail to gain insights into factors affecting clinical pathways and outcomes. The Joint Working Project is an example of how the NHS and pharmaceutical industry can work together to facilitate the delivery of projects that are valuable to the NHS and patients. Supplementary material Supplementary material is available at European Heart Journal – Quality of Care and Clinical Outcomes online. Supplementary Material Supplementary Data Click here for additional data file. Acknowledgements The e-registry was enabled by a Joint Working Agreement between AstraZeneca UK Ltd, NHS Greater Glasgow & Clyde and Golden Jubilee National Hospital and was supported by NHS Scotland, the University of Glasgow and AstraZeneca UK. The authors would like to specifically acknowledge the following members of the project team for their support: Karen Fairbrother, Karen Ross, Alan Foster, Marion Flood, Roma Armstrong, Jim Christie, Stewart Hatrick, Brian Lawson.

Hospital and was supported by NHS Scotland, the University of Glasgow and AstraZeneca UK. The authors would like to specifically acknowledge the following members of the project team for their support: Karen Fairbrother, Karen Ross, Alan Foster, Marion Flood, Roma Armstrong, Jim Christie, Stewart Hatrick, Brian Lawson. Funding AstraZeneca UK Ltd, NHS Greater Glasgow and Clyde and the Golden Jubilee Foundation supported this project. Clinical Training Fellowship from the British Heart Foundation (FS/15/54/31639 to K.M.). Conflict of interest: B.F., S.S., and T.M. are employed by AstraZeneca UK Ltd, a biopharmaceutical company that manufactures drugs for the treatment of cardiovascular disease. C.B., A.M., R.Z., and C.M. are employed by the University of Glasgow which received grants from AstraZeneca in support of this project. I.F. reports to receive research funding from AstraZeneca UK Ltd. K.M. has no potential conflicts of interest. Based on a contract with the University of Glasgow, C.B. has acted as a consultant and speaker for AstraZeneca UK Ltd.

Introduction Hypertrophic cardiomyopathy (HCM) is the most common cause of sudden death in young adults with a prevalence of 0.2%.1 When asymmetrical septal hypertrophy leads to left ventricular outflow tract obstruction (LVOTO) with associated systolic anterior motion of the anterior mitral valve leaflet, this confers a diagnosis of the sub-phenotype hypertrophic obstructive cardiomyopathy (HOCM, also known as ‘obstructed HCM’), which occurs in around 70% of HCM sufferers.1 Alongside contributing to mortality through multiple mechanisms, including heart failure and malignant arrhythmia, LVOTO produces significant morbidity, inducing symptoms of chest pain, breathlessness, exertion intolerance, light-headedness, and syncope. Management of symptomatic LVOTO is initially pharmacological but interventions are available in the form of surgical septal myectomy or percutaneous alcohol septal ablation, both carrying risks of complication. Failure, intolerance and reluctance with pharmacological and interventional treatment of LVOTO led to interest in the use of right ventricular pacing as an alternative method for gradient reduction and resultant symptomatic improvement.2,3 A number of studies, both randomized and observational, have investigated the effect of dual chamber pacing in HOCM. We systematically analysed these to quantify the effect on LVOTO, left ventricular systolic function, symptoms and functional status.

Management of symptomatic LVOTO is initially pharmacological but interventions are available in the form of surgical septal myectomy or percutaneous alcohol septal ablation, both carrying risks of complication. Failure, intolerance and reluctance with pharmacological and interventional treatment of LVOTO led to interest in the use of right ventricular pacing as an alternative method for gradient reduction and resultant symptomatic improvement.2,3 A number of studies, both randomized and observational, have investigated the effect of dual chamber pacing in HOCM. We systematically analysed these to quantify the effect on LVOTO, left ventricular systolic function, symptoms and functional status. Methods We carried out a meta-analysis of studies that evaluated right ventricular pacing in HOCM in accordance with PRISMA guidelines.4 We prospectively registered this meta-analysis on the PROSPERO international register of systematic reviews (registration number CRD42017062165).5

A number of studies, both randomized and observational, have investigated the effect of dual chamber pacing in HOCM. We systematically analysed these to quantify the effect on LVOTO, left ventricular systolic function, symptoms and functional status. Methods We carried out a meta-analysis of studies that evaluated right ventricular pacing in HOCM in accordance with PRISMA guidelines.4 We prospectively registered this meta-analysis on the PROSPERO international register of systematic reviews (registration number CRD42017062165).5 Search strategy We searched PubMed and the Cochrane Central Register of Controlled Trials for any studies [randomized controlled trials (RCTs), non-RCTs, and uncontrolled observational studies] published in the English language where adults with HOCM underwent atrial-synchronous right ventricular pacing. Studies were included if prospectively determined clinically relevant outcomes were reported: left ventricular outflow tract gradient (LVOTg), New York Heart Association (NYHA) functional status, left ventricular ejection fraction, exercise duration, and peak oxygen uptake during cardiopulmonary exercise testing. Systematic reviews were examined for references to relevant studies. Any discrepancies were resolved by consensus. The full search strategy is reported in the Supplementary material online. The study protocol was drafted by A.D.A. and revised by all co-authors. Preliminary search and eligibility analysis was performed by K.C. and Y.A. F.d.V. and A.D.A. performed an independent preliminary search.

Search strategy We searched PubMed and the Cochrane Central Register of Controlled Trials for any studies [randomized controlled trials (RCTs), non-RCTs, and uncontrolled observational studies] published in the English language where adults with HOCM underwent atrial-synchronous right ventricular pacing. Studies were included if prospectively determined clinically relevant outcomes were reported: left ventricular outflow tract gradient (LVOTg), New York Heart Association (NYHA) functional status, left ventricular ejection fraction, exercise duration, and peak oxygen uptake during cardiopulmonary exercise testing. Systematic reviews were examined for references to relevant studies. Any discrepancies were resolved by consensus. The full search strategy is reported in the Supplementary material online. The study protocol was drafted by A.D.A. and revised by all co-authors. Preliminary search and eligibility analysis was performed by K.C. and Y.A. F.d.V. and A.D.A. performed an independent preliminary search. Inclusion and exclusion criteria We considered all studies of right ventricular pacing in HOCM. Studies were eligible if they recruited patients with an elevated gradient (>30 mmHg).1 We excluded case reports, studies of biventricular pacing, or studies where pacing occurred was delivered in combination with other invasive gradient reduction interventions such as septal ablation or myomectomy.

lar pacing in HOCM. Studies were eligible if they recruited patients with an elevated gradient (>30 mmHg).1 We excluded case reports, studies of biventricular pacing, or studies where pacing occurred was delivered in combination with other invasive gradient reduction interventions such as septal ablation or myomectomy. Endpoints The primary efficacy endpoint was change in gradient (ΔLVOTg). The secondary efficacy outcomes were change in symptomatic and functional status measured by NYHA class, exercise time, or peak oxygen uptake, from baseline to follow-up. Detailed inclusion criteria for endpoints are found in the Supplementary material online. Follow-up It was anticipated that the search would reveal trials of varying follow-up duration. We prospectively determined that ΔLVOTg would be analysed according to the following groups of mean follow-up durations: immediate (<12 h), short-term (12 h to 6 months), medium-term (>6 months to <2 years), and long-term (>2 years). This would allow testing for progressive change in LVOTg due to remodelling over time.6 Randomized studies were pre-specified to be analysed separately from observational studies with comparison between RCTs and non-randomized studies of similar follow-up duration.

erm (>6 months to <2 years), and long-term (>2 years). This would allow testing for progressive change in LVOTg due to remodelling over time.6 Randomized studies were pre-specified to be analysed separately from observational studies with comparison between RCTs and non-randomized studies of similar follow-up duration. Data extraction and analysis A.D.A., K.C., W.J.K., and F.d.V. performed data extraction. J.P.H. performed meta-analysis and designed and carried out statistical methodology with contributions from H.Y.J., L.C., D.P.F., Z.I.W., and M.J.S.S. The statistical programming environment R with the metafor package was used for all statistical analysis. Random-effects meta-analyses were performed using the restricted maximum likelihood estimator. For ordinal categorical outcomes (NYHA), Agresti's generalized odds ratios (ORs) were first calculated for each trial before meta-analysis.7,8 Interactions between groups were assessed using a mixed effects meta-analytical model with the variable in question as a moderator. We used the I2 statistic to assess heterogeneity. Included RCTs were assessed using the Cochrane risk of bias tool.

tios (ORs) were first calculated for each trial before meta-analysis.7,8 Interactions between groups were assessed using a mixed effects meta-analytical model with the variable in question as a moderator. We used the I2 statistic to assess heterogeneity. Included RCTs were assessed using the Cochrane risk of bias tool. Results We identified 604 study reports (flowchart in Figure 1), of which 34 were eligible for inclusion for at least one pre-specified outcome of interest, comprising 1135 patients.6,9–40 There were four RCTs,20,24,28,31 all blinded crossover trials, and 30 observational studies.6,9–19,21–23,25–27,29,30,32–41 Baseline characteristics are in Table 1 and design features are in Table 2. Most studies reported data at multiple follow-up durations.6,10,11,13–17,20,23,24,27,31,34–38,40–42 Mean age was 55.5 years and mean baseline unpaced LVOTg was 78.9 mmHg. Figure 1 Flow chart for study selection. Table 1 Characteristics of included studies—baseline values

Results We identified 604 study reports (flowchart in Figure 1), of which 34 were eligible for inclusion for at least one pre-specified outcome of interest, comprising 1135 patients.6,9–40 There were four RCTs,20,24,28,31 all blinded crossover trials, and 30 observational studies.6,9–19,21–23,25–27,29,30,32–41 Baseline characteristics are in Table 1 and design features are in Table 2. Most studies reported data at multiple follow-up durations.6,10,11,13–17,20,23,24,27,31,34–38,40–42 Mean age was 55.5 years and mean baseline unpaced LVOTg was 78.9 mmHg. Figure 1 Flow chart for study selection. Table 1 Characteristics of included studies—baseline values Authors Year N Age (years) Male (%) Baseline NYHA Baseline LVEF (%) Baseline LVOTg (mmHg) Javidgonbadi et al.a 2017 88 55 ± 18 48 2.3 ± 0.6 69 (16) 64 (66) Jurado Román et al. 2016 82 66 (range 22–88) 38 NA 73 ± 11 95 ± 37 Krejci et al. 2013 24 50 ± 17 NA 2.7 ± 0.5 70 ± 9 82 ± 46 Lucon et al. 2013 51 59 ± 14 47 2.7 ± 0.6 64 ± 8 79 ± 36 Yue-Cheng et al. 2013 37 52 ± 21 54 2.6 ± 0.8 64 ± 12 62 ± 11 Knyshov et al. 2013 49 38 ± 21 47 1.9 ± 0.8 NA 84 ± 15 Galve et al. 2010 50 62 ± 11 52 3.1 ± 0.3 76 ± 10 86 ± 29 Minami et al. 2010 24 52 ± 16 50 NA NA 89 ± 38 Sandìn et al. 2009 72 64 ± 14 38 2.6 ± 0.5 67 ± 10 87 (IQR 61.5–115.2) Binder et al. 2008 66 67 41 2.7 ± 0.7 NA 66 ± 36 Topilski et al. 2006 25 71 ± 12 48 3.2 ± 0.8 NA 92 ± 28 Hozumi et al. 2006 14 55 ± 16 79 NA 66 ± 6b 24 ± 12b Megevand et al.b 2005 18 47 NA 2.4 NA 82 ± 35 Dimitrow et al. 2004 19 47 ± 16 52 3.2 ± 0.9 NA 77 ± 25 Mickelsen et al. 2004 11 69 ± 10 82 NA NA 96 ± 21 Betocchi et al. 2002 21 45 ± 15 52 3.1 ± 0.4 NA 77 ± 37 Achterberg et al. 2002 7 52 ± 13 43 3.1 ± 0.5 NA 88 ± 13 Sant’Anna et al. 1999 9 47 ± 15 33 2.3 ± 0.5 NA 92 ± 22 Park et al. 1999 10 62 ± 13 50 3.5 ± 0.5 NA 83 ± 44 Sakai et al. 1999 12 55 ± 8 58 2.3 ± 0.5 NA 106 ± 47 Maron et al. 1999 44 53 ± 17 46 NA NA 82 ± 33 Pak et al. 1998 5 48 ± 10 60 3 81 ± 8 67 ± 33 Simantirakis et al. 1998 8 56 ± 7 63 NA NA 70 ± 18 Nishimura et al. 1997 19 59 ± 13 53 2.9 ± 0.4 NA 76 ± 61c Gadler et al. 1997 22 68 ± 14 27 3 ± 0.6 NA 86 ± 40 Kappenberger et al. 1997 83 53 (range 32–87) 60 2.6 ± 0.5 NA 70 ± 24 Slade et al. 1996 52 48 ± 18 61 2.7 ± 0.6 NA 78 ± 31 Nishimura et al.d 1996 21 58 ± 16 50 NA NA 73 ± 45 Gadler et al. 1996 22 65 ± 12 47 2.9 ± 0.6 NA 96 ± 33 Fananapazir et al. 1994 84 49 ± 16 50 3.2 ± 0.5 NA 96 ± 41 McAreavey et al. 1992 18 48 ± 14 44 3.3 ± 0.5 NA 94 ± 47 Jeanrenaud et al. 1992 13 56 ± 14 69 NA NA 82 ± 41 Fananapazir et al. 1992 44 49 ± 14 50 3.4 ± 0.5 NA 64 ± 7 McDonald et al. 1988 11 51 ± 15 55 3 ± 0.6 NA 43 ± 25 Values for age, NYHA, EF, and LVOTg are mean ± standard deviation unless otherwise stated. Values for male are percentages. NA if not reported.

.5 NA 94 ± 47 Jeanrenaud et al. 1992 13 56 ± 14 69 NA NA 82 ± 41 Fananapazir et al. 1992 44 49 ± 14 50 3.4 ± 0.5 NA 64 ± 7 McDonald et al. 1988 11 51 ± 15 55 3 ± 0.6 NA 43 ± 25 Values for age, NYHA, EF, and LVOTg are mean ± standard deviation unless otherwise stated. Values for male are percentages. NA if not reported. NYHA, New York Heart Association class; LVEF, left ventricular ejection fraction; LVOTg, left ventricular outflow tract gradient. a LVOTg and LVEF data for this trial are reported as median (interquartile range). b Baseline LVEF and LVOTg in this study is immediately after pacing is switched off after period of pacing (rather than prior to pacing initiation as performed in the other studies). c Value reported from echocardiogram; 87 ± 54 on cardiac catheterization. d This study contains the data from the acute haemodynamic protocol of Nishimura et al.31 Table 2 Characteristics of included studies—study design Authors Year Study type Optimal AV delay selection description Optimal AV delay methods Longest follow-up AV delay (ms) Javidgonbadi et al. 2017 Observational single-arm (retrospective) ‘set under ECG control to ensure abolition of spontaneous conduction and then evaluated by echocardiography to obtain maximal LVOT gradient reduction without deterioration of diastolic filling’ ECG Full capture TTE LVOTg Mitral filling pattern

Authors Year Study type Optimal AV delay selection description Optimal AV delay methods Longest follow-up AV delay (ms) Javidgonbadi et al. 2017 Observational single-arm (retrospective) ‘set under ECG control to ensure abolition of spontaneous conduction and then evaluated by echocardiography to obtain maximal LVOT gradient reduction without deterioration of diastolic filling’ ECG Full capture TTE LVOTg Mitral filling pattern 16 ± 8 years NA Jurado Román et al. 2016 Observational single-arm (retrospective) ‘highest LVOTg reduction without excessive shortening of the ventricular filling time, as indicated by the minimal deterioration in the qualitative morphology of the mitral filling pattern on echocardiography’ TTE LVOTg Mitral filling Median 8.5 years (range 1–18 years) 120 ± 16 Krejci et al. 2013 Observational signal-arm (retrospective) ‘set under ECG control to ensure full capture stimulation without the presence of spontaneous or fused contractions. In most patients, AV intervals were optimized under echocardiographic control so that LVOTg was reduced, while the stroke volume was not significantly affected’ ECG Full capture TTE LVOTg Stroke Volume

Median 8.5 years (range 1–18 years) 120 ± 16 Krejci et al. 2013 Observational signal-arm (retrospective) ‘set under ECG control to ensure full capture stimulation without the presence of spontaneous or fused contractions. In most patients, AV intervals were optimized under echocardiographic control so that LVOTg was reduced, while the stroke volume was not significantly affected’ ECG Full capture TTE LVOTg Stroke Volume 101 ± 49 months NA Lucon et al. 2013 Observational single-arm (retrospective) ‘longest interval associated with complete ventricular capture, at rest and during exercise. Radiofrequency modification of the AV junction was performed in patients whose short spontaneous PR interval precluded the complete capture of the ventricles. In patients whose P wave duration was ≥120 ms, a third lead was placed in the coronary sinus and connected to a biatrial DDD pacemaker to resynchronize the atria’ ECG Full capture Rest and exercise Invasive AVNA biatrial pacing 11.5 years (range 0.4–21.8) NA Yue-Cheng et al. 2013 Observational single-arm (retrospective) ‘AV delay (was) adjusted to 90–180 ms in order to ensure the ratio of ventricular pacing was more than 98%’ Device Full capture 4 years 120 ± 21 Knyshov et al. 2013 Observational single-arm (retrospective) ‘acute haemodynamic study with the real-time direct measurement of LVOTg during temporary pacing test in AAI, VDD, and DDD modes with different AV delays’ Invasive LVOTga

11.5 years (range 0.4–21.8) NA Yue-Cheng et al. 2013 Observational single-arm (retrospective) ‘AV delay (was) adjusted to 90–180 ms in order to ensure the ratio of ventricular pacing was more than 98%’ Device Full capture 4 years 120 ± 21 Knyshov et al. 2013 Observational single-arm (retrospective) ‘acute haemodynamic study with the real-time direct measurement of LVOTg during temporary pacing test in AAI, VDD, and DDD modes with different AV delays’ Invasive LVOTga 68 ± 6.6 months Range 45–120 (s), 85–180 (p) Galve et al. 2010 Observational single-arm (prospective) ‘The optimal AV interval was defined as that obtaining a complete ventricular capture both at rest and during exercise’ ECG/device Full capture (rest and exercise) 5 ± 2.9 years NA Minami et al. 2010 Observational single-arm (prospective) ‘producing the lowest LVOTg without compromise of aortic pressure’ Invasive LVOTg, Aorta Immediate 70 ± 30 Sandìn et al. 2009 Observational single-arm (retrospective) ‘method consisted of modifying the AV pacing interval and assessing the appearance of acute changes in the LVOTg, as well as the transmitral filling curves. The curve that achieved the largest gradient decrease without excessive shortening of the filling time was chosen’ TTE LVOTg Mitral filling

Immediate 70 ± 30 Sandìn et al. 2009 Observational single-arm (retrospective) ‘method consisted of modifying the AV pacing interval and assessing the appearance of acute changes in the LVOTg, as well as the transmitral filling curves. The curve that achieved the largest gradient decrease without excessive shortening of the filling time was chosen’ TTE LVOTg Mitral filling >1 yearb NA Binder et al. 2008 Observational single-arm (retrospective) Not stated NA 3.7 ± 3 years NA Topilski et al. 2006 Observational single-arm (prospective) ‘To determine the optimized AVI, the AVI was set at 50 ms less than the native PR interval and increased in 25-ms steps, with five different AVIs. The AVIs were tested during sinus rhythm (VDD) and atrial pacing rates of 60 and 80 b.p.m. (DDD). At each AVI and heart rate combination, the LVOT gradient was measured. To achieve haemodynamic steady state, 15 min elapsed between pacemaker programming and the measurement of LVOTg. Systolic cuff blood pressure was used as an estimate of peak systolic aortic pressure. The AVI was programmed at the value with minimal LVOTg not associated with systolic arterial pressure reduction.’ TTE LVOTg Non-invasive: cuff BP 68 ± 34 months 106 ± 30 Hozumi et al. 2006 Observational single-arm (prospective) ‘optimized in individual patients to achieve the lowest LVOTg’ TTE LVOTg

>1 yearb NA Binder et al. 2008 Observational single-arm (retrospective) Not stated NA 3.7 ± 3 years NA Topilski et al. 2006 Observational single-arm (prospective) ‘To determine the optimized AVI, the AVI was set at 50 ms less than the native PR interval and increased in 25-ms steps, with five different AVIs. The AVIs were tested during sinus rhythm (VDD) and atrial pacing rates of 60 and 80 b.p.m. (DDD). At each AVI and heart rate combination, the LVOT gradient was measured. To achieve haemodynamic steady state, 15 min elapsed between pacemaker programming and the measurement of LVOTg. Systolic cuff blood pressure was used as an estimate of peak systolic aortic pressure. The AVI was programmed at the value with minimal LVOTg not associated with systolic arterial pressure reduction.’ TTE LVOTg Non-invasive: cuff BP 68 ± 34 months 106 ± 30 Hozumi et al. 2006 Observational single-arm (prospective) ‘optimized in individual patients to achieve the lowest LVOTg’ TTE LVOTg 7.4 ± 2.1 years 120 ± 31 Megevand et al.c 2005 Observational single-arm (prospective) ‘the longest interval that captured the ventricle and induced the greatest reduction in outflow gradient without compromising haemodynamics (during cardiac catheterization)’ Invasive LVOTg, haemodynamics 4.1 years (range 1–10) Median 60 Dimitrow et al. 2004 Observational single-arm (prospective) ‘insure fully paced ventricular activation’ ECG Full capture 6 months NA Mickelsen et al. 2004 Randomized single-blinded crossover trial ‘DDD with an AV interval at 30 ms (DDD30)’d Device Fixed 30 ms

7.4 ± 2.1 years 120 ± 31 Megevand et al.c 2005 Observational single-arm (prospective) ‘the longest interval that captured the ventricle and induced the greatest reduction in outflow gradient without compromising haemodynamics (during cardiac catheterization)’ Invasive LVOTg, haemodynamics 4.1 years (range 1–10) Median 60 Dimitrow et al. 2004 Observational single-arm (prospective) ‘insure fully paced ventricular activation’ ECG Full capture 6 months NA Mickelsen et al. 2004 Randomized single-blinded crossover trial ‘DDD with an AV interval at 30 ms (DDD30)’d Device Fixed 30 ms 1 month 30 Betocchi et al. 2002 Cohort (prospective non-randomized controlled observational study) ‘Italian cohort: AV interval was chosen as the one associated with the smallest gradient without a decrease in systolic blood pressure. UK cohort: AV interval was chosen as the one associated with the largest width of the QRS complex on the electrocardiograms’ TTE LVOTg BP ECG Full capture 1 year 89 ± 16 Achterberg et al. 2002 Observational single-arm (prospective) ‘programmed between 50 ms and 100 ms to ensure continuous ventricular capture’ Device Fixed 50–100ms 2.3 ± 1.1 years 50–100 Sant’Anna et al. 1999 Observational single-arm (prospective) ‘lowest LVOTg’ TTE LVOTg 6 months NA Park et al. 1999 Observational single-arm (prospective) ‘echocardiographic guidance to obtain maximal reduction in LVOTg’ TTE LVOTg 12 ± 11 months 82 ± 17 (s), 93 ± 19 (p) Sakai et al. 1999 Observational single-arm (prospective) ‘produced the minimum LVOTg’ Invasive LVOTg, Aorta

2.3 ± 1.1 years 50–100 Sant’Anna et al. 1999 Observational single-arm (prospective) ‘lowest LVOTg’ TTE LVOTg 6 months NA Park et al. 1999 Observational single-arm (prospective) ‘echocardiographic guidance to obtain maximal reduction in LVOTg’ TTE LVOTg 12 ± 11 months 82 ± 17 (s), 93 ± 19 (p) Sakai et al. 1999 Observational single-arm (prospective) ‘produced the minimum LVOTg’ Invasive LVOTg, Aorta 1 year NA Maron et al. 1999 Randomized double-blinded crossover trial ‘longest interval which captured the ventricle and induced greatest reduction in LVOTg without compromising haemodynamics (i.e. decreasing blood pressure 30 mmHg), after testing a range of AV intervals’ Invasive LVOTg 1 yearg 85 ± 35 (s) Pak et al. 1998 Observational single-arm (prospective) ‘longest value that still yielded optimal ventricular pre-excitation as judged by QRS duration’ ECG Full capture Immediate NA Simantirakis et al. 1998 Observational single-arm (prospective) ‘longest AV delay that produced a QRS complex of the same width as that seen in VVI pacing’ ECG Full capture 1 year 80 ± 23 (s) Nishimura et al. 1997 Randomized double-blinded crossover trial ‘producing the lowest left ventricular outflow tract gradient without a significant decrease in aortic pressure or increase in left atrial pressure’ Invasive LVOTg, BP, LAP 3 months 71 ± 21 Gadler et al. 1997 Observational single-arm (prospective) ‘most pronounced reduction of left ventricular outflow tract gradient without any decrease in total mitral flow’ TTE LVOTg Mitral filling pattern

1 year 80 ± 23 (s) Nishimura et al. 1997 Randomized double-blinded crossover trial ‘producing the lowest left ventricular outflow tract gradient without a significant decrease in aortic pressure or increase in left atrial pressure’ Invasive LVOTg, BP, LAP 3 months 71 ± 21 Gadler et al. 1997 Observational single-arm (prospective) ‘most pronounced reduction of left ventricular outflow tract gradient without any decrease in total mitral flow’ TTE LVOTg Mitral filling pattern 12 ± 9 months 64 ± 17 (s) Kappenberger et al.e 1997 Randomized double-blinded crossover trial ‘full ventricular capture on the ECG without a drop in aortic pressure’ ECG Full capture Invasive 3 months 61 ± 23 (s) Slade et al. 1996 Observational single-arm (prospective) ‘Shortest sensed AV delay not associated with haemodynamic deterioration, defined as a reduction in mean aortic pressure or cardiac output of >10%’ Invasive BP 11 ± 11 months Median 65 (range 25–125) (s) Nishimura et al. 1996 Observational single-arm (prospective) ‘longest AV interval in which there is full ventricular activation by the pacemaker without fusion complexes on the ECG’ ECG Full capture Immediate NA Gadler et al. 1996 Observational single-arm (prospective) ‘resulting in the greatest reduction of LVOTg without reducing the integral of the A and E waves’ TTE LVOTg Mitral filling pattern Immediate 60–80 (s) Fananapazir et al.f 1994 Observational single-arm (prospective) ‘longest interval that permitted ventricular pre-excitation (maximum widening of the QRS complex during the exercise tests)’ ECG Full capture (exercise)

Immediate NA Gadler et al. 1996 Observational single-arm (prospective) ‘resulting in the greatest reduction of LVOTg without reducing the integral of the A and E waves’ TTE LVOTg Mitral filling pattern Immediate 60–80 (s) Fananapazir et al.f 1994 Observational single-arm (prospective) ‘longest interval that permitted ventricular pre-excitation (maximum widening of the QRS complex during the exercise tests)’ ECG Full capture (exercise) 2.3 ± 0.8 years 120 ± 9 (s) McAreavey et al. 1992 Observational single-arm (prospective) Not stated NA 12 weeks NA Jeanrenaud et al. 1992 Observational single-arm (prospective) ‘best reduction in LVOTg without drop in mean aortic pressure’ Invasive LVOTg 44 ± 11 63 ± 18 Fananapazir et al. 1992 Observational single-arm (prospective) ‘longest AV delay that allowed for maximal pre-excitation (widest-paced QRS duration)’ ECG Full capture 1.5–3 months 115 ± 17 (s) McDonald et al. 1988 Observational single-arm (prospective) ‘highest value that maintained ventricular capture at maximum exercise’ ECG Full capture (exercise)

44 ± 11 63 ± 18 Fananapazir et al. 1992 Observational single-arm (prospective) ‘longest AV delay that allowed for maximal pre-excitation (widest-paced QRS duration)’ ECG Full capture 1.5–3 months 115 ± 17 (s) McDonald et al. 1988 Observational single-arm (prospective) ‘highest value that maintained ventricular capture at maximum exercise’ ECG Full capture (exercise) 1 h 90 (range 50–150) Values are represented as mean ± standard deviation unless otherwise stated. Study type refers to arms of trials fulfilling inclusion criteria. Single-arm studies refer to studies where there is either one arm or where other arms are not control groups with conventional/medical/non-interventional therapy (instead they are alternative therapies). Optimal AV delay selection is the description of optimal AV delay selection in the wording of the source text with changes to wording made only to paraphrase and abbreviate. Optimal AV delay methods refer to the techniques used in determining the optimal AV delay. AV delay refers to the AV delay used in the trial; where the AV delay is specifically stated to be sensed AV delay, this is acknowledged by (s), and where paced (p). a Unclear from publication whether invasive or echocardiographic but wording suggests invasive measurement. b Precise duration of follow-up not stated but longer than 1 year. c Non-responders in initial acute study did not undergo implantation of pacemaker. d This study also included an arm where pacing was optimised to peak aortic flow but LVOTg response to this was not reported for all patients (only for responders and non-responders).

b Precise duration of follow-up not stated but longer than 1 year. c Non-responders in initial acute study did not undergo implantation of pacemaker. d This study also included an arm where pacing was optimised to peak aortic flow but LVOTg response to this was not reported for all patients (only for responders and non-responders). e Long-term follow-up data. f Long-term follow-up of Fananapazir et al.13 g Crossover periods were for 3 months but after the crossover study 6 months of DDD pacing occurred. Risk of bias Trial quality for the four RCTs, assessed by the Cochrane risk of bias assessment tool, is shown in Table 3. Three were rated as low risk of bias due to double blinding but the fourth was single blinded. Lack of randomization and lack of blinding put all 31 observational studies at risk of bias. Table 3 Risk of bias assessment—randomized controlled trials

Risk of bias Trial quality for the four RCTs, assessed by the Cochrane risk of bias assessment tool, is shown in Table 3. Three were rated as low risk of bias due to double blinding but the fourth was single blinded. Lack of randomization and lack of blinding put all 31 observational studies at risk of bias. Table 3 Risk of bias assessment—randomized controlled trials Authors Random sequence generation Allocation concealment Blinding of participants and personnel Blinding of outcome assessment Incomplete outcome data Selective reporting Overall quality Kappenberger et al. Low risk Low risk Low risk Low risk Low risk Low risk Low risk Nishimura et al. Uncertain Uncertain Low risk Low risk Low risk Low risk Low risk Maron et al. Uncertain Uncertain Low risk Low risk Low risk Low risk Low risk Mickelsen et al. Uncertain Uncertain High risk High risk Low risk Low risk High risk Left ventricular outflow tract obstruction Thirty-two studies reported mean change in LVOTg (ΔLVOTg) from baseline for at least one follow-up duration. The results are shown in Figure 2 for observational studies, Figure 3 for RCTs, and summarized in Table 4. Figure 2 Effect of right ventricular pacing on left ventricular outflow tract gradient at immediate (<12 h), short-term (12 h to 6 months), medium-term (>6 months to <2 years), and long-term (at least 2 years) follow-up in non-randomized studies. Figure 3 Effect of right ventricular pacing on left ventricular outflow tract gradient at short-term follow-up (1–3 months) in crossover randomized controlled trials. Table 4 Results

Figure 2 Effect of right ventricular pacing on left ventricular outflow tract gradient at immediate (<12 h), short-term (12 h to 6 months), medium-term (>6 months to <2 years), and long-term (at least 2 years) follow-up in non-randomized studies. Figure 3 Effect of right ventricular pacing on left ventricular outflow tract gradient at short-term follow-up (1–3 months) in crossover randomized controlled trials. Table 4 Results Trials (n) Patients (n) Study typea Follow-up duration Pooled result 95% confidence interval P-value I 2 heterogeneity P-value for heterogeneity Percentage change in LVOT gradient from baseline 9 234 Obs Immediate (<12 h) −40.8% −29.8 to −51.9 <0.0001 74.9% (high) <0.0001 10 243 Obs Short-term (12 h to 6 months) −54.3% −44.1 to −64.6 <0.0001 39.9% (moderate) 0.12 11 369 Obs Medium-term (>6 months to <2 years) −51.5% −44.5 to −58.4 <0.0001 10.8% (low) 0.3 16 644 Obs Long-term (at least 2 years) −66.8% −56.4 to −77.1 <0.0001 49.9% (moderate) 0.01 4 115 RCT Short-term (1–3 months) −35% −23.2 to 46.9 <0.0001 0% (low) 0.75 Odds ratio for improved NYHA class from baseline 9 388 Obs All follow-up durations 8.39 4.39 to 16.04 <0.0001 74.9% (high) <0.0001 3 137 RCT All follow-up durations 1.82 0.96 to 3.44 0.066 81.7% (high) 0.0042 Obs, observational studies (non-randomized); RCT, randomized controlled crossover trials. In the four RCTs (follow-up ranged from 1 months to 3 months), pacing reduced LVOTg by 35%. The observational studies reported slightly larger LVOTg reductions (Table 4).

Trials (n) Patients (n) Study typea Follow-up duration Pooled result 95% confidence interval P-value I 2 heterogeneity P-value for heterogeneity Percentage change in LVOT gradient from baseline 9 234 Obs Immediate (<12 h) −40.8% −29.8 to −51.9 <0.0001 74.9% (high) <0.0001 10 243 Obs Short-term (12 h to 6 months) −54.3% −44.1 to −64.6 <0.0001 39.9% (moderate) 0.12 11 369 Obs Medium-term (>6 months to <2 years) −51.5% −44.5 to −58.4 <0.0001 10.8% (low) 0.3 16 644 Obs Long-term (at least 2 years) −66.8% −56.4 to −77.1 <0.0001 49.9% (moderate) 0.01 4 115 RCT Short-term (1–3 months) −35% −23.2 to 46.9 <0.0001 0% (low) 0.75 Odds ratio for improved NYHA class from baseline 9 388 Obs All follow-up durations 8.39 4.39 to 16.04 <0.0001 74.9% (high) <0.0001 3 137 RCT All follow-up durations 1.82 0.96 to 3.44 0.066 81.7% (high) 0.0042 Obs, observational studies (non-randomized); RCT, randomized controlled crossover trials. In the four RCTs (follow-up ranged from 1 months to 3 months), pacing reduced LVOTg by 35%. The observational studies reported slightly larger LVOTg reductions (Table 4). Meta-regression showed progressively greater gradient reductions at longer follow-up durations, by an average of 5.2% per month [confidence interval (CI) 2.5–7.9, P = 0.0001]. We, therefore, compared gradient effect size of RCTs with observational studies at similar follow-up times (denoted short-term). The observational studies reported a gradient reduction, which was 18.6% greater than RCTs (CI 1.3–36, P = 0.0351).

Meta-regression showed progressively greater gradient reductions at longer follow-up durations, by an average of 5.2% per month [confidence interval (CI) 2.5–7.9, P = 0.0001]. We, therefore, compared gradient effect size of RCTs with observational studies at similar follow-up times (denoted short-term). The observational studies reported a gradient reduction, which was 18.6% greater than RCTs (CI 1.3–36, P = 0.0351). New York Heart Association status As pre-specified, because there were too few studies reporting NYHA, we did not sub-divide them by follow-up duration. The results are displayed in Figures 4 and 5 and summarized in Table 4. Figure 4 Effect of right ventricular pacing on New York Heart Association class at short-term follow-up (1–3 months) in non-randomized studies. Figure 5 Effect of right ventricular pacing on New York Heart Association class at short-term follow-up (1–3 months) in crossover randomized controlled trials. In RCTs, there was a trend towards improved NYHA class with pacing (OR 1.82, CI 0.96–3.44; P = 0.066) but with considerable heterogeneity (I2 = 81.7%) as a result of one trial showing a particularly prominent favourable effect. In contrast, almost all of the observational studies (8/9) reported marked improvement in NYHA class giving a combined OR of 8.39 (CI 4.39–16.04, P < 0.0001). This OR is over 450% of the OR for RCTs: ratio of ORs 4.54 (CI 1.61–12.82, P = 0.0042).

In RCTs, there was a trend towards improved NYHA class with pacing (OR 1.82, CI 0.96–3.44; P = 0.066) but with considerable heterogeneity (I2 = 81.7%) as a result of one trial showing a particularly prominent favourable effect. In contrast, almost all of the observational studies (8/9) reported marked improvement in NYHA class giving a combined OR of 8.39 (CI 4.39–16.04, P < 0.0001). This OR is over 450% of the OR for RCTs: ratio of ORs 4.54 (CI 1.61–12.82, P = 0.0042). Exercise and systolic function For each of the other variables (ejection fraction, exercise duration, and peak oxygen uptake) fewer than five studies reported the data require, and therefore, meta-analysis was not conducted. Reported changes in ejection fraction ranged from +3% to −11%, and in exercise duration from +0.3 min to +3.1 min. There was a single RCT report of peak oxygen uptake change, −0.1 mL/kg/min. Discussion This is the first meta-analysis assessing the role of right ventricular pacing as a treatment for LVOTO in HOCM. We found that, in blinded RCTs, pacing reduces LVOTg and shows a non-significant trend to reduce NYHA class. Unblinded observational studies report very much larger symptomatic effects, suggesting an unintentional bias much larger than any genuine effect of the pacing. Echocardiographic gradient assessments, which appear less vulnerable to this bias, suggest a progressive enhancement of the therapeutic effect with the passage of time.

observational studies report very much larger symptomatic effects, suggesting an unintentional bias much larger than any genuine effect of the pacing. Echocardiographic gradient assessments, which appear less vulnerable to this bias, suggest a progressive enhancement of the therapeutic effect with the passage of time. Left ventricular outflow tract gradient reduction The meta-analysed RCT data strongly supports the concept that pacing reduces measured LVOTg. There are broadly four mechanisms by which pacing can have an influence. Right ventricular pacing causes incoordination of ventricular activation which may reduce the driving force of ventricular ejection. The altered ventricular activation sequence attenuates the tendency of the left ventricular outflow tract (LVOT) lumen to become very small during systole. Atrioventricular (AV) sequential pacing alters ventricular filling through a change in AV delay, which impacts on ventricular ejection. Finally, knowledge that the patient is receiving a treatment believed to decrease LVOTg may cause an unintended bias of the echocardiographer to try less hard to find a high gradient.

Atrioventricular (AV) sequential pacing alters ventricular filling through a change in AV delay, which impacts on ventricular ejection. Finally, knowledge that the patient is receiving a treatment believed to decrease LVOTg may cause an unintended bias of the echocardiographer to try less hard to find a high gradient. Randomization with allocation concealment (‘blinding’) is the most effective approach to reduce unintended bias when evaluating therapies. Under blinded conditions each of the trials individually showed that pacing reduced gradient. The measured effect was much larger in the (unblinded) observational studies, −55% rather than −35% (P = 0.0351 for difference in study designs). While it is not certain that the much larger effect sizes reported by unblinded studies is due to unintended bias, it is difficult to imagine that the patient groups or procedural characteristics were so markedly different between the two study designs. Symptoms without blinding The blinded RCTs showed an encouraging trend towards a statistically significant reduction in NYHA class. A very much larger reduction in NYHA class was reported by the unblinded observational studies. The pooled estimates were so far apart their CIs did not overlap. Indeed, the point estimates for eight out of nine observational studies were for a greater NYHA reduction than even the highest upper limit of the CI of any individual RCT.

rger reduction in NYHA class was reported by the unblinded observational studies. The pooled estimates were so far apart their CIs did not overlap. Indeed, the point estimates for eight out of nine observational studies were for a greater NYHA reduction than even the highest upper limit of the CI of any individual RCT. We conclude from this that there may be a symptomatic benefit, but that unblinded study design provides no useful information on it. We cannot even use the unblinded symptomatic relief data to compare different studies in order to select pacing approaches for future blinded trials. This is because the great majority of the unblinded symptomatic relief appears to be bias. Therefore, the differences between the unblinded symptom-relief effects reported by different studies will be dominated by differences in the amount of bias rather than clinically meaningful differences in protocol. The origin of bias in unblinded observational studies investigating subjective outcomes, such as NYHA, may be due to the anticipation of effect by patients. However, the clinicians that rate patients’ NYHA status, who are also aware of treatment allocation, may also contribute to bias with their own expectations of successful, or, indeed, unsuccessful, treatment.

studies investigating subjective outcomes, such as NYHA, may be due to the anticipation of effect by patients. However, the clinicians that rate patients’ NYHA status, who are also aware of treatment allocation, may also contribute to bias with their own expectations of successful, or, indeed, unsuccessful, treatment. Clinical implications Knowing whether AV sequential right ventricular pacing is beneficial in HOCM is important because many patients with HOCM require a defibrillator and the modern era offers two changes from the traditional dual-chamber, transvenous defibrillator. First, omitting the atrial lead can reduce complication rates43 but it prevents AV sequential pacing. Second, subcutaneous defibrillators are now available, which are easier to remove than trans-venous defibrillators, but do not have any pacing function.

ges from the traditional dual-chamber, transvenous defibrillator. First, omitting the atrial lead can reduce complication rates43 but it prevents AV sequential pacing. Second, subcutaneous defibrillators are now available, which are easier to remove than trans-venous defibrillators, but do not have any pacing function. If AV sequential right ventricular pacing genuinely helps patients, then dual chamber, transvenous defibrillators would be preferable over both single chamber transvenous and subcutaneous defibrillators. Although the blinded RCTs show only a non-significant trend to reduction in NYHA class, the point estimate of the pooled effect size is an OR of 1.82. Roughly speaking, this means patients are twice as likely to feel better with pacing switched on rather than off. This is a potentially meaningful clinical benefit and merits further investigation through further blinded RCTs. Such RCTs need not be expensive or resource-intensive, recruiting patients undergoing de novo implantation. They could be conducted in patients who already have a defibrillator, with two randomized periods (pacing on vs. off) and appropriately blinded evaluation. A 2012 Cochrane review2 of observational and randomized studies concluded, as we do, that the existing observational studies are of low quality due to inadequate blinding but they did not quantify this effect as we have done. They also call for high quality trials to investigate the potential for a true symptomatic benefit.

2 Cochrane review2 of observational and randomized studies concluded, as we do, that the existing observational studies are of low quality due to inadequate blinding but they did not quantify this effect as we have done. They also call for high quality trials to investigate the potential for a true symptomatic benefit. The clear evidence of gradient reduction by pacing should not be assumed to prove that symptoms also improve. This is because pacing (i) reduces the force of contraction by inducing incoordination, (ii) reduces the impingement of the LVOT during systole, and (iii) necessarily alters filling since the paced activation must begin before the native activation would have otherwise occurred. The symptomatic effect will therefore be a result of not only changes in LVOT calibre but also ventricular filling and ejection. The choice of AV delay may be important. Applying a very short AV delay almost always reduces stroke volume, even if systolic LVOT calibre is increased. It is unclear how to programme the AV delay of a pacemaker in a patient with HOCM, to provide a net advantage. Each of the studies has taken a different approach, which may have contributed to the lack of translation of gradient reduction to symptomatic benefit, and more work on this is required. The TRICHAMPION trial is underway, which is examining the role of AV nodal ablation, for complete control of AV delay, with biventricular pacing. Importantly, not all symptoms contributing to NYHA status in HCM are due to LVOTO, some maybe due to the cardiomyopathic process itself or comorbidities, and thus unaffected by pacing.

MPION trial is underway, which is examining the role of AV nodal ablation, for complete control of AV delay, with biventricular pacing. Importantly, not all symptoms contributing to NYHA status in HCM are due to LVOTO, some maybe due to the cardiomyopathic process itself or comorbidities, and thus unaffected by pacing. The meta-regression of gradient with respect to follow-up duration suggests a progressive enlargement of the therapeutic effect with longer follow-up across all published data, as has been noted before within single studies.34,40 It is not yet clear what the mechanism for such a progression might be. One possibility is progressive changes in the structure of the ventricle. However, it should be remembered that long-term right ventricular pacing is known to cause left ventricular function to deteriorate, and therefore, a long-term reduction in gradient should not automatically be assumed to be beneficial. Limitations There were only four RCTs, although all were blinded. Studies mostly did not report lead position or baseline QRS morphology, preventing separate analysis of the apical and septal positions or the presence of pre-existing bundle branch block or meta-regression for these variables. The apical lead position may be expected to result in more dyssynchronous activation and thus greater ‘beneficial’ LVOTg reduction, but this may be offset, or even superceded, by dyssynchrony-related reduction in myocardial performance. This has importance in procedure technique as achieving an apical lead position in severe hypertrophy can be technically challenging.

in more dyssynchronous activation and thus greater ‘beneficial’ LVOTg reduction, but this may be offset, or even superceded, by dyssynchrony-related reduction in myocardial performance. This has importance in procedure technique as achieving an apical lead position in severe hypertrophy can be technically challenging. Many studies reported the mean baseline (unpaced) and the mean follow-up (paced) LVOTg and NYHA class with a categorical description of the P-value (e.g. ‘P < 0.05’) for the statistical test for differences between them. The most useful value, however, would be the mean intra-patient change in LVOTg along with its CI, standard error, standard deviation, or precise P-value. This has limited the precision with which we can calculate the CI for the pooled estimate for the change in LVOTg and NYHA, but has not affected the point-estimate. Conclusions The blinded RCTs show that AV sequential right ventricular pacing reduces LVOTg in HOCM, and shows a trend to benefit in symptoms that is not statistically significant but of a potentially clinically meaningful size. This may have implications for the choice of defibrillator to implant in HOCM. More research is needed into appropriate selection of AV delay. More blinded RCTs on symptom relief are required and they need not be very demanding or expensive.

not statistically significant but of a potentially clinically meaningful size. This may have implications for the choice of defibrillator to implant in HOCM. More research is needed into appropriate selection of AV delay. More blinded RCTs on symptom relief are required and they need not be very demanding or expensive. Unblinded observational studies report substantially larger gradient reductions and very much larger symptom reductions. The difference between the effect sizes reported by the two study designs (RCT vs. observational) is so large that observational data are of uncertain value in progressing the field. This is particularly so for symptoms, where the unblinded effect sizes appear to be scaled up by 450%. Supplementary Material qcz006_Supplementary_Materials Click here for additional data file. Acknowledgements We are grateful to Professor Ingegerd Östman-Smith and Dr Davood Javidgonbadi at Sahlgrenska University Hospital, University of Gothenburg, Sweden for providing parametric data for change in LVOT gradient in their study. Funding This work was supported by the National Institute for Health Research (NIHR) Biomedical Research Centre at Imperial College Healthcare NHS Trust and Imperial College London; Wellcome Trust (212183/Z/18/Z to J.P.H.); FS/14/27/30752 to M.J.S.S., FS/15/53/31615 to D.K., FS 04/079 to D.P.F.; and FS/13/44/30291 to Z.I.W. are supported by the British Heart Foundation. Conflict of interest: none declared.

Hypertrophic obstructive cardiomyopathy (HOCM) is the most common inherited cardiomyopathy, affecting approximately 1 in 500 individuals. The male predominance of the condition varies from 51% to 91%, suggesting other factors (i.e. environment, sex hormones, and epigenetics) affect the phenotype.1 Women with HOCM tend to be more symptomatic, present later in life, are more likely to have left ventricular outflow tract obstruction, and have greater mortality when < 50 years of age.2 Because the selection of treatment is based on symptom presentation, it is unclear if there is a sex bias in applying the criteria and/or outcomes independent of selection bias, and whether females’ benefit more from a particular therapy. Thus, an a priori protocol to determine if there were sex differences in selection of treatment and outcomes for HOCM was created for a systematic review to predefine population criteria, description of interventions, and comparisons of the outcomes of interest of three treatments for HOCM: surgical myectomy (SM), alcohol septal ablation (ASA), and dual chamber pacing (DDD) according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA).3 Electronic databases (MEDLINE, EMBASE, Cochrane Central Register of Controlled Trials, and Scopus) were searched for studies of a minimum of 5 adults who underwent SM, ASA, or DDD as a primary procedure from inception in 1946 to 30 December 2015. The detailed search strategy, list of studies included and discussion are reported in the Supplementary material online.

Central Register of Controlled Trials, and Scopus) were searched for studies of a minimum of 5 adults who underwent SM, ASA, or DDD as a primary procedure from inception in 1946 to 30 December 2015. The detailed search strategy, list of studies included and discussion are reported in the Supplementary material online. Sixty-three studies were included (Figure 1) reporting on 4586 patients: 1852 (40.4%) were male, 1780 (38.8%) were female, 954 (20.4%) were unidentifiable by sex. Of the total number of patients, 2212 (48.2%) underwent ASA, 1920 (41.9%) underwent SM and 454 (9.8%) underwent DDD. Of the 63 studies, 11 articles did not report sex in basic demographics, or grouped all treatments together, such that numbers of each sex by treatment could not be determined. Where sex was reported, females made up 847 (49.6%) of patients in ASA studies, 770 (48.5%) of patients in SM, and 163 (48.0%) of patients in DDD studies. Figure 1 Search strategy identifying articles for inclusion in the systematic review.

Sixty-three studies were included (Figure 1) reporting on 4586 patients: 1852 (40.4%) were male, 1780 (38.8%) were female, 954 (20.4%) were unidentifiable by sex. Of the total number of patients, 2212 (48.2%) underwent ASA, 1920 (41.9%) underwent SM and 454 (9.8%) underwent DDD. Of the 63 studies, 11 articles did not report sex in basic demographics, or grouped all treatments together, such that numbers of each sex by treatment could not be determined. Where sex was reported, females made up 847 (49.6%) of patients in ASA studies, 770 (48.5%) of patients in SM, and 163 (48.0%) of patients in DDD studies. Figure 1 Search strategy identifying articles for inclusion in the systematic review. Only 1 case series of 18 patients (9 males) treated by DDD reported outcomes by sex.4 In that study, there was no difference in mean gradient reduction following DDD pacing: males −58.5 (25.5) mmHg vs. females −55.7 (19.3) mmHg (P = 0.82). Similarly, reduction in New York Heart Association functional class did not differ by sex. None of the other studies stratified any of the baseline characteristics of patients by sex and there were minimal outcome data stratified by other confounders such as age and disease severity. Therefore, subgroup analyses based on sex and other patient characteristics that are prognostic effect modifiers were not possible.

of the other studies stratified any of the baseline characteristics of patients by sex and there were minimal outcome data stratified by other confounders such as age and disease severity. Therefore, subgroup analyses based on sex and other patient characteristics that are prognostic effect modifiers were not possible. A patient’s sex, age, stage of disease, and other comorbidities will influence choice of treatment and outcomes. Therefore, critical to evaluations of outcomes in treatment modalities is the accurate reporting of these characteristics. Sex is a basic biological variable that should be included in reporting of clinical outcomes even if the study is not powered to show a sex difference. In the USA, the 1993 Revitalization Act required inclusion of women in clinical studies but not in the reporting of data by sex. As medicine embraces a precision, personalized approach, reporting and analysis of data by sex and other important patient characteristics will inform the practice so that treatment approaches maximize patient outcomes. Requiring such reporting in future studies would accelerate the knowledge base to better inform patient selection and treatment strategies.5 Supplementary material Supplementary material is available at European Heart Journal—Quality of Care and Clinical Outcomes online. Supplementary Material Supplementary Data Click here for additional data file. Acknowledgements This letter was written in partial fulfilment of requirements for the Mayo Clinic Graduate School of Medicine Biomedical Engineering (BME) 6855 Tutorial in Cardiovascular Physiology course.

Supplementary material is available at European Heart Journal—Quality of Care and Clinical Outcomes online. Supplementary Material Supplementary Data Click here for additional data file. Acknowledgements This letter was written in partial fulfilment of requirements for the Mayo Clinic Graduate School of Medicine Biomedical Engineering (BME) 6855 Tutorial in Cardiovascular Physiology course. Funding This publication was made possible by support from the Mayo Clinic Graduate School of Medicine, NIH P50 AG044170 (to V.M.M.) and the Clinical and Translational Science Award Grant Number UL1 TR000135, supporting the Mayo Clinic Center for Clinical and Translational Science (CCaTS), from the National Center for Advancing Translational Sciences (NCATS), a component of NIH (The contents of this article are solely the responsibility of the authors and do not necessarily represent the official view of the NIH.). C.A.A.C. and V.K.S. are supported by NIH HL 65176. C.A.A.C. is supported by the American Heart Association (Award number 17POST33400211). Conflict of interest: none declared.

Introduction Peripheral artery disease (PAD) has been recognized as a major contributor to the cardiovascular (CV) health burden.1,2 Peripheral artery disease is a highly prevalent atherosclerotic syndrome affecting approximately 20% of people over 60 years of age in Sweden, and estimates [assessed using the ankle-brachial index (ABI)] have shown a recent increase in prevalence worldwide (of 23% during the last decade).3,4 Peripheral artery disease patients are at high risk of experiencing major CV events (MACE), which are associated with substantial impairment in quality of life and increased morbidity rates.5–8 Thus, PAD is associated with a substantial economic burden both in terms of prevention and treatment of MACE and when managing lower limb-related symptoms and procedures.2 Previous studies have found that PAD patients are even more costly than patients with coronary artery disease (CAD) or cerebrovascular disease (CVD), having a 2-year cumulative cost of nearly USD 12 000, where half of the hospitalization costs are limb-related and half are due to treatment of MACE.2 Despite the high prevalence of PAD, very few studies have investigated the long-term use of resources and costs after diagnosis. In addition, the relationship between costs related to PAD and total healthcare costs requires clarification.

he hospitalization costs are limb-related and half are due to treatment of MACE.2 Despite the high prevalence of PAD, very few studies have investigated the long-term use of resources and costs after diagnosis. In addition, the relationship between costs related to PAD and total healthcare costs requires clarification. As PAD is associated with high risk of MACE and mortality, with an increasing trend over time, the costs to healthcare in the long-term should also be acknowledged. In this observational study, we investigated CV outcome and long-term CV resource use and total healthcare costs for patients, before and after diagnosis of PAD in a Swedish nationwide setting. Methods Overview In this observational cohort study, we retrieved data from three mandatory Swedish nationwide registries: the Swedish National Patient Register (NPR), the Swedish Prescribed Drug Register (SPDR), and the Swedish Cause of Death Register. The Swedish NPR covers more than 99% of all somatic and psychiatric hospital discharges, with inpatient admission and discharge dates, and also main and secondary diagnoses according to the International Statistical Classification of Diseases and Related Health Problems, 10th Revision [ICD-10].8 The SPDR has data on all prescription medications dispensed by all pharmacies in Sweden.9 Individual patient-level data from the registers were linked using the mandatory and unique Swedish personal identification numbers, which were subsequently replaced with study identification numbers before further data processing.

he SPDR has data on all prescription medications dispensed by all pharmacies in Sweden.9 Individual patient-level data from the registers were linked using the mandatory and unique Swedish personal identification numbers, which were subsequently replaced with study identification numbers before further data processing. The study protocol was reviewed and approved by the regional ethics committee of the University of Gothenburg, Sweden (reference number: 649-14). Linkage of data was performed by the Swedish National Board of Health and Welfare. The linked database was managed by the Institute of Medicine at the Sahlgrenska Academy, Gothenburg, Sweden. Population All patients with a first time primary or secondary diagnosis of PAD in a hospital setting (as inpatient or outpatient) [ICD-10 I70.0 (atherosclerosis of aorta), I70.2 (atherosclerosis of arteries of extremities), or I73.9 (claudicatio intermittens)] between 2006 and 2013 were included. The index date was defined as the date of the first recorded PAD diagnosis during the specified observation period. Follow-up ended when a patient died or at the end of the observational period (January 2014). In Sweden, the diagnosis of PAD in a hospital setting is normally based on the medical history and on results of a clinical vascular examination including the ABI test. Baseline characteristics and data on medication use were retrieved from the NPR and SPDR registers. The population was stratified by age and risk profile at index date: Patients aged <65 years Patients aged 65–75 years Patients aged >75 years

Population All patients with a first time primary or secondary diagnosis of PAD in a hospital setting (as inpatient or outpatient) [ICD-10 I70.0 (atherosclerosis of aorta), I70.2 (atherosclerosis of arteries of extremities), or I73.9 (claudicatio intermittens)] between 2006 and 2013 were included. The index date was defined as the date of the first recorded PAD diagnosis during the specified observation period. Follow-up ended when a patient died or at the end of the observational period (January 2014). In Sweden, the diagnosis of PAD in a hospital setting is normally based on the medical history and on results of a clinical vascular examination including the ABI test. Baseline characteristics and data on medication use were retrieved from the NPR and SPDR registers. The population was stratified by age and risk profile at index date: Patients aged <65 years Patients aged 65–75 years Patients aged >75 years Patients with none of the following comorbidities in their previous medical history were defined as low-risk patients: diabetes mellitus, myocardial infarction (MI), stroke, heart failure, and chronic renal dysfunction. Patients with one or more of the following comorbidities were defined as high-risk patients: diabetes mellitus, MI, stroke, heart failure, or chronic renal dysfunction.

Patients with none of the following comorbidities in their previous medical history were defined as low-risk patients: diabetes mellitus, myocardial infarction (MI), stroke, heart failure, and chronic renal dysfunction. Patients with one or more of the following comorbidities were defined as high-risk patients: diabetes mellitus, MI, stroke, heart failure, or chronic renal dysfunction. Clinical outcomes The primary endpoint of MACEs was a composite of hospitalization with a main diagnosis of non-fatal MI (ICD-10: I21), non-fatal IS (ICD-0: I63-I64), or CV death (ICD-10 codes I00–I99). Lower limb revascularization was defined as an open or endovascular procedure as captured in NPR based on procedure codes (see Supplementary Material online). Resource use Data on hospitalizations and outpatient care visits were collected from the NPR. In cases where a subsequent hospitalization occurred without a calendar day between the discharge date and the new admission date, a single episode of hospitalization was recorded. When a patient had both a primary and a secondary diagnosis, the primary diagnosis defined the event type. Resource use associated with CV disease included hospitalizations, outpatient care visits, and drug use. All non-procedural lower limb-related events were included in the category ‘CV events’. Lower limb procedures included only invasive procedures for treatment of PAD. Non-CV-related care included all hospitalizations, outpatient care visits, and drug use that were not related to a diagnosis of CV as defined in ICD-10.

. All non-procedural lower limb-related events were included in the category ‘CV events’. Lower limb procedures included only invasive procedures for treatment of PAD. Non-CV-related care included all hospitalizations, outpatient care visits, and drug use that were not related to a diagnosis of CV as defined in ICD-10. The Prescribed Drug Register included data on dispensed, prescribed drugs in terms of substance, formulation, dose, and date of administration. Cardiovascular drugs included drugs in the ATC class C: anti-platelets, warfarin, statins, NOACs, nitrates, and anti-hypertensives. Non-CV drugs were defined as all drugs not included in the ATC class C. The major items of resource use and unit costs are listed in Supplementary material online, Table S4a–d. Unit costs Each recorded hospitalization and outpatient care visit was assigned a 2015 diagnosis-related group (DRG) weight, which was multiplied by the most recent 2015 cost per weight.11 In cases of missing DRG codes in the 2015 DRG catalogue, older DRG catalogues were used to apply the correct weight. If DRG codes recorded before 2015 had been stratified into several DRG codes in the 2015 DRG catalogue, a weighted average of these weights was applied. Irrespective of the year in which the DRG code was recorded, all costs were multiplied by the most recent cost per weight. The daily cost of a drug was calculated by multiplying the average dose by the most recent retail price available.12

Unit costs Each recorded hospitalization and outpatient care visit was assigned a 2015 diagnosis-related group (DRG) weight, which was multiplied by the most recent 2015 cost per weight.11 In cases of missing DRG codes in the 2015 DRG catalogue, older DRG catalogues were used to apply the correct weight. If DRG codes recorded before 2015 had been stratified into several DRG codes in the 2015 DRG catalogue, a weighted average of these weights was applied. Irrespective of the year in which the DRG code was recorded, all costs were multiplied by the most recent cost per weight. The daily cost of a drug was calculated by multiplying the average dose by the most recent retail price available.12 All costs were converted to euros using an average 2015 exchange rate, according to the European Central Bank: 1 euro (EUR) = 9.35 Swedish crowns (SEK). Analysis Baseline characteristics are presented as mean and standard deviation for continuous variables and absolute and relative frequencies for categorical variables. Follow-up data were collected from the time of the index diagnosis of PAD until death or the end of follow-up. The frequency and proportion of patients with the primary composite endpoint were assessed and a Kaplan–Meier analysis was performed to estimate the cumulative probability of the primary composite endpoint during study follow-up. If one patient had several events, only the first was used in the survival model. Results are presented as hazard ratios (HRs) and 95% confidence intervals (CIs).

oint were assessed and a Kaplan–Meier analysis was performed to estimate the cumulative probability of the primary composite endpoint during study follow-up. If one patient had several events, only the first was used in the survival model. Results are presented as hazard ratios (HRs) and 95% confidence intervals (CIs). Resource use was calculated for each year, i.e. 1 year before initial PAD diagnosis, the year after being diagnosed with PAD (starting from the hospital admission date, or the date recorded for the outpatient visit when the PAD diagnosis was established), and the 5 years that followed. Patients contributed to a particular year of analysis if they died during the year or had a full year of exposure. Thus, a patient dying after 1.5 years of follow-up contributed to Year 2 whereas patients who were censored at 1.5 years due to no more follow-up time did not. Mean healthcare costs per patient per year were estimated by applying unit costs to the corresponding resource use items. If a patient had both a PAD CV-related diagnosis and a lower limb procedure performed at the same visit, the costs were reported as being lower limb-related. Costs were differentiated into CV-related, non-CV-related, and lower limb-related and presented as subgroups stratified by a combination of risk profile and age. Statistical analyses were performed using SAS version 9.3 (SAS Institute Inc., Cary, NC, USA) and R version 3.2.3.

Mean healthcare costs per patient per year were estimated by applying unit costs to the corresponding resource use items. If a patient had both a PAD CV-related diagnosis and a lower limb procedure performed at the same visit, the costs were reported as being lower limb-related. Costs were differentiated into CV-related, non-CV-related, and lower limb-related and presented as subgroups stratified by a combination of risk profile and age. Statistical analyses were performed using SAS version 9.3 (SAS Institute Inc., Cary, NC, USA) and R version 3.2.3. Results Overall, 141 266 patients with a diagnosis of PAD were identified, 66 189 of whom had their first PAD diagnosis established during the observation period and were included in the study. Peripheral artery disease was mainly diagnosed at hospital outpatient visits (71%), and was the main reason for hospital contact in 77% of the patients. Mean length of follow-up was 2.8 years, with a maximum of 8 years, resulting in a total of 184 614 patient-years of follow-up.

n period and were included in the study. Peripheral artery disease was mainly diagnosed at hospital outpatient visits (71%), and was the main reason for hospital contact in 77% of the patients. Mean length of follow-up was 2.8 years, with a maximum of 8 years, resulting in a total of 184 614 patient-years of follow-up. The youngest and oldest patient groups with a high risk of CV had different profiles. Compared with subjects over 75 years of age, a higher proportion of subjects less than 65 years old were men (69% vs. 50%), had diabetes (71% vs. 53%), and had renal insufficiency (11% vs. 4%), whereas cancer (9% vs. 23%) and stroke (16% vs. 29%) were more prevalent in older patients. Statin use was more common in the youngest patients than in the oldest (75% vs. 39%), who in turn used more analgesics (49% vs. 70%, Table 1). A higher proportion of older women (over 75 years old) were categorized as being low-risk (61%) than women aged 75 years or younger (47%). Table 1 Description of analysis population after being diagnosed with peripheral artery disease

in the oldest (75% vs. 39%), who in turn used more analgesics (49% vs. 70%, Table 1). A higher proportion of older women (over 75 years old) were categorized as being low-risk (61%) than women aged 75 years or younger (47%). Table 1 Description of analysis population after being diagnosed with peripheral artery disease Age <65 high-risk Age <65 low-risk Age 65–75 high-risk Age 65–75 low-risk Age 75+ high-risk Age 75+ low-risk Total n = 5050 n = 5752 n = 10 733 n = 9908 n = 21 068 n = 13 678 n = 66 189 Age (SD) 59.5 (3.8) 59.3 (3.9) 70.4 (3.1) 70.1 (3.1) 84.0 (5.2) 83.4 (5.3) 75.6 (10.3) Gender (Female) 1567 (31.0) 2510 (43.6) 3719 (34.7) 4914 (49.6) 10 595 (50.3) 8297 (60.7) 31 602 (47.7) Aorta aneurysm 163 (3.2) 258 (4.5) 712 (6.6) 708 (7.1) 951 (4.5) 629 (4.6) 3421 (5.2) Diabetes 3594 (71.2) 0 (0.0) 6977 (65.0) 0 (0.0) 9840 (46.7) 0 (0.0) 20 411 (30.8) Hypertension 4284 (84.8) 2708 (47.1) 9585 (89.3) 6326 (63.8) 18 086 (85.8) 9397 (68.7) 50 386 (76.1) Myocardial infarction 1409 (27.9) 0 (0.0) 3189 (29.7) 0 (0.0) 6391 (30.3) 0 (0.0) 10 989 (16.6) Angina pectoris 1553 (30.8) 474 (8.2) 3700 (34.5) 1129 (11.4) 6754 (32.1) 1913 (14.0) 15 523 (23.5) Ischaemic stroke 794 (15.7) 0 (0.0) 2433 (22.7) 0 (0.0) 6040 (28.7) 0 (0.0) 9267 (14.0) Heart failure 1141 (22.6) 0 (0.0) 3303 (30.8) 0 (0.0) 10 464 (49.7) 0 (0.0) 14 908 (22.5) Atrial fibrillation 636 (12.6) 205 (3.6) 2495 (23.2) 808 (8.2) 8823 (41.9) 2256 (16.5) 15 223 (23.0) Major organ specific bleedings 433 (8.6) 231 (4.0) 1088 (10.1) 553 (5.6) 2941 (14.0) 1161 (8.5) 6407 (9.7) Chronic renal insufficiency 548 (10.9) 0 (0.0) 862 (8.0) 0 (0.0) 951 (4.5) 0 (0.0) 2361 (3.6) Chronic obstructive pulmonary disease 428 (8.5) 325 (5.7) 1448 (13.5) 961 (9.7) 2310 (11.0) 1027 (7.5) 6499 (9.8) Cancer 449 (8.9) 563 (9.8) 1822 (17.0) 1750 (17.7) 4883 (23.2) 3013 (22.0) 12 480 (18.9) Anti-platelets 3711 (73.5) 3689 (64.1) 7974 (74.3) 6767 (68.3) 14 893 (70.7) 8677 (63.4) 45 711 (69.1) Clopidogrel 731 (14.5) 333 (5.8) 1463 (13.6) 638 (6.4) 2166 (10.3) 769 (5.6) 6100 (9.2) Low dose aspirin 3525 (69.8) 3564 (62.0) 7455 (69.5) 6453 (65.1) 13 856 (65.8) 8233 (60.2) 43 086 (65.1) Warfarin 479 (9.5) 243 (4.2) 1694 (15.8) 608 (6.1) 3843 (18.2) 1169 (8.5) 8036 (12.1) Statins 3793 (75.1) 3383 (58.8) 7816 (72.8) 6168 (62.3) 8985 (42.6) 5301 (38.8) 35 446 (53.6) Anti-hypertensives 4429 (87.7) 3095 (53.8) 9938 (92.6) 7007 (70.7) 19 842 (94.2) 10 908 (79.7) 55 219 (83.4) Anti-diabetics 3263 (64.6) 17 (0.3) 6159 (57.4) 19 (0.2) 7915 (37.6) 20 (0.1) 17 393 (26.

8.2) 1169 (8.5) 8036 (12.1) Statins 3793 (75.1) 3383 (58.8) 7816 (72.8) 6168 (62.3) 8985 (42.6) 5301 (38.8) 35 446 (53.6) Anti-hypertensives 4429 (87.7) 3095 (53.8) 9938 (92.6) 7007 (70.7) 19 842 (94.2) 10 908 (79.7) 55 219 (83.4) Anti-diabetics 3263 (64.6) 17 (0.3) 6159 (57.4) 19 (0.2) 7915 (37.6) 20 (0.1) 17 393 (26. 3) Analgesics 2478 (49.1) 2124 (36.9) 5742 (53.5) 3957 (39.9) 14 851 (70.5) 8017 (58.6) 37 169 (56.2) All data are n (%) unless stated otherwise. SD, standard derivation. The overall 1-year cumulative incidence rates of the primary composite CV endpoint (MI, stroke, or CV death) and all-cause death were 16.6% and 21.1%, respectively. In patients who were 75 years old or younger, the 1-year cumulative incidence rate for the primary composite endpoint was 12.2% in high-CV-risk patients and 4.0% in low-CV-risk patients. Corresponding figures for patients over 75 years of age were 31.4% in high-CV-risk patients and 14.7% in low-CV-risk patients (Figure 1). Figure 1 Kaplan–Meier estimate of the risk of the composite primary endpoint (myocardial infarction, ischaemic stroke, or cardiovascular death) in different age and risk categories.

In patients who were 75 years old or younger, the 1-year cumulative incidence rate for the primary composite endpoint was 12.2% in high-CV-risk patients and 4.0% in low-CV-risk patients. Corresponding figures for patients over 75 years of age were 31.4% in high-CV-risk patients and 14.7% in low-CV-risk patients (Figure 1). Figure 1 Kaplan–Meier estimate of the risk of the composite primary endpoint (myocardial infarction, ischaemic stroke, or cardiovascular death) in different age and risk categories. Procedures performed In total, 23 481 lower limb revascularization procedures were performed during the study period. The cumulative incidence rate of lower limb revascularization procedures for the full study population was 23.2 (95% CI 22.9–23.5) at 1 year after being diagnosed with PAD. The cumulative probability of lower limb revascularization was 20.1 (95% CI 19.8–20.4) at 6 months and 27.6 (27.2–27.9) at 3 years (see Supplementary material online, Table S2). A larger proportion of high-risk patients underwent amputations, whereas the proportion of patients who underwent lower limb revascularizations was more similar across the low- and high-CV-risk populations and age categories (see Supplementary material online, Table S5).

s (see Supplementary material online, Table S2). A larger proportion of high-risk patients underwent amputations, whereas the proportion of patients who underwent lower limb revascularizations was more similar across the low- and high-CV-risk populations and age categories (see Supplementary material online, Table S5). Pattern of resource use One year before diagnosis of PAD, the mean total number of contacts per patient (i.e. hospitalization and outpatient care visits) was 4.05, with outpatient visits being the main reason for contact (mean number: 3.21) (Table 2). In the year following diagnosis of PAD, the mean total number of contacts increased to 6.36, with outpatient visits being the main reason for contact (mean number: 4.99). During the year after diagnosis of PAD, the mean number of CV-related hospitalizations and outpatient visits was 2.30, with PAD being the main reason for contact. The mean number of lower limb procedure contacts was 0.38 in the year after diagnosis of PAD, which became reduced to 0.04 in the subsequent years. For the CV-related long-term drug therapy [such as low-dose aspirin, angiotensin-converting enzyme (ACE) inhibitors, and statins], the average number of days on drug continued to be higher from the second year after the year of being diagnosed with PAD compared with the year before the PAD diagnosis.

Pattern of resource use One year before diagnosis of PAD, the mean total number of contacts per patient (i.e. hospitalization and outpatient care visits) was 4.05, with outpatient visits being the main reason for contact (mean number: 3.21) (Table 2). In the year following diagnosis of PAD, the mean total number of contacts increased to 6.36, with outpatient visits being the main reason for contact (mean number: 4.99). During the year after diagnosis of PAD, the mean number of CV-related hospitalizations and outpatient visits was 2.30, with PAD being the main reason for contact. The mean number of lower limb procedure contacts was 0.38 in the year after diagnosis of PAD, which became reduced to 0.04 in the subsequent years. For the CV-related long-term drug therapy [such as low-dose aspirin, angiotensin-converting enzyme (ACE) inhibitors, and statins], the average number of days on drug continued to be higher from the second year after the year of being diagnosed with PAD compared with the year before the PAD diagnosis. Healthcare costs The mean annual total cost of healthcare in the year before the diagnosis of PAD was €6577, of which €1710 (26%) were CV event-related hospitalization costs and outpatient visits and €3748 (57%) were non-CV-related hospitalization costs and outpatient visits. Drug therapy was responsible for 17% of the total.

47–0.76) Warfarin at baseline 0.45 0.001 2.54 0.030 0.57 0.013 0.58 0.175 0.73 0.037 (0.27–0.73) (1.10–5.90) (0.36–0.89) (0.26–1.28) (0.54–0.98) Antiplatelet use 0.98 0.943 1.44 0.089 1.05 0.778 1.31 0.395 1.14 0.281 (0.62–1.55) (0.95–2.20) (0.73–1.52) (0.71–2.42) (0.90–1.44) Vascular disease = coronary artery disease. CrCl, creatinine clearance; HR, hazard ratio; CI, confidence interval; TIA, transient ischaemic attack. aThromboembolism, major haemorrhage, and all-cause death. bVersus CrCl ≥80 mL/min. Predictive ability of renal function for events The c-indices of CrCl values were 0.61 (95% CI 0.56–0.66, P < 0.001) for thromboembolism, 0.60 (95% CI 0.55–0.65, P < 0.001) for major haemorrhage, 0.75 (95% CI 0.71–0.79, P < 0.001) for all-cause death, 0.76 (95% CI 0.69–0.83, P < 0.001) for cardiovascular death, and 0.67 (95% CI 0.64–0.70, P < 0.001) for composite events. The sensitivity and specificity of cut-off CrCl points for events are shown in Supplementary material online, Table S5. Although the sensitivity of CrCl <30 mL/min was less than 30%, the specificity was more than 90% for all events except for major haemorrhage. The sensitivity of CrCl <50 mL/min ranged from 37% to 69%, but the specificity was around 75% for all events (see Supplementary material online, Table S5).

are costs The mean annual total cost of healthcare in the year before the diagnosis of PAD was €6577, of which €1710 (26%) were CV event-related hospitalization costs and outpatient visits and €3748 (57%) were non-CV-related hospitalization costs and outpatient visits. Drug therapy was responsible for 17% of the total. During the year after PAD diagnosis, there was a 90% increase in the mean total costs for all patient age and risk groups, totalling €12 549. Thirty per cent of this was attributed to CV-related hospitalizations and outpatient visits (€3824), with PAD-related follow-up being the main reason for hospital attendance (Table 2). Also, the number of lower limb-related invasive procedures increased during this year, with a total mean cost of €3201. Non-CV-related costs were not substantially different from those in the year before the diagnosis of PAD. Table 2 Resource use pattern over time, year 1 being first year after peripheral artery disease diagnosis 1 year prior to PAD diagnosis Year after diagnosis 1 2 3 4 5 6 Number of patients 66,189 53,024 42,032 32,547 24,338 17,610 11.938 Hospitalizations CV related care 0,27   0,47 0,18 0,17 0,15 0,15 0,14 Lower limb procedures 0,03   0,35 0,07 0,05 0,04 0,04 0,04 Non-CV related care 0,54   0,55 0,4 0,38 0,37 0,35 0,35 Outpatient care visits CV related care 0,32   1,83 0,53 0,4 0,36 0,35 0,35 Lower limb procedures 0   0,03 0,01 0 0 0 0 Non-CV related care 2,89   3,13 2,58 2,41 2,37 2,34 2,28

1 year prior to PAD diagnosis Year after diagnosis 1 2 3 4 5 6 Number of patients 66,189 53,024 42,032 32,547 24,338 17,610 11.938 Hospitalizations CV related care 0,27   0,47 0,18 0,17 0,15 0,15 0,14 Lower limb procedures 0,03   0,35 0,07 0,05 0,04 0,04 0,04 Non-CV related care 0,54   0,55 0,4 0,38 0,37 0,35 0,35 Outpatient care visits CV related care 0,32   1,83 0,53 0,4 0,36 0,35 0,35 Lower limb procedures 0   0,03 0,01 0 0 0 0 Non-CV related care 2,89   3,13 2,58 2,41 2,37 2,34 2,28 Pharmaceuticals Anti-platelets 169   245 238 238 239 237 235 Clopidogrel 14   30 24 24 24 25 26 Low dose ASA 155   226 220 219 220 218 215 Anticoagulants 20   23 24 24 24 24 26 Statins 105   183 179 182 184 186 184 Anti-hypertensives 270   281 281 282 282 283 284 Anti-diabetics 77   79 79 79 79 79 77 Analgesics 79   101 87 83 81 80 80 CV related care, lower limb procedures and non-CV related care resource utilization are reported in mean numbers of contacts for hospitalisations and outpatient care visits. Drug usage are reported in mean number of days (DDD). The mean total healthcare cost decreased from the second year after diagnosis of PAD and onwards, with lower mean total annual costs (€5750) than the year before PAD diagnosis. However, lower limb-related procedure costs remained higher throughout the study period, with a mean total annual cost of €728. The mean annual CV-related cost was €1140 after the first year of being diagnosed with PAD.

s of PAD and onwards, with lower mean total annual costs (€5750) than the year before PAD diagnosis. However, lower limb-related procedure costs remained higher throughout the study period, with a mean total annual cost of €728. The mean annual CV-related cost was €1140 after the first year of being diagnosed with PAD. High-risk CV patients had higher total healthcare costs than low-risk CV patients after diagnosis of PAD, the mean annual costs being €7439 and €4063, respectively. Also, the mean CV-related hospitalization cost was higher in the high-CV-risk group than in the low-risk CV group: €1442 as opposed to €838. After patients were diagnosed with PAD, CV drug treatment contributed least to healthcare costs in all the years studied (mean annual cost: €200). Both CV drugs and non-CV drugs showed a similar trend, with a higher observed cost in high-risk patients.

High-risk CV patients had higher total healthcare costs than low-risk CV patients after diagnosis of PAD, the mean annual costs being €7439 and €4063, respectively. Also, the mean CV-related hospitalization cost was higher in the high-CV-risk group than in the low-risk CV group: €1442 as opposed to €838. After patients were diagnosed with PAD, CV drug treatment contributed least to healthcare costs in all the years studied (mean annual cost: €200). Both CV drugs and non-CV drugs showed a similar trend, with a higher observed cost in high-risk patients. High-risk patients had higher costs associated with lower limb-related procedures (mean total: £3952) than low-risk patients (mean total: €2605)―and for amputation in particular (€1703 vs. €629) (Figure 3). The selected CV-related costs were high in all risk groups and age categories, with a mean for all groups of €2071. In all patients, PAD-related costs (not including limb-related procedures) were the greatest costs within the selected CV category (52%), with coronary events and stroke (32%), and heart failure (13%) being observed as the other major CV cost drivers. Also, in the years that followed, total PAD-related costs remained the most important cost contributor among the different CV-related costs, although there was a shift in PAD costs to a larger proportion of limb procedure-related costs over time (Figure 4).

ilure (13%) being observed as the other major CV cost drivers. Also, in the years that followed, total PAD-related costs remained the most important cost contributor among the different CV-related costs, although there was a shift in PAD costs to a larger proportion of limb procedure-related costs over time (Figure 4). After being diagnosed with PAD, lower limb procedure-related costs were an annual major cost driver in the study population over time (mean: €728), with lower limb revascularizations being the main cost contributor (mean: €474) (Figure 4). The difference in lower limb procedure costs in high-risk and low-risk patients was mainly caused by the fact that there were more amputations in the high-risk CV population (Figure 3 and see Supplementary material online, Table S5). Discussion One-third of the PAD population was over 75 years of age and was categorized as high-risk, but even among patients aged less than 75 years, more than 50% could be classified as high-risk, with diabetes and a history of coronary events being the most prevalent comorbidities. Within a year after diagnosis of PAD, more than one in five patients died and one in six experienced a MACE. Compared with patients surviving an MI, PAD patients had a significantly higher 1-year mortality risk (21.1% vs. 13.2%) and showed a comparable CV risk (16.6% vs. 18.3%).13,14

g the most prevalent comorbidities. Within a year after diagnosis of PAD, more than one in five patients died and one in six experienced a MACE. Compared with patients surviving an MI, PAD patients had a significantly higher 1-year mortality risk (21.1% vs. 13.2%) and showed a comparable CV risk (16.6% vs. 18.3%).13,14 In high-risk patients, the 1-year risk of CV events was increased three-fold for those less than 75 years old and doubled for those over 75 years, as compared with PAD patients without risk factors. The resource use and pattern of costs was associated with age and underlying risk, with the latter being the most important determinant of costs, as has also been observed in MI patients.15 This study reports only hospitalization costs (including hospital-based outpatient visits), but other drivers of the total healthcare costs for these patients, as for example nursing home and primary healthcare costs were not included. Furthermore, wider data on community-care and patients' own costs and productivity impacts are not included.

hospitalization costs (including hospital-based outpatient visits), but other drivers of the total healthcare costs for these patients, as for example nursing home and primary healthcare costs were not included. Furthermore, wider data on community-care and patients' own costs and productivity impacts are not included. Costs of hospitalizations and outpatient visits related to PAD were the greatest of the CV-related costs, particularly during the year after PAD was diagnosed. However, non-CV-related hospitalizations were the largest cost contributor overall, being approximately twice as frequent in Year 2 after PAD diagnosis, with five times as many outpatient care visits, as compared with CV-related visits. Interestingly, although the PAD population has a well-recognized high risk of CV, the major part of the hospitalization costs for PAD patients (including outpatient visits) is not related to CV diseases―with, for example, costs associated with diabetes and chronic renal insufficiency being larger cost contributors (Figure 3). It may not be relevant to focus only on CV-related risk prevention separately, but it is perhaps better to have a broader view when assessing risk and potential interventions for this patient population.

r example, costs associated with diabetes and chronic renal insufficiency being larger cost contributors (Figure 3). It may not be relevant to focus only on CV-related risk prevention separately, but it is perhaps better to have a broader view when assessing risk and potential interventions for this patient population. Despite generally having a higher CV baseline risk and more CV events than younger patients, patients over 75 years of age generally had lower CV-related costs. This might be explained by the fact that a lower proportion of elderly patients undergo expensive invasive heart related procedures as percutaneous coronary intervention or coronary artery bypass grafting in Sweden.16 Also, the lower limb procedure-related costs, especially for amputations, and non-CV-related costs were substantially higher in the youngest age group (< 65 years), which may have been attributable to the high prevalence of diabetes (71%). The total annual CV-related costs―excluding lower limb procedure costs―for PAD patients during long-term follow-up were higher than they are for MI patients, with mean of €1945 per patient as opposed to approximately €1700–1800 per patient,15 an effect of the progressive, chronic nature of PAD.

Despite generally having a higher CV baseline risk and more CV events than younger patients, patients over 75 years of age generally had lower CV-related costs. This might be explained by the fact that a lower proportion of elderly patients undergo expensive invasive heart related procedures as percutaneous coronary intervention or coronary artery bypass grafting in Sweden.16 Also, the lower limb procedure-related costs, especially for amputations, and non-CV-related costs were substantially higher in the youngest age group (< 65 years), which may have been attributable to the high prevalence of diabetes (71%). The total annual CV-related costs―excluding lower limb procedure costs―for PAD patients during long-term follow-up were higher than they are for MI patients, with mean of €1945 per patient as opposed to approximately €1700–1800 per patient,15 an effect of the progressive, chronic nature of PAD. Not surprisingly, the contributors to CV-associated costs are somewhat different in the MI and PAD populations. Myocardial infarction patients have more recurrent MIs, while PAD patients have more recurrent PAD manifestations with relatively fewer MIs. This is supported by the observation that the PAD-related costs due to hospitalizations and outpatient visits were the main contributors to CV-related costs for all patient categories, contributing to more than 50% of the CV-related costs in first year after diagnosis of PAD. In total, approximately 23 500 revascularizations were performed, and the majority within the first 6 months, which would explain the decline in PAD-related costs over time.

butors to CV-related costs for all patient categories, contributing to more than 50% of the CV-related costs in first year after diagnosis of PAD. In total, approximately 23 500 revascularizations were performed, and the majority within the first 6 months, which would explain the decline in PAD-related costs over time. Lower limb-related procedure costs were a significant overall cost contributor at Year 1, both for low- and high-CV-risk patients (Figures 2–4), but they decreased over time to be comparable with other studies where PAD procedure-related costs constitute only a modest fraction.1 However, costs associated with amputations are higher in the high-CV-risk groups than in low-CV-risk patients, whereas the costs of lower limb revascularization are more similar in the different patient groups. This might be related to the inherently worse limb prognosis in patients with PAD in combination with diabetes, cardiac failure, or kidney failure, even when successful lower limb revascularization procedures are undertaken, due to having more severe lesions. Figure 2 Annual costs per patient prior to and after peripheral artery disease (PAD), by cost category, age, and risk. cardiovascular (CV)-related: includes all ICD-10 CV ‘I’ diagnoses except PAD-related costs in combination with lower limb procedures. If a PAD patient had a hospitalization with a PAD diagnosis ‘I’ and a lower limb procedure, then the cost for this visit is reported as being lower limb procedure-related. Non-CV-related: all costs except costs related to CV (ICD-10 ‘I’).

‘I’ diagnoses except PAD-related costs in combination with lower limb procedures. If a PAD patient had a hospitalization with a PAD diagnosis ‘I’ and a lower limb procedure, then the cost for this visit is reported as being lower limb procedure-related. Non-CV-related: all costs except costs related to CV (ICD-10 ‘I’). Figure 3 Mean costs per patient during the first year after diagnosis of peripheral artery disease, by selected cost category, age, and risk. Coronary events: myocardial infarction and unstable angina pectoris. PAD-related, peripheral arterial disease (follow-up, not including lower limb procedures); COPD, chronic obstructive pulmonary disease. Figure 4 Cumulative cost during follow-up of selected diagnoses and procedures. Ischaemic event-related: myocardial infarction, unstable angina pectoris, and stroke. Peripheral arterial disease follow-up (non-procedural): follow-up of peripheral arterial disease, not including lower limb procedure. It is difficult to compare healthcare costs due to differences in study design and healthcare systems, but our data on total costs for the combination of CV-related and lower limb-related procedures are comparable to what has been reported previously for PAD patients in France and Germany,17 but they are lower than data from the USA.2

to compare healthcare costs due to differences in study design and healthcare systems, but our data on total costs for the combination of CV-related and lower limb-related procedures are comparable to what has been reported previously for PAD patients in France and Germany,17 but they are lower than data from the USA.2 Cardiovascular-related drug costs contributed least among the cost categories investigated. This is partly explained by the fact that most drugs given in association with CV disease today are generic, and have a low acquisition cost. Another contributing factor may be the still uncommon use of cardioprotective medications in PAD. The present study had some limitations. Firstly, we did not have access to data describing the extent and severity of PAD, which may have an impact on the cost of treatment. Furthermore, the resource use and costs were divided into CV-related and non-CV-related, with a rather narrow definition of CV-related hospitalizations and outpatient care visits. A hospitalization was assigned an ICD-10 circulatory system diagnosis as the primary diagnosis to be categorized as CV, excluding CV-related hospitalization costs when attributable as for example a secondary diagnosis. As a registry data-based analysis, the study relied on ICD-10 codes for morbidity data, so the possibility of coding errors cannot be completely ruled out.

ystem diagnosis as the primary diagnosis to be categorized as CV, excluding CV-related hospitalization costs when attributable as for example a secondary diagnosis. As a registry data-based analysis, the study relied on ICD-10 codes for morbidity data, so the possibility of coding errors cannot be completely ruled out. These data, however, provide a comprehensive description of the outcome, use of healthcare resources, and costs over time for all patients with a hospital diagnosis of PAD in a longitudinal, nationwide setting. These results provide information that will be useful for future healthcare planning and allocation of resources. Conclusions Data from this nationwide study showed that almost 50% of PAD patients aged below 75 years who were diagnosed in a hospital setting had additional CV risk factors. One in five patients died within a year after PAD diagnosis. The presence of additional risk factors other than age was the main driver for both CV-related and non-CV-related costs. Peripheral artery disease-related costs including hospitalizations and outpatient care visits were the main contributory CV-related costs in the first year after diagnosis of PAD. Also, lower limb procedure-related costs were initially high, and remained so during subsequent follow-up of these patients. Although the PAD population has a well-recognized high-CV risk, the major proportion of hospitalization costs for PAD patients are not related to CV disease. Healthcare systems will need to consider preventive strategies and optimize costs of prevention in the growing PAD population.

equent follow-up of these patients. Although the PAD population has a well-recognized high-CV risk, the major proportion of hospitalization costs for PAD patients are not related to CV disease. Healthcare systems will need to consider preventive strategies and optimize costs of prevention in the growing PAD population. Supplementary material Supplementary material is available at European Heart Journal – Quality of Care and Clinical Outcomes online. Supplementary Material Supplementary Tables Click here for additional data file. Acknowledgements We thank Urban Olsson at Statisticon for being data manager in this study. English language assistance was provided by Alistair Kidd (Good Written English GWE AB). Funding The study was sponsored by AstraZeneca. Conflict of interest: P.H., T.K., and S.J. are employed by AstraZeneca. M.T. is employed at Statisticon, of which AstraZeneca is a client. J.N, B.K, M.F and B.S report no conflict of interest.

Introduction Each year, >10 000 Norwegians have a percutaneous coronary intervention (PCI) after either a myocardial infarction (MI) or angina. Antiplatelet therapy with 12 months use of the antiplatelet drug clopidogrel has, for decades, been considered the standard treatment after a PCI in order to prevent MI and death. Recently, two new antiplatelet drugs, prasugrel and ticagrelor, have been introduced. Both of these two drugs have proved to be efficacious with regards to, for example, MIs, but some concerns have also been raised with regards to increased risk of bleeding. 1 , 2 Hence, there is uncertainty as to the risk–benefit trade-off. Prices of the drugs are also an issue. Clopidogrel is no longer patented and prices have decreased considerably. The more newly developed prasugrel and ticagrelor are considerably more expensive; for instance, these cost >20 times as much in the UK ( http://www.nice.org.uk/guidance/ta236/resources/ta236-acute-coronary-syndromes-ticagrelor-costing-template ), and in Norway, these are >4 times as expensive ( http://legemiddelverket.no/Blaa_resept_og_pris/Helseoekonomiske%20rapporter/Documents/2012-2011/Brilique_Akutt-Koronarsyndrom_2011.pdf ). Similar differences are seen all across Europe. In 2014, >36 000 patients in Norway used either clopidogrel, prasugrel, or ticagrelor. Of these, 71% used clopidogrel, 24% used ticagrelor, and 5% used prasugrel. 3 Hence, it seems that clopidogrel still is the preferred antiplatelet therapy among doctors.

Each year, >10 000 Norwegians have a percutaneous coronary intervention (PCI) after either a myocardial infarction (MI) or angina. Antiplatelet therapy with 12 months use of the antiplatelet drug clopidogrel has, for decades, been considered the standard treatment after a PCI in order to prevent MI and death. Recently, two new antiplatelet drugs, prasugrel and ticagrelor, have been introduced. Both of these two drugs have proved to be efficacious with regards to, for example, MIs, but some concerns have also been raised with regards to increased risk of bleeding. 1 , 2 Hence, there is uncertainty as to the risk–benefit trade-off. Prices of the drugs are also an issue. Clopidogrel is no longer patented and prices have decreased considerably. The more newly developed prasugrel and ticagrelor are considerably more expensive; for instance, these cost >20 times as much in the UK ( http://www.nice.org.uk/guidance/ta236/resources/ta236-acute-coronary-syndromes-ticagrelor-costing-template ), and in Norway, these are >4 times as expensive ( http://legemiddelverket.no/Blaa_resept_og_pris/Helseoekonomiske%20rapporter/Documents/2012-2011/Brilique_Akutt-Koronarsyndrom_2011.pdf ). Similar differences are seen all across Europe. In 2014, >36 000 patients in Norway used either clopidogrel, prasugrel, or ticagrelor. Of these, 71% used clopidogrel, 24% used ticagrelor, and 5% used prasugrel. 3 Hence, it seems that clopidogrel still is the preferred antiplatelet therapy among doctors. Given the risk–benefit trade-off generated by the fact that each of the three available drugs are significantly better than at least one of the other drugs, and considerable cost difference, there is great uncertainty with regards to which of the three antiplatelet drugs offer the greatest value for money. Although some cost-effectiveness analyses have been conducted comparing each of the new drugs with clopidogrel, 4 no health economic evaluation has compared the cost-effectiveness of all these drugs with each other.

at uncertainty with regards to which of the three antiplatelet drugs offer the greatest value for money. Although some cost-effectiveness analyses have been conducted comparing each of the new drugs with clopidogrel, 4 no health economic evaluation has compared the cost-effectiveness of all these drugs with each other. Our objective was to compare the cost-effectiveness of different antiplatelet drugs for patients who have undergone PCI. Methods We modified a previously developed probabilistic Markov model (MOCCA—Model Of Cost-effectiveness of CArdiac Disease) to fit the current research question. 5 The model applies a lifelong healthcare payer perspective after a PCI operation, including risk of MI, major bleeding, new revascularization (PCI or coronary artery bypass graft), and death ( Figure 1 ). Figure 1 Model structure. The model was built to model half-year cycles from the age of 50 onwards to 105 years old. In our base case analysis, we analysed 60 year olds until aged 100, hence 80 cycles. Within-cycle correction was done using Simpsons 1/3rd methods, as recommended by Elbasha and Chhatwal. 6 Efficacy data of prasugrel and ticagrelor compared with clopidogrel were based on the two licensing phase III randomized controlled trials including 13 608 and 18 624 participants, respectively. 1 , 2 Outcomes included significant reductions in risk of MI for both drugs, increased risk of bleeding and reduced risk of revascularization with prasugrel, and reduced overall mortality with ticagrelor (

n the two licensing phase III randomized controlled trials including 13 608 and 18 624 participants, respectively. 1 , 2 Outcomes included significant reductions in risk of MI for both drugs, increased risk of bleeding and reduced risk of revascularization with prasugrel, and reduced overall mortality with ticagrelor ( Table 1 ). Due to different reporting in trials, effect on revascularization in our model was based on data on urgent target vessel revascularization from TRITON Thrombolysis in Myocardial Infarction Study Group (TIMI) 38 and recurrent ischaemia from PLATO invasive, as has also been done in a meta-analysis of these drugs. 7 Table 1 Effect of new drugs compared with clopidogrel Hazard ratio Confidence interval Distribution Source Prasugrel vs. clopidogrel Urgent target vessel revascularization 0.66 0.54–0.81 Log-normal (−0.4155, 0.1034) TRITON TIMI-38 1 Death from cardiovascular causes 0.89 0.70–1.12 Log-normal (−0.1165, 0.1199) TRITON TIMI-38 1 Non-fatal MI 0.76 0.67–0.85 Log-normal (−0.2744, 0.0607) TRITON TIMI-38 1 Major TIMI bleeding 1.32 1.03–1.68 Log-normal (0.2776, 0.1248) TRITON TIMI-38 1 Ticagrelor vs. clopidogrel Recurrent ischaemia 0.93 0.82–1.05 Log-normal (−0.0726, 0.0631) PLATO invasive 2 Death from cardiovascular causes 0.79 0.69–0.91 Log-normal (−0.2357, 0.0706) PLATO invasive 2 Non-fatal MI 0.84 0.75–0.95 Log-normal (−0.1744, 0.0603) PLATO invasive 2 Major TIMI bleeding 1.03 0.93–1.15 Log-normal (0.0296, 0.0542) PLATO invasive 2

t ischaemia 0.93 0.82–1.05 Log-normal (−0.0726, 0.0631) PLATO invasive 2 Death from cardiovascular causes 0.79 0.69–0.91 Log-normal (−0.2357, 0.0706) PLATO invasive 2 Non-fatal MI 0.84 0.75–0.95 Log-normal (−0.1744, 0.0603) PLATO invasive 2 Major TIMI bleeding 1.03 0.93–1.15 Log-normal (0.0296, 0.0542) PLATO invasive 2 Costs of all three antiplatelet drugs are based on current prices from the Norwegian Medicines Agency: 8 €207 per year for clopidogrel (300 mg loading dose, thereafter 75 mg per day), €509 per year for prasugrel (60 mg loading dose, thereafter 10 mg per day), and €817 per year for ticagrelor (180 mg loading dose, thereafter 90 mg per day). Costs of treatment [acute myocardial infarction (AMI), revascularization and bleeding] were based on items from the original publication, 5 but all costs of services were based on fees for 2015 ( Table A2 ). We have chosen a Norwegian healthcare sector perspective for our analyses, as recently suggested in a Norwegian white paper on prioritizing. 9 Costs are converted into Euros (€) using average rate of 2014 (€1 = NOK 8.35). Effectiveness is in our base case analysis measured as life years gained (LYG), which is one of the two acceptable measures in Norwegian cost-effectiveness analyses. 10 The other acceptable measure of effect in Norwegian health economic evaluations is quality-adjusted life years (QALYs). All weights were based on EQ-5D data, health states as derived in a Norwegian study by Pettersen et al.11 and events as used in a previous Norwegian economic evaluation. 12

-effectiveness analyses. 10 The other acceptable measure of effect in Norwegian health economic evaluations is quality-adjusted life years (QALYs). All weights were based on EQ-5D data, health states as derived in a Norwegian study by Pettersen et al.11 and events as used in a previous Norwegian economic evaluation. 12 Incremental cost-effectiveness ratios (ICERs) are regarded as cost-effective if below NOK 588 000 (€70 000) per LYG or QALYs, as recommended by the Norwegian Directorate of Health. 13 Recently, it was suggested in a white paper that the Norwegian threshold should rather be NOK 250 000 (€30 000) to better be in line with the opportunity cost. 9 We compared our results also with this suggested threshold. All costs and health benefits were discounted at 4% as recommended in Norwegian guidelines for health economic evaluations. 10 We incorporated probability distributions on all parameters in the model that were considered uncertain ( Table 1 , A1, and A2 ). Choice of distribution types was based largely on knowledge of how data of different types are usually distributed, as recommended by Briggs et al . 14 Uncertainty within each distribution was based on confidence intervals, where available. All results shown are based on probabilistic analyses, meaning that we sampled 10 000 times from all probability distributions in the model and calculated means of model outcomes based on these samples.

mmended by Briggs et al . 14 Uncertainty within each distribution was based on confidence intervals, where available. All results shown are based on probabilistic analyses, meaning that we sampled 10 000 times from all probability distributions in the model and calculated means of model outcomes based on these samples. When modelling disease progression with different health states, there is often a possibility of double counting. In our model, and in the trials we based the difference in effectiveness on, there are possibilities that patients were classified as having both MI and revascularization or both MI and death. In addition to our base case simulation, we therefore performed scenario analyses with effect on fewer outcomes in the model. We did analyses eliminating either effect on revascularization or mortality or both, due to the possibility of overlap between these outcomes and AMI. Risk of revascularization and bleeding was assumed to be the same in the first half year and in later half-year periods due to lack of data. Since these risks are likely to be lower in later half-year periods after the initial PCI, we performed separate analyses with these probabilities being 0 in all years post the first year. We also conducted analyses where the model was run for only 5 years to explore whether results differ between a model based mostly on data (5 years) and a model based on assumptions with regards to extrapolation. To explore whether cost-effectiveness would be different across ages, we also performed simulations for 50-, 70- and 80-year olds.

We also conducted analyses where the model was run for only 5 years to explore whether results differ between a model based mostly on data (5 years) and a model based on assumptions with regards to extrapolation. To explore whether cost-effectiveness would be different across ages, we also performed simulations for 50-, 70- and 80-year olds. We performed one deterministic sensitivity analysis reducing the prices of the drugs not considered cost-effective in our base case analysis, to see how low these prices had to be for the drugs to be cost-effective. Results Sixty-year-old patients undergoing PCI had a life expectancy of 11.96 years (discounted) if treated with clopidogrel the first year. The treatment with prasugrel increased the life expectancy to 12.32 years, while ticagrelor resulted in 12.70 years. Prasugrel was cost-effective compared with clopidogrel, and ticagrelor was cost-effective compared with both prasugrel and clopidogrel ( Table 2 ). Table 2 Lifetime costs and effects (incrementals compared with strategy above) Lifetime cost (€) Incremental cost (€) Life expectancy Life years gained Incremental cost-effectiveness ratio Clopidogrel 19 929 11.96 Prasugrel 22 649 2720 12.32 0.36 7505 €/LYG Ticagrelor 25 612 2963 12.70 0.38 7820 €/LYG Monte Carlo simulations show that at an assumed Norwegian cost-effectiveness threshold of EUR 70 000 per life year gained, 76, 24, and 0.1% of simulations indicated that ticagrelor, prasugrel, and clopidogrel were cost-effective, respectively (

Lifetime cost (€) Incremental cost (€) Life expectancy Life years gained Incremental cost-effectiveness ratio Clopidogrel 19 929 11.96 Prasugrel 22 649 2720 12.32 0.36 7505 €/LYG Ticagrelor 25 612 2963 12.70 0.38 7820 €/LYG Monte Carlo simulations show that at an assumed Norwegian cost-effectiveness threshold of EUR 70 000 per life year gained, 76, 24, and 0.1% of simulations indicated that ticagrelor, prasugrel, and clopidogrel were cost-effective, respectively ( Figure 2 ). If we rather compared with the suggested threshold of EUR 30 000, these percentages would be 72, 27, and 0.4%. The cost-effectiveness acceptability frontier gives that clopidogrel is cost-effective for cost-effectiveness thresholds lower than €7505 per LYG, prasugrel is cost-effective for threshold between €7505 per LYG and €7820 per LYG, and that ticagrelor is cost-effective for threshold above €7820 per LYG ( Figure 3 ). Figure 2 Scatter plot of simulations (incrementals compared with clopidogrel). Figure 3 Cost-effectiveness acceptability frontier. When including only efficacy data on MI and bleeding, prasugrel dominated the other two drugs with both lowest costs and highest effectiveness. When including efficacy on revascularization as well as bleeding and MI, clopidogrel was the most effective and least costly. When including efficacy only on death, bleeding, and MI, ticagrelor was the most effective and cost-effective. When excluding the probability of bleeding and revascularization after the first year, analyses gave that ticagrelor was the most effective and cost-effective.

pidogrel was the most effective and least costly. When including efficacy only on death, bleeding, and MI, ticagrelor was the most effective and cost-effective. When excluding the probability of bleeding and revascularization after the first year, analyses gave that ticagrelor was the most effective and cost-effective. In analyses of expected value of perfect information on parameters, we found that the relative effect on death of ticagrelor and prasugrel compared with clopidogrel was the only parameter with any value of gathering new evidence. The per person expected value of partial perfect information was €3220 and €580 for prasugrel and ticagrelor, respectively, indicating that for each person in Norway who experience a PCI, up to €3220 could be used to reduce the decision uncertainty, if this was completely removed by the research. In sensitivity analyses without discounting health, life expectancy with clopidogrel, prasugrel, and ticagrelor was 17.52, 18.23, and 19.05, respectively. Applying discounted costs to these life expectancies gave ICERs of 3840 €/LYG of prasugrel compared with clopidogrel and 3620 €/LYG of ticagrelor compared with prasugrel. When incorporating health-related quality of life in the model, expected discounted remaining QALYs were 9.54, 9.82, and 10.12 for clopidogrel, prasugrel, and ticagrelor, respectively. The cost-utility of prasugrel compared with clopidogrel then becomes 9531 €/QALY. Ticagrelor compared with prasugrel gives a cost-utility of 9987 €/QALY.

health-related quality of life in the model, expected discounted remaining QALYs were 9.54, 9.82, and 10.12 for clopidogrel, prasugrel, and ticagrelor, respectively. The cost-utility of prasugrel compared with clopidogrel then becomes 9531 €/QALY. Ticagrelor compared with prasugrel gives a cost-utility of 9987 €/QALY. We explored the impact of reducing price of clopidogrel and prasugrel. Not even reducing prices to €0,- would make prasugrel or clopidogrel cost-effective, because the main differences in costs between the strategies are those resulting from reduction of clinical events. In simulations of 50-, 70-, and 80 year-olds, we also found that ticagrelor was cost-effective compared with the other two drugs. Probabilities of ticagrelor being cost-effective were 76, 78, and 79% for 50-, 70-, and 80-year olds, respectively. Probability of clopidogrel being cost-effective was <0.3% in all cases. We performed a separate simulation with only 5-year time horizon. In these analyses, ticagrelor was also the most cost-effective option, but the probability of being cost-effective was lower at 61%. Discussion Our base case analysis based on efficacy data on four different outcomes showed that ticagrelor is clearly cost-effective compared with prasugrel and clopidogrel. Simulations gave a 77% that ticagrelor is cost-effective, given the current evidence base. Considering that only 24% of users of these drugs use ticagrelor, it may be time for a change in practice.

data on four different outcomes showed that ticagrelor is clearly cost-effective compared with prasugrel and clopidogrel. Simulations gave a 77% that ticagrelor is cost-effective, given the current evidence base. Considering that only 24% of users of these drugs use ticagrelor, it may be time for a change in practice. Interpretation of the analyses did not change for different suggested cost-effectiveness thresholds. Results were also extremely robust with regards to choice of starting point for the model. Having a PCI at 50, 60, 70, or 80 did not change conclusions; ticagrelor is cost-effective and clopidogrel is clearly not. In a recent review of cost-effectiveness, four studies were found that evaluated prasugrel vs. clopidogrel and two studies that evaluated ticagrelor vs. clopidogrel. 15 None of these six evaluations had more than a 2-year perspective, so our 40-year perspective is probably not all that comparable, yet all these analyses found the analysed new drug to be cost-effective compared with clopidogrel. Cost-effectiveness analyses are usually reckoned to not be transferable across jurisdictions, although recent evidence may indicate otherwise. 16 That taken into account, in addition to our analyses showing that ticagrelor is cost-effective even if the prices of the two other drugs were €0,- are clear indications that ticagrelor is likely to be cost-effective regardless of jurisdiction.

le across jurisdictions, although recent evidence may indicate otherwise. 16 That taken into account, in addition to our analyses showing that ticagrelor is cost-effective even if the prices of the two other drugs were €0,- are clear indications that ticagrelor is likely to be cost-effective regardless of jurisdiction. Remaining life expectancy for patients who have recently undergone a PCI is reasonable to assume to be shorter than for the general population. Our analyses indicate a remaining life expectancy of 17.5 years with current treatment strategy (clopidogrel) for 60-year olds, compared with an average of 24.5 years for the general population at the same age. This indicates that our model gives results around what was reasonable to assume. Limitations We used a cycle length of 6 months in our model analyses. In this way, we distinguished clearly between those events patients are at risk of getting during the first days and months after a PCI, compared with risks later on. We acknowledge that a shorter cycle length is even more accurate, and might have been preferable. We chose a 40-year perspective when modelling 60-year-old patients who had undergone a PCI. By the end of this period, 99.99% would be dead; hence, all relevant aspects should have been included, as for instance recommended by SMDM-ISPOR. 17 On the other hand, modelling such a long period is largely based on assumptions. We, therefore, conducted a separate simulation with only a 5-year perspective, although results differ, conclusions do not; ticagrelor is still the most cost-effective.

ould have been included, as for instance recommended by SMDM-ISPOR. 17 On the other hand, modelling such a long period is largely based on assumptions. We, therefore, conducted a separate simulation with only a 5-year perspective, although results differ, conclusions do not; ticagrelor is still the most cost-effective. A limitation of our analysis is the lack of longer follow-up data on revascularization and bleeding. Here, we assumed that rates were similar in later periods compared with the first half year.

ould have been included, as for instance recommended by SMDM-ISPOR. 17 On the other hand, modelling such a long period is largely based on assumptions. We, therefore, conducted a separate simulation with only a 5-year perspective, although results differ, conclusions do not; ticagrelor is still the most cost-effective. A limitation of our analysis is the lack of longer follow-up data on revascularization and bleeding. Here, we assumed that rates were similar in later periods compared with the first half year. When creating models of disease progression, choice of which health states to include is essential and may very well influence results. In our model, we included AMI, revascularization, bleeding, and death, mainly because these trials have indicated that the most pronounced differences between the drugs are on these outcomes. We might also have incorporated other health states, for instance stroke, but chose not to do so, because P -values of 0.22 and 0.93 indicate that if there is a difference, it is probably not that substantial. 1 , 2 In analyses exploring whether eliminating any of the included health states would affect the results, we found that all three drugs evaluated could be considered cost-effective, depending on which health states that was included. All scenarios analysed here can be justified based on reasoning related to double counting of events and hence be used by pharmaceutical companies when applying for reimbursement. We therefore urge governmental institutions appraising reimbursement applications to be aware of the potential impact of structural uncertainty when making decisions on reimbursement.

d based on reasoning related to double counting of events and hence be used by pharmaceutical companies when applying for reimbursement. We therefore urge governmental institutions appraising reimbursement applications to be aware of the potential impact of structural uncertainty when making decisions on reimbursement. The present model is based solely on life years as health outcome. In other model of patients receiving PCI, outcomes such as avoided revascularizations and QALYs have also been presented. We did not use revascularizations as outcome because it would not capture the time aspect. QALYs could have been used, because there might be a lower QALY with bleeding compared with AMI. This impact would, however, be only for a limited time after each episode, making the impact on the ICER minimal. Conclusions Ticagrelor is clearly cost-effective compared with prasugrel and clopidogrel for a Norwegian setting. Results can be easily modified by pharmaceutical companies who want to prove their drug being the most cost-effective. Funding The study was performed as part of regular work at the Oslo University Hospital. Funding to pay the Open Access publication charges for this article was provided by Oslo University Hospital through University of Oslo. Conflict of interest: D.A. reports personal fees from Astra-Zeneca, personal fees from Sanofi-Aventis, outside the submitted work. Appendix Table A1 Probabilities and proportions Probability Expectation Distribution details Source AMI first 6 months after primary PCI 0.0639 Beta ( r = 823, n = 12 880)

Funding The study was performed as part of regular work at the Oslo University Hospital. Funding to pay the Open Access publication charges for this article was provided by Oslo University Hospital through University of Oslo. Conflict of interest: D.A. reports personal fees from Astra-Zeneca, personal fees from Sanofi-Aventis, outside the submitted work. Appendix Table A1 Probabilities and proportions Probability Expectation Distribution details Source AMI first 6 months after primary PCI 0.0639 Beta ( r = 823, n = 12 880) SCAAR 18 Revascularisation first 6 months after primary PCI 0.01978 Beta ( r = 232, n = 11 730) WDHR 19 Bleeding first 6 months after primary PCI 0.00068 Beta ( a = 11.620, b = 17 027) Meta-analysis 20 Death first 6 months 0.0316 Log-normal (1.6190, 0.0569) a SCAAR 18 AMI in later half-year periods 0.01614 Beta ( r = 507, n = 31 409) SCAAR 18 Revascularisation in later half-year periods Assumed same as first half-year period Bleeding in later half-year periods Assumed same as first half-year period Death in later half-year periods 0.01028 Log-normal(0.4856, 0.0489) a SCAAR 18 Mortality during PCI operation 0.005 Beta ( a = 30.710, b = 6111.3) Feiring Heart Clinic 21 Mortality during CABG operation 0.008 Beta ( a = 37.567, b = 4658.3) Feiring Heart Clinic 21 Bleeding mortality 0.00193 Beta ( r = 26, n = 13 457) NOKC report 22 a Distribution of a rate ratio which is multiplied by mortality rate to give age-dependence. Table A2 Cost items

SCAAR 18 Revascularisation in later half-year periods Assumed same as first half-year period Bleeding in later half-year periods Assumed same as first half-year period Death in later half-year periods 0.01028 Log-normal(0.4856, 0.0489) a SCAAR 18 Mortality during PCI operation 0.005 Beta ( a = 30.710, b = 6111.3) Feiring Heart Clinic 21 Mortality during CABG operation 0.008 Beta ( a = 37.567, b = 4658.3) Feiring Heart Clinic 21 Bleeding mortality 0.00193 Beta ( r = 26, n = 13 457) NOKC report 22 a Distribution of a rate ratio which is multiplied by mortality rate to give age-dependence. Table A2 Cost items Cost item Expectation Distribution details Source Cost per half year of ACE inhibitor 1110.06 No NoMA 8 Cost per ground ambulance turn-out 11000.00 Gamma (4, 0.00036) NorCaD 23 Cost per half year of ASA use 145.13 No NoMA 8 Cost per half year of beta-blocker 341.76 No NoMA 8 Half year cost of clopidogrel 865.64 No NoMA 8 Cost per loading dose of clopidogrel (300 mg) 47.07 No NoMA 8 Cost per DRG-point 35127.00 No DRG price list 24 Cost per GP lab test with ECG and cholesterol 141.00 No GP tariff 25 Cost per GP visit 289.00 No GP tariff 25 Cost per half year of prasugrel 3413.13 No NoMA 8 Cost per loading dose of prasugrel (60 mg) 112.14 No NoMA 8 Cost per half year of statin 338.48 No NoMA 8 Half year cost og ticagrelor treatment 2124.53 No NoMA 8 Cost per loading dose of ticagrelor (180 mg) 23.27 No NoMA 8 Number of GP lab tests per half year with ECG and cholesterol 1.0 Gamma (4, 4) NorCaD 23 Number of GP visits per half year when asymptomatic 1.0 Gamma (4, 4) NorCaD 23 Number of GP visits with STEMI at intervention hospital 1.0 Gamma (4, 4) NorCaD 23 Number of ambulances when STEMI at intervention hospital 1.0 Gamma(4, 4) NorCaD 23 Number of DRG 112e when STEMI at intervention hospital 0.5 Beta ( r = 50, n = 100)

rial online, Table S5. Although the sensitivity of CrCl <30 mL/min was less than 30%, the specificity was more than 90% for all events except for major haemorrhage. The sensitivity of CrCl <50 mL/min ranged from 37% to 69%, but the specificity was around 75% for all events (see Supplementary material online, Table S5). Discussion The major findings of the present study were as follows. First, since patients with renal impairment were characterized as high risk for both thromboembolism and major haemorrhage, there were significant trends of incidence of clinical events and composite events among the four CrCl groups. Event rates increased along with a decrease in CrCl values, with the exception of thromboembolism. Second, after adjustment for possible confounders, lower CrCl values were independently associated with clinical adverse events and composite events, but not with major haemorrhage. Finally, warfarin use was associated with a lower risk of composite events with CrCl values <80 mL/min.

of GP visits per half year when asymptomatic 1.0 Gamma (4, 4) NorCaD 23 Number of GP visits with STEMI at intervention hospital 1.0 Gamma (4, 4) NorCaD 23 Number of ambulances when STEMI at intervention hospital 1.0 Gamma(4, 4) NorCaD 23 Number of DRG 112e when STEMI at intervention hospital 0.5 Beta ( r = 50, n = 100) NorCaD 23 Number of DRG 112f when STEMI at intervention hospital 0.5 Beta ( r = 50, n = 100) NorCaD 23 Number of ambulances with STEMI at other hospitals 2.8 Gamma (2.8, 1) NorCaD 23 Number of DRG 112e when STEMI at other hospitals 0.45 Beta ( r = 45, n = 100) NorCaD 23 Number of DRG 112f when STEMI at other hospitals 0.45 Beta ( r = 45, n = 100) NorCaD 23 Number of DRG 121 when STEMI at other hospitals 0.5 Beta ( r = 50, n = 100) NorCaD 23 Number of DRG 122 when STEMI at other hospitals 1.4 Gamma (4.2, 3) NorCaD 23 Number of GP visits when STEMI without PCI facilities 1.0 No NorCaD 23 Weight for DRG 112e 1.19 Gamma (4, 3.3613) DRG price list 24 Weight for DRG 112f 1.46 Gamma (4, 2.7397) DRG price list 24 Weight for DRG 121 1.51 Gamma (4, 2.6490) DRG price list 24 Figure A1 Cost-effectiveness acceptability curve.

Introduction Atrial fibrillation (AF) is a common arrhythmia and a strong risk factor for cardiogenic thromboembolism.1,2 Anticoagulation therapy with vitamin K antagonists (VKA), mainly warfarin, is able to reduce the risk of cardiogenic thromboembolism by 60–70%.3,4 Although direct oral anticoagulants (DOACs) are currently available for the prevention of ischaemic stroke and systemic embolism in patients with non-valvular AF (NVAF), warfarin is still indicated in those with severe renal impairment. The use of DOACs is contraindicated for patients with creatinine clearance (CrCl) values of <30 mL/min for dabigatran and <15 mL/min for factor Xa inhibitors (i.e. rivaroxaban, apixaban, and edoxaban).5 Renal impairment itself is a risk factor for stroke or all-cause mortality in patients with AF6–9 as well as in the general population.10,11 In addition, warfarin therapy in patients with renal impairment is not always safe due to the increased risk of bleeding.7,12 Reports on the association between renal impairment and clinical outcomes are still limited in Japanese patients with NVAF.13,14 Therefore, a post hoc analysis was performed using our prospective observational data of the J-RHYTHM Registry in order to investigate the influence of renal function on thromboembolism, major haemorrhage, and mortality in Japanese patients with NVAF. Although estimated glomerular filtration rate (eGFR) has been adopted for the definition of chronic kidney disease15 and is widely used for the evaluation of renal function in a clinical practice, CrCl was used in this subanalysis at the physicians’ convenience, since renal function is determined using CrCl values for dose adjustments of DOACs.5

ar filtration rate (eGFR) has been adopted for the definition of chronic kidney disease15 and is widely used for the evaluation of renal function in a clinical practice, CrCl was used in this subanalysis at the physicians’ convenience, since renal function is determined using CrCl values for dose adjustments of DOACs.5 Methods Study design of the J-RHYTHM Registry The J-RHYTHM Registry was conducted as a prospective observational study to investigate the optimal anticoagulation therapy with warfarin in Japanese patients with AF.16 The study design and baseline patient characteristics have been reported elsewhere.16,17 Briefly, the study protocol conformed to the Declaration of Helsinki and was approved by the ethics committee of each participating institution. A consecutive series of outpatients with AF of any type were enrolled from 158 institutions without any exclusion criterion regarding renal function. All participants gave written informed consent at the time of enrolment. All treatment strategies including the selection of an oral anticoagulant were determined at the discretion of the treating cardiologists. Patients with valvular AF (mechanical valve replacement and mitral stenosis) were excluded from this subanalysis. Patients were followed up for 2 years or until the occurrence of an event, whichever occurred first. Primary endpoints were defined as thromboembolism including symptomatic ischaemic stroke, transient ischaemic attack (TIA), and systemic embolic events; major haemorrhage including intracranial haemorrhage, gastrointestinal haemorrhage, and other haemorrhages requiring hospitalization; or all-cause and cardiovascular death. The composite of thromboembolism, major haemorrhage, and all-cause death, whichever occurred first was also evaluated. The diagnostic criteria for each event have been described elsewhere.16,17

rrhage, gastrointestinal haemorrhage, and other haemorrhages requiring hospitalization; or all-cause and cardiovascular death. The composite of thromboembolism, major haemorrhage, and all-cause death, whichever occurred first was also evaluated. The diagnostic criteria for each event have been described elsewhere.16,17 Anticoagulation intensity was determined using the international normalized ratio (INR) of prothrombin time in patients receiving warfarin, and the time in therapeutic range (TTR) was determined using the method developed by Rosendaal et al.18 The target INR level was set at 1.6–2.6 for elderly patients aged ≥70 years and at 2.0–3.0 for patients aged <70 years according to Japanese guidelines.19 Using the data of age, gender, body weight, and serum creatinine value at the time of enrolment, CrCl was calculated by the Cockcroft–Gault formula,20 i.e. CrCl (mL/min) = (140 − age) × (body weight in kg) × (0.85 if female)/(72 × serum creatinine in mg/dL). Patients were divided into four groups based on CrCl values following previous phase III trials of DOACs;21,22 CrCl <30 mL/min, 30–49.9 mL/min, 50–79.9 mL/min, and ≥80 mL/min.

ed by the Cockcroft–Gault formula,20 i.e. CrCl (mL/min) = (140 − age) × (body weight in kg) × (0.85 if female)/(72 × serum creatinine in mg/dL). Patients were divided into four groups based on CrCl values following previous phase III trials of DOACs;21,22 CrCl <30 mL/min, 30–49.9 mL/min, 50–79.9 mL/min, and ≥80 mL/min. Statistical analysis Data are presented as mean ± one standard deviation. The statistical significance of differences in mean values was analysed using Student’s t-test or analysis of variance, as appropriate. Frequencies of parameters or events were compared using the χ2 test or Fisher’s exact test, as appropriate. Kaplan–Meier curves were used to compare time to events with log-rank tests. A Cox proportional hazards model was used to investigate the influence of renal function on events. Hazard ratios (HRs) and 95% confidence interval (CI) of the groups with CrCl <80 mL/min were calculated with CrCl ≥80 mL/min as a reference. Explanatory variables for multivariate analysis were adopted from well-known risk factors, i.e. the components of the CHA2DS2-VASc score [congestive heart failure, hypertension, age ≥75 years, diabetes mellitus, and history of stroke or TIA, vascular disease (coronary artery disease), age 65–74 years, and female sex]23 plus the use of warfarin and antiplatelet medication. HRs for each end point and composite events were also determined using CrCl values as a continuous variable. The predictive ability of CrCl for each event was evaluated with the c-index. The sensitivity and specificity of cut-off CrCl points for each event were also obtained from the receiver operating characteristic curves. Two-tailed P-values of <0.05 were considered statistically significant. All statistical analyses were performed with SPSS software version 23.0 (IBM Corporation, Armonk, NY, USA).

The sensitivity and specificity of cut-off CrCl points for each event were also obtained from the receiver operating characteristic curves. Two-tailed P-values of <0.05 were considered statistically significant. All statistical analyses were performed with SPSS software version 23.0 (IBM Corporation, Armonk, NY, USA). Results Among the 7937 patients with AF who were enrolled in the J-RHYTHM Registry,17 421 (5.3%) patients with valvular AF were excluded and 110 (1.5%) patients were lost to follow-up. Of the remaining 7406 patients with NVAF, 1354 patients without baseline CrCl data were excluded due to missing of serum creatinine in 828 and/or body weight in 974 patients. Ultimately, a total of 6052 patients with CrCl values at baseline constituted the study group. Baseline patient characteristics and medications Baseline patient characteristics and medications of the four CrCl groups are shown in Table 1. Prevalence of cardiomyopathy and diabetes mellitus, and systolic blood pressure were comparable, but other variables showed significant differences among the groups. In particular, age and prevalence of a history of stroke or TIA were higher in the groups with CrCl <30 and 30–49.9 mL/min. There were significant trends for CrCl values among the four groups by the study design. Risk scores for both thromboembolism (CHADS224 and CHA2DS2-VASc scores23) and bleeding (HAS-BLED score25) were the highest in the CrCl <30 mL/min group (Table 1). Table 1 Baseline patient characteristics and medications

9.9 mL/min. There were significant trends for CrCl values among the four groups by the study design. Risk scores for both thromboembolism (CHADS224 and CHA2DS2-VASc scores23) and bleeding (HAS-BLED score25) were the highest in the CrCl <30 mL/min group (Table 1). Table 1 Baseline patient characteristics and medications Creatinine clearance values (mL/min) <30 30–49.9 50–79.9 ≥80 P for trend (n = 356) (n = 1201) (n = 2686) (n = 1809) Age, years 79.8 ± 7.9 77.7 ± 6.4 70.7 ± 7.0 61.1 ± 9.1 <0.001 Female 181 (50.8) 481 (40.0) 764 (28.4) 320 (17.7) <0.001 Body weight, kg 49.6 ± 10.6 54.2 ± 9.4 61.5 ± 9.3 71.5 ± 13.5 <0.001 Renal function Serum creatinine, mg/dL 2.2 ± 1.9 1.1 ± 0.3 0.9 ± 0.2 0.8 ± 0.2 <0.001 CrCl, mL/min 21.8 ± 6.6 41.3 ± 5.6 64.7 ± 8.3 100.8 ± 21.4 <0.001 Type of atrial fibrillation Paroxysmal 117 (32.9) 400 (33.3) 1036 (38.6) 729 (40.3) <0.001 Persistent 40 (11.2) 144 (12.0) 375 (14.0) 287 (15.9) Permanent 199 (55.9) 657 (54.7) 1275 (47.4) 793 (43.8) Comorbidities Coronary artery disease 77 (21.6) 180 (15.0) 276 (10.3) 143 (7.9) <0.001 Cardiomyopathy 45 (12.6) 100 (8.3) 230 (8.6) 172 (9.5) 0.652 HCM 20 (5.6) 46 (3.8) 107 (4.0) 58 (3.2) 0.060 DCM 25 (7.0) 54 (4.5) 123 (4.6) 114 (6.3) 0.298 COPD 5 (1.4) 43 (3.6) 53 (2.0) 15 (0.8) <0.001 Hyperthyroidism 7 (2.0) 14 (1.2) 36 (1.3) 45 (2.5) 0.029 Risk factors for stroke Heart failure 223 (62.6) 482 (40.1) 666 (24.8) 163 (9.0) <0.001 Hypertension 234 (65.7) 785 (65.4) 1650 (61.4) 1020 (56.4) <0.001 Age (≥75 years) 282 (79.2) 866 (72.1) 841 (31.3) 98 (5.4) <0.001 Diabetes mellitus 81 (22.8) 242 (20.1) 492 (18.3) 64 (3.5) 0.147 Stroke/TIA 65 (18.3) 218 (18.2) 385 (14.3) 68 (3.8) <0.001 CHADS2 score 2.7 ± 1.2 2.3 ± 1.2 1.7 ± 1.2 1.2 ± 1.1 <0.001 CHA2DS2-VASc score 4.4 ± 1.4 3.9 ± 1.4 2.9 ± 1.6 3.0 ± 1.5 <0.001 HAS-BLED score 2.2 ± 1.0 1.9 ± 0.9 1.6 ± 0.9 1.0 ± 1.0 <0.001 Heart rate, /min 72.8 ± 12.8 72.9 ± 13.4 72.0 ± 13.0 74.3 ± 13.3 0.694 Systolic BP, mmHg 123.4 ± 19.5 125.3 ± 17.3 126.7 ± 15.9 125.4 ± 18.1 0.105 Diastolic BP, mmHg 67.5 ± 12.3 70.8 ± 11.6 73.9 ± 19.2 68.2 ± 12.1 <0.001 Warfarin 322 (90.4) 1064 (88.6) 2364 (88.0) 1538 (85.0) <0.001 Dosage, mg/day 2.1 ± 0.9 2.4 ± 1.0 2.9 ± 1.1 3.3 ± 1.2 <0.001 INR 1.89 ± 0.52 1.92 ± 0.53 1.90 ± 0.47 1.90 ± 0.49 0.003 TTRa, % 61.7 ± 26.5 69.8 ± 24.8 61.9 ± 28.3 47.5 ± 29.5 <0.001 (n = 299) (n = 997) (n = 2251) (n = 1449) Antiplatelet 123 (34.6) 387 (32.2) 700 (26.1) 400 (22.1) <0.001 Aspirin 100 (28.1) 320 (26.6) 609 (22.7) 364 (20.1) <0.001 Others 35 (9.8) 106 (8.8) 162 (6.

INR 1.89 ± 0.52 1.92 ± 0.53 1.90 ± 0.47 1.90 ± 0.49 0.003 TTRa, % 61.7 ± 26.5 69.8 ± 24.8 61.9 ± 28.3 47.5 ± 29.5 <0.001 (n = 299) (n = 997) (n = 2251) (n = 1449) Antiplatelet 123 (34.6) 387 (32.2) 700 (26.1) 400 (22.1) <0.001 Aspirin 100 (28.1) 320 (26.6) 609 (22.7) 364 (20.1) <0.001 Others 35 (9.8) 106 (8.8) 162 (6. 0) 58 (3.2) <0.001 Warfarin + antiplatelet 99 (27.8) 298 (24.8) 500 (18.6) 257 (14.2) <0.001 Data are number of patients (%) or mean ± SD. CrCl, creatinine clearance; HCM, hypertrophic cardiomyopathy; DCM, dilated cardiomyopathy; COPD, chronic obstructive pulmonary disease; CHADS2, congestive heart failure, hypertension, age ≥75 years, diabetes mellitus, and history of stroke or TIA; CHA2DS2-VASc, CHADS2 components plus vascular disease (coronary artery disease), age 65–74 years, and female sex; HAS-BLED, hypertension (systolic BP ≥140 mmHg), abnormal renal/liver function, stroke, bleeding history or predisposition, labile INR (episodes of INR ≥3.5), elderly (age >65 years), drugs (use of antiplatelets)/alcohol concomitantly; TIA, transient ischaemic attack; BP, blood pressure; INR, international normalized ratio of prothrombin time; TTR, time in therapeutic range. aTarget INR was 2.0–3.0 (<70 years) or 1.6–2.6 (≥70 years).

CrCl, creatinine clearance; HCM, hypertrophic cardiomyopathy; DCM, dilated cardiomyopathy; COPD, chronic obstructive pulmonary disease; CHADS2, congestive heart failure, hypertension, age ≥75 years, diabetes mellitus, and history of stroke or TIA; CHA2DS2-VASc, CHADS2 components plus vascular disease (coronary artery disease), age 65–74 years, and female sex; HAS-BLED, hypertension (systolic BP ≥140 mmHg), abnormal renal/liver function, stroke, bleeding history or predisposition, labile INR (episodes of INR ≥3.5), elderly (age >65 years), drugs (use of antiplatelets)/alcohol concomitantly; TIA, transient ischaemic attack; BP, blood pressure; INR, international normalized ratio of prothrombin time; TTR, time in therapeutic range. aTarget INR was 2.0–3.0 (<70 years) or 1.6–2.6 (≥70 years). There were significant differences in some baseline characteristics between patients with and without CrCl data; however, age, body weight, baseline INR values, or TTR did not differ between the four groups (see Supplementary material online, Table S1). Baseline characteristics of patients stratified by CrCl and warfarin use at baseline are shown in Supplementary material online, Table S2.

een patients with and without CrCl data; however, age, body weight, baseline INR values, or TTR did not differ between the four groups (see Supplementary material online, Table S1). Baseline characteristics of patients stratified by CrCl and warfarin use at baseline are shown in Supplementary material online, Table S2. Event rates and renal function Two-year event rates in the four groups are shown in Table 2. There was a significant trend for the four events, and consequently, for composite events among the four CrCl groups (P < 0.001 for trend). The CrCl <30 mL/min group showed the highest event rates among the four CrCl groups except for thromboembolic events. The Kaplan–Meier curves for each endpoint are shown in Figure 1. Significant differences in the event-free rates of thromboembolism, major haemorrhage, all-cause death, and cardiovascular death among the four groups were revealed by the log-tank test (Figure 1). This was also true for composite events (Figure 2). The CrCl <30 mL/min group showed the lowest event-free rate for composite events. There were no significant differences in the 2-year event rates of all events between patients with and without CrCl data (see Supplementary material online, Table S3). Table 2 Two-year event rates in creatinine clearance groups

igure 2). The CrCl <30 mL/min group showed the lowest event-free rate for composite events. There were no significant differences in the 2-year event rates of all events between patients with and without CrCl data (see Supplementary material online, Table S3). Table 2 Two-year event rates in creatinine clearance groups Creatinine clearance values (mL/min) <30 30–49.9 50–79.9 ≥80 P for trend (n = 356) (n = 1201) (n = 2686) (n = 1809) Thromboembolism 7 (2.0%) 33 (2.7%) 53 (2.0%) 16 (0.9%) 0.001 Cerebral infarction 5 26 42 15 Transient ischaemic attack 1 4 2 1 Systemic embolism 1 3 9 0 Major haemorrhage 13 (3.7%) 35 (2.9%) 48 (1.8%) 24 (1.3%) <0.001 Intracranial 5 14 20 7 Gastrointestinal 6 10 15 9 Others 2 11 13 8 All-cause death 41 (11.5%) 59 (4.9%) 47 (1.7%) 13 (0.7%) <0.001 Cardiovascular death 15 (4.2%) 23 (1.9%) 10 (0.4%) 7 (0.4%) <0.001 Composite eventsa 61 (17.1%) 126 (10.5%) 147 (5.5%) 53 (2.9%) <0.001 Data are number of patients (%). aThromboembolism, major haemorrhage, and all-cause death. Figure 1 Kaplan–Meier curves for thromboembolism (A), major haemorrhage (B), all-cause death (C), and cardiovascular death (D). CrCl, creatinine clearance (mL/min). Figure 2 Kaplan–Meier curves for the composite of thromboembolism, major haemorrhage, and all-cause death. CrCl, creatinine clearance (mL/min).

aThromboembolism, major haemorrhage, and all-cause death. Figure 1 Kaplan–Meier curves for thromboembolism (A), major haemorrhage (B), all-cause death (C), and cardiovascular death (D). CrCl, creatinine clearance (mL/min). Figure 2 Kaplan–Meier curves for the composite of thromboembolism, major haemorrhage, and all-cause death. CrCl, creatinine clearance (mL/min). A comparison of the 2-year event rates between patients with and without warfarin treatment at the time of events or at the end of the follow-up period is shown for each CrCl group in Supplementary material online, Table S4. The Kaplan–Meier curves for composite events were compared between patients with and without warfarin treatment (Figure 3). The event-free rate was higher in patients with warfarin than in those without warfarin for each CrCl group, except for the CrCl ≥80 mL/min group. Figure 3 Kaplan–Meier curves for the composite of thromboembolism, major haemorrhage, and all-cause death with the status of warfarin use at the time of events or at the end of the follow-up period in the four CrCl groups. CrCl, creatinine clearance (mL/min).

A comparison of the 2-year event rates between patients with and without warfarin treatment at the time of events or at the end of the follow-up period is shown for each CrCl group in Supplementary material online, Table S4. The Kaplan–Meier curves for composite events were compared between patients with and without warfarin treatment (Figure 3). The event-free rate was higher in patients with warfarin than in those without warfarin for each CrCl group, except for the CrCl ≥80 mL/min group. Figure 3 Kaplan–Meier curves for the composite of thromboembolism, major haemorrhage, and all-cause death with the status of warfarin use at the time of events or at the end of the follow-up period in the four CrCl groups. CrCl, creatinine clearance (mL/min). Multivariate analysis of the influence of renal function on events When unadjusted, lower CrCl values were associated with adverse clinical events (Table 3). When adjusted for possible confounders, this was also true except for major haemorrhage (Table 4). As expected, warfarin use was associated with as increased risk of major haemorrhage, and with a decreased risk of thromboembolism, all-cause death, and composite events (Table 4). P-values for interaction between warfarin use and CrCl were 0.407 for thromboembolism, 0.863 for major haemorrhage, 0.157 for all-cause death, 0.869 for cardiovascular death, and 0.630 for composite events, indicating no interaction between warfarin use and CrCl for any event. When baseline CrCl values were used as a continuous variable, adjusted risk of thromboembolism (HR 1.011, 95% CI 1.000–1.021, P = 0.040), all-cause death (HR 1.028, 95% CI 1.019–1.037, P < 0.001), cardiovascular death (HR 1.030, 95% CI 1.014–1.045, P < 0.001), and composite events (HR 1.014, 95% CI 1.009–1.020, P < 0.001) increased significantly for every 1-mL/min decrease in CrCl. Although the risk of major haemorrhage increased significantly for every 1-mL/min decrease in CrCl values (HR 1.013, 95% CI 1.006–1.021, P = 0.001) in unadjusted model, it became insignificant after adjustment for confounding factors (HR 1.002, 95% CI 0.993–1.011, P = 0.669). Table 3 Univariate analysis of influence of creatinine clearance on events

orrhage increased significantly for every 1-mL/min decrease in CrCl values (HR 1.013, 95% CI 1.006–1.021, P = 0.001) in unadjusted model, it became insignificant after adjustment for confounding factors (HR 1.002, 95% CI 0.993–1.011, P = 0.669). Table 3 Univariate analysis of influence of creatinine clearance on events Thromboembolism Major haemorrhage All-cause death Cardiovascular death Composite eventsa HR P-value HR P-value HR P-value HR P-value HR P-value (95% CI) (95% CI) (95% CI) (95% CI) (95% CI) CrCl <30 mL/min 2.41 0.052 3.00 0.001 17.77 <0.001 12.19 <0.001 6.41 <0.001 (0.99–5.86) (1.53–5.90) (9.52–33.2) (4.44–9.27) (4.97–29.9) CrCl 30–49.9 mL/min 3.22 <0.001 2.22 0.003 7.15 <0.001 5.20 <0.001 3.73 <0.001 (1.77–5.85) (1.31–3.74) (3.92–13.0) (2.23–12.1) (2.71–5.14) CrCl 50–79.9 mL/min 2.66 0.004 1.34 0.246 2.48 <0.001 0.98 0.968 1.90 <0.001 (1.29–3.95) (0.82–2.19) (1.34–4.58) (0.37–2.58) (1.39–2.60) CrCl ≥80 mL/min Reference — Reference — Reference — Reference — Reference — CrCl, creatinine clearance; HR, hazard ratio; CI, confidence interval; TIA, transient ischaemic attack. aThromboembolism, major haemorrhage, and all-cause death. Table 4 Multivariate analysis for variables associated with clinical events

Thromboembolism Major haemorrhage All-cause death Cardiovascular death Composite eventsa HR P-value HR P-value HR P-value HR P-value HR P-value (95% CI) (95% CI) (95% CI) (95% CI) (95% CI) CrCl <30 mL/min 2.41 0.052 3.00 0.001 17.77 <0.001 12.19 <0.001 6.41 <0.001 (0.99–5.86) (1.53–5.90) (9.52–33.2) (4.44–9.27) (4.97–29.9) CrCl 30–49.9 mL/min 3.22 <0.001 2.22 0.003 7.15 <0.001 5.20 <0.001 3.73 <0.001 (1.77–5.85) (1.31–3.74) (3.92–13.0) (2.23–12.1) (2.71–5.14) CrCl 50–79.9 mL/min 2.66 0.004 1.34 0.246 2.48 <0.001 0.98 0.968 1.90 <0.001 (1.29–3.95) (0.82–2.19) (1.34–4.58) (0.37–2.58) (1.39–2.60) CrCl ≥80 mL/min Reference — Reference — Reference — Reference — Reference — CrCl, creatinine clearance; HR, hazard ratio; CI, confidence interval; TIA, transient ischaemic attack. aThromboembolism, major haemorrhage, and all-cause death. Table 4 Multivariate analysis for variables associated with clinical events Thromboembolism Major haemorrhage All-cause death Cardiovascular death Composite eventsa HR P-value HR P-value HR P-value HR P-value HR P-value (95% CI) (95% CI) (95% CI) (95% CI) (95% CI) CrCl <30 mL/minb 1.69 0.309 1.37 0.441 6.44 <0.001 3.99 0.017 2.99 <0.001 (0.62–4.62) (0.62–3.03) (3.03–13.7) (1.28–12.4) (1.91–4.67) CrCl 30–49.9 mL/minb 2.27 0.029 1.10 0.775 3.14 0.002 2.29 0.126 1.94 0.001 (1.09–4.72) (0.58–2.09) (1.54–6.41) (0.79–6.61) (1.31–2.88) CrCl 50–79.9 mL/minb 1.99 0.030 0.91 0.747 1.73 0.111 0.71 0.523 1.40 0.058 (1.07–3.72) (0.53–1.58) (0.88–3.38) (0.25–2.04) (0.99–1.99) Heart failure 1.14 0.545 1.44 0.063 2.90 <0.001 5.98 <0.001 1.80 <0.001 (0.75–1.73) (0.98–2.12) (2.06–4.08) (3.06–11.71) (1.46–2.22) Hypertension 1.04 0.309 1.46 0.070 0.66 0.011 0.54 0.024 0.95 0.602 (0.70–1.54) (0.97–2.19) (0.48–0.91) (0.31–0.92) (0.77–1.17) Age ≥75 years 1.67 0.113 2.65 0.004 2.80 0.002 2.03 0.201 2.37 <0.001 (0.88–3.16) (1.36–5.17) (1.44–5.44) (0.69–6.03) (1.62–3.45) Diabetes mellitus 2.00 0.444 1.12 0.602 1.17 0.397 1.57 0.135 1.15 0.241 (0.75–1.90) (0.73–1.72) (0.81–1.69) (0.87–2.83) (0.91–1.47) Stroke/TIA 1.56 0.064 1.60 0.030 1.52 0.029 1.18 0.645 1.55 <0.001 (0.97–2.50) (1.05–2.46) (1.04–2.20) (0.59–2.36) (1.22–1.97) Vascular diseases 0.92 0.791 1.14 0.622 1.76 0.005 1.58 0.185 1.34 0.036 (0.50–1.71) (0.68–1.91) (1.19–2.61) (0.80–3.10) (1.02–1.77) Age 65–74 years 0.99 0.961 1.59 0.144 1.39 0.331 1.41 0.521 1.31 0.143 (0.53–1.82) (0.85–2.98) (0.72–2.70) (0.49–4.06) (0.91–1.89) Female 0.80 0.303 0.60 0.026 0.48 <0.001 0.80 0.448 0.59 <0.001 (0.52–1.23) (0.38–0.94) (0.33–0.71) (0.44–1.44) (0.47–0.76) Warfarin at baseline 0.45 0.001 2.54 0.030 0.57 0.013 0.58 0.175 0.73 0.037 (0.27–0.73) (1.10–5.90) (0.36–0.89) (0.26–1.28) (0.54–0.98) Antiplatelet use 0.98 0.943 1.44 0.089 1.05 0.778 1.31 0.395 1.14 0.281 (0.62–1.55) (0.95–2.20) (0.73–1.52) (0.71–2.42) (0.90–1.44) Vascular disease = coronary artery disease.

the exception of thromboembolism. Second, after adjustment for possible confounders, lower CrCl values were independently associated with clinical adverse events and composite events, but not with major haemorrhage. Finally, warfarin use was associated with a lower risk of composite events with CrCl values <80 mL/min. Renal impairment and thromboembolism Although renal dysfunction is not included in well-known risk scores for thromboembolism in patients with NVAF such as the CHADS224 or CHA2DS2-VASc scores,23 it appears to be a potent risk factor for stroke in patients with AF6–9 as well as in the general population.10,11 Several mechanisms have been proposed to underlie the increased thromboembolic event rates in patients with AF and renal dysfunction, including impaired function of the left atrial appendage, endothelial damage, coagulation abnormalities, activation of the renin-angiotensin-aldosterone system, chronic inflammation, and others.9 Previously, incorporation of renal dysfunction was proposed to enable improved risk stratification of thromboembolism in patients with AF, i.e. the R2CHADS2 score,26 however, renal dysfunction is not always detected as a risk factor for thromboembolism.27–30

tensin-aldosterone system, chronic inflammation, and others.9 Previously, incorporation of renal dysfunction was proposed to enable improved risk stratification of thromboembolism in patients with AF, i.e. the R2CHADS2 score,26 however, renal dysfunction is not always detected as a risk factor for thromboembolism.27–30 In a Japanese cohort study, the Fushimi AF Registry,14 CrCl <30 mL/min was associated with the highest HR for stroke/systemic embolic events. However, this was not the case in the present study, in which the HR of CrCl values of <30 mL/min showed only marginal significance for thromboembolic events in the univariate analysis (P = 0.052, Table 3), though the P-value exceeded 0.30 in the multivariate analysis (Table 4). Nevertheless, CrCl values of 30–49.9 mL/min and 50–79.9 mL/min were independently associated with thromboembolic events in our cohort. The sample size, CHADS2 score, and CHA2DS2-VASc score of patients with CrCl <30 mL/min were comparable between the Fushimi AF registry14 and the present study. Some of the patients with CrCl <30 mL/min in our cohort died before thromboembolic events occurred.

ndently associated with thromboembolic events in our cohort. The sample size, CHADS2 score, and CHA2DS2-VASc score of patients with CrCl <30 mL/min were comparable between the Fushimi AF registry14 and the present study. Some of the patients with CrCl <30 mL/min in our cohort died before thromboembolic events occurred. Previous studies7,31,32 showed that warfarin was effective in reducing rates of stroke and thromboembolism in patients with AF and chronic kidney disease. This was also true for CrCl values of 30–49.9 mL/min and 50–79.9 mL/min in the present study (see Supplementary material online, Table S4). In the Fushimi AF Registry,14 however, oral anticoagulants, mainly warfarin, were not associated with lower event rates of thromboembolism as a whole, and the effects of oral anticoagulants were not determined for each CrCl group. In our cohort, anticoagulation therapy with warfarin was performed frequently in more than 85% of patients and TTR was approximately 60%. Consequently, the preventive effect of warfarin for thromboembolism was evident as we previously reported.33 In the Fushimi AF Registry, anticoagulants were prescribed in only half of patients with AF and anticoagulation intensity was suboptimal.34 These differences in anticoagulant use might explain inconsistent results between the Fushimi AF Registry and the present registry, although both registries were conducted in the same country Japan.

Registry, anticoagulants were prescribed in only half of patients with AF and anticoagulation intensity was suboptimal.34 These differences in anticoagulant use might explain inconsistent results between the Fushimi AF Registry and the present registry, although both registries were conducted in the same country Japan. Renal impairment and major haemorrhage Since renal impairment is known as a potent risk factor for bleeding complications in both patients with AF7,12,30,35 and the general population,29 it is included in the HAS-BLED score.25 Causes of the increased risk of haemorrhagic events in patients with renal impairment include platelet dysfunction, altered von Willebrand factor, and others.9 However, in the present study, renal impairment was not an independent predictor for major haemorrhage (Table 4), although the event-free rate of major haemorrhage differed among the four CrCl groups according to the log-rank test for the Kaplan–Meier curves (Figure 1), and the unadjusted HRs for major haemorrhage in patients with renal impairment were significantly high (Table 3). It was also true when CrCl values were used as a continuous variable. This finding also differed from that in the Fushimi AF Registry,14 in which CrCl values of <30 mL/min were independently associated with major haemorrhage. This is possibly attributable to differences in study design, patient characteristics, and the frequency and control status of anticoagulation therapy between our cohort and that of the Fushimi AF Registry.14 Haemorrhagic complications could have been strongly associated with other confounding factors such as age, a history of stroke or TIA, sex, and warfarin use in our cohort (Table 4).

characteristics, and the frequency and control status of anticoagulation therapy between our cohort and that of the Fushimi AF Registry.14 Haemorrhagic complications could have been strongly associated with other confounding factors such as age, a history of stroke or TIA, sex, and warfarin use in our cohort (Table 4). Anticoagulation therapy with warfarin in patients with renal impairment has also shown to increase the risk of major haemorrhage,7 especially in older adults.12,35 However, conflicting results have been reported.8,31 In our cohort, although the incidence rate of major haemorrhage increased in tandem with a decrease in CrCl values, it did not differ significantly between patients with and without warfarin treatment for each CrCl value group (see Supplementary material online, Table S4). Renal impairment and composite events Renal impairment is also an established risk factor of all-cause and cardiovascular mortality in Japanese patients with AF13,14 and in the general population.10,11 In the present study, this association was confirmed, and even moderate renal impairment, i.e. CrCl 30–49.9 mL/min, was shown to be an independent predictor for all-cause mortality. The c-indices of CrCl for all-cause and cardiovascular death were higher than those for thromboembolism and major haemorrhage, indicating that the predictive ability of CrCl for mortality was superior to that for thromboembolism and major haemorrhage.

n, was shown to be an independent predictor for all-cause mortality. The c-indices of CrCl for all-cause and cardiovascular death were higher than those for thromboembolism and major haemorrhage, indicating that the predictive ability of CrCl for mortality was superior to that for thromboembolism and major haemorrhage. Previous studies showed that warfarin treatment was associated with a decreased rate of all-cause death in patients with AF and chronic kidney disease.31,36 In our cohort, warfarin treatment was associated with lower all-cause mortality in groups with CrCl <80 mL/min. Higher rates of all-cause death contributed to the higher rates of composite events in the CrCl <30 mL/min and 30–49.9 mL/min groups in the present study, in accordance with a previous finding.8 Warfarin treatment was associated with a higher event-free rate for composite events as compared with no warfarin treatment for patients with CrCl <80 mL/min, particularly for those with CrCl <30 mi/min (Figure 3), a finding consistent with a previous finding.27 However, this specific benefit of warfarin treatment was not observed in the CrCl ≥80 mL/min group in our cohort; this could be due to the lower overall event rates in that CrCl group even without warfarin treatment.

particularly for those with CrCl <30 mi/min (Figure 3), a finding consistent with a previous finding.27 However, this specific benefit of warfarin treatment was not observed in the CrCl ≥80 mL/min group in our cohort; this could be due to the lower overall event rates in that CrCl group even without warfarin treatment. Limitation The present study had several limitations. First, this study was a post hoc analysis of an observational study and was therefore hypothesis-generating in nature. Mechanisms underlying the increased event rates among patients with lower CrCl values could not be determined. Second, the participants were recruited from only 158 institutions in Japan. Most of the participating physicians specialized in cardiology and in the management of cardiac arrhythmias. Therefore, these results cannot necessarily be extrapolated to the general Japanese population with NVAF. Third, for determination of renal function, CrCl was not directly measured using 24-h urine creatinine excretion but was estimated by the Cockcroft–Gault formula at baseline.20 Owing to missing data on serum creatinine, body weight, or both, 1354 (18.3%) patients were excluded from the present analysis. In addition, neither urinary protein concentration nor the aetiologies of renal impairment were determined in the present study. The serial changes in CrCl values over the follow-up period also went undetermined. Fourth, TTR values differed significantly among the four CrCl groups with the CrCl ≥80 mL/min group showing the lowest TTR value. However, the event rates were lowest in the CrCl ≥80 mL/min group; therefore, the variable TTR values could not have affected the present results. The efficacy of warfarin was determined by the status of its use at the time of events or at the end of the follow-up period, as in our previous subanalysis.37 Finally, when the J-RHYTHM Registry was started in 2009, DOACs were not approved for clinical use in Japan. Therefore, the effects of DOACs on the relationship between renal function and adverse clinical events were not clarified in the present analysis.

he end of the follow-up period, as in our previous subanalysis.37 Finally, when the J-RHYTHM Registry was started in 2009, DOACs were not approved for clinical use in Japan. Therefore, the effects of DOACs on the relationship between renal function and adverse clinical events were not clarified in the present analysis. Conclusions Lower CrCl values were independently associated with thromboembolism, all-cause death, and cardiovascular death, but not with major haemorrhage in Japanese patients with NVAF. Warfarin treatment was associated with lower rates of composite events among patients with lower CrCl values. Supplementary material Supplementary material is available at European Heart Journal—Quality of Care and Clinical Outcomes online. Supplementary Material Supplementary Tables Click here for additional data file. Acknowledgements We would like to thank all investigators of the J-RHYTHM Registry listed in references 16 and 17. Funding The J-RHYTHM Registry is registered at the University Hospital Medicine Information Network (UMIN) Clinical Trials Registry (UMIN000001569) and was supported by a grant from the Japan Heart Foundation (12080025). This research was partially supported by the Practical Research Project for Life-Style related Diseases including Cardiovascular Diseases and Diabetes Mellitus from the Japan Agency for Medical Research and Development (AMED) (15656344).

00001569) and was supported by a grant from the Japan Heart Foundation (12080025). This research was partially supported by the Practical Research Project for Life-Style related Diseases including Cardiovascular Diseases and Diabetes Mellitus from the Japan Agency for Medical Research and Development (AMED) (15656344). Conflict of interest: E.K. received remuneration from Ono Pharmaceutical. H.A. received remuneration from Daiichi-Sankyo. H.I. received remuneration from Daiichi-Sankyo, Bayer Healthcare, Boehringer Ingelheim, and Bristol-Myers Squibb. K.O. received research funding from Boehringer Ingelheim and Daiichi-Sankyo and remuneration from Boehringer Ingelheim, Bayer Healthcare, Daiichi-Sankyo, and Pfizer. T.Y. received research funding from Daiichi-Sankyo, Bayer Healthcare, Tanabe-Mitsubishi, Ono Pharmaceutical, and Bristol-Meyers Squibb and remuneration from Daiichi-Sankyo, Pfizer, Bayer Healthcare, Bristol-Myers Squibb, Boehringer Ingelheim, Eisai, and Ono Pharmaceutical. H.O. received remuneration from Daiichi-Sankyo.

Introduction Pregnancy is considered a ‘cardiovascular stress test’1 that can unmask underlying cardiovascular disease through progressive haemodynamic changes2 which may lead to clinical decompensation and potentially life-threatening disease. While hypertension, sepsis, and haemorrhage remain the leading global causes of maternal death,3 cardiac disease has become the leading cause of maternal mortality in high income countries in recent years4,5 and has gained increasing relevance even in low and middle-income countries.6,7 Myocardial infarction (MI) is a major cause of morbidity and mortality for pregnant and postpartum women. A recent review of cause-specific maternal mortality in Canada showed coronary artery disease to be the leading etiology.4 In the recent UK Confidential Enquiries, ischaemic heart disease and maternal MI were the most common causes of maternal mortality due to cardiac disease.5

tality for pregnant and postpartum women. A recent review of cause-specific maternal mortality in Canada showed coronary artery disease to be the leading etiology.4 In the recent UK Confidential Enquiries, ischaemic heart disease and maternal MI were the most common causes of maternal mortality due to cardiac disease.5 Despite this rising risk of maternal mortality and morbidity due to MI, there is an unclear understanding of the true incidence of pregnancy-associated MI (PAMI). In order to allocate resources and design appropriate programs, it is critical to understand the true burden of disease due to PAMI. Understanding the interaction between morbidity and mortality as related to pregnancy is also critical to informing maternal health programmes. Declines in overall maternal mortality observed with improvements in health care have been matched with an increasing proportion of cases of morbidity; a phenomenon described as the obstetric transition.8 In order to address questions related to maternal morbidity, the World Health Organization (WHO) convened a Maternal Morbidity Working Group (MMWG) to address the knowledge gaps regarding maternal morbidity.9,10 As part of this process, the MMWG identified the need for systematic reviews to research current estimates and evidence of maternal conditions of high priority. To date, there have been no systematic reviews that summarize the population-based data on pregnancy-associated MI. Our primary objective for this systematic review and meta-analysis was to characterize the overall incidence of PAMI. We also sought to determine the overall incidence of maternal mortality as well as the case fatality rates due to PAMI.

matic reviews that summarize the population-based data on pregnancy-associated MI. Our primary objective for this systematic review and meta-analysis was to characterize the overall incidence of PAMI. We also sought to determine the overall incidence of maternal mortality as well as the case fatality rates due to PAMI. Methods Search strategy We performed this systematic review in accordance with the MOOSE guidelines for reporting of observational studies.11 We identified all potentially relevant articles by searching MEDLINE (1946 through March 2016), EMBASE (1980 through March 2016), PubMed (1960-2016), CENTRAL (Cochrane Central Register of Controlled Trials), and Web of Science Core Collection (1899–2015). This search was enhanced by scanning the bibliographies of identified articles and review articles, as well as by reviewing the conference proceedings from the first four Cardiac Problems in Pregnancy Congresses (2010, Valencia, Spain; 2012, Berlin, Germany; 2014, Venice, Italy; 2016, Las Vegas, USA). We also explored relevant data available in the grey literature by searching publications and reports available via national organizations including Statistics Canada; Health Canada; the Centre for Maternal and Child Health (CEMACH, UK) and the Centers for Disease Control (CDC, USA). There were no language or date restrictions.

xplored relevant data available in the grey literature by searching publications and reports available via national organizations including Statistics Canada; Health Canada; the Centre for Maternal and Child Health (CEMACH, UK) and the Centers for Disease Control (CDC, USA). There were no language or date restrictions. As per the strategy recommended for systematic reviews of observational studies,11 we searched the electronic databases using three comprehensive search themes (regarding the population/exposure, outcomes and study design) which were then combined using the Boolean operator ‘AND’. This detailed search strategy is outlined in Supplementary material online, Appendix S1. Study selection We included population-based single and multi-centre cohort studies (reporting on the incidence of MI and/or mortality and case fatality rates due to MI in pregnant or postpartum women) as well as case control studies (reporting on case fatality rates). Studies describing only MI occurring outside of pregnancy and the postpartum period (defined as up to 6 weeks after delivery) were excluded. We excluded case reports and case series, since they cannot provide measures of pregnancy-associated MI incidence. Two of the authors (PG and MN) independently screened all abstracts for eligibility. The full text articles of eligible abstracts were then independently reviewed by both authors for inclusion. Discrepancies were resolved by consensus between the two reviewers.

not provide measures of pregnancy-associated MI incidence. Two of the authors (PG and MN) independently screened all abstracts for eligibility. The full text articles of eligible abstracts were then independently reviewed by both authors for inclusion. Discrepancies were resolved by consensus between the two reviewers. Data extraction and quality assessment Data were independently extracted by PG and MN, with disagreements resolved by consensus. The following data were extracted: author, publication year, study sample size, cohort demographics, study setting, total number of pregnancies, number of PAMIs, timing of MI (antepartum defined as conception until admission for delivery, peripartum defined as 24–48 h around delivery, postpartum defined as up to 6 weeks after delivery) and number of deaths due to PAMI. The primary outcome was the incidence of PAMI, determined as the number of PAMIs divided by the number of pregnancies. We also determined the rates of maternal mortality (defined as number of maternal deaths/number of total pregnancies) and case fatality (defined as number of maternal deaths/number of PAMI).

e to PAMI. The primary outcome was the incidence of PAMI, determined as the number of PAMIs divided by the number of pregnancies. We also determined the rates of maternal mortality (defined as number of maternal deaths/number of total pregnancies) and case fatality (defined as number of maternal deaths/number of PAMI). Study quality was assessed using an assessment tool based on general quality criteria developed for observational studies12 and tailored to our research question. Our quality assessment addressed whether the research question was clearly stated, the study sample selection was clearly described, study subject demographics were clearly described, inclusion and exclusion criteria were pre-specified and applied uniformly to all participants, MI diagnosis criteria were clearly defined and potentially important baseline risk factors that may influence MI incidence were identified.13 These quality elements were adjudicated as good, if clearly described and given a score of 1, or as poor if unclear, inadequately or not described and given a score of 0. These quality factors were evaluated individually and aggregated for a total quality score that could range from zero to six (out of a total score of six).

quality elements were adjudicated as good, if clearly described and given a score of 1, or as poor if unclear, inadequately or not described and given a score of 0. These quality factors were evaluated individually and aggregated for a total quality score that could range from zero to six (out of a total score of six). Data synthesis and analysis The incidence of pregnancy-associated MI, along with 95% CIs, was identified for each study. For small proportions (when the numerator is too small), the calculated lower limit of a confidence interval may fall below zero based on a Gaussian distribution. To ensure that all CIs were between 0 and 1, the Wilson score interval was calculated using a binomial distribution.14 This has been shown to be suitable for studies with small sample size and/or extreme probabilities.15 Weights for the individual studies were calculated using the inverse of the variance method. To obtain a pooled estimate of the incidence of PAMI, a fixed effect model (using the method of Mantel and Haenszel) was initially performed. The Q-statistic was calculated to assess for significant heterogeneity between the included studies. If significant heterogeneity was observed using the fixed effect model, we performed a meta-analysis using a random effects model (using the method of DerSimonian & Laird) to obtain a pooled estimate of PAMI. We used meta-regression to explore potential factors associated with heterogeneity that might affect the incidence of PAMI. These were defined a priori and included country/region of study, time period of study, timing of PAMI and study quality (individual quality factors and total quality score).

led estimate of PAMI. We used meta-regression to explore potential factors associated with heterogeneity that might affect the incidence of PAMI. These were defined a priori and included country/region of study, time period of study, timing of PAMI and study quality (individual quality factors and total quality score). All analyses were performed using STATA 13.0 (Statacorp, College Station, TX). A two-sided P-value less than 0.05 was considered statistically significant for meta-analyses and a P-value of less than 0.1 was considered statistically significant during meta-regression. Results Identification of studies Figure

led estimate of PAMI. We used meta-regression to explore potential factors associated with heterogeneity that might affect the incidence of PAMI. These were defined a priori and included country/region of study, time period of study, timing of PAMI and study quality (individual quality factors and total quality score). All analyses were performed using STATA 13.0 (Statacorp, College Station, TX). A two-sided P-value less than 0.05 was considered statistically significant for meta-analyses and a P-value of less than 0.1 was considered statistically significant during meta-regression. Results Identification of studies Figure 1 shows the study selection process. Following title/abstract screening, 51 papers remained for full text review. Of these, 17 studies comprising 75 570 508 women, met the inclusion criteria.4,5,16–29 Of the 34 excluded studies, seven reported on non-original data, 13 were not population-based, five were non-consecutive case series, three reported on cohorts with a history of prior ischemia or cardiomyopathy and six reported only on specific subtypes of MIs. Six studies which analyzed data from the same database–the US-based Nationwide Inpatient Sample (NIS) of the Healthcare Cost and Utilization Project (HCUP)–were identified,19,22,30–33 with overlapping years of study. In order to avoid duplication of data sampling, three studies were selected for inclusion in our analysis (covering the years of 2000–2002, 2004–2006, and 2008–2010), and a sensitivity analysis was performed evaluating the effects of substituting the excluded studies in their place (see Supplementary material online, Appendix S2). Such substitutions did not significantly alter the primary or secondary outcomes reported.

is (covering the years of 2000–2002, 2004–2006, and 2008–2010), and a sensitivity analysis was performed evaluating the effects of substituting the excluded studies in their place (see Supplementary material online, Appendix S2). Such substitutions did not significantly alter the primary or secondary outcomes reported. Figure 1 Article flow diagram. Of the 17 included articles, 13 reported on the primary outcome of PAMI incidence,16–18,21–27,29,31,33 eight of which also reported on mortality and/or case-fatality rates. The other four studies4,5,20,28 did not describe the incidence of PAMI but reported on mortality (total 12 studies) and/or case fatality (total 9 studies) rates due to PAMI. Details of included studies Table 1 shows details of the studies that met our inclusion criteria and reported on the incidence, maternal mortality and/or case fatality related to PAMI. The periods of study ranged from 196020 to 2011,34 and the study sample sizes ranged from 358 87421 to 12 670 12631 patients. Of the 17 study cohorts, eight were based in the United States (USA),16,17,20,22,23,26,31,33 three in Canada,4,25,29 two in the United Kingdom (UK),5,18 two in the Netherlands,21,28 one in Sweden,27 and one in Taiwan.24 No studies from low and middle income countries (LMICs) were identified. All included studies acquired their samples from either national or regional patient registries/databases or data collected from multiple centres. Table 1 Characteristics of included studies and their outcomes

one in Sweden,27 and one in Taiwan.24 No studies from low and middle income countries (LMICs) were identified. All included studies acquired their samples from either national or regional patient registries/databases or data collected from multiple centres. Table 1 Characteristics of included studies and their outcomes Study characteristics Outcomes Study Study setting Sample size Country Study period Pregnancy timing Outcomes reported MI incidence (per 100 000 pregnancies) Mortality rate (per 100 000 pregnancies) Case-fatality rate (%) Bateman et al. 2013(a) Medicaid database 854 823 USA 2000–2007 Antepartum postpartum MI incidence 7.60 (5.97–9.69) — — Bateman et al. 2013(b) Database 2 233 630 USA 2007–2011 Peripartum MI incidence; mortality; case-fatality 2.60 (2.01–3.36) 0.13 (0.05–0.40) 6.25 (1.31–17.20) Bush et al. 2013 OKOSS database 3 444 507 UK 2005–2010 Antepartum postpartum MI incidence; mortality; case-fatality 0.72 (0.49–1.07) 0.00 (0.00–0.11) 0.00 (0.00–30.85) Cantwell et al. 2011 Maternal mortality enquiry 2 294 372 UK 1994–2008 Antepartum peripartum postpartum Mortatlity — 0.48 (0.27–0.87) — Grotegut et al. 2014 National inpatient sample 12 628 746 USA 2008–2010 Peripartum postpartum MI incidence 3.05 (2.74–3.35) — — Hibbard 1975 State database 3 194 000 USA 1960–1968 Antepartum peripartum postpartum Mortality — 0.28 (0.15–0.54) — Huisman et al. 2013 Multicenter 358 874 Netherlands 2004–2006 Peripartum postpartum MI incidence; mortality; case-fatality 2.79 (1.51–5.13) 0.28 (0.05–1.58) 10 (0.25–44.50) James et al. 2006 National inpatient sample 12 595 624 USA 2000–2002 Antepartum postpartum MI incidence; mortality; case-fatality 6.82 (6.38–7.29) 0.35 (0.26–0.47) 5.12 (3.75–6.82) Kuklina et al. 2010 National inpatient sample 12 670 176 USA 2004–2006 Peripartum postpartum MI incidence 5.77 (5.49–6.05) — — Ladner et al. 2005 State database 5 393 228 USA 1991–2000 Antepartum peripartum postpartum MI incidence; mortality; case-fatality 2.80 (2.39–3.28) 0.22 (0.13–0.39) 7.95 (4.17–13.47) Lin et al. 2011 Registry 1 132 064 Taiwan 1999–2003 Peripartum MI incidence 2.92 (2.08–4.09) — — Macarthur et al. 2006 CIHI discharge database 10 032 375 Canada 1970–1998 Peripartum MI incidence; mortality; case-fatality 1.14 (0.95–1.36) 0.02 (0.01–0.07) 1.75 (0.21–6.19) Mulla et al. 2015 Hospital discharge database 1 573 740 USA 2004–2007 Antepartum peripartum postpartum MI incidence; mortality; case-fatality 6.54 (5.40–7.94) 0.64 (0.34–1.17) 9.71 (4.75–17.13) Rusen et al.

da 1970–1998 Peripartum MI incidence; mortality; case-fatality 1.14 (0.95–1.36) 0.02 (0.01–0.07) 1.75 (0.21–6.19) Mulla et al. 2015 Hospital discharge database 1 573 740 USA 2004–2007 Antepartum peripartum postpartum MI incidence; mortality; case-fatality 6.54 (5.40–7.94) 0.64 (0.34–1.17) 9.71 (4.75–17.13) Rusen et al. 2004 National registry 1 054 828 Canada 1997–2000 Antepartum peripartum postpartum Mortatlity; case fatality — 0.38 (0.15–0.98) 12.90 (3.63–29.83) Salonen Ros et al. 2001 National registry 1 003 489 Sweden 1987–1995 Antepartum peripartum postpartum MI incidence 0.60 (0.27–1.30) — — Schutte et al. 2010 Maternal mortality enquiry 2 557 208 Netherlands 1993–2005 Antepartum peripartum postpartum Mortality — 0.20 (0.08–0.46) — Wen et al. 2005 CIHI discharge database 2 548 824 Canada 1991–2000 Peripartum MI incidence; mortality; case-fatality 1.22 (0.86–1.73) 0.08 (0.02–0.29) 6.45 (0.79–21.42) MI, myocardial infarction. Study quality assessment Table

2004 National registry 1 054 828 Canada 1997–2000 Antepartum peripartum postpartum Mortatlity; case fatality — 0.38 (0.15–0.98) 12.90 (3.63–29.83) Salonen Ros et al. 2001 National registry 1 003 489 Sweden 1987–1995 Antepartum peripartum postpartum MI incidence 0.60 (0.27–1.30) — — Schutte et al. 2010 Maternal mortality enquiry 2 557 208 Netherlands 1993–2005 Antepartum peripartum postpartum Mortality — 0.20 (0.08–0.46) — Wen et al. 2005 CIHI discharge database 2 548 824 Canada 1991–2000 Peripartum MI incidence; mortality; case-fatality 1.22 (0.86–1.73) 0.08 (0.02–0.29) 6.45 (0.79–21.42) MI, myocardial infarction. Study quality assessment Table 2 displays the quality assessment of the included studies. The mean total quality score for the included studies was 4.6 out of six, with a range of three to six (indicating a range from moderate to excellent study quality). For all studies the research question, MI definition, and inclusion/exclusion criteria were clearly described. Seven studies had poor/inadequate description of subject demographics for their sample, such as the age and ethnicity of women included in the study. There was inconsistent and variable reporting of the prevalence of potentially important risk maternal factors such as body mass index, medical comorbidities (such as diabetes and hypertension) and pregnancy complications that may influence PAMI incidence. Table 2 Quality indicators of included studies

ed in the study. There was inconsistent and variable reporting of the prevalence of potentially important risk maternal factors such as body mass index, medical comorbidities (such as diabetes and hypertension) and pregnancy complications that may influence PAMI incidence. Table 2 Quality indicators of included studies Authoryear Research question described Sample described Demographics described Inclusion/exclusion criteria MI clearly defined Risk factors described Overall quality score Hibbard et al. 1975 Yes No No Yes Yes No 3 Salonen Ros et al. 2001 Yes Yes No Yes Yes No 4 Rusen et al. 2004 Yes No No Yes Yes No 3 Ladner et al. 2005 Yes No Yes Yes Yes Yes 5 Wen et al. 2005 Yes Yes No Yes Yes No 4 James et al. 2006 Yes Yes Yes Yes Yes Yes 6 Macarthur et al. 2006 Yes Yes Yes Yes Yes No 5 Schutte et al. 2010 Yes Yes Yes Yes Yes No 5 Kuklina et al. 2010 Yes Yes Yes Yes Yes No 5 Cantwell et al. 2011 Yes Yes Yes Yes Yes No 5 Lin et al. 2011 Yes No No Yes Yes No 3 Bateman et al. 2013 (a) Yes No No Yes Yes No 3 Bateman et al. 2013 (b) Yes Yes Yes Yes Yes No 5 Bush et al. 2013 Yes Yes Yes Yes Yes Yes 6 Huisman et al. 2013 Yes Yes Yes Yes Yes Yes 6 Grotegut et al. 2014 Yes No No Yes Yes Yes 4 Mulla et al. 2015 Yes Yes Yes Yes Yes Yes 6 MI, myocardial infarction.

No No Yes Yes No 3 Bateman et al. 2013 (a) Yes No No Yes Yes No 3 Bateman et al. 2013 (b) Yes Yes Yes Yes Yes No 5 Bush et al. 2013 Yes Yes Yes Yes Yes Yes 6 Huisman et al. 2013 Yes Yes Yes Yes Yes Yes 6 Grotegut et al. 2014 Yes No No Yes Yes Yes 4 Mulla et al. 2015 Yes Yes Yes Yes Yes Yes 6 MI, myocardial infarction. Incidence of myocardial infarction in pregnancy Thirteen studies reported on the primary outcome of pregnancy-associated MI incidence, including a total cohort of 66 470 100 pregnancies. The reported incidence of pregnancy-associated MI ranged from 0.6027 to 7.6017 per 100 000 pregnancies (Figure 2). When individual studies were combined in meta-analysis, there was significant heterogeneity: Q-statistic, P < 0.001; I2 = 98.5%, P < 0.001. The pooled estimate of pregnancy-associated MI using a random effects model was 3.34 per 100 000 pregnancies (95% CI: 2.09–4.58 per 100 000 pregnancies; Figure 2). Figure 2 Forest plot of incidence of pregnancy-associated myocardial infarction.

Incidence of myocardial infarction in pregnancy Thirteen studies reported on the primary outcome of pregnancy-associated MI incidence, including a total cohort of 66 470 100 pregnancies. The reported incidence of pregnancy-associated MI ranged from 0.6027 to 7.6017 per 100 000 pregnancies (Figure 2). When individual studies were combined in meta-analysis, there was significant heterogeneity: Q-statistic, P < 0.001; I2 = 98.5%, P < 0.001. The pooled estimate of pregnancy-associated MI using a random effects model was 3.34 per 100 000 pregnancies (95% CI: 2.09–4.58 per 100 000 pregnancies; Figure 2). Figure 2 Forest plot of incidence of pregnancy-associated myocardial infarction. In meta-regression two factors were associated with significant heterogeneity: country/region of study (meta-regression P = 0.006) and study period (meta-regression P = 0.04) (Table 3). In stratified analysis, studies conducted in the USA reported the highest incidence proportions of pregnancy-associated MI (4.87/100 000 pregnancies), while studies done in Canada (1.15/100 000 pregnancies) and Europe (0.84/100 000 pregnancies) reported the lowest proportions. Similarly, in stratified analysis the studies examining cohorts between 2000 and 2009 reported the highest overall incidence of pregnancy-associated MI (4.39/100 000 pregnancies). Meta-regression evaluating differences by timing of PAMI during pregnancy (reported in 9 of the 13 included studies) did not explain the observed heterogeneity (meta-regression P = 0.54), with similar rates of PAMI described during the antepartum (1.68/100 000 pregnancies), peripartum (1.10/100 000 pregnancies) and postpartum periods (1.11/100 000 pregnancies). A sensitivity analysis restricted to the five papers which reported on risk factors for PAMI18,22,23,25,26 revealed an incidence and range of PAMI which was very similar to the entire cohort of studies (3.56 events/100 000 pregnancies, range 0.73–6.82). Finally, none of the ‘study quality’ assessment components, nor the aggregate study quality score, were identified as significant sources of heterogeneity (P = 0.84, see Table 3). Table 3 Stratified meta-analysis and meta-regression of pregnancy-associated MI incidence and maternal mortality by methodological and clinical source

one of the ‘study quality’ assessment components, nor the aggregate study quality score, were identified as significant sources of heterogeneity (P = 0.84, see Table 3). Table 3 Stratified meta-analysis and meta-regression of pregnancy-associated MI incidence and maternal mortality by methodological and clinical source Potential variable Cohort stratifications MI incidence (per 100 000 pregnancies) P-value Maternal mortality (per 100 000 pregnancies) P-value Country/region Canada 1.15 (0.96–1.34) 0.006 0.06 (–0.04–0.15) 0.29 Europe 0.84 (0.29–1.40) 0.19 (–0.04–0.42) Taiwan 2.92 (1.91–3.92) N/A USA 4.87 (3.42–6.33) 0.28 (0.17–0.38) Study period 1970–1989 0.92 (0.40–1.44) 0.04 0.13 (–0.12–0.39) 0.38 1990–1999 2.27 (1.06–3.49) 0.17 (0.08–0.26) 2000–2009 4.39 (2.56–6.23) 0.28 (0.07–0.49) Study quality Range 3–6 (0–6) Variable 0.84 Variable 0.94 Pregnancy timing Antepartum 1.68 (–0.14–3.50) 0.54 N/A Peripartum 1.10 (0.43–1.77) N/A N/A Postpartum 1.11 (0.00–2.22) N/A MI, myocardial infarction.

4 0.13 (–0.12–0.39) 0.38 1990–1999 2.27 (1.06–3.49) 0.17 (0.08–0.26) 2000–2009 4.39 (2.56–6.23) 0.28 (0.07–0.49) Study quality Range 3–6 (0–6) Variable 0.84 Variable 0.94 Pregnancy timing Antepartum 1.68 (–0.14–3.50) 0.54 N/A Peripartum 1.10 (0.43–1.77) N/A N/A Postpartum 1.11 (0.00–2.22) N/A MI, myocardial infarction. Maternal mortality and case-fatality due to pregnancy-associated myocardial infarction Twelve studies reported on the secondary outcome of mortality rate due to PAMI, with a total cohort of 47 281 210 pregnancies. The rate of maternal mortality due to MI ranged from 0.0018 to 0.6426 per 100 000 pregnancies (Figure 3). In meta-analysis, there was significant heterogeneity: Q-statistic, P < 0.001; I2 = 84.4%, P < 0.001. The pooled estimate of mortality rate due to pregnancy-associated MI using a random effects model was 0.20 per 100 000 pregnancies (95% CI: 0.10–0.29 per 100 000 pregnancies; Figure 3). In meta-regression, one factor was identified which may have contributed to the heterogeneity seen: country/region of study (meta-regression P = 0.29, P-value for USA-based studies vs. Canada/Europe combined = 0.09) (Table 3). In stratified analysis, studies conducted in the USA reported the highest rates of maternal mortality due to pregnancy-associated MI (0.28/100 000 pregnancies), while studies done in Canada (0.06/100 000 pregnancies) and Europe (0.19/100 000 pregnancies) reported lower rates. Examining cohorts by period of study did not explain the observed heterogeneity (meta-regression P = 0.38), although studies examining cohorts between 2000 and 2009 reported higher overall rates of maternal mortality due to PAMI (0.28/100 000 pregnancies) relative to earlier cohorts (0.13–0.17/100 000 pregnancies). None of the ‘study quality’ assessment components, nor the aggregate study quality score, were identified as significant sources of heterogeneity (P = 0.94, see Tables 2 and 3). Timing of MI during pregnancy was not consistently reported in the studies reporting on maternal mortality due to PAMI and was not evaluated as a cause of heterogeneity.

assessment components, nor the aggregate study quality score, were identified as significant sources of heterogeneity (P = 0.94, see Tables 2 and 3). Timing of MI during pregnancy was not consistently reported in the studies reporting on maternal mortality due to PAMI and was not evaluated as a cause of heterogeneity. Figure 3 Forest plot of incidence of maternal mortality due to pregnancy-associated myocardial infarction. Nine studies (all cohort studies) reported on the secondary outcome of case-fatality rate due to PAMI, with a total cohort of 39 235  630 pregnancies. The case-fatality rate due to maternal MI ranged from 0.00%18 to 12.90%4 (Figure 4). When individual studies were combined in a meta-analysis, there was no significant heterogeneity: Q-statistic, P = 0.29; I2 = 23.7%, P =0.233. The case-fatality rates did not vary significantly by country/region or by period of study. The pooled estimate of case-fatality due to pregnancy-associated MI using a fixed effect model was 5.03% (95% CI: 3.78–6.27%; Figure 4). Figure 4 Forest plot of case-fatality rate due to pregnancy-associated myocardial infarction.

Nine studies (all cohort studies) reported on the secondary outcome of case-fatality rate due to PAMI, with a total cohort of 39 235  630 pregnancies. The case-fatality rate due to maternal MI ranged from 0.00%18 to 12.90%4 (Figure 4). When individual studies were combined in a meta-analysis, there was no significant heterogeneity: Q-statistic, P = 0.29; I2 = 23.7%, P =0.233. The case-fatality rates did not vary significantly by country/region or by period of study. The pooled estimate of case-fatality due to pregnancy-associated MI using a fixed effect model was 5.03% (95% CI: 3.78–6.27%; Figure 4). Figure 4 Forest plot of case-fatality rate due to pregnancy-associated myocardial infarction. Discussion Our study is the first systematic review and meta-analysis characterizing the incidence of pregnancy-associated MI. We determined an estimated pooled incidence of PAMI of 3.34 (2.09–4.58) per 100 000 pregnancies, based on 13 population-based studies reporting on over 66 million pregnancies in six countries. The maternal mortality rate due to PAMI among this global population of women was 0.20 (0.10–0.29) per 100 000 pregnancies, with a case fatality of 5.03% (3.78–6.27%). Our pooled estimate of the incidence of PAMI falls within the range of recently reported population-based studies. The most commonly referenced studies from the USA. in recent years reported rates of maternal MI of 2.8–6.8/100 000 pregnancies,22,23 and a recent multicenter study from the UK identified a rate of maternal MI of 0.72/100 000.18 However, there was significant heterogeneity in the incidence of pregnancy-associated MI across the population-based studies identified. Given the few number of countries included in the review and the fact that no data from LMIC countries was identified, it is unlikely that there is one ‘true’ global incidence of PAMI. Rather, differences over time and between regions will reflect true variation in this important maternal health outcome. The summary estimate of 3.34 PAMI events per 100 000 pregnancies is therefore a ‘weighted average’ of PAMI incidence. We explored the potential sources of heterogeneity by meta-regression, in which we observed a significantly higher reported incidence of maternal MI among U.S-based study cohorts (incidence of PAMI: 4.98/100 000 pregnancies) and in one study from Taiwan (2.92/100 000 pregnancies). This is much higher than the rates observed in similar developed countries (Canada and Europe) of 0.84–1.15 MIs/100 000 pregnancies.

significantly higher reported incidence of maternal MI among U.S-based study cohorts (incidence of PAMI: 4.98/100 000 pregnancies) and in one study from Taiwan (2.92/100 000 pregnancies). This is much higher than the rates observed in similar developed countries (Canada and Europe) of 0.84–1.15 MIs/100 000 pregnancies. The reasons for this apparent difference in national rates of PAMI are speculative, including differential access to medical care or population-based differences in the risk status of the maternal populations studied. Although the populations studied were from HIC, the nature of the study populations, particularly with reference to maternal age, racial mix, obesity, and other medical comorbidities which might predispose to MI, were poorly elucidated in the included studies (as various maternal comorbidities and pregnancy complications were inconsistently described in only 5 of 13 studies) due to their database methodology. As such, we were not able to assess whether the American maternal population varies with respect to known MI risk factors or comorbidities relative to other regions.

s (as various maternal comorbidities and pregnancy complications were inconsistently described in only 5 of 13 studies) due to their database methodology. As such, we were not able to assess whether the American maternal population varies with respect to known MI risk factors or comorbidities relative to other regions. It is also possible that reporting tendencies may have biased these findings. In the US-based studies, several of which were based on the National Inpatient Sample of the US-Healthcare Cost and Utilization Project (HCUP) Database, a higher sensitivity to report on PAMI may have occurred. This finding of higher maternal risk in American women, however, is concordant with recent estimates of maternal mortality (up until 2015)35 which revealed a higher overall maternal mortality rate in the USA (14 per 100 000 pregnancies) relative to Canada (7/100 000), the UK (9/100 000), the Netherlands (7/100 000) and Taiwan (14/100 000).36 The influence of improved identification and documentation of maternal deaths on estimates which are intended to be internationally comparable remains under debate. The country/region-specific rates of maternal MI identified in our study mirror the regional differences in maternal mortality–thus providing supportive evidence that these differences are a real phenomenon.

tation of maternal deaths on estimates which are intended to be internationally comparable remains under debate. The country/region-specific rates of maternal MI identified in our study mirror the regional differences in maternal mortality–thus providing supportive evidence that these differences are a real phenomenon. We identified that pregnancy-associated myocardial infarction is becoming more common, paralleling the widespread phenomenon of increasing mean maternal age due to delayed maternity,37 as well as the global rise in obesity and metabolic syndrome.38 Increasing maternal age also confers a higher likelihood of accumulating medical comorbidities such as hypertension, diabetes, and dyslipidemia, thereby increasing cardiovascular risk. This increasing incidence of maternal MI was noted in the recent study by Ladner et al.,23 which reported that the rate of maternal MI in California increased from 1 in 73 400 to 1 in 24 600 between the years 1991–2000. Similarly, in the latest periodic review on maternal mortality in the UK, maternal mortality dropped slightly relative to the previous triennium (after rising progressively over the previous 9 years) but maternal mortality due to cardiac disease continued to rise (from 1.01 deaths per million pregnancies in 1985–1987 to 2.31 deaths per million pregnancies in 2006–2008).5

mortality in the UK, maternal mortality dropped slightly relative to the previous triennium (after rising progressively over the previous 9 years) but maternal mortality due to cardiac disease continued to rise (from 1.01 deaths per million pregnancies in 1985–1987 to 2.31 deaths per million pregnancies in 2006–2008).5 We found that the rate of maternal MI appears to be highest antepartum (1.68/100 000 pregnancies), with the subsequent incidence equally distributed across the peripartum (1.10/100 000) and postpartum (1.11/100 000) periods. Although pregnancy stage was not identified as a significant source of heterogeneity in our study, it is noteworthy that the antepartum interval is long (up to 40 weeks) relative to the peripartum (1–2 days) and postpartum (6 weeks) periods. This supports the clinical impression that while many maternal MIs may occur during the antepartum period, the latter two intervals confer the greatest day-to-day risk of pregnancy-associated MI. Mechanisms for an increased ‘day to day’ risk during the peripartum and postpartum periods might include the increased metabolic demands on the heart during labour, the use of uterotonic medications,16 as well as a dramatic return of cardiac preload following delivery. A recent case series suggested that the mechanism of maternal MI may also vary by the timing of the events, with atherosclerosis and thrombosis being most common among antepartum MI events while coronary artery dissection is the most common mechanism for postpartum MIs.39 Studies such as this are subject to publication bias, however, and may not reflect the true spectrum of pathology manifested in an obstetric population. The mechanism of MI was not consistently reported in these population-based database studies, and as such whether the mechanism of pregnancy-associated MI truly varies with the timing in pregnancy will require detailed evaluation of consecutive events in a large cohort/population and application of standard diagnostic criteria.

nism of MI was not consistently reported in these population-based database studies, and as such whether the mechanism of pregnancy-associated MI truly varies with the timing in pregnancy will require detailed evaluation of consecutive events in a large cohort/population and application of standard diagnostic criteria. One of the strengths of our study is the inclusion of multiple population-based databases to synthesize all the existing worldwide published data supplemented by the search of multiple relevant national data sources. We included studies from six countries on three continents and limited our work to studies of general obstetric populations, rather than high-risk groups, in order to make the results widely generalizable. Despite the wide search, we were unable to identify any studies from LMICs. As we believe that PAMI is a significant global cause of maternal morbidity and mortality, more direct data from LMICs would be very useful to guide planning and resource allocation for maternal cardiac care in all regions; especially in light of reports of increasing incidence of non-communicable diseases. Our study has several other limitations. The rate of PAMI may have been overestimated in some population-based studies using large databases which rely on discharge diagnostic coding for the relevant (primary) outcome. Although this methodology has been validated as useful for population-based evaluation of rare outcomes,40 and the specificity of the diagnosis of myocardial infarction in database analyses may be as high as 97%,41 it is possible that the incidence of PAMI was overestimated in database studies relative to multicenter studies (in which the outcomes were more adjudicated).18 On the other hand, not all studies that we included reported on MI incidence during all time periods of pregnancy (antepartum, peripartum, and postpartum). If the timing-specific rates of MI were combined, we might arrive at a higher overall incidence of PAMI of up to 3.89/10 000 pregnancies (Table 3). Another limitation is that there is little detail regarding the mechanism of and risk factors for PAMI in the large databases required to evaluate the rare event of pregnancy-associated MI. Finally, a number of American studies that we initially identified were based on the same Nationwide Inpatient Sample database, resulting in overlapping cohorts of patients and necessitating the exclusion of studies with otherwise useful data.

ge databases required to evaluate the rare event of pregnancy-associated MI. Finally, a number of American studies that we initially identified were based on the same Nationwide Inpatient Sample database, resulting in overlapping cohorts of patients and necessitating the exclusion of studies with otherwise useful data. Conclusions Our study provides a comprehensive and global estimate of the incidence, mortality and case fatality rates of pregnancy-associated myocardial infarction. We identified regional/national differences and an increasing incidence of PAMI over time. Given the ongoing trend of increasing maternal age, as well as the rising prevalence of obesity and diabetes, further attention to and research regarding this population is needed. This is especially true in the context of the Sustainable Development Goals (SDG) which seek to end preventable maternal mortality.42 Both the Ending Preventable Maternal Mortality Strategy and the UN Secretary General’s Global Strategy for Women’s, Children’s, and Adolescent’s Health bring further nuance and detail to the vision of maternal health and stress the need for improved data and improving health systems.43,44 This is particularly relevant for maternal health, among which data on health outcomes and co-morbidities have been poorly documented or scarce. Research and collection of empirical data on maternal and perinatal outcomes should be strengthened, especially in LMICs. Supplementary material Supplementary material is available at European Heart Journal–Quality of Care and Clinical Outcomes online.

Conclusions Our study provides a comprehensive and global estimate of the incidence, mortality and case fatality rates of pregnancy-associated myocardial infarction. We identified regional/national differences and an increasing incidence of PAMI over time. Given the ongoing trend of increasing maternal age, as well as the rising prevalence of obesity and diabetes, further attention to and research regarding this population is needed. This is especially true in the context of the Sustainable Development Goals (SDG) which seek to end preventable maternal mortality.42 Both the Ending Preventable Maternal Mortality Strategy and the UN Secretary General’s Global Strategy for Women’s, Children’s, and Adolescent’s Health bring further nuance and detail to the vision of maternal health and stress the need for improved data and improving health systems.43,44 This is particularly relevant for maternal health, among which data on health outcomes and co-morbidities have been poorly documented or scarce. Research and collection of empirical data on maternal and perinatal outcomes should be strengthened, especially in LMICs. Supplementary material Supplementary material is available at European Heart Journal–Quality of Care and Clinical Outcomes online. Funding Funding was provided by the Bill and Melinda Gates Foundation and the WHO's Department of Reproductive Health and Research. Conflict of interest: The authors alone are responsible for the views expressed in this article and they do not necessarily represent the views, decisions or policies of the institutions with which they are affiliated.

Funding Funding was provided by the Bill and Melinda Gates Foundation and the WHO's Department of Reproductive Health and Research. Conflict of interest: The authors alone are responsible for the views expressed in this article and they do not necessarily represent the views, decisions or policies of the institutions with which they are affiliated. Supplementary Material Supplementary Data Click here for additional data file.

Introduction Aortic stenosis is the most common valvular disease in the Western World, affecting 1 in 8 individuals over the age of 75 years. The incidence of functionally important disease is rising in line with the ageing population, providing challenges for conventional valve replacement surgery.1 Patients over 80 years old undergoing elective cardiac surgery have more operative complications and a 10% mortality rate at 30 days; therefore, decisions around intervention in older patients are complex.2 Transcatheter aortic valve implantation (TAVI) has become a widespread and viable alternative for patients considered high risk for conventional surgery. Population modelling suggests in excess of 91 000 people fall into this category across North America each year.1 The Society of Thoracic Surgeons (STS)3 and EuroSCORE4 tools are often used to guide treatment based on the predicted risk of poor outcomes, but these scoring systems have not been designed or formally tested in TAVI populations. The application of such scores in elderly patients suitable for conventional surgery has also been questioned.5,6 Many believe that a holistic approach through frailty assessment may improve the decision-making process.

r outcomes, but these scoring systems have not been designed or formally tested in TAVI populations. The application of such scores in elderly patients suitable for conventional surgery has also been questioned.5,6 Many believe that a holistic approach through frailty assessment may improve the decision-making process. Frailty is a multimodal concept describing loss of strength, endurance, and physiological reserve across multiple systems that increases vulnerability for developing dependency or death.7 It becomes more common with age but is a very distinct concept of biological rather than chronological years; indeed, the majority of individuals over 85 years old are not frail. Common models focus on the development of a phenotype or the gradual accumulation of deficits over time, but there is no clear consensus on the best form of measurement.7–9 Within non-cardiac surgical cohorts, frailty is predictive of mortality, post-operative complications, and institutionalization.10–13 It is plausible that such measures applied to high-risk patients undergoing TAVI may improve the discrimination of current risk assessment tools for important patient outcomes. In this systematic review, we evaluate the effect of preoperative frailty on important patient outcomes after TAVI.

institutionalization.10–13 It is plausible that such measures applied to high-risk patients undergoing TAVI may improve the discrimination of current risk assessment tools for important patient outcomes. In this systematic review, we evaluate the effect of preoperative frailty on important patient outcomes after TAVI. Methods Search strategy We conducted a systematic literature review of Medline, EMBASE, and CINAHL databases between 1 January 2000 and 1 June 2015 using the key search terms of frailty (and its synonyms) and TAVI (and its synonyms) (see Supplementary material online, Appendix). Reference and forward citation searching via the Web of Science (Thomson Reuters) was performed on papers meeting the criteria for inclusion. Hand-searching using the primary search terms was performed within the three most commonly identified journals from the initial search. This was repeated using the Google Scholar search engine. Eligibility criteria We included any primary peer-reviewed paper where a measure of frailty was defined by the authors prior to TAVI, and where this was related to at least one of the predefined post-TAVI outcomes. No other assessments were adjudicated to represent frailty unless stipulated as a determinant of frailty by the authors of a study. No restrictions were placed on the age of study participants, specific vascular route, or operator technique by which TAVI was performed. Results in all languages were considered, using translation services where required to adjudicate eligibility.

nless stipulated as a determinant of frailty by the authors of a study. No restrictions were placed on the age of study participants, specific vascular route, or operator technique by which TAVI was performed. Results in all languages were considered, using translation services where required to adjudicate eligibility. The primary outcome was all-cause mortality after TAVI, either reported in the short (≤30 days) or long term (>30 days). Secondary outcomes comprised procedural complications as defined by the Valve Academic Research Consortium (VARC) standardized endpoint definitions. These include cardiovascular mortality, myocardial infarction, major stroke, bleeding, acute kidney injury requiring dialysis, and numerous other vascular complications.14 Any measures of functional capacity or patient independence after TAVI were sought as secondary outcomes where the relationship to a pre-TAVI frailty measure was presented. Review articles and non-peer-reviewed material (such as conference proceedings and poster abstracts) were excluded.

other vascular complications.14 Any measures of functional capacity or patient independence after TAVI were sought as secondary outcomes where the relationship to a pre-TAVI frailty measure was presented. Review articles and non-peer-reviewed material (such as conference proceedings and poster abstracts) were excluded. Data extraction All extracted abstracts and full-text articles meeting the inclusion criteria were assessed between three researchers (A.A., A.V., and C.H.), such that two people independently reviewed each submission. Disagreements were resolved by consensus including the third reviewer. For each study meeting the inclusion criteria, a standardized data extraction form was developed to record study design, TAVI population demographics, assessed risk of the population (STS and EuroSCORE), specific frailty measure, length to follow-up, and any data related to the primary and/or secondary outcomes. Where the relationship between frailty and outcome was qualitatively but not quantitatively expressed, primary authors were contacted in an attempt to gain additional primary data. Where the same study appeared to be reported across more than one article, only the most complete submission was included, with the aim of maximizing the volume of frailty data included.

ome was qualitatively but not quantitatively expressed, primary authors were contacted in an attempt to gain additional primary data. Where the same study appeared to be reported across more than one article, only the most complete submission was included, with the aim of maximizing the volume of frailty data included. Quality and bias assessment No validated quality assessment tool has been widely established to assess observational studies that are not designed to directly compare two groups. The Newcastle–Ottawa scale was used to provide a structured assessment of sample selection (four points), comparability (two points), and outcomes (three points).15 This gives a maximum score of 9 points. Studies were independently assessed by two reviewers and disagreement resolved by consensus: ≥7 points considered high quality for frailty reporting and <7 moderate or low quality. Publication bias was assessed in the primary end point with the greatest number of studies by creating a funnel plot and using Egger's regression test.16 We then corrected for asymmetry using the trim-and-fill method to determine an adjusted effect size.17

h quality for frailty reporting and <7 moderate or low quality. Publication bias was assessed in the primary end point with the greatest number of studies by creating a funnel plot and using Egger's regression test.16 We then corrected for asymmetry using the trim-and-fill method to determine an adjusted effect size.17 Data synthesis and analysis All included studies were observational cohorts with respect to frailty. Meta-analysis was performed when at least three studies reported a comparable end point to generate a meta-estimate. Given the wide number of frailty tools available, significant heterogeneity was expected across the studies, and therefore a random-effects model (maximum likelihood approach) was chosen to calculate summary effect estimates.18 Statistical analysis was performed using the metafor statistical package within R version 3.1.3 (http://www.r-project.org) and GraphPad Prism version 6.0 (GraphPad Software, San Diego, CA, USA). A P-value of <0.05 was considered statistically significant.

od approach) was chosen to calculate summary effect estimates.18 Statistical analysis was performed using the metafor statistical package within R version 3.1.3 (http://www.r-project.org) and GraphPad Prism version 6.0 (GraphPad Software, San Diego, CA, USA). A P-value of <0.05 was considered statistically significant. Results Search results and patient characteristics We identified 2623 abstracts from our initial search, resulting in 54 articles for full-text review to assess eligibility. Ten studies from Europe and North America met the full inclusion criteria (Figure 1). These comprised 4592 patients undergoing TAVI in whom a frailty measure was made prior to surgery. The mean age was 80–86 years, 34–53% of participants were men, and the STS-predicted 30-day mortality rates where available were between 6.3 and 16.6%. In those studies detailing the access route chosen for TAVI, the femoral approach was the most common, although this ranged from 47 to 100% of cases. The proportion of TAVI patients identified as frail varied greatly across the included studies, from 5 to 83% (Table 1). Table 1 Contextual details of included studies

%. In those studies detailing the access route chosen for TAVI, the femoral approach was the most common, although this ranged from 47 to 100% of cases. The proportion of TAVI patients identified as frail varied greatly across the included studies, from 5 to 83% (Table 1). Table 1 Contextual details of included studies Author, year Country Definition of frailty n Mean age (years) Male gender (%) Proportion frail (%) TAVI access route 30-day mortality (%) 1-year mortality (%) Ewe, 201020 The Netherlands/Italy Fried criteria based on gait speed, grip strength, weight loss, physical activity, and exhaustion 147 80 43 32.7 Femoral 51%, apical 49% 6.8 15.0 Stortecky, 201235 Switzerland Frailty index based on geriatric assessment of cognition, nutrition, timed get-up-and-go, ADLs, and disability. Scored 0–7 with ≥3 considered frail 100 84 40 49 Femoral 85%, apical 14%, subclavian 1% 8.0 19.0 Rodés-Cabau, 201236 Canada Subjective assessment of multidisciplinary team 339 81 45 25.1 Femoral 48%, apical 52% 10.6 – Kamga, 201319 Belgium ISAR score (self-reported functional dependence, recent hospitalization, impaired memory, difficulties with vision and polypharmacy) SHERPA score (age, ADLs, cognitive decline, falls, and self-perceived health) 30 86 53 ISAR: 83.3% moderate or high risk SHERPA: 73.3% moderate or high risk Femoral 100% – 26.7 Zahn, 201337 Germany Presumed subjective assessment (limited detail) 1318 82 42 17.7 Femoral 88%, apical 9%, subclavian 2%, aortic 1% – 19.9 Puls, 201438 Germany Katz index of ADLs (score <6 frail) 300 82 34 48 Femoral 47%, apical 53% 11.3 28 Seiffert, 201439 Germany Subjective assessment guided by CHSA clinical frailty scale29 score ≥6 347a 81 52 4.6 – – 24.2 Capodanno, 201440 Italy Geriatric status scale based on ADLs, cognition, continence, and mobility. Scored 0–3 with ≥2 labelled frail 1256b 82 42 24.4 – 6.1 – Debonnaire, 201541 The Netherlands/Italy Presumed subjective assessment 511 82 38 19.2 Femoral 52%, apical 48% 5.7 15.7 Green, 201542 USA Frailty score composed of serum albumin, grip strength, gait speed, and ADLs. Scored between 0 and 12 with ≥6 considered frail 244 86 52 45.1 Femoral 49%, others presumed apical 8.6 23.5 ADLs, activities of daily living.

ly Presumed subjective assessment 511 82 38 19.2 Femoral 52%, apical 48% 5.7 15.7 Green, 201542 USA Frailty score composed of serum albumin, grip strength, gait speed, and ADLs. Scored between 0 and 12 with ≥6 considered frail 244 86 52 45.1 Femoral 49%, others presumed apical 8.6 23.5 ADLs, activities of daily living. Observed mortality data refer to the whole study population including frail and non-frail individuals. aOnly the Bonn subgroup that received frailty assessment considered from this multicentre study. bOnly the development cohort of this study included. The validation data set does not contain frailty related outcome data. Figure 1 Flow diagram of reviewed studies. Definitions of frailty Frailty was identified by authors as either subjective (four studies) or objective (six studies). Subjective frailty was based on the judgement of a clinical team without reporting use of a specific tool. Objective frailty was determined by use of a tool specifically with the purpose of defining frailty, such as activity of daily living assessments, comprehensive geriatric assessment, and frailty indices. With the exception of one small study of 30 patients by Kamga et al.,19 frailty data were available as a dichotomized variable when related to outcomes, even where it had been measured on a continuous scale.

frailty, such as activity of daily living assessments, comprehensive geriatric assessment, and frailty indices. With the exception of one small study of 30 patients by Kamga et al.,19 frailty data were available as a dichotomized variable when related to outcomes, even where it had been measured on a continuous scale. Frailty and mortality Four studies (n = 1900) reported frailty (using objective measures) and early (≤30 days) mortality after TAVI (Table 2 and Figure 2), identifying greater than doubling of the risk of early death amongst patients identified as frail (HR 2.35, 95% CI 1.78–3.09, P < 0.001). All papers reported unadjusted univariate analyses for the association between frailty and mortality. There was no significant heterogeneity between studies (I2 = 0%, P = 0.33). Table 2 Early (≤30 days) outcomes related to frailty in included studies

patients identified as frail (HR 2.35, 95% CI 1.78–3.09, P < 0.001). All papers reported unadjusted univariate analyses for the association between frailty and mortality. There was no significant heterogeneity between studies (I2 = 0%, P = 0.33). Table 2 Early (≤30 days) outcomes related to frailty in included studies Author, year Outcome(s) related to frailty Adjustment Effect estimatea Lower 95% CI Upper 95% CI P-value Stortecky, 201235 30-day MACCE Nil 4.78 0.96 23.77 0.05 30-day MAACE (per unit increase in frailty index) Nil 1.66 1.14 2.44 0.01 30-day all-cause mortality Nil 8.33 0.99 70.48 0.03 30-day all-cause mortality (per unit increase in frailty index) Nil 2.18 1.32 3.61 0.002 Puls, 201438 All-cause mortality Nil 3.05 1.4 5.7 0.003 Procedural myocardial infarction Nil 1.08 0.15 7.59 0.94 Procedural major stroke Nil 0.98 0.41 2.33 0.95 Procedural TIA Nil 1.08 0.07 17.16 0.95 Life-threatening or disabling bleeding Nil 0.86 0.45 1.62 0.63 Major bleeding Nil 2.17 0.84 5.62 0.11 Minor bleeding Nil 1.50 1.05 2.16 0.03 Renal failure requiring dialysis Nil 2.01 1.09 3.70 0.02 Capodanno, 201440 All-cause mortality Nil 2.09 1.30 3.37 0.003 Green, 201542 All-cause mortality Nil 1.34 0.59 3.04 0.48 Cardiovascular mortality Nil 1.22 0.47 3.14 0.68 Major stroke Nil 0.61 0.06 6.63 0.68 Major bleeding Nil 1.74 0.69 4.42 0.24 Major vascular complications Nil 1.42 0.49 4.11 0.52 Permanent pacemaker insertion Nil 1.02 0.46 2.26 0.97 Renal failure requiring dialysis Nil 1.57 0.60 4.07 0.36 MAACE, major adverse cardiovascular and cerebral events.

Nil 1.22 0.47 3.14 0.68 Major stroke Nil 0.61 0.06 6.63 0.68 Major bleeding Nil 1.74 0.69 4.42 0.24 Major vascular complications Nil 1.42 0.49 4.11 0.52 Permanent pacemaker insertion Nil 1.02 0.46 2.26 0.97 Renal failure requiring dialysis Nil 1.57 0.60 4.07 0.36 MAACE, major adverse cardiovascular and cerebral events. aWhere not presented directly by authors, relative risk ratios calculated from two-by-two tables for those with and without frailty. Figure 2 Risk of early (≤30 days after TAVI) and late (>30 days) mortality in studies suitable for meta-analysis ordered by date of publication. Summary meta-estimate calculations based on random-effects model analysis.

aWhere not presented directly by authors, relative risk ratios calculated from two-by-two tables for those with and without frailty. Figure 2 Risk of early (≤30 days after TAVI) and late (>30 days) mortality in studies suitable for meta-analysis ordered by date of publication. Summary meta-estimate calculations based on random-effects model analysis. Seven studies (n = 3159) quantified the relationship between frailty and late mortality >30 days after TAVI, with every study completing at least 1 year of follow-up (Table 3 and Figure 2). All reported an increased risk of death amongst frail patients, with an overall effect size of HR 1.63 (95% CI 1.34–1.97, P < 0.001). The was only marginally increased by restricting analysis to studies undertaking adjustment for potential confounders (5 studies, HR 1.85, 95% CI 1.34–2.55, P < 0.001) or including only studies of higher quality for frailty reporting (4 studies, HR 1.79, 95% CI 1.28–2.50, P < 0.001). There was moderate heterogeneity (I2 = 66%, P = 0.01), which was reduced by performing a sensitivity analysis by the type of frailty measure used (Figure 3 and Supplementary material online, Figure S1). The mortality risk for frail patients was greater amongst those studies using an objective measure (HR 2.63, 95% CI 1.87–3.70, P < 0.001) rather than subjective assessment (HR 1.42, 95% CI 1.28–1.59, P < 0.001). Table 3 Late (≥30 days) outcomes related to frailty in included studies

mentary material online, Figure S1). The mortality risk for frail patients was greater amongst those studies using an objective measure (HR 2.63, 95% CI 1.87–3.70, P < 0.001) rather than subjective assessment (HR 1.42, 95% CI 1.28–1.59, P < 0.001). Table 3 Late (≥30 days) outcomes related to frailty in included studies Author, year Outcome(s) related to frailty Adjustment Effect estimatea Lower 95% CI Upper 95% CI P-value Ewe, 201020 MACCE defined as composite of death, nonfatal stroke, heart failure, or nonfatal myocardial infarction (mean follow-up of 9.1 months) Logistic EuroSCORE, peripheral vascular disease, previous CABG, baseline LVEF 4.20 2.00 8.84 <0.001 Stortecky, 201235 1-year MACCE Nil 4.89 1.64 14.6 0.003 1-year MACCE STS score 4.17 1.37 12.72 0.01 1-year MACCE Logistic EuroSCORE 4.48 1.48 13.53 0.01 1-year MACCE (per unit increase in frailty index) Nil 1.80 1.33 2.45 <0.001 1-year all-cause mortality Nil 3.68 1.21 11.19 0.02 1-year all-cause mortality STS score 2.93 0.93 9.24 0.07 1-year all-cause mortality Logistic EuroSCORE 3.29 1.06 10.15 0.04 1-year all-cause mortality (per unit increase in frailty index) Nil 1.80 1.31 2.47 <0.001 Rodés-Cabau, 201236 All-cause mortality (mean follow-up of 42 ± 15 months) Atrial fibrillation, cerebrovascular disease, COPD, eGFR, pulmonary hypertension 1.41 1.02 1.96 0.034 Late all-cause mortality (excluding mortality within 30 days of TAVI) Age, atrial fibrillation, COPD, eGFR 1.52 1.07 2.17 0.021 Kamga, 201319 1-year all-cause mortality (per 1 unit increase in SHERPA score) Unclear but likely gender, BMI, pulmonary hypertension, diabetes 2.74 1.39 5.39 0.004 Zahn, 201337 1-year mortality Nil 1.50 1.19 1.89 <0.001 Puls, 201438 All-cause mortality (median follow-up of 537 days) Age and sex 2.67 1.7 4.3 <0.0001 Seiffert, 201439 1-year mortality Age and sex 1.41 1.23 1.63 <0.001 Debonnaire, 201541 1-year mortality Nil 1.29 0.80 2.06 0.29 Green, 201542 1-year all-cause mortality (frailty dichotomized) Nil 2.18 1.27 3.75 0.005 1-year all-cause mortality (frailty dichotomized) Stepwise inclusion of variablesb with entry/stay criteria of 0.1/0.1 and a maximum of one covariate for every 10 events 2.5 1.40 4.35 0.002 1-year all-cause mortality (per unit increase in frailty score) Nil 1.12 1.02 1.22 0.01 Poor outcome (death or poor quality of lifec) at 6 months Stepwise inclusion of variablesb as above 2.21 1.09 4.46 0.03 Poor outcome (death or poor quality of lifec) at 1 year Stepwise inclusion of variablesb as above 2.40 1.14 5.05 0.0

ar all-cause mortality (per unit increase in frailty score) Nil 1.12 1.02 1.22 0.01 Poor outcome (death or poor quality of lifec) at 6 months Stepwise inclusion of variablesb as above 2.21 1.09 4.46 0.03 Poor outcome (death or poor quality of lifec) at 1 year Stepwise inclusion of variablesb as above 2.40 1.14 5.05 0.0 2 MACCE, major adverse cardiovascular and cerebral events; CABG, coronary artery bypass grafting; LVEF, left ventricular ejection fraction; COPD, chronic obstructive pulmonary disease; eGFR, estimated glomerular filtration rate; BMI, body mass index; TIA, transient ischaemic attack; STS, Society of Thoracic Surgeons. aWhere not presented directly by authors, relative risk ratios calculated from two-by-two tables for those with and without frailty. bCandidate variables: age, sex, body mass index, access route, STS score, diabetes, hypertension, angina, heart failure, New York Heart Association Class IV, coronary artery disease, previous coronary angioplasty, previous CABG, cerebrovascular disease, peripheral vascular disease, previous balloon aortic valvuloplasty, permanent pacemaker, renal disease, liver disease, chronic pulmonary disease, aortic valve mean gradient, ejection fraction, moderate or severe mitral regurgitation. cPoor quality of life defined as Kansas City Cardiomyopathy Questionnaire Overall Summary score <45 or a decrease of ≥10 points on serial testing before and after TAVI.

bCandidate variables: age, sex, body mass index, access route, STS score, diabetes, hypertension, angina, heart failure, New York Heart Association Class IV, coronary artery disease, previous coronary angioplasty, previous CABG, cerebrovascular disease, peripheral vascular disease, previous balloon aortic valvuloplasty, permanent pacemaker, renal disease, liver disease, chronic pulmonary disease, aortic valve mean gradient, ejection fraction, moderate or severe mitral regurgitation. cPoor quality of life defined as Kansas City Cardiomyopathy Questionnaire Overall Summary score <45 or a decrease of ≥10 points on serial testing before and after TAVI. Figure 3 Risk of late (>30 days after TAVI) mortality amongst frail patients. Summary meta-estimates presented grouped by type of frailty assessment used (subjective vs. objective), adjustment for confounders (unadjusted vs. adjusted) and study quality with regard to frailty reporting (high vs. low). All summary meta-estimate calculations based on random-effects model analysis. Individual study level data are presented in Supplementary material online, Figure S1.

t used (subjective vs. objective), adjustment for confounders (unadjusted vs. adjusted) and study quality with regard to frailty reporting (high vs. low). All summary meta-estimate calculations based on random-effects model analysis. Individual study level data are presented in Supplementary material online, Figure S1. Five studies provided the absolute number of deaths by frailty status allowing combined incidence estimations. This calculation totalled 3629 TAVI patients (24.6% frail) followed for the equivalent of 2717 patient years. Amongst those with frailty, 34 deaths/100 patient years were observed, against 19 deaths/100 patient years in non-frail individuals (Table 4). Two studies could not be included in the meta-analysis due to frailty being reported as a continuous variable,19 or because only a composite end point of MACCE (major adverse cardiovascular or cerebrovascular event) rather than all-cause mortality was reported.20 However, both studies did report significant associations of frailty with poorer outcomes including late mortality. Table 4 Comparisons of mortality in frail and non-frail patients after TAVI

osite end point of MACCE (major adverse cardiovascular or cerebrovascular event) rather than all-cause mortality was reported.20 However, both studies did report significant associations of frailty with poorer outcomes including late mortality. Table 4 Comparisons of mortality in frail and non-frail patients after TAVI Author, year Zahn, 201337 Puls, 201438 Capodanno, 201440 Debonnair, 201541 Green, 201542 Overall Frail (n) 233 144 306 98 110 891 Frail deaths (n) 70 80 30 20 36 236 Non-frail (n) 1085 156 950 413 134 2738 Non-frail deaths (n) 217 37 47 60 21 382 Follow-up period Mean 12.9 months Median 537 days 30 days 1 year (censored) 1 year (censored) – Frail years of follow-up 250 212 25 98 110 695 Non-frail years of follow-up 1166 230 78 413 134 2021 Death rate/100 frail patient years 28 38 120 20 33 34 Death rate/100 non-frail patient years 19 16 60 15 16 19 Significance value for difference between bold values: P<0.001.

0 days 1 year (censored) 1 year (censored) – Frail years of follow-up 250 212 25 98 110 695 Non-frail years of follow-up 1166 230 78 413 134 2021 Death rate/100 frail patient years 28 38 120 20 33 34 Death rate/100 non-frail patient years 19 16 60 15 16 19 Significance value for difference between bold values: P<0.001. Frailty and VARC outcomes There was wide variation in the reporting of secondary outcomes across the included studies, with only three studies reporting comparable outcomes in relation to frailty. Meta-analysis of these end points was therefore not possible. VARC outcome measures ≤30 days after TAVI were reported in relation to frailty status in only two of the included studies, totalling 544 patients (Table 2). Both used objective tools, and reported increased effect estimates for the risk of major bleeding and renal failure requiring dialysis in frail patients, but only the latter complication reached significance in the paper by Puls et al. (OR 2.23, 95% CI 1.12–4.47, P = 0.02). Both studies reported no increase in the risk of stroke amongst frail individuals after TAVI.

d effect estimates for the risk of major bleeding and renal failure requiring dialysis in frail patients, but only the latter complication reached significance in the paper by Puls et al. (OR 2.23, 95% CI 1.12–4.47, P = 0.02). Both studies reported no increase in the risk of stroke amongst frail individuals after TAVI. Quality and risk of bias Six studies met our frailty-defined criteria for high quality (Newcastle–Ottowa scale score ≥7), and four were considered moderate or low in quality (see Supplementary material online, Table S1). No study scored maximum points. All those considered of lower quality did not include adjustment for potential confounders of the relationship between frailty and outcomes. Publication bias was observed amongst the seven studies reporting late mortality (Egger's test for asymmetry P = 0.02). Adjustment by the trim-and-fill method (see Supplementary material online, Figure S2 funnel plot) had no effect on the size estimate, which remained statistically significant (HR 1.59, 95% CI 1.33–1.90, P < 0.001 vs. HR 1.63, 95% CI 1.34–1.97, P < 0.001 before adjustment).

rtality (Egger's test for asymmetry P = 0.02). Adjustment by the trim-and-fill method (see Supplementary material online, Figure S2 funnel plot) had no effect on the size estimate, which remained statistically significant (HR 1.59, 95% CI 1.33–1.90, P < 0.001 vs. HR 1.63, 95% CI 1.34–1.97, P < 0.001 before adjustment). Discussion In this systematic review and meta-analysis, we explored the relationship between pre-procedure frailty and outcomes after TAVI in 10 studies from Europe and North America comprising 4592 patients. We have made several important observations. First, the measurement of frailty detects a population at double the risk of both early and late mortality after TAVI. Second, using objective measures of frailty appears to identify an even more vulnerable group than ‘end-of-the-bed’ subjective assessment. However, it is worth acknowledging that such subjective frailty assessment still provides important discrimination of risk within a population already considered at ‘high risk’ for conventional surgery. Third, VARC complication rates in relation to frailty status are not well reported, with only very limited data to suggest increased risk of dialysis requirement and bleeding risk in frail patients. However, these observations were not suitable for meta-analysis and are subject to competing risk bias from the increased early mortality observed amongst those with frailty.

status are not well reported, with only very limited data to suggest increased risk of dialysis requirement and bleeding risk in frail patients. However, these observations were not suitable for meta-analysis and are subject to competing risk bias from the increased early mortality observed amongst those with frailty. A recent review by Puri et al.21 has emphasized the potential value of frailty assessment in TAVI candidates. Through the process of systematic review and meta-analysis, we have further clarified the growing body of research in this area and have numerically quantified the mortality risk of frailty identified by both objective and subjective measures. Established methods for determining those most likely to benefit from TAVI over medical management or conventional surgical aortic valve replacement are lacking. The PARTNER randomized controlled trial of high-risk severe aortic stenosis patients, demonstrated improved survival with TAVI, but 43% of patients had still died within 2 years of intervention compared with 68% with standard medical care. The stroke rate of 13.8% in the TAVI cohort was also more than double that of medically managed patients,22–25 although rates are falling as procedural techniques improve.26 TAVI as an intervention may therefore have population-level survival benefits over medical management, but the severe aortic stenosis population is heterogeneous and individual risk is likely to vary greatly.

double that of medically managed patients,22–25 although rates are falling as procedural techniques improve.26 TAVI as an intervention may therefore have population-level survival benefits over medical management, but the severe aortic stenosis population is heterogeneous and individual risk is likely to vary greatly. Mortality prediction using traditional risk assessment tools such as the STS mortality score and logistic EuroSCORE was commonly reported amongst the reviewed papers. It is possible to directly compare these figures to observe early (≤30 days) mortality in six of the included studies (see Supplementary material online, Table S2). This comparison highlights the poor correlation of predictive scores with actual outcomes in this population, which is perhaps unsurprising given these tools were developed in younger cohorts excluding TAVI. Others have also identified the weakness of existing risk scores.5,6 It is noteworthy that these predictive algorithms only provide prognostic estimates for early surgical outcomes, which may not be the most important end point after TAVI. In such complex older patients approaching the end of life, quality of life after intervention may be more important than survival or avoidance of procedural complications. A systematic review by Kim et al.27 of function and quality of life after TAVI reported mixed patient outcomes, with improvements in physical function amongst survivors not matched by changes in psychological and general health measures.

tion may be more important than survival or avoidance of procedural complications. A systematic review by Kim et al.27 of function and quality of life after TAVI reported mixed patient outcomes, with improvements in physical function amongst survivors not matched by changes in psychological and general health measures. Frailty has gained traction within surgical and cardiovascular literature as a potential metric for the currently unmeasured risk of older patients undergoing complex interventions.10–13 Whilst this may be seen as positive for the holistic care of older patients, there is wide variation in definitions and measurement. In this review, the six studies that sought to objectively measure frailty each used different tools, varying from functional scales to composite scores including nutrition, cognition, and mobility. In the absence of trial data with randomization based on frailty, it is not possible to infer which elements of these measures will carry the most prognostic weight. However, it is notable that all the tools used included some estimation of participation in activities of daily living. It is possible that such measures are particularly sensitive to procedural risk in severe aortic stenosis populations as impairments may reflect established heart failure at the time of consideration for TAVI.

it is notable that all the tools used included some estimation of participation in activities of daily living. It is possible that such measures are particularly sensitive to procedural risk in severe aortic stenosis populations as impairments may reflect established heart failure at the time of consideration for TAVI. There remains no consensus on the optimum approach to frailty assessment. The majority of studies included in this review considered frailty as a dichotomized variable for the purpose of outcome analysis. This reflects the phenotypic model of frailty and is perhaps attractive as a simple clinical concept.8 However, forcing a continuous variable into a binary form limits the consideration of a ‘pre-frail’ status and may be open to criticism for the potentially arbitrary nature of the threshold used to define frailty. Dichotomous phenotypic frailty assessment may also suffer from saturation amongst the highest-risk populations and therefore provide limited discrimination compared with an index of deficits.28 A formal frailty index, such as that first described by Rockwood et al.,29 may better reflect the accumulation of markers of frailty over time. Three of the included studies do present some outcome data per unit change in the chosen frailty index, but given the differences in the structure of these scales, meta-estimation of a combined effect size was not possible or logical.

ckwood et al.,29 may better reflect the accumulation of markers of frailty over time. Three of the included studies do present some outcome data per unit change in the chosen frailty index, but given the differences in the structure of these scales, meta-estimation of a combined effect size was not possible or logical. Although the included studies comprise 4592 patients undergoing TAVI, there are even larger published population registries in America, the UK, France, Germany, Italy, and Belgium. Unfortunately, there is currently no systematic measurement of frailty within any of these cohorts of consecutive patients.30–34 It is likely that these registries will be used to produce future TAVI-specific surgical risk assessment tools similar to STS and EuroSCORE, and therefore inclusion of frailty measurement would provide a valuable opportunity to test effectiveness in large populations.

any of these cohorts of consecutive patients.30–34 It is likely that these registries will be used to produce future TAVI-specific surgical risk assessment tools similar to STS and EuroSCORE, and therefore inclusion of frailty measurement would provide a valuable opportunity to test effectiveness in large populations. Limitations Several limitations of our review should be considered. First, there are no studies randomized by frailty status, and so it is likely that patient selection in the observational cohort studies included in our meta-analysis was already influenced by underlying and unmeasured frailty. This is inevitable given the nature of TAVI as a treatment reserved for high-risk aortic stenosis patients requiring valve replacement. Whilst this selection bias may limit interpretation of frailty measurement in a broader aortic stenosis population, the results are representative of real-world TAVI cohorts. Studies evaluating frailty and outcomes in patients referred for TAVI, but in whom the procedure was felt too high risk by their multidisciplinary team, would be informative, but to our knowledge, no such studies have been reported. Second, we have only included studies where frailty was defined by the researchers. It is possible that other data exist including similar measurements without specific use of the term frailty. However, such studies would be less likely to report outcomes directly related to these measures without acknowledging the concept of frailty. Third, the meta-estimate for early mortality is based on a small number of studies, without adjustment for potential confounders. We were limited by the infrequent reporting of standardized VARC complications in relation to frailty status, and these interpretations are open to competing risk bias. Therefore, whilst the observations of the effect of frailty on early outcomes are important, further work is required in this area. It is in this light that the addition of objective frailty measures to ongoing large TAVI registries would be helpful.

ty status, and these interpretations are open to competing risk bias. Therefore, whilst the observations of the effect of frailty on early outcomes are important, further work is required in this area. It is in this light that the addition of objective frailty measures to ongoing large TAVI registries would be helpful. Conclusions We demonstrate that frailty is associated with poorer early and late outcomes in TAVI patients. Objective frailty tools identify an even more vulnerable population at greater than double the late mortality risk of non-frail patients. There is currently a lack of consistency in frailty measures and clarity in reporting against standardized early VARC outcomes. Given the ongoing uncertainty in appropriate patient selection for TAVI, randomized controlled trials should consider including patients based on an objective assessment of frailty status. Supplementary material Supplementary material is available at European Heart Journal – Quality of Care and Clinical Outcomes online. Funding A.A. is supported by a Clinical Research Fellowship from Chest, Heart and Stroke Scotland (RES/Fell/A163), and N.L.M. is supported by an Intermediate Clinical Research Fellowship from the British Heart Foundation (FS/10/024/28266). Conflict of interest: none declared. Supplementary Material Supplementary Data Click here for additional data file.

Funding A.A. is supported by a Clinical Research Fellowship from Chest, Heart and Stroke Scotland (RES/Fell/A163), and N.L.M. is supported by an Intermediate Clinical Research Fellowship from the British Heart Foundation (FS/10/024/28266). Conflict of interest: none declared. Supplementary Material Supplementary Data Click here for additional data file. Acknowledgements We would like to acknowledge the support of our certified librarian Sheila Fisken in the preparation of the search strategy. A.A., A.M., and S.S. are members or associated members of the University of Edinburgh Centre for Cognitive Ageing and Cognitive Epidemiology, part of the cross council Lifelong Health and Wellbeing Initiative, who provided systematic review training.

‘Mrs Jones, based on your risk factors for having a heart attack, I recommend that we start you on a statin’. ‘No, thank you, doctor, I’ve read too many scary things about those drugs on the internet. Plus, I worry that some in your profession make these recommendations for reasons of personal financial gain. I also found that online’. Undoubtedly, the majority of cardiologists have had conversations just like this, urging a patient to take a statin, powerful cholesterol-lowering drugs with robust mortality benefit. Part of the reason these oftentimes ‘no brainer’ recommendations are rejected derives from widely disseminated incorrect information which vastly over-states the risks of these drugs. (Of course, like anything in life, statin use is not entirely risk-free; their application should always entail a thoughtful analysis of risks vs. benefits.) Most patients do not recognize that the benefits of statin use are invisible (‘I didn’t have a heart attack or stroke this past year’.), whereas the small and typically reversible risks (e.g. muscle pain) are readily apparent. Many patients who would benefit from statin use do not take them. Cardiovascular disease is the no. 1 killer of both men and women around the world. Robust scientific advances, published in the pages of our journals, have fostered significant improvements that benefit individuals and society. Yet, cardiovascular disease continues to transform itself, emerging in new forms, such as heart failure. The struggle has shifted to new battlefields.

women around the world. Robust scientific advances, published in the pages of our journals, have fostered significant improvements that benefit individuals and society. Yet, cardiovascular disease continues to transform itself, emerging in new forms, such as heart failure. The struggle has shifted to new battlefields. These successes derive from an armamentarium of powerful tools—medicines and devices—and awareness of lifestyle-related hazards, such as high blood pressure, high cholesterol, and smoking. Sadly, however, we do not take full advantage of the tools at our disposal. One significant cause of suboptimal utilization of our prodigious tool chest is medical misinformation hyped through the internet, television, chat rooms, and social media. In many instances, celebrities, activists, and politicians convey false information; not uncommonly, authors with purely venal motives participate.

One significant cause of suboptimal utilization of our prodigious tool chest is medical misinformation hyped through the internet, television, chat rooms, and social media. In many instances, celebrities, activists, and politicians convey false information; not uncommonly, authors with purely venal motives participate. We can point to numerous other examples, including the entirely unfounded concerns regarding vaccinations. The notion MMR (measles, mumps, rubella) vaccination causes autism was based on a single, flawed study, long since refuted, and its publication retracted. Seventeen much larger and properly controlled studies have proven otherwise. Nevertheless, the internet shouts unfounded warnings. Once again, celebrities, actors, activists, and politicians with no specific knowledge or training use their fame to promote a message that causes serious harm. Individuals who are neither physicians nor scientists, but often with a specific agenda, have outsized influence over our lives. They dispute scientific evidence without ever having studied it.1 Recognizing that it is impossible to prove ‘never’, scientists appropriately couch their statements in statistical terms, which may come across to the public as equivocation. The nuanced voices of scientists often do not resonate with the public as much as the strident alarms sounded by people of fame, speaking in absolute terms.

that it is impossible to prove ‘never’, scientists appropriately couch their statements in statistical terms, which may come across to the public as equivocation. The nuanced voices of scientists often do not resonate with the public as much as the strident alarms sounded by people of fame, speaking in absolute terms. Further, scientists are appropriately sceptical, as any individual scientist or study can be wrong. Yet, science ultimately self-corrects. When a scientist gets it wrong, as happens, people sometimes vilify the entire, self-correcting scientific enterprise. We trust aeronautical science when we board an aeroplane; we trust the science buried within our cell phones; we trust mechanical engineering science when we cross a bridge; yet, many are uniquely sceptical of biological science. Sadly, we cannot exclude that some in the professions of science and medicine act based on motives driven by financial considerations; incomplete declarations of potential conflict of interest persist.2 Recent examples of dramatic price hikes for important medications have reinforced this notion. Indeed, many physicians have had conversations with patients who believe that our recommendations stem, at least in part, from the prospect of personal financial gain. We, the editors-in-chief of the major cardiovascular scientific journals around the globe, sound the alarm that human lives are at stake. Pointing to the two examples elaborated above, people who decline to use a statin when recommended by their doctor, or parents who withhold vaccines from their children, put lives in harm’s way.

ors-in-chief of the major cardiovascular scientific journals around the globe, sound the alarm that human lives are at stake. Pointing to the two examples elaborated above, people who decline to use a statin when recommended by their doctor, or parents who withhold vaccines from their children, put lives in harm’s way. The media must do a better job. It is unacceptable to posit false equivalents in these discussions, often done to foster debate and controversy. It is easy to find a rogue voice but inappropriate to suggest that voice carries the same weight as that emerging from mainstream science. (We can easily point to examples outside the medical domain, as well, such as climate change, evolution, nutraceuticals, and GMO foods where false equivalents are frequently posited.) Furthermore, recent evidence suggests that misinformation travels faster through social networks than truth.3 We must work to enhance science literacy in our world; one place to start is by doing a better job of teaching the scientific method in our schools so that the lay public is aware that science is accomplished in fits and starts, but in the end, gets it right.

ormation travels faster through social networks than truth.3 We must work to enhance science literacy in our world; one place to start is by doing a better job of teaching the scientific method in our schools so that the lay public is aware that science is accomplished in fits and starts, but in the end, gets it right. Purveyors of social media must be responsible for the content they disseminate. It is no longer acceptable to hide behind the cloak of ‘platform’. We, as editors, are charged with evaluating the validity of the science presented to us for possible publication, and we work hard to fulfil this heady responsibility. Recognizing that lives are at stake, we reach out to thought-leading experts to evaluate the veracity of each report we receive. Here, we challenge social media to do the same, to leverage the ready availability of science-conversant expertise before disseminating content that may not be reliable. Without exaggeration, significant harm, to society and individuals, derives from the wanton spread of medical misinformation. It is high time that this stop, and we lay at the feet of the purveyors of internet and social media content the responsibility to fix this. Conflict of interest: P.G. Camici is consultant for Servier. R.S. Rosenson reports research grants to his institutions from Akcea, Amgen, Astra Zeneca, Medicines Company and Regeneron. R.S. Rosenson reports speaking engagements at Amgen and Kowa, research consulting for Akcea and Regeneron, royalties from UpToDate, Inc. and stock holdings in MediMergent. All other authors have nothing to disclose.

ts research grants to his institutions from Akcea, Amgen, Astra Zeneca, Medicines Company and Regeneron. R.S. Rosenson reports speaking engagements at Amgen and Kowa, research consulting for Akcea and Regeneron, royalties from UpToDate, Inc. and stock holdings in MediMergent. All other authors have nothing to disclose. Footnotes A complete list of all journals publishing this article, along with links to the individual articles, can be found online at https://www.ahajournals.org/circ/medical-misinformation

Introduction The population of older adults is growing, and intrinsic susceptibility to aortic stenosis is high with this new demographic scenery. Improved decision making is necessary for the expanding population of those eligible for transcatheter aortic valve implantation (TAVI).1 Commonly used risk scores for mortality and morbidity in coronary heart surgery, like the Society of Thoracic Surgeons risk score (STS score) and European System for Cardiac Operative Risk Evaluation (EuroSCORE), are based on age and comorbidity.2,3 However, by omitting frailty, sensitivity of these scores to predict adverse events in the oldest population is limited.4–6 Frailty, a condition frequent in older adults, is defined as a state of impaired physiologic reserve and decreased resistance to stressors which increase the risk of an adverse outcome.7,8 Frailty status enhances prognostic sensitivity for patients with multiple heart conditions including acute coronary disease, stable angina, heart failure, 9–11 and TAVI.12,13 A recent systematic review confirmed the relationship between frailty and mortality in the TAVI population, with a more than doubled risk [hazard ratio (HR) 2.35] of early (≤30 days) death in frail patients, and a 1.63 HR of later death.14 Although TAVI has been assessed to be cost-effective compared with medical treatment,15 this is undermined by early mortality after TAVI. As the population of older adults expands, it is important to select patients who will benefit most from the intervention to best justify its expense.16

d a 1.63 HR of later death.14 Although TAVI has been assessed to be cost-effective compared with medical treatment,15 this is undermined by early mortality after TAVI. As the population of older adults expands, it is important to select patients who will benefit most from the intervention to best justify its expense.16 Both US and European guidelines recommend the use of a Heart Team in decision making prior to treatment for severe, symptomatic aortic stenosis.17,18 In addition to the assessments by the interventional cardiologist, cardiac surgeons and imaging specialists, the guidelines recommend a frailty assessment to evaluate cognition and physical function using validated checklists.17 However, it is not described in detail who should perform and evaluate the frailty assessment and which tools to use.1,17,19 Recently, Afilalo et al.20 demonstrated that the essential frailty toolset (EFT) outperformed other frailty scores in predicting 1-year mortality in TAVI patients. Nonetheless the authors emphasized that the EFT is primary a screen for frailty. Once patients are identified by the EFT, further geriatric assessment (GA) is recommended. This demands a more thorough clinical evaluation. We developed a novel frailty score that provides additional information, based on a comprehensive GA. In this study, we show the utility of this novel GA frailty score to predict 2-year mortality, showcasing its powerful prognostic value.

assessment (GA) is recommended. This demands a more thorough clinical evaluation. We developed a novel frailty score that provides additional information, based on a comprehensive GA. In this study, we show the utility of this novel GA frailty score to predict 2-year mortality, showcasing its powerful prognostic value. Methods Study design A prospective, observational cohort study with 2-year follow-up and inclusion of elective TAVI patients from 2011 to 2015. The study was approved by the Regional Committee for Medical Research Ethics (REK 2010/2936-6 and 2013/1310). All participants signed an informed consent before assessment. Participants Patients with severe and symptomatic aortic stenosis accepted for TAVI were recruited from a tertiary university hospital in Western Norway serving a population of 1.1 million. All patients were recruited the day before the procedure and were assessed by a Heart Team consisting of a cardiac surgeon, an interventional cardiologist and an imaging specialist. Based on the evaluation, patients were all turned down for open heart surgery due to comorbidity and/or high EuroSCORE. The recruitment period lasted from February 2011 to April 2015. From February 2011 to September 2013, 65 patients ≥80 years also participating in a concomitant study of delirium were included.21 From October 2013 to April 2015, 82 patients ≥70 years were included (Figure 1). Age was then adjusted to 70 years as frailty was assessed to be important also in this younger group. Exclusion criteria were declined consent or inability to understand and/or speak Norwegian.

oncomitant study of delirium were included.21 From October 2013 to April 2015, 82 patients ≥70 years were included (Figure 1). Age was then adjusted to 70 years as frailty was assessed to be important also in this younger group. Exclusion criteria were declined consent or inability to understand and/or speak Norwegian. Figure 1 Patient recruitment flowchart. Severe aortic stenosis was defined as maximal Doppler velocity across the aortic valve ≥4 m/s, a mean gradient ≥40 mmHg or an aortic valve area <1 cm2 (indexed area <0.6 cm2/m2) and concomitant clinical symptoms indicating severe aortic stenosis.

nswering yes to the question ‘have you lost weight during the past year’), (ii) one point if the patient is unable to rise from a chair without using their arms (This was tested by L.S.P.E./E.S., not reported by the patients.), and (iii) one point if the patient answers no to the question ‘Do you feel full of energy?’. Comorbidity was assessed by Charlson comorbidity index. This is a weighted index based both on the numbers of diseases and the seriousness of each disease. A score of 1 is assigned for myocardial infarction, congestive heart failure, dementia, etc., while the highest score of 6 is given to metastatic solid tumours and AIDS. In the original paper describing the index, Charlson et al.32 recommends a high cut-off of 2 or 3 if the mortality in the disease under study is high, and we chose a cut-off ≥3. Psychological health was assessed by the Hospital Anxiety and Depression Scale (HADS),33 with seven questions on anxiety and seven on depression. Each question ranges from 0 to 3. Summing up the anxiety and depression subscales, we get total HADS, of which a cut-off ≥15 was used to identify symptoms of anxiety and/or depression.33,34 The modified essential frailty toolset The Essential Frailty Toolset (EFT) is a brief four-item (chair rise, cognition, haemoglobin, and serum albumin) frailty scale that predicts morbidity and mortality after TAVI.20 Afilalo recommends applying this scale as a screening tool.20 In this study, we aimed to compare the GA frailty score to the EFT.

oncomitant study of delirium were included.21 From October 2013 to April 2015, 82 patients ≥70 years were included (Figure 1). Age was then adjusted to 70 years as frailty was assessed to be important also in this younger group. Exclusion criteria were declined consent or inability to understand and/or speak Norwegian. Figure 1 Patient recruitment flowchart. Severe aortic stenosis was defined as maximal Doppler velocity across the aortic valve ≥4 m/s, a mean gradient ≥40 mmHg or an aortic valve area <1 cm2 (indexed area <0.6 cm2/m2) and concomitant clinical symptoms indicating severe aortic stenosis. Development of a novel frailty score The GA frailty score was developed based on a comprehensive GA which includes cognition, instrumental activity of daily living, nutrition, physical frailty, comorbidity, and psychological health.22,23 The method for developing this score is described by Harrell.24 In this method expert clinicians assign severity points to each condition and sum the points in a total score. Three geriatricians (A.W.S., A.H.R., and E.S.) and one cardiologist (J.E.N.) independently ranked the clinical severity of signs within each potentially important domain. The suggestions were sent to the first author who developed a combined frailty score based on the different proposals.24 All cut-off values in this combined score were based on previous studies.13,22 The researchers then agreed on the GA frailty score, a 0–9 point numeric scale with 8 validated geriatric variables (Table 1). The score was finalized before the statistical analysis were performed. Table 1 Geriatric assessment tools used in the novel frailty score, along with the corresponding scoring scheme

ies.13,22 The researchers then agreed on the GA frailty score, a 0–9 point numeric scale with 8 validated geriatric variables (Table 1). The score was finalized before the statistical analysis were performed. Table 1 Geriatric assessment tools used in the novel frailty score, along with the corresponding scoring scheme Domain Cut-off Points Cognition MMSE ≥27 0 MMSE 20–26 1 MMSE <20 2 Instrumental activity of living NEADL ≤43 1 Nutrition BMI <20.5 1 Energy level SOF index Low energy 1 Weight loss SOF indexa Weight loss 1 Limb strength SOF index Chair stand (not able) 1 Comorbidity Charlson comorbidity index ≥3 1 Psychological factors HADS (total score) ≥ 15 1 Total Maximum score 9 The total score is calculated by adding the different domain scores. BMI, body mass index; HADS, Hospital Anxiety and Depression Scale; MMSE, Mini Mental Status Examination; NEADL, Nottingham Extended Activity of Daily Living Scale; SOF Study of Osteoporotic Fractures index. a Modified from the original SOF; see ‘Measurements’ section for details. Measurements Novel frailty score Baseline data were collected by L.S.P.E. and E.S. All baseline examinations were performed the day before the procedure. Cognition was assessed by the Mini Mental Status Examination (MMSE).25 It has a range from 0 to 30, with higher scores indicating better cognition. Different cut-offs are reported, and we chose a weighted score with one point for possible cognitive impairment/mild dementia and two points for probable dementia.13,26

re. Cognition was assessed by the Mini Mental Status Examination (MMSE).25 It has a range from 0 to 30, with higher scores indicating better cognition. Different cut-offs are reported, and we chose a weighted score with one point for possible cognitive impairment/mild dementia and two points for probable dementia.13,26 Instrumental activities of daily living was measured by Nottingham Extended Activities of Daily Living scale (NEADL),27 a 22-item questionnaire assessing mobility, kitchen, domestic, and leisure activities. Each item has a score from 0 to 3, and the items are added to a total score from 0 to 66, with a higher score indicating better functioning. A cut-off ≤43 suggests that the patient is dependent, and studies have shown that this predicts complications and mortality after elective surgery in older patients.28,29 Nutrition was assessed by the body mass index (BMI) and the weight question of modified Study of Osteoporotic Fractures index. The cut-off value for BMI was based on the nutritional risk screening 2002, a screening instrument for nutritional risk.30

Instrumental activities of daily living was measured by Nottingham Extended Activities of Daily Living scale (NEADL),27 a 22-item questionnaire assessing mobility, kitchen, domestic, and leisure activities. Each item has a score from 0 to 3, and the items are added to a total score from 0 to 66, with a higher score indicating better functioning. A cut-off ≤43 suggests that the patient is dependent, and studies have shown that this predicts complications and mortality after elective surgery in older patients.28,29 Nutrition was assessed by the body mass index (BMI) and the weight question of modified Study of Osteoporotic Fractures index. The cut-off value for BMI was based on the nutritional risk screening 2002, a screening instrument for nutritional risk.30 Physical frailty was assessed by a modified version (patients self-reported weight loss past year, not measured as in the original index) of the Study of Osteoporotic Fractures Index (mSOF index).31 This validated index has a maximum of three points: (i) One point if the patient has >5% weight loss the previous year (Since, we only had baseline characteristics, the patients were given one point if answering yes to the question ‘have you lost weight during the past year’), (ii) one point if the patient is unable to rise from a chair without using their arms (This was tested by L.S.P.E./E.S., not reported by the patients.), and (iii) one point if the patient answers no to the question ‘Do you feel full of energy?’.

lty toolset The Essential Frailty Toolset (EFT) is a brief four-item (chair rise, cognition, haemoglobin, and serum albumin) frailty scale that predicts morbidity and mortality after TAVI.20 Afilalo recommends applying this scale as a screening tool.20 In this study, we aimed to compare the GA frailty score to the EFT. However, for the first 62 patients in our study, we only had information on success/failure of chair rise, not on the number of seconds it took to complete the chair rises. Therefore, when calculating the EFT for these patients, we assigned 0 points if they completed five sit-to-stand repetitions without using arms (chair rises) and 2 points if they failed to complete all five chair rises. We refer to this modified methodology for the EFT as the modified Essential Frailty Toolset (mEFT). This might give some patients one point lower total score (i.e. the patients who used ≥15 s to perform chair rise). For three patients, we missed serum albumin values, and the mEFT was thus calculated for 139 patients. Follow-up measurements Two-year all-cause mortality has been stated as a clinically relevant outcome for TAVI candidates and was the primary outcome of this study.1 The Valve Academic Research Consortium-2 (VARC 2) consensus document6 recommends the use of composite endpoints after TAVI, and we report this for the first 6 months.

easurements Two-year all-cause mortality has been stated as a clinically relevant outcome for TAVI candidates and was the primary outcome of this study.1 The Valve Academic Research Consortium-2 (VARC 2) consensus document6 recommends the use of composite endpoints after TAVI, and we report this for the first 6 months. Power analysis The initial power analysis was based on categorizing the patients into three groups, a fit group, an intermediate group and a frail group, with 25% in the frail group.22,35,36 To achieve a power of 80% with a 5% level of significance, power calculations showed that we needed a total of 140 patients. To account for dropouts, we included 5% more, a total of 147 patients. In order to make the frailty score more applicable in clinical practice, we ultimately dichotomized it into frail and non-frail (fit and intermediate). In addition, we analysed frailty as a continuous score, which increases the statistical power.

s. To account for dropouts, we included 5% more, a total of 147 patients. In order to make the frailty score more applicable in clinical practice, we ultimately dichotomized it into frail and non-frail (fit and intermediate). In addition, we analysed frailty as a continuous score, which increases the statistical power. Statistical analyses We present the data as means and standard deviations (SDs), counts and percentages, or proportions and Hazard Ratios (HRs) with 95% confidence intervals (CIs), as appropriate. To assess whether the new frailty score could predict mortality within 2 years, and also when adjusted for other common predictors, we fitted Cox regression models with Firth’s correction. Firth’s correction provides reduced bias when there are few events (deaths) compared with the number of predictors. The regression models included frailty score as a continuous predictor (unadjusted model/trend test), or frailty score, age, gender, and logistic EuroSCORE as predictors (adjusted model). We also fit a similar adjusted model with frailty score as a dichotomized variable. We present time to death stratified by frailty score (continuous or dichotomized ) using Kaplan–Meier plots.

predictor (unadjusted model/trend test), or frailty score, age, gender, and logistic EuroSCORE as predictors (adjusted model). We also fit a similar adjusted model with frailty score as a dichotomized variable. We present time to death stratified by frailty score (continuous or dichotomized ) using Kaplan–Meier plots. The Receiver Operating Characteristic (ROC) curve37 was examined to find cut-off values for the dichotomized GA frailty score. We reported the Area Under the Curve (AUC) as a summary measure. We found two cut-off values with an estimated high sensitivity and specificity, and chose the one (≥4) emphasizing specificity over sensitivity. Confidence intervals for the sensitivity and specificity were calculated using the Wilson (score) method.38 Some patients had missing data for a few of the questions in the HADS and NEADL questionnaires. Where it was unambiguous on which side of the cut-off the total score would fall on, we used the data for these patients; otherwise, the patients were excluded. For one secondary analysis (based on the mEFT frailty scale), there were additional missing data. For all analyses, we report the number of observations used. Statistical analyses were performed using IBM SPSS Statistics 24 and R version 3.5.0.39 Cox regression with Firth’s correction was performed using R ‘coxphf’ package40 version 1.13, and the ROC and AUC calculations were performed using the R ‘pROC’ package version 1.12.1.41

Some patients had missing data for a few of the questions in the HADS and NEADL questionnaires. Where it was unambiguous on which side of the cut-off the total score would fall on, we used the data for these patients; otherwise, the patients were excluded. For one secondary analysis (based on the mEFT frailty scale), there were additional missing data. For all analyses, we report the number of observations used. Statistical analyses were performed using IBM SPSS Statistics 24 and R version 3.5.0.39 Cox regression with Firth’s correction was performed using R ‘coxphf’ package40 version 1.13, and the ROC and AUC calculations were performed using the R ‘pROC’ package version 1.12.1.41 Results Baseline data General characteristics A total of 147 patients with severe and symptomatic aortic stenosis were included. Of these, 142 patients had enough data so that the frailty score could be computed (Figure 1). Of the 142 patients, 54% were women, mean age was 83 years (SD 4), five patients were less than 75 years old and three patients were 90 years or older. The oldest patient in the study was 95 years old. More than half of the patients lived with their spouse.

h data so that the frailty score could be computed (Figure 1). Of the 142 patients, 54% were women, mean age was 83 years (SD 4), five patients were less than 75 years old and three patients were 90 years or older. The oldest patient in the study was 95 years old. More than half of the patients lived with their spouse. Geriatric characteristics More than half of the patients (56%) did not have significant cognitive disturbance, a MMSE of 27 or higher. The others (44%) had possible cognitive impairment, but for most of them (89%) probably mild cognitive impairment or mild dementia (MMSE 20–26). Most patients (82%) had a NEADL score above 43, suggesting they were independent in activities of daily living. Few patients (13%) had low BMI (below 20.5 kg/m2); however, 52 (37%) patients had a reported weight loss during the last year. Sixty-one (43%) of the patients had a high score of ≥3 on the Charlson comorbidity scale. Cardiovascular characteristics Almost all patients 127/135 (missing data on seven patients) had an indexed aortic valve area below 0.6 cm2/m2. Logistic EuroSCORE was below 10 in 18% and over 20 in 30% of the patients. Half of the patients had New York Heart Association III or IV at the time of the procedure (Table 2). Table 2 Patient baseline characteristics (n = 142)

127/135 (missing data on seven patients) had an indexed aortic valve area below 0.6 cm2/m2. Logistic EuroSCORE was below 10 in 18% and over 20 in 30% of the patients. Half of the patients had New York Heart Association III or IV at the time of the procedure (Table 2). Table 2 Patient baseline characteristics (n = 142) Mean or count SD or (proportion) Characteristics Age, years 83.4 4.0 Women 76 (54%) Living alone 60 (42%) Education Primary school 88 (62%) High school 33 (23%) University 21 (15%) Geriatric characteristics Cognition MMSE 26.3 3.3 MMSE ≥27 80 (56%) MMSE 20–26 55 (39%) MMSE <20 7 (5%) Activities of daily living NEADL ≤43 121 (82%) Nutrition BMI 25.0 3.9 BMI <20.5 19 (13%) SOF index Weight lossa 52 (37%) Low energy 58 (41%) Unable to chair stand 42 (30%) Comorbidity Charlson comorbidity index 2.53 1.3 Charlson comorbidity index ≥3 61 (43%) Psychological factors HADS ≥15 17 (12%) Cardiovascular characteristics Logistic EuroSCORE 17 8.7 Aortic valve area index, cm2/m2 b 0.4 0.12 Mean aortic valve gradient, mmHgc 47.6 14.4 Left ventricular ejection fraction 56.4 11 NYHA ≥III 67/134 (50%) Previous myocardial infarction 34 (24%) CABG 31 (22%) Permanent pacemaker 12 (9%) Atrial fibrillation 45 (32%) Pulmonary hypertension 45/139 (32%) Cerebral vascular disease 16 (11%) Comorbidity COPD 31 (22%) Kidney failure; creatinine >177 µmol/Ld 5 (4%) MMSE, Mini Mental Status Examination; NEADL, Nottingham Extended Activities of Daily Living Scale; BMI, Body Mass Index; SOF Index, Study of Osteoporotic Fractures Index; HADS, Hospital Anxiety and Depression Scale; NYHA, New York Heart Association Functional Classification of Heart Failure, Range From I-IV, Most Severe Symptoms at IV; CABG, Coronary Artery Bypass Grafting; COPD, Chronic Obstructive Pulmonary Disease.

I, Body Mass Index; SOF Index, Study of Osteoporotic Fractures Index; HADS, Hospital Anxiety and Depression Scale; NYHA, New York Heart Association Functional Classification of Heart Failure, Range From I-IV, Most Severe Symptoms at IV; CABG, Coronary Artery Bypass Grafting; COPD, Chronic Obstructive Pulmonary Disease. a Modified from the original SOF index; see ‘Measurements’ section for details. b Missing data on seven patients. c Missing data on two patients. d As reported in the PARTNER study; creatinine >2 mg/dL (177 µmol/L).1 Follow-up No patients were lost to follow-up. Mortality and morbidity Fifteen patients (11%) had died within 2 years, 11 of cardiovascular causes and four of non-cardiovascular causes. There was a high degree of early device success, with 141/142 (99.3%) valves in the correct position with good valve function. Early (≤30 days) mortality was seen in four patients (2.8%). Moderate to severe prosthetic valve regurgitation and stroke occurred within 6 months in 12.7% and 4.8% of the patients, respectively (Table 3). Table 3 Composite endpoints according to Valve Academic Research Consortium-2 consensus documenta criteria

nction. Early (≤30 days) mortality was seen in four patients (2.8%). Moderate to severe prosthetic valve regurgitation and stroke occurred within 6 months in 12.7% and 4.8% of the patients, respectively (Table 3). Table 3 Composite endpoints according to Valve Academic Research Consortium-2 consensus documenta criteria Total (N = 142) Percent Device success Absence of immediate procedural mortalityb 142 100 Correct positioning 141 99.3 Intended performance of the prosthetic heart valvec 141 99.3 No moderate or severe prosthetic valve regurgitationd 135 95.1 Early safety(at 30 days) All-cause mortality 4 2.8 All stroke(disabling or non-disabling) in hospitale 4 2.8 Life-threatening or disabling bleeding 8 5.6 Acute kidney injury Stage 2 or 3f 3 2.1 Coronary artery obstruction requiring intervention 1 0.7 Major vascular complication 6 4.2 Valve-related dysfunction requiring intervention 1 0.7 Clinical efficacy(30 days–6 months) All-cause mortality 6 4.2 All stroke(disabling or non-disabling) 3 2.1 Requiring hospitalizations for valve-related symptoms or worsening congestive heart failure 15 10.6 NYHA class III or IVg 9/136 6.6 Time-related valve safety Structural valve deterioration Valve-related dysfunction (mean aortic valve gradient ≥20 mmHg) and/ or moderate or severe prosthetic valve regurgitationh 18/141 12.7 Requiring repeat procedure (TAVI or SAVR) 1 0.7 Prosthetic valve endocarditis 1 0.7 Trombo-embolic events (e.g. stroke) 7 4.8 VARC bleeding (life threatening/disabling bleeding or major bleeding) and unless clearly unrelated to valve therapy (e.g. trauma) 28 19.7 SAVR, surgical aortic valve replacement; TAVI, transcatheter aortic valve implantation; NYHA New York Heart Association.

endocarditis 1 0.7 Trombo-embolic events (e.g. stroke) 7 4.8 VARC bleeding (life threatening/disabling bleeding or major bleeding) and unless clearly unrelated to valve therapy (e.g. trauma) 28 19.7 SAVR, surgical aortic valve replacement; TAVI, transcatheter aortic valve implantation; NYHA New York Heart Association. a The Valve Academic Research Consortium (VARC)-2 consensus document (see references). b Immediate or consequent death ≤72 h post-procedure. c No prosthesis patient mismatch and mean aortic valve gradient <20 mmHg or peak velocity <3 m/s. d After TAVI procedure at index hospitalization. e Assessment of stroke at index. All strokes verified by imaging (CT or MRI). f Evaluation of acute kidney injury is based on serum creatinine, we miss data on urine output. g New York Heart Association (NYHA), missing data on six patients. h Follow-up at 6 months, missing data on one patient. Frailty and mortality The distribution of frailty scores and the corresponding 2-year mortality is shown in Table 5. Based on the dichotomized GA frailty score, 34 patients (24%) were characterized as frail (score ≥4).

g New York Heart Association (NYHA), missing data on six patients. h Follow-up at 6 months, missing data on one patient. Frailty and mortality The distribution of frailty scores and the corresponding 2-year mortality is shown in Table 5. Based on the dichotomized GA frailty score, 34 patients (24%) were characterized as frail (score ≥4). The Cox analyses showed that the continuous GA frailty score predicted mortality within 2 years, with an estimated HR of 1.79 (95% CI: 1.34–2.36, P < 0.001), i.e. an estimated 79% increase in hazard for a unit increase in GA frailty score. This predictive power also remained (HR = 1.75, 95% CI: 1.28–2.42, P < 0.001) when adjusting for age, gender, and logistic EuroSCORE (Table 4). A test of the proportional hazard assumption did not find any problems with the model (P = 0.77). Table 4 Cox regression (with Firth’s correction) (n = 142) Unadjusted Adjusted HR 95% CI P-value HR 95% CI P-value Age, years 1.16 1.00–1.37 0.04 1.16 1.01–1.37 0.04 Male gender 1.01 0.37–2.71 0.99 2.14 0.68–6.93 0.19 Logistic EuroSCORE 1.06 1.01–1.11 0.02 1.04 0.99–1.08 0.13 GA frailty score 1.79 1.34–2.36 0.001 1.75 1.28–2.42 <0.001 CI, confidence interval; HR, hazard ratio. Hazard ratio estimates (with 95% confidence intervals) for death within 2 years of transcatheter aortic valve implantation. The hazard ratios show the increase in hazard for a unit increase in age (years), logistic EuroSCORE and/or GA frailty score, and for males compared with females.

Unadjusted Adjusted HR 95% CI P-value HR 95% CI P-value Age, years 1.16 1.00–1.37 0.04 1.16 1.01–1.37 0.04 Male gender 1.01 0.37–2.71 0.99 2.14 0.68–6.93 0.19 Logistic EuroSCORE 1.06 1.01–1.11 0.02 1.04 0.99–1.08 0.13 GA frailty score 1.79 1.34–2.36 0.001 1.75 1.28–2.42 <0.001 CI, confidence interval; HR, hazard ratio. Hazard ratio estimates (with 95% confidence intervals) for death within 2 years of transcatheter aortic valve implantation. The hazard ratios show the increase in hazard for a unit increase in age (years), logistic EuroSCORE and/or GA frailty score, and for males compared with females. The corresponding results for the dichotomous GA frailty score were HR = 5.35 (95% CI: 1.99–15.3, P = 0.001) (unadjusted) and HR = 4.91 (95% CI: 1.79–14.2, P = 0.002) (adjusted). The ROC curve (Figure 2) illustrates that a frailty score cut-off at ≥4 predicts 2 year mortality with a specificity of 80% (95% CI: 73–86%) and a sensitivity of 60% (95% CI: 36–80%). The AUC was 0.81 (95% CI: 0.71–0.90). Figure 2 Receiver operator characteristics curve for geriatric assessment frailty score (0–9) and 2 year mortality (n = 142). The area under the curve is 0.81 (95% confidence interval: 0.71–0.90). Figure 3 Kaplan–Meier plot of 2 years survival after transcatheter aortic valve implantation (n = 142). The coloured bands indicate 95% confidence intervals.

Figure 2 Receiver operator characteristics curve for geriatric assessment frailty score (0–9) and 2 year mortality (n = 142). The area under the curve is 0.81 (95% confidence interval: 0.71–0.90). Figure 3 Kaplan–Meier plot of 2 years survival after transcatheter aortic valve implantation (n = 142). The coloured bands indicate 95% confidence intervals. None of the patients with a frailty score of 0 or 1 were dead after 2 years, and none of the patients had a frailty score of 8 or 9. In general, the higher the frailty score, the higher the risk of dying within 2 years (Table 5). Table 5 Distribution of geriatric assessment frailty score and mortality within each frailty score (n = 142) Frailty score Count Prop. (%) Cum. prop. (%) Deathsa Mortalitya (%) 0 15 11 11 0 0 1 26 18 29 0 0 2 39 27 56 2 5 3 28 20 76 4 14 4 20 14 90 4 20 5 10 7 97 3 30 6 2 1 99 2 100 7 2 1 100 0 0 8 0 0 100 — — 9 0 0 100 — — Cum., cumulative; Prop., proportion. a Deaths within 2 years after TAVI. When adjusting for mEFT along with age, gender, and logistic EuroSCORE, the continuous GA frailty score were no longer a statistically significant predictor (HR = 1.36, 95% CI: 0.87–2.21, P = 0.18, n = 139), and neither were any of the other variables (including mEFT). Discussion In this prospective observational study, we found that a novel GA frailty score could predict 2-year all-cause mortality in TAVI patients declined for open heart surgery by a Heart Team. After 2 years, there were no deaths in the cohort with very low (0 or 1) frailty score.

When adjusting for mEFT along with age, gender, and logistic EuroSCORE, the continuous GA frailty score were no longer a statistically significant predictor (HR = 1.36, 95% CI: 0.87–2.21, P = 0.18, n = 139), and neither were any of the other variables (including mEFT). Discussion In this prospective observational study, we found that a novel GA frailty score could predict 2-year all-cause mortality in TAVI patients declined for open heart surgery by a Heart Team. After 2 years, there were no deaths in the cohort with very low (0 or 1) frailty score. Standard risk scores like EuroSCORE and STS score are insufficient for predicting adverse events in the older adult,11,13,42 and a frailty assessment adds information which increases predictability. There is an ongoing discussion regarding the definition of frailty and whether to include cognition, psychological factors, and comorbidity.43 The GA frailty score was developed to provide information for a better decision making prior to TAVI. This study adds to previous work highlighting the need for a more thorough evaluation of the individual patient based on a comprehensive GA. This can provide significant decision-making support for the interventional cardiologist or surgeon. The purpose of the score is not to screen all TAVI candidates, as this will be too time-consuming. Rather it should be used in patients in whom a simpler screening44 has revealed potential obstacles for TAVI. In this setting, an assessment solely on physical frailty would not be sufficient, in part due to decline in physical performance related to severe aortic stenosis. An approach with an initial basic screening for frailty and a selective thorough assessment by a geriatrician has been advocated.44,45

obstacles for TAVI. In this setting, an assessment solely on physical frailty would not be sufficient, in part due to decline in physical performance related to severe aortic stenosis. An approach with an initial basic screening for frailty and a selective thorough assessment by a geriatrician has been advocated.44,45 Patients categorized as frail might still be eligible for TAVI. All patients should be involved in a shared decision process regarding their treatment, but for patients where there is doubt whether the procedure is beneficial, it is especially important. Previous studies have underlined the importance of exploring patients’ perspectives.46,47 Asking the question ‘What do you hope to accomplish by having your valve repaired?’ might capture what is most important to patients.48 The decision to offer TAVI should in the end be made by the interventional cardiologist or cardiac surgeon performing TAVI, based on an analysis of benefit vs. risk, taking into account symptoms, comorbidity, patient perspective, procedural risk, and frailty. We suggest the geriatrician to be an important collaborator in this analysis. If TAVI is offered despite frailty, the treatment team should be prepared for a higher risk of complications, including delirium.21 Ideally, detecting frailty should lead to additional pre-, per-, and post-operative support.11 The GA frailty score provides delineation of specific aspects of frailty that can be addressed (e.g. nutritional supply if undernourished, treatment for depression).49,50 We do not have enough evidence to recommend specific exercise before TAVI in order to improve frailty status.

e-, per-, and post-operative support.11 The GA frailty score provides delineation of specific aspects of frailty that can be addressed (e.g. nutritional supply if undernourished, treatment for depression).49,50 We do not have enough evidence to recommend specific exercise before TAVI in order to improve frailty status. This study confirms the clinical relevance of frailty assessment prior to TAVI.12,13,17 The GA frailty score evaluating cognition, independence in daily life, nutrition, physical frailty, comorbidity, and psychological health, give a thorough and comprehensive assessment of the patient. A high GA frailty score ≥4 indicates a reduced 2-year survival (Figure 3). However, we do not advocate a strict cut-off where TAVI is not offered. Knowledge of the (0–9 based) GA frailty score should lead to a careful final evaluation by the TAVI team, and should involve weighting frailty, technical challenges, exploring patient preferences, and symptom burden before offering TAVI. The geriatrician can contribute to the heart team as a frailty expert.

ot offered. Knowledge of the (0–9 based) GA frailty score should lead to a careful final evaluation by the TAVI team, and should involve weighting frailty, technical challenges, exploring patient preferences, and symptom burden before offering TAVI. The geriatrician can contribute to the heart team as a frailty expert. Strengths of the study This is a prospective study, with potentially fewer sources of bias and higher quality of data than a retrospective study would have. In Norway, deaths of all patients are automatically registered in the patients’ electronic journal. Our primary outcome is therefore complete. We also have high completeness in the rest of our data, and importantly, no patients are lost to follow-up for the primary endpoint. The variables included in the GA frailty score were determined before the statistical analysis, eliminating the risks associated with a purely data-driven analysis. Finally, our risk score was reliable in patients already excluded from surgery due to comorbidity.

patients are lost to follow-up for the primary endpoint. The variables included in the GA frailty score were determined before the statistical analysis, eliminating the risks associated with a purely data-driven analysis. Finally, our risk score was reliable in patients already excluded from surgery due to comorbidity. Limitations of the study Survival with benefit after 2 years is advocated as a relevant clinical endpoint, and it would have strengthened the study if we also assessed quality of life in the patient 2 years after the procedure.1 However, there are limitations to soft endpoints, and in order to simplify the interpretation of the frailty score, we chose to focus mainly on prediction of mortality. Some items were self-reported and not performance based, introducing some subjectivity to the index; however, previous studies have showed for all the selected self-report items to be markers of frailty.7,31 This is a single-centre study, and the results might not be transferable to any other centre, although they are probably comparable to other European centres of the same size. The study population changed during the study. Initially, inclusion consisted of patients ≥80 years, but was later expanded to include all patients ≥70 years. This was partly due to a shift in the general TAVI population, but also a growing awareness that frailty is a complex phenomenon where age is only one contributing factor.7 The partial lack of data used in calculating the EFT score reduces the precision of the score somewhat. And finally, the small sample size (especially the few number of deaths) is a limitation, particularly for calculating the sensitivity of the dichotomized frailty score in predicting 2-year mortality. Before recommending the GA frailty scale, it needs to be validated in an independent population.

of the score somewhat. And finally, the small sample size (especially the few number of deaths) is a limitation, particularly for calculating the sensitivity of the dichotomized frailty score in predicting 2-year mortality. Before recommending the GA frailty scale, it needs to be validated in an independent population. Conclusions In patients declined for open heart surgery, an 8-element frailty score based upon GA can identify patients less likely to benefit from TAVI. Patients with a frailty score ≥4 had significantly higher 2-year mortality. We believe the novel GA frailty score has clinical relevance and may be a useful tool for heart teams in decision making for TAVI. Acknowledgements We would like to thank the participants in this study. Funding Grants were received from Grieg Foundation, Department of Heart Disease, Haukeland University Hospital, Kavli Research Centre for Geriatrics and Dementia, Haraldsplass Deaconess Hospital, and from the Western Norway Regional Health Authority. Conflict of interest: none declared.

Introduction Remote monitoring (RM) of implantable cardioverter-defibrillators (ICDs) and cardiac resynchronization therapy defibrillators (CRT-Ds) has become widely available and is recommended by the HRS/EHRA expert consensus panels from 2008 on.1,2 State-of-the-art RM technology allows automatic transmission of data related to device function, arrhythmias, and other physiological parameters without active participation of the patient.2–9 Remote monitoring enables safe reduction of the in-person follow-up (FU) burden for patient and clinician convenience and a substantial shortening of the delay between a clinically actionable event detected by the device and the clinical reaction.2,4,6,10

gical parameters without active participation of the patient.2–9 Remote monitoring enables safe reduction of the in-person follow-up (FU) burden for patient and clinician convenience and a substantial shortening of the delay between a clinically actionable event detected by the device and the clinical reaction.2,4,6,10 The potential influence of RM on patient outcomes has been investigated in randomized controlled trials (RCTs) employing a variety of RM technologies in various healthcare models.11–13 To date, only the IN-TIME trial8 demonstrated a significant impact of implant-based RM on hard clinical endpoints, a composite clinical score for heart failure patients14 (primary endpoint of the study) and mortality. In eight other RCTs included in a dedicated meta-analysis (2015)15 and in three recent trials (2016–17),16–18 RM failed to improve patient outcomes. However, a pooled patient-level analysis of the TRUST,4 ECOST,19 and IN-TIME8 results (2010–14) with the same RM system, confirmed the benefit reported in the IN-TIME and suggested that it was driven by the prevention of heart failure exacerbation.20 The question arises why the results with this particular RM system differ from those of other systems.

nalysis of the TRUST,4 ECOST,19 and IN-TIME8 results (2010–14) with the same RM system, confirmed the benefit reported in the IN-TIME and suggested that it was driven by the prevention of heart failure exacerbation.20 The question arises why the results with this particular RM system differ from those of other systems. It is generally plausible that a clinical effect of RM can be conferred only if relevant information on the patient’s medical status is received in time and if it leads to important therapeutic changes.21 The present article reports details of the information flow (content and information speed) and workflow in the IN-TIME trial. Its aim is to discuss whether a difference in these aspects across studies may be responsible for different clinical outcomes, with IN-TIME being an ‘outlier’ not because of chance or a faulty analysis or interpretation, but due to differences in study set-up and RM technology.

speed) and workflow in the IN-TIME trial. Its aim is to discuss whether a difference in these aspects across studies may be responsible for different clinical outcomes, with IN-TIME being an ‘outlier’ not because of chance or a faulty analysis or interpretation, but due to differences in study set-up and RM technology. Methods IN-TIME study design and results The IN-TIME study design and main results have been published elsewhere.8,11 In brief, the study enrolled 716 patients with chronic heart failure, New York Heart Association (NYHA) Class II/III symptoms, ejection fraction ≤35%, no permanent atrial fibrillation (AF), and an indication for dual-chamber ICD or CRT-D treatment. After a run-in phase of 1 month, patients were randomly assigned to either automatic, daily RM in addition to optimal care (n = 333) or optimal care without RM (n = 331). After 12 months of FU, the prevalence of a worsened composite clinical score, combining all-cause death, overnight hospitalization for heart failure, increase in NYHA class, and worsening in patient’s global self-assessment,14 was 18.9% in the RM group vs. 27.2% in the control group (P = 0.013).8 The Kaplan–Meier estimate of all-cause mortality was 3.4% vs. 8.7% (P = 0.004).

site clinical score, combining all-cause death, overnight hospitalization for heart failure, increase in NYHA class, and worsening in patient’s global self-assessment,14 was 18.9% in the RM group vs. 27.2% in the control group (P = 0.013).8 The Kaplan–Meier estimate of all-cause mortality was 3.4% vs. 8.7% (P = 0.004). The proprietary Biotronik Home Monitoring® (HM) technology (Biotronik SE & Co. KG, Berlin, Germany) was used. Everyday, the implanted devices sent data wirelessly to the Home Monitoring Service Center (HMSC) via a patient device situated at the patient’s bedside. The only patient activity required was to connect once the patient device to a power line. The kind of transmitted data is summarized in Table 1. The HMSC processes incoming data and posts them as trend graphs and tables on a secure website for treating physicians. Furthermore, it sends e-mails (‘HM-alerts’) when criteria, which can be adapted online for individual patient, are met. In IN-TIME, HM data were transmitted irrespective of the randomization group but were concealed for control patients until study completion. Four patients (0.6%) with poor data transmission during the run-in phase were excluded from randomization. Table 1 Data received in the Home Monitoring Service Center on a daily basis

N-TIME, HM data were transmitted irrespective of the randomization group but were concealed for control patients until study completion. Four patients (0.6%) with poor data transmission during the run-in phase were excluded from randomization. Table 1 Data received in the Home Monitoring Service Center on a daily basis Trend data Bradycardia and CRT Pacing statistics for all pacing channels Atrioventricular conduction statistics Percentage of CRT pacing HF monitor and physiological parameters Mean atrial heart rate Mean ventricular heart rate Mean ventricular heart rate at rest Mean number of VES per hour Atrial heart rate variability Patient activity in hours per day Thoracic impedancea Atrial tachyarrhythmia Atrial burden (percent of the day in AF) Mean and maximum ventricular rate during AF SVT episode (counter) Mode switch (counter) Ventricular tachyarrhythmia VT1, VT2, and VF zone detected (counter) ATP and shock therapies, started/successful (counters) Ineffective shock with maximum energy (counter) Lead related RA, RV, and LV shock impedance RA amplitude (mean) RV, LV amplitude (minimum) RV, LVa pacing threshold Technical Actual device programming setting Battery and technical status IEGM episode (typically one per day) VT1, VT2, VF, or SVT episode Atrial monitoring episode Periodic IEGM All data are transmitted on a daily basis and displayed as trends or listings.

ude (mean) RV, LV amplitude (minimum) RV, LVa pacing threshold Technical Actual device programming setting Battery and technical status IEGM episode (typically one per day) VT1, VT2, VF, or SVT episode Atrial monitoring episode Periodic IEGM All data are transmitted on a daily basis and displayed as trends or listings. AF, atrial fibrillation; ATP, antitachycardia pacing; CRT, cardiac resynchronization therapy; HF, heart failure; IEGM, intracardiac electrogram; LV, left ventricular; RA, right atrial; RV, right ventricular; SVT, supraventricular tachyarrhythmia; VES, ventricular extrasystoles; VF, ventricular fibrillation; VT1/VT2, slow/fast ventricular tachycardia. a Available in one-third of implanted study devices.

AF, atrial fibrillation; ATP, antitachycardia pacing; CRT, cardiac resynchronization therapy; HF, heart failure; IEGM, intracardiac electrogram; LV, left ventricular; RA, right atrial; RV, right ventricular; SVT, supraventricular tachyarrhythmia; VES, ventricular extrasystoles; VF, ventricular fibrillation; VT1/VT2, slow/fast ventricular tachycardia. a Available in one-third of implanted study devices. Home Monitoring information flow in IN-TIME is summarized in Figure 1. In contrast to normal HM routines, HM-alerts from the HMSC were sent to a central monitoring unit (CMU) located at an investigational site in Leipzig.8 The CMU had standard operating procedures which defined study specific alerts, mostly based on HM-alerts (Table 2). Study nurses, supported by local physicians, screened all patients’ data during normal office working hours (Mondays–Fridays), alerted sites and required confirmation of alert receipt. Investigators could ask for modification of alerts for specific patients. The CMU was established to tailor the alerts specifically for the study, to enhance sites’ awareness of alerts, and to record alerts and receipt confirmations systematically. The CMU staff did not assist in diagnostic or therapeutic decisions, except that they checked intracardiac electrogram snapshots for suspicious content. Table 2 Alerts sent to investigational sites by the central monitoring unit in IN-TIME: definitions and occurrencesa

d alerts and receipt confirmations systematically. The CMU staff did not assist in diagnostic or therapeutic decisions, except that they checked intracardiac electrogram snapshots for suspicious content. Table 2 Alerts sent to investigational sites by the central monitoring unit in IN-TIME: definitions and occurrencesa Alert type Device/HMSC: alert setting CMU activity in a patient with HMSC alert CMU: condition for SOP compatible event Number of alerts (number of patients) Atrial tachyarrhythmia 109 (65) First onset of AF for >30 s in patients with no prior documented AF Device: IEGM for atrial monitoring episode Check IEGM for atrial monitoring episode in patients with no prior documented AF IEGM shows true AF 31 (31) HMSC: sends alert Long atrial arrhythmia episode (>6 h) with high-ventricular rate (>120 b.p.m.) HMSC: alert for atrial burden >6 h with ventricular rate >120 b.p.m. Check IEGM Ventricular rate >120 b.p.m. None Daily AF burden >50% on 7 consecutive days after a period of 4 weeks with burden <25% for patients with known AF HMSC: alert for atrial burden >6 h Check trend graph Atrial burden ≥50% on 7 consecutive days 20 (16) Atrial arrhythmia in the VT zone HMSC: VT zone detected Check IEGM Suspicious IEGM 1 (1) First onset of AF for >30 s in patients with prior documented AF Device: IEGM for atrial monitoring episode Check IEGM for atrial monitoring episode Not predefined 22 (22) HMSC: alert for atrial monitoring episode Atrial episode HMSC: alert for atrial monitoring episode Check IEGM for atrial monitoring episode Not predefined 22 (14) Atrial burden >50% for more or less than 1 week HMSC: alert for atrial burden ≥6 h Check trend graph Not predefined 13 (8) Ventricular tachyarrhythmia or ICD shock 56 (42) ≥3 VT2/VF episodes over 48 h HMSC: alerts for VT2 and VF detection Check IEGM for appropriateness and count true VT/VF episodes in 48 h ≥3 episodes within 48 h 7 (6) First ICD shock in spontaneous episode HMSC: alert for VT and VF detections Check if the episode was induced or spontaneous Spontaneous episodes with shock 22 (22) First occurrence of slow VT (<150 b.p.m.) HMSC: alert for VT1 (monitoring zone) Check IEGM if true VT True VT <150 b.p.m. 8 (8) VT/VF detection with suspicious IEGM (e.g.

k in spontaneous episode HMSC: alert for VT and VF detections Check if the episode was induced or spontaneous Spontaneous episodes with shock 22 (22) First occurrence of slow VT (<150 b.p.m.) HMSC: alert for VT1 (monitoring zone) Check IEGM if true VT True VT <150 b.p.m. 8 (8) VT/VF detection with suspicious IEGM (e.g. sensing issue or inappropriate detection) HMSC: alert for VT and VF detections Check IEGM Suspicious IEGM 2 (2) <3 VT/VF episodes over 48 h HMSC: alert for VT and VF detections Check IEGM Not predefined 18 (14) Ineffective ICD shock HMSC: ineffective shock Check IEGM Not predefined 2 (2) VT and low CRT pacing HMSC: alert for VT and low CRT Check IEGM Not predefined 1 (1) Low percentage of CRT pacing 91 (35) CRT pacing <80% over 48 h HMSC: alert for CRT pacing <80% Check CRT pacing on the day before CRT pacing <80% on two consecutive days 81 (30) CRT pacing <80% combined with IEGM HMSC: alert for CRT pacing <80% Check IEGM Suspicious IEGM 3 (3) CRT pacing <80% over 24 h HMSC: alert for CRT pacing <80% Check trend graph Not predefined 8 (7) CRT pacing <80% and ventricular or atrial arrhythmia HMSC: alert for CRT pacing <80% Check IEGM Not predefined 2 (2) Ventricular extrasystoles 54 (46) Visible upward trend in VES/h over 7 days Not programmable Check trend graph weekly Visible trend 39 (34) VES >110 per hour for more than 10 days Not programmable Check trend graph Not predefined 15 (15) Physiological trends 2 (2) Visible downward trend in patient activity over 7 days Not programmable Check rend graph weekly Visible trend 1 (1) Rising mean heart rate Not programmable Check trend graph Not predefined 1 (1) Lead related measurements and IEGMs 98 (58) Suspicious IEGMb HMSC: alert for any new IEGM Check IEGM Suspicious IEGM 25 (18) RA amplitude HMSC: programmable Check trend graph <0.5 mV 13 (9) RV amplitude HMSC: programmable Check trend graph <2.0 mV 22 (18) RV pacing threshold HMSC: programmable Check trend graph Safety margin <1 V 11 (10) RV impedance HMSC: programmable Check trend graph <250 or >1500 Ohm 1 (1) LV amplitude HMSC: programmable Check trend graph <2.0 mV None LV impedance HMSC: programmable Check trend graph <250 or >1500 Ohm 2 (2) LV pacing threshold HMSC: programmable Check trend graph Safety margin <1 V 12 (8) Shock impedance HMSC: programmable Check trend graph <30 or >125 Ohm 13 ( 5) Gaps in data transmission 818 (241) Missing HM messages for 3 days HMSC: alert for missing messages for 3 days Check if the patient is known to be on holidays or i

pacing threshold HMSC: programmable Check trend graph Safety margin <1 V 12 (8) Shock impedance HMSC: programmable Check trend graph <30 or >125 Ohm 13 ( 5) Gaps in data transmission 818 (241) Missing HM messages for 3 days HMSC: alert for missing messages for 3 days Check if the patient is known to be on holidays or i n hospital If the patient is not known to be absent 818 (241) Technical alerts None Elective replacement indicator HMSC: programmable None Technical alert HMSC: programmable None Sums of numbers of alerts do not match completely because few alerts are contained in more than one category, and three alerts contained no description of the content. CMU, central monitoring unit; ICD, implantable cardioverter-defibrillator; HM, Home Monitoring; SOP, standard operating procedure defined in the study protocol; VT, ventricular tachycardia; AF, atrial fibrillation; CRT, cardiac resynchronization therapy; IEGM, intracardiac electrogram; LV, left ventricular; RA, right atrial; RV, right ventricular; VES, ventricular extrasystoles; VF, ventricular fibrillation; VT1/VT2, slow/fast ventricular tachycardia. a During the randomized period in the RM group. b T-wave oversensing, far-field atrial sensing of ventricular activity, or other suspected sensing problem. Figure 1 Sketch of the remote monitoring information flow and interaction between parties in IN-TIME. CMU, clinical monitoring unit; HM, Home Monitoring.

CMU, central monitoring unit; ICD, implantable cardioverter-defibrillator; HM, Home Monitoring; SOP, standard operating procedure defined in the study protocol; VT, ventricular tachycardia; AF, atrial fibrillation; CRT, cardiac resynchronization therapy; IEGM, intracardiac electrogram; LV, left ventricular; RA, right atrial; RV, right ventricular; VES, ventricular extrasystoles; VF, ventricular fibrillation; VT1/VT2, slow/fast ventricular tachycardia. a During the randomized period in the RM group. b T-wave oversensing, far-field atrial sensing of ventricular activity, or other suspected sensing problem. Figure 1 Sketch of the remote monitoring information flow and interaction between parties in IN-TIME. CMU, clinical monitoring unit; HM, Home Monitoring. If the investigators considered the content of the alert clinically relevant, they were supposed to contact the patient and conduct a structured interview on the patient’s overall condition, weight change, and drug compliance. After gaps in RM transmission, this interview was also conducted to confirm the patient’s status after a period without monitoring data. Per patient-year of FU, the CMU sent 4.0 alerts to the sites, the sites contacted the patient 2.1 times, and 0.3 additional FUs or contacts to other physicians were arranged.8 Investigators were suggested to react according to current guidelines but the therapeutic consequences were not followed.

thout monitoring data. Per patient-year of FU, the CMU sent 4.0 alerts to the sites, the sites contacted the patient 2.1 times, and 0.3 additional FUs or contacts to other physicians were arranged.8 Investigators were suggested to react according to current guidelines but the therapeutic consequences were not followed. Data analysis The present analysis of IN-TIME data aims at evaluating the transmission performance of HM, describing the CMU performance, and estimating delays from alerts to FU visits. The following aspects are addressed as follows: Establishing HM transmission: the time from post-implant hospital discharge to first HM transmission described by the Kaplan–Meier method, in all patients with successful implantation including some not who were not randomized at 1 month. Transmission performance: the number of days with HM message divided by the total days between randomization and study termination. Also the length of transmission gaps was analysed. Decline of transmission during the study: a linear fit of the share of patients with a HM message as a function of time after randomization, weighted with the number of patients in the study. Delay from an event until the information is received in the HMSC: an estimation based on the distribution of the time to the next successful transmission for all days between randomization and study termination in patients randomized to RM.

Decline of transmission during the study: a linear fit of the share of patients with a HM message as a function of time after randomization, weighted with the number of patients in the study. Delay from an event until the information is received in the HMSC: an estimation based on the distribution of the time to the next successful transmission for all days between randomization and study termination in patients randomized to RM. Working time compliance of the CMU: we estimated whether the CMU worked on every working day. For this, we selected a period of 2 years when most patients were included in the study and calculated the mean number of CMU alerts per day and the number of 7 days of the week (Mondays, Tuesdays, …, Sundays) without any alert. The Poisson-distribution predicts an expected number of days without any alert from the mean number of alerts per day, assuming that alerts were independent of each other. If the CMU did not comply with their working time rules, a higher than predicted number of days without alerts would be expected.

without any alert. The Poisson-distribution predicts an expected number of days without any alert from the mean number of alerts per day, assuming that alerts were independent of each other. If the CMU did not comply with their working time rules, a higher than predicted number of days without alerts would be expected. Delay from alert to patient contact and follow-up: the delay from a HM alert to a patient contact was recorded in the study documentation. However, the delay from a HM alert to related FU visits was not available because the study data do not allow an assignment of FUs to alerts. We, therefore, estimated alert-to-FU delays without any a priori assumptions on relations between FUs and alerts by a mathematical model. It used as inputs 1222 alerts and 1289 FUs captured on case report forms or evident in the HMSC as dates of reprogramming in 280 patients who had alerts in the RM group. For each patient, we calculated the set of time intervals between each alert and each FU. We show the relative number of FUs per day in the 8 weeks after the alert, as mean and standard deviation, normalized to the FU rate between 14 and 100 days after the alert. Statistical methods Home Monitoring performances had non-normal distributions and were compared with the Mann–Whitney U-test. We used a linear regression model to calculate the decline of HM transmission after randomization. A P-value <0.05 was considered statistically significant. The analysis was conducted with the R 3.3 statistical software (R Development Core Team, Vienna, Austria).

ibutions and were compared with the Mann–Whitney U-test. We used a linear regression model to calculate the decline of HM transmission after randomization. A P-value <0.05 was considered statistically significant. The analysis was conducted with the R 3.3 statistical software (R Development Core Team, Vienna, Austria). Results Transmission performance of home monitoring Of 702 patients discharged from hospital with a study device implanted, 41.6% had HM transmission established already 1 day after discharge, 68.8% on Day 3, and 95.5% on Day 30. In the RM group, HM messages were received on 82.2% of all patient-days, or on 83.1% out-of-hospital days after randomization. Patient-individually, the median percentage of days with a message was higher in the RM group [87.8%, interquartile range (IQR): 78.6–93.4%] than in the control group (85.4%, IQR: 71.8–91.8%) (P = 0.003). In the RM group, 79.9% of the patients transmitted messages on 75% or more of all study days. There were 29.5 transmission interruptions per patient-year in the RM group with an average length of 2.1 days, but only 2.3 interruptions per year were longer than 3 days. In the control group, the total number of interruptions was similar (29.2 per patient-year), but interruptions longer than 3 days were more frequent (2.8 per patient-year). A decline of transmission performance during the study was observed in both randomization groups (Figure 2). The linear fit line fell by an absolute of 3.3%, from 84.7% (randomization) to 81.4% (365 days), in the RM group, and by 10.1%, from 83.4% to 73.3%, in the control group.

There were 29.5 transmission interruptions per patient-year in the RM group with an average length of 2.1 days, but only 2.3 interruptions per year were longer than 3 days. In the control group, the total number of interruptions was similar (29.2 per patient-year), but interruptions longer than 3 days were more frequent (2.8 per patient-year). A decline of transmission performance during the study was observed in both randomization groups (Figure 2). The linear fit line fell by an absolute of 3.3%, from 84.7% (randomization) to 81.4% (365 days), in the RM group, and by 10.1%, from 83.4% to 73.3%, in the control group. Figure 2 Percentage of patients with home monitoring data transmission on any given day between randomization and study termination. Black line denotes the RM group and red line denotes the control group. The straight lines are the linear fits. The linear regression model shows that the decline was statistically significant in both groups (P < 0.001). The estimated delay between a medical event and the information being received in the HMSC was 1 day in 83.1% of all cases. On Days 2 and 3, 91.2% and 94.3% of all events were available, respectively. Definition and occurrence of alerts sent by the central monitoring unit

Figure 2 Percentage of patients with home monitoring data transmission on any given day between randomization and study termination. Black line denotes the RM group and red line denotes the control group. The straight lines are the linear fits. The linear regression model shows that the decline was statistically significant in both groups (P < 0.001). The estimated delay between a medical event and the information being received in the HMSC was 1 day in 83.1% of all cases. On Days 2 and 3, 91.2% and 94.3% of all events were available, respectively. Definition and occurrence of alerts sent by the central monitoring unit Table 2 summarizes CMU alerts. For example, the first AF episode was reported in 31 patients without history of AF. Twenty alerts were sent for daily AF burden ≥50% on 7 consecutive days. A first ICD shock triggered by spontaneous ventricular tachyarrhythmia was reported in 22 patients, whereas arrhythmia ‘storm’ (≥3 ventricular episodes over 48 h) led to seven alerts. Low percentage of CRT pacing over two consecutive days was reported 83 times. Beyond alerts predefined in the study protocol, the CMU sent self-initiatively notifications for a first AF episode in patients with known AF history and for some other conditions (Table 2).

entricular episodes over 48 h) led to seven alerts. Low percentage of CRT pacing over two consecutive days was reported 83 times. Beyond alerts predefined in the study protocol, the CMU sent self-initiatively notifications for a first AF episode in patients with known AF history and for some other conditions (Table 2). Working time compliance of the central monitoring unit In a period of 104 weeks between 1 July 2008 and 28 June 2010, on average 113 patients (between 73 and 140) were followed by the CMU. For these patients, the CMU sent 938 alerts, or 1.29 alerts per day. No alerts were sent on 12.5% of all Mondays (13/104), on 30.0% of all Tuesdays–Fridays (Tuesdays 29/104, Wednesdays 30/104, Thursdays 31/104, and Fridays 35/104), and on 85.1% of all Saturdays and Sundays (Saturdays 86/104 and Sundays 91/104) in this period. The data indicate that the CMU did not work during most weekends, as defined per study protocol, and that a backlog of alerts remained for Mondays. The Poisson-distribution predicts no alerts on 27.5% of all days, which is close to the result for Tuesdays–Fridays, indicating good working time compliance.

n this period. The data indicate that the CMU did not work during most weekends, as defined per study protocol, and that a backlog of alerts remained for Mondays. The Poisson-distribution predicts no alerts on 27.5% of all days, which is close to the result for Tuesdays–Fridays, indicating good working time compliance. Delay from alert to patient contact and follow-up (estimated) Patients were contacted after a median delay of 1 day (IQR 0–6 days). Figure 3 shows that the number of FUs is clearly increased by a factor of two in the week after the alert, but only very slightly in the second week, suggesting that FUs, which were conducted as a consequence of HM information took place in most cases within a week of the alert. A sketch of some performance figures is given in Figure 4. Figure 3 Increase of follow-ups after alerts. Based on 1222 alerts and 1289 follow-ups in 280 patients who had alerts in the RM group. We calculated the set of time intervals between each alert and all follow-ups following in the same patient. We show the relative number of follow-ups per day in the 8 weeks after the alert, as mean and standard deviation, normalized to the follow-up rate between 14 and 100 days after the alert. Note that the number of follow-ups is increased in the week after the alert, but not later. This suggests that most follow-ups resulting as a consequence of home monitoring information took place within a week of the alert.

nd standard deviation, normalized to the follow-up rate between 14 and 100 days after the alert. Note that the number of follow-ups is increased in the week after the alert, but not later. This suggests that most follow-ups resulting as a consequence of home monitoring information took place within a week of the alert. Figure 4 Sketch of the IN-TIME information- and workflow with some performance characteristics. HM, Home Monitoring; Mo, Monday; Fr, Friday; Su, Sunday; FU, follow-up; asap, as soon as possible; n.a., not available; IQR, interquartile range. Discussion To understand whether differences in clinical outcomes of implant-based RM in different studies can be attributed to differences in study set-up, we analysed the information flow and workflow in the IN-TIME trial. Three main possible sources of heterogeneity between studies will be discussed: content of RM messages, information speed and completeness, and workflow in response to RM messages.

in different studies can be attributed to differences in study set-up, we analysed the information flow and workflow in the IN-TIME trial. Three main possible sources of heterogeneity between studies will be discussed: content of RM messages, information speed and completeness, and workflow in response to RM messages. Content of remote monitoring messages In the IN-TIME study, a multitude of different triggers was used to generate alerts. When treating physicians examined the patients’ data, they had access to up-to-date data trends for most variables stored in the device statistics, including arrhythmia burden, heart rates, patient activity, intracardiac electrograms, and pacemaker timing statistics. These data may help with the decision if an alert, mostly derived from a single variable, is appropriate. The possibility to modify the alert criteria online without reprogramming the implanted device may reduce inappropriate alerts and increase the physician’s willingness to act.

rams, and pacemaker timing statistics. These data may help with the decision if an alert, mostly derived from a single variable, is appropriate. The possibility to modify the alert criteria online without reprogramming the implanted device may reduce inappropriate alerts and increase the physician’s willingness to act. In other studies, the number of possible alerts was seemingly limited by study design (thoracic impedance only12) or by the RM system (e.g. five alerts available: thoracic impedance, atrial burden, ventricular rate during AF, high number of shocks per episode, and all therapies exhausted).22,23 It is unclear from the publications of these studies which up-to-date data were available when an alert was received to review the patient’s condition and decide about the clinical need for a patient contact. At least in one study, a patient contact and a manual data download was required after automatic alerts to allow for RM data review.12 In REM-HF, no alerts were programmed but a complete data set was checked after every transmission (scheduled once per week).13

and decide about the clinical need for a patient contact. At least in one study, a patient contact and a manual data download was required after automatic alerts to allow for RM data review.12 In REM-HF, no alerts were programmed but a complete data set was checked after every transmission (scheduled once per week).13 Speed and completeness The results on HM transmission in IN-TIME show that HM is suited for long-term monitoring in consecutive unselected ICD/CRT-D recipients. Within 1 month of implantation, 95% of all patients had successfully sent a first message. Less than 1% of patients were excluded from the study due to inability to establish HM transmission. Only 3.3% of surviving patients in the RM group who sent messages in the beginning failed to do so after 1 year. The fact that messages were still received on more than 70% of all patient-days in control group patients after being left alone for 1 year suggests that the transmission is robust. However, the early reaction to interruptions of the transmission, which is only possible with frequent scheduled transmission, exerts a significant effect on transmission performance. In REM-HF, 58%, 66%, and 62% of the patients successfully transmitted 75% or more of the scheduled weekly messages, at 6, 12, and 24 months, respectively.13 The corresponding figure from IN-TIME is 80%; this is higher, but the main difference is that messages were scheduled daily, not weekly. Since no alerts were transmitted in REM-HF, changes in the patient’s status were detected only in the next scheduled transmission.

y messages, at 6, 12, and 24 months, respectively.13 The corresponding figure from IN-TIME is 80%; this is higher, but the main difference is that messages were scheduled daily, not weekly. Since no alerts were transmitted in REM-HF, changes in the patient’s status were detected only in the next scheduled transmission. In the CONNECT study, 180 of 575 clinical events resulted in a transmitted alert (31%).6 The devices used in this study could transmit each type of alert only once between FUs. Other studies reported a better rate of alerts transmission: 76% (OptiLink)16 and 88% (MORE-CARE).17 In the RM system used in IN-TIME, alerts are not generated in the device but in the service centre. Because most alerts are based on a comparison between the recent and the previous transmission, alerts will not be lost even if the transmission on that day fails. About 83% of all events detected in the RM group were available in the HMSC within 1 day, and approximately 94% were available within 3 days. The median delay until the contact to the patient—if a contact took place—was 1 day. The delay between the events and their availability in the RM service centre has not been reported from other trials. The median delay between alerts and their reviewing 1.4 days in EVOLVO7 and 3 days in MORE-CARE.24

ble within 3 days. The median delay until the contact to the patient—if a contact took place—was 1 day. The delay between the events and their availability in the RM service centre has not been reported from other trials. The median delay between alerts and their reviewing 1.4 days in EVOLVO7 and 3 days in MORE-CARE.24 Workflow The IN-TIME study protocol required checking of the transmitted HM data in the CMU on all working days. Although the chain from medical event to clinical action cannot be reconstructed exactly, the data indicate that the CMU fulfilled its duties. It worked on all days between Mondays and Fridays, with stand-in for holidays and sick leave. The vast majority of CMU alerts were compliant with the predefined rules, but some went beyond, such as the first AF episode in patients with known history of AF. Whether this was by oversight or on purpose, the information can be medically relevant. The feedback to clinical sites on transmission gaps exerted a significant effect on transmission performance, as evidenced by better transmission rate in the RM group vs. the control group by absolute 8.1% late in the study. The analysis of the temporal relationship between alerts and FUs indicated that most FUs took place within a week of the alert. The information speed and completeness (of alerts) is an obvious requirement for this result, but it is per se not sufficient without a daily check of the RM data. We are not aware of publications of workflow details from other studies, such as estimates of a delay from alert to FU.

ok place within a week of the alert. The information speed and completeness (of alerts) is an obvious requirement for this result, but it is per se not sufficient without a daily check of the RM data. We are not aware of publications of workflow details from other studies, such as estimates of a delay from alert to FU. Influence of remote monitoring on outcome With the presented data, we cannot prove that the difference between the results of IN-TIME and other studies has its cause in different study set-ups or the RM system characteristics. However, RM can influence outcome only through early appraisal of relevant medical information by the physician. We have shown that in IN-TIME, a considerable list of alerts was used; that the vast majority of alerts was available within 1 day, together with a current set of diagnostic data comparable to that in the device’s memory; and that the investigators managed to see most patients within a week of an event, if they decided that this was indicated. Comparable results are not (yet?) reported for other studies, so we are not able to describe differences with the required precision. However, several performance characteristics reported here are inherently connected to the technology of the RM system used in IN-TIME, especially the ability to transmit the complete data set daily.

sults are not (yet?) reported for other studies, so we are not able to describe differences with the required precision. However, several performance characteristics reported here are inherently connected to the technology of the RM system used in IN-TIME, especially the ability to transmit the complete data set daily. A recent analysis of several clinical endpoints from the IN-TIME and ECOST studies has suggested that daily HM improves only heart failure events.20 The IN-TIME study was the first study to demonstrate clinical benefit of implant-based RM, but it has been overseen that it was also the first study to use a heart-failure specific primary endpoint.8,14 If one observes new onset AF, asymptomatic ventricular tachycardia, increasing frequency of ventricular extrasystoles, or decreasing percentage of CRT pacing in a heart failure patient, we believe that it is very plausible that a deterioration of the patient’s clinical status can be prevented if the patient is seen within a few days. The lack of an appropriate alert trigger, a failure to transmit the alert and accompanying data or a significant lag before transmission, the inability to judge the patient’s status from RM data, or the failure to contact the patient without delay may be the reason for the failure to improve clinical outcome. Too many of these details are unknown for too many trials to dismiss IN-TIME as an outlier.

mpanying data or a significant lag before transmission, the inability to judge the patient’s status from RM data, or the failure to contact the patient without delay may be the reason for the failure to improve clinical outcome. Too many of these details are unknown for too many trials to dismiss IN-TIME as an outlier. The observation that the clinical benefit of RM is restricted to heart failure events suggests that it may be reasonable to establish a monitoring unit especially for heart failure patients, possibly in co-operation between device and heart failure clinics. IN-TIME had no more than 140 patients under HM surveillance at any time, thus, even medium size implanting centres may establish an efficient monitoring team. Conclusion The difference between the IN-TIME result and other outcome studies may be caused by differences in content of transmitted data, speed and completeness of transmission, and workflow to contact the patient when needed. Both for clinical routine and for future studies, we suggest to establish processes to assure a high transmission compliance in the long term, to use a wide array of medical data to trigger alerts and to judge if a patient contact is needed, and to establish processes allowing to see the patient within less than 1 week after a medically relevant event. Publications of work- and information flow details from other studies—whether successful or not—would be valuable to inform such planning. Acknowledgements The authors would like to thank Dejan Danilovic for critical reading and editorial assistance.

Conclusion The difference between the IN-TIME result and other outcome studies may be caused by differences in content of transmitted data, speed and completeness of transmission, and workflow to contact the patient when needed. Both for clinical routine and for future studies, we suggest to establish processes to assure a high transmission compliance in the long term, to use a wide array of medical data to trigger alerts and to judge if a patient contact is needed, and to establish processes allowing to see the patient within less than 1 week after a medically relevant event. Publications of work- and information flow details from other studies—whether successful or not—would be valuable to inform such planning. Acknowledgements The authors would like to thank Dejan Danilovic for critical reading and editorial assistance. Funding The study was funded by Biotronik SE & Co. KG, Berlin, Germany. The funder contributed to the study design but had not role in data collection. D.H. was head of the Clinical Monitoring Unit. J.S. was the responsible biostatistician of the present results. Conflict of interest: C.G. reports minor travel support and lecture fees from Biotronik. C.S. reports grants from BIOTRONIK. J.C.N. reports grants from the Novo Nordisk Foundation. S.P.H. reports personal fees from Pfizer/BMS. J.S. is an employee of Biotronik. T.L. reports personal fees from Biotronik. G.H. received research grants through the Heart Centre Leipzig from Abbott and Boston Scientific. And all other authors have nothing to disclose.

Introduction Cardiovascular disease (CVD) is the number one cause of death worldwide, accounting for nearly 18 million of global deaths in 2013, of which approximately 7.4 million were related to coronary heart disease (CHD) and 6.7 million were due to stroke.1,2 Cardiovascular disease burden is rapidly growing due to the increase of major risk factors such as obesity, hypertension, and Type 2 diabetes.3,4 The global economic burden of CVD, including cost of screening, primary prevention, secondary prevention, acute hospital care, and lost productivity, is anticipated to reach 1044 billion US dollar in 2030.5 Patients at high-risk of experiencing cardiovascular (CV) events are defined as those with CHD, or other forms of arteriosclerotic disease, diabetes, and multiple risk factors that confer a 10-year risk for CHD >20% (estimated by Framingham risk scores).6 Elevated levels of low-density lipoprotein cholesterol (LDL-C) has been shown to be a key risk factor of CVD and the treatment of hypercholesterolaemia represents an important strategy to reduce new CV events as well as mortality.7 The evidence to support the effectiveness of statin therapy in secondary prevention is well proven, where high-risk patients are recommended more intensive statin regimens.8–11 Recent meta-analyses of data from clinical trials evaluating therapies specifically designed to lower LDL-C, including statins, ezetimibe, and PCSK9 inhibitors, have shown that reducing LDL-C reduces the risk of CV events proportional to the absolute achieved reduction in LDL-C.12,13

more intensive statin regimens.8–11 Recent meta-analyses of data from clinical trials evaluating therapies specifically designed to lower LDL-C, including statins, ezetimibe, and PCSK9 inhibitors, have shown that reducing LDL-C reduces the risk of CV events proportional to the absolute achieved reduction in LDL-C.12,13 Established CVD is associated with a higher risk for recurrent CV events following the first event.14–16 The majority of data informing CV-specific event rates in patients with existing CVD is derived from clinical trial populations. While such data is informative, clinical trials may underestimate event rates for a variety of reasons including disproportionate recruitment from high performing academic centres, higher quality of care received by clinical trial participants, or potential patient selection bias.17,18 Given the large number of patients who are living with CVD, there is a need to evaluate the CV risk of this patient-group outside of large clinical trials. Further, the availability and future emergence of highly potent therapies that may have clinical value in this population make development of a predictive model in this Subgroup of patients of clinical value. The study ‘Further Cardiovascular Outcomes Research With PCSK9 Inhibition in Subjects With Elevated Risk’ (FOURIER) was a double-blinded, randomized, placebo-controlled, multicentre study assessing the impact of additional LDL-C reduction on major CV events when evolocumab is used in combination with statin therapy in patients with clinically evident CVD.19 The current study aimed to address current gaps in the existing literature by estimating the rate of CV events in the real world among patients meeting FOURIER-like population criteria.

L-C reduction on major CV events when evolocumab is used in combination with statin therapy in patients with clinically evident CVD.19 The current study aimed to address current gaps in the existing literature by estimating the rate of CV events in the real world among patients meeting FOURIER-like population criteria. Methods Study design A retrospective population‐based cohort study was conducted using Swedish national-based population registers. The study included adult patients, 40–85 years of age, at high risk of CV events receiving statin therapy who were followed for the occurrence of subsequent CV events. The study period ranged from 1 January 2001 to 31 December 2015. Patients were followed until either CV event, death, or end of the study period.

sters. The study included adult patients, 40–85 years of age, at high risk of CV events receiving statin therapy who were followed for the occurrence of subsequent CV events. The study period ranged from 1 January 2001 to 31 December 2015. Patients were followed until either CV event, death, or end of the study period. Data sources Patient-level data from the (i) National Patient Register, (ii) Cause of Death Register, and (iii) Prescription Drug Register, were linked together by the Swedish National Board of Health and Welfare using unique personal identifiers.20–23 The three national registers are mandatory to report to and are associated with a high degree of completeness.20,21 Therefore, the registers enable complete nation-wide coverage of the Swedish population. The Prescription Drug Register includes data on all prescriptions filled at pharmacies, including drug type, dispensing date, dose and pack size. Information on diagnoses, hospitalizations, surgical and non-surgical procedures, and outpatient specialist visits were collected from the National Patient Register for the complete observable period for each patient. The Cause of Death Register provided confirmed dates of death together with cause of death, allowing for removing patients from the analyses when they were no longer under observation and for identifying deaths as either CV-related or non-CV-related. Ethical approval was obtained from the regional ethical review board in Stockholm on 3 March 2016; reference number 2016/456-31/2. Individual patient informed consent is not required for register studies on retrospective data in Sweden and was therefore not collected.

ying deaths as either CV-related or non-CV-related. Ethical approval was obtained from the regional ethical review board in Stockholm on 3 March 2016; reference number 2016/456-31/2. Individual patient informed consent is not required for register studies on retrospective data in Sweden and was therefore not collected. Patient population Three study cohorts were identified and followed separately for outcomes; (i) the atherosclerotic cardiovascular disease (ASCVD) prevalent cohort which included patients with myocardial infarction (MI), ischaemic stroke (IS), or peripheral artery disease (PAD) who met the FOURIER-like criteria as of 1 July 2006 (index date); (ii) the MI incident cohort which included patients with an incident MI who met the FOURIER-like criteria at the time of the MI event (variable index dates from 1 July 2006 to 31 December 2014); (iii) the IS incident cohort which included patients with an incident IS who met the FOURIER-like criteria at the time of the IS event (variable index dates from 1 July 2006 to 31 December 2014).

I who met the FOURIER-like criteria at the time of the MI event (variable index dates from 1 July 2006 to 31 December 2014); (iii) the IS incident cohort which included patients with an incident IS who met the FOURIER-like criteria at the time of the IS event (variable index dates from 1 July 2006 to 31 December 2014). The three study cohorts were identified in the Swedish dataset by applying the FOURIER-like inclusion criteria within a real-world clinical practice setting. All three cohorts needed to fulfil the FOURIER-like criteria at index date, including; age between 40 and 85 at index date; one or more (≥1) filled statin prescriptions of moderate and/or high-intensity during the 1-year period prior to the index date; at least one (≥1) major risk factor or two (≥2) minor risk factors, as defined in Table 1. Statin dose intensity was defined in accordance with the American College of Cardiology and the American Heart Association (ACC/AHA) guidelines for cholesterol treatment (Table 2).7 While not included in the ACC/AHA guidelines, simvastatin 80 mg was defined as high-intensity based on its expected LDL-C reduction of nearly 50%.24 Table 1 Risk factors included in the FOURIER-like inclusion criteria

The three study cohorts were identified in the Swedish dataset by applying the FOURIER-like inclusion criteria within a real-world clinical practice setting. All three cohorts needed to fulfil the FOURIER-like criteria at index date, including; age between 40 and 85 at index date; one or more (≥1) filled statin prescriptions of moderate and/or high-intensity during the 1-year period prior to the index date; at least one (≥1) major risk factor or two (≥2) minor risk factors, as defined in Table 1. Statin dose intensity was defined in accordance with the American College of Cardiology and the American Heart Association (ACC/AHA) guidelines for cholesterol treatment (Table 2).7 While not included in the ACC/AHA guidelines, simvastatin 80 mg was defined as high-intensity based on its expected LDL-C reduction of nearly 50%.24 Table 1 Risk factors included in the FOURIER-like inclusion criteria Major risk factors—one or more (≥1) required Diabetes (Type 1 or Type 2) Age ≥65 years and ≤85 years at time of index date If qualifying with MI; the MI event occurred during 6 months prior to the index date, or the patient has a history of IS and/or PAD in the baseline period If qualifying with IS; the IS event occurred during 6 months prior to index date, or the patient has a history of MI and/or PAD in the baseline period Minor risk factors—two or more (≥2) required Coronary revascularization with no history of MI Coronary artery disease Metabolic syndrome MI, myocardial infarction; IS, ischaemic stroke; PAD, peripheral artery disease. Table 2 Statin dose intensity

Major risk factors—one or more (≥1) required Diabetes (Type 1 or Type 2) Age ≥65 years and ≤85 years at time of index date If qualifying with MI; the MI event occurred during 6 months prior to the index date, or the patient has a history of IS and/or PAD in the baseline period If qualifying with IS; the IS event occurred during 6 months prior to index date, or the patient has a history of MI and/or PAD in the baseline period Minor risk factors—two or more (≥2) required Coronary revascularization with no history of MI Coronary artery disease Metabolic syndrome MI, myocardial infarction; IS, ischaemic stroke; PAD, peripheral artery disease. Table 2 Statin dose intensity Intensity Statin type and strength for daily doses Moderate-intensity statin Fluvastatin 40 mg twice daily Fluvastatin XL 80 mg Lovastatin 40 mg Pravastatin 40 mg and 80 mg Simvastatin 20–40 mg Atorvastatin 10 mg and 20 mg Rosuvastatin 5 mg and 10 mg Pitavastatin 2–4 mg High-intensity statin Atorvastatin 40–80 mg Rosuvastatin 20 mg and 40 mg Simvastatin 80 mg Patients were excluded if they had a new MI or IS event within 4 weeks after the index date, had a known history of haemorrhagic stroke, or were recipient of any major organ transplant. Data used in the FOURIER trial criteria, which were not available in Swedish registers included LDL-C, high-density lipoprotein cholesterol (HDL-C), triglycerides, smoking information, New York Heart Association (NYHA) class, and other lab data such as blood pressure, estimated glomerular filtration rate, and creatinine kinase levels.

FOURIER trial criteria, which were not available in Swedish registers included LDL-C, high-density lipoprotein cholesterol (HDL-C), triglycerides, smoking information, New York Heart Association (NYHA) class, and other lab data such as blood pressure, estimated glomerular filtration rate, and creatinine kinase levels. The study cohorts were not mutually exclusive. Patients could thus be included in both the prevalent and an incident cohort. In addition, patients could be included in both incident cohorts if they had a MI and an IS during follow-up, if the FOURIER-like criteria were fulfilled at the time of the index event. Each new CV event during follow-up, i.e. each outcome event, was counted in the appropriate cohort.

prevalent and an incident cohort. In addition, patients could be included in both incident cohorts if they had a MI and an IS during follow-up, if the FOURIER-like criteria were fulfilled at the time of the index event. Each new CV event during follow-up, i.e. each outcome event, was counted in the appropriate cohort. Study outcomes The study outcomes were composites of CV events occurring during follow-up. The hard major adverse cardiovascular events (MACE) composite was defined as MI, IS, or CV death; the ASCVD composite was defined as MI, IS, unstable angina (UA), coronary revascularization (coronary artery bypass grafting or percutaneous coronary intervention), or CV death. CV events were defined as hospitalizations with a primary ICD-10 diagnosis for the included outcome events: MI, UA, IS, or CV death, or as coronary revascularization using procedure codes. Demographics and clinical characteristics, including age at index, gender, follow-up time, diabetes, hypertension, medication use, Charlson comorbidity index, and history of CV, was assessed during the baseline period. CV history during baseline included MI, UA, IS, transient ischaemic attack (TIA), heart failure (HF), and coronary revascularization.

teristics, including age at index, gender, follow-up time, diabetes, hypertension, medication use, Charlson comorbidity index, and history of CV, was assessed during the baseline period. CV history during baseline included MI, UA, IS, transient ischaemic attack (TIA), heart failure (HF), and coronary revascularization. A minimum of 30 days was required between outcome events of the same type to be considered as separate events. If a MI or UA was followed by a coronary revascularization, a minimum of 30 days was required to have passed for the revascularization procedure to be considered as a separate event. CV death was defined as death from any CV-related cause (ICD-10-CM codes I00 to I78), or as death within 30 days of hospitalization due to a CV event. Non-CV death was defined as death from any other cause than CV-related one.

ired to have passed for the revascularization procedure to be considered as a separate event. CV death was defined as death from any CV-related cause (ICD-10-CM codes I00 to I78), or as death within 30 days of hospitalization due to a CV event. Non-CV death was defined as death from any other cause than CV-related one. Statistical analysis The baseline period was defined as the 5-year period prior to the index date and was used to observe demographic and clinical characteristics as well as the patient selection criteria. Statin-use was observed during the 1-year period prior to index date and patients were followed until either CV event, lost to follow-up, or end of study. Patient characteristics were assessed during the baseline period and presented as mean and standard deviation for continuous variables, and absolute numbers (n) or proportions (%) for categorical variables. Rates of incident CV events were calculated by dividing the number of first events by the person-years of follow-up until the event, death, or end of follow-up, expressed per 100 person-years. Cardiovascular event rates for multiple events were calculated as the total number of events divided by the total follow-up time, until either death or end of follow-up (31 December 2015). All conditions in the register data were identified based on the ICD-10 coding system, which includes diagnoses, and the KVÅ coding system, which includes medical procedures. All data management and statistical analysis was performed using MySQL and Stata 14 (StataCorp LP, College Station, TX, USA).

Statistical analysis The baseline period was defined as the 5-year period prior to the index date and was used to observe demographic and clinical characteristics as well as the patient selection criteria. Statin-use was observed during the 1-year period prior to index date and patients were followed until either CV event, lost to follow-up, or end of study. Patient characteristics were assessed during the baseline period and presented as mean and standard deviation for continuous variables, and absolute numbers (n) or proportions (%) for categorical variables. Rates of incident CV events were calculated by dividing the number of first events by the person-years of follow-up until the event, death, or end of follow-up, expressed per 100 person-years. Cardiovascular event rates for multiple events were calculated as the total number of events divided by the total follow-up time, until either death or end of follow-up (31 December 2015). All conditions in the register data were identified based on the ICD-10 coding system, which includes diagnoses, and the KVÅ coding system, which includes medical procedures. All data management and statistical analysis was performed using MySQL and Stata 14 (StataCorp LP, College Station, TX, USA). Results The number of patients meeting the study inclusion criteria was 54 992 for the ASCVD prevalent cohort, 45 895 for the MI incident cohort, and 36 134 for the IS incident cohort (Table 3). The overall high ASCVD risk study population had a mean age of >70 years, at least 60% were men, and the majority of all patients had hypertension. Few patients (3–10%) were receiving high-intensity statin therapy at baseline, while 25% of patients in the MI incident cohort and 11% of patients in the IS incident cohort received a high-intensity statin on the first filled prescription date following index event. Approximately half of patients (ASCVD prevalent cohort: 43.8%; MI incident cohort: 54.0%; IS incident cohort: 40.0%) experienced a CV event of either MI, UA, IS, coronary revascularization, or CV death, during a mean follow-up of 7.3 years, 3.9 years, and 3.7 years in the ASCVD prevalent cohort, the MI incident cohort, and the IS incident cohort, respectively (Table 4). A large proportion of patients had more than one (≥2) CV event during follow-up; the hard MACE composite: 13.8% (ASCVD prevalent cohort), 12.2% (MI incident cohort), 10.1% (IS incident cohort); the ASCVD composite: 16.1% (ASCVD prevalent cohort), 19.2% (MI incident cohort), and 10.9% (IS incident cohort). The most frequent CV event within both outcome composites was CV death (Table 5).

g follow-up; the hard MACE composite: 13.8% (ASCVD prevalent cohort), 12.2% (MI incident cohort), 10.1% (IS incident cohort); the ASCVD composite: 16.1% (ASCVD prevalent cohort), 19.2% (MI incident cohort), and 10.9% (IS incident cohort). The most frequent CV event within both outcome composites was CV death (Table 5). Table 3 Patient characteristics at baseline

g follow-up; the hard MACE composite: 13.8% (ASCVD prevalent cohort), 12.2% (MI incident cohort), 10.1% (IS incident cohort); the ASCVD composite: 16.1% (ASCVD prevalent cohort), 19.2% (MI incident cohort), and 10.9% (IS incident cohort). The most frequent CV event within both outcome composites was CV death (Table 5). Table 3 Patient characteristics at baseline ASCVD prevalent cohort (n = 54 992) MI incident cohort (n = 45 895) IS incident cohort (n = 36 134) Mean or proportion SD or count Mean or proportion SD or count Mean or proportion SD or count Age (years) 72.5 8.3 71.0 9.6 72.9 8.7 Gender, male (%) 63.2 34 754 66.9 30 704 60.1 21 719 CV history (%) History of MI 68.2 37 480 100a 45 895 14.7 5321 History of UA 17.9 9841 12.1 5546 5.3 1903 History of IS 34.6 19 024 8.7 3981 100a 36 134 History of HF 20.8 11 445 27.1 12 413 17.5 6331 History of TIA 5.1 2827 4.5 2052 11.2 4048 History of CABG/PCI 8.9 4884 21.4 9819 3.4 1230 Charlson comorbidity index 2.5 1.8 3.2 2.2 3.00 2.1 Follow-up length (years) 7.3 3.0 3.9 2.7 3.7 2.6 Diabetes (%) 36.4 20 014 42.9 19 699 39.1 14 140 Hypertension (%) 98.0 53 916 99.6 45 724 96.8 34 961 Chronic kidney disease (%) 2.6 1447 7.1 3261 4.9 1772 High-intensity statin at index (%) 3.3 1814 10.6 4847 7.7 2792 High-intensity statin as first filled prescription following index event (%) — — 25.1 11 516 10.9 3947 Anti-thrombotic medication (%) 48.1 26 438 51.5 23 640 59.1 21 365 Anti-hypertensive medication (%) 47.2 25 976 51.9 23 823 57.4 20 726 ASCVD, atherosclerotic cardiovascular disease; CABG/PCI, coronary artery bypass grafting/percutaneous coronary intervention; IS, ischaemic stroke; MI, myocardial infarction; SD, standard deviation; UA, unstable angina.

1 26 438 51.5 23 640 59.1 21 365 Anti-hypertensive medication (%) 47.2 25 976 51.9 23 823 57.4 20 726 ASCVD, atherosclerotic cardiovascular disease; CABG/PCI, coronary artery bypass grafting/percutaneous coronary intervention; IS, ischaemic stroke; MI, myocardial infarction; SD, standard deviation; UA, unstable angina. a Including the index event (myocardial infarction/ischaemic stroke). Table 4 Frequency of cardiovascular events during follow-up within the ASCVD composite and the hard MACE composite Event count ASCVD prevalent cohort MI incident cohort IS incident cohort ASCVD composite Hard MACE composite ASCVD composite Hard MACE composite ASCVD composite Hard MACE composite N Percentage N Percentage N Percentage N Percentage N Percentage N Percentage Total with ≥1 incident CV event 24 100 43.8 22 427 40.8 24 806 54.0 18 021 39.3 14 470 40.0 14 039 38.9 1 15 222 27.7 14 825 27.0 15 997 34.9 12 414 27.0 10 546 29.2 10 401 28.8 2 5756 10.5 5197 9.5 5415 11.8 3562 7.8 2840 7.9 2686 7.4 3 1892 3.4 1563 2.8 1942 4.2 1230 2.7 755 2.1 683 1.9 ≥4 1230 2.2 842 1.5 1452 3.2 815 1.8 329 0.9 269 0.7 The ASCVD composite includes myocardial infarction, unstable angina, ischaemic stroke, revascularization procedures (coronary artery bypass grafting/percutaneous coronary intervention), and cardiovascular death. The hard MACE composite includes myocardial infarction, ischaemic stroke, and cardiovascular death. ASCVD, atherosclerotic cardiovascular disease; IS, ischaemic stroke; MACE, major adverse cardiovascular event; MI, myocardial infarction; N, number of patients.

N Percentage N Percentage N Percentage N Percentage N Percentage N Percentage Total with ≥1 incident CV event 24 100 43.8 22 427 40.8 24 806 54.0 18 021 39.3 14 470 40.0 14 039 38.9 1 15 222 27.7 14 825 27.0 15 997 34.9 12 414 27.0 10 546 29.2 10 401 28.8 2 5756 10.5 5197 9.5 5415 11.8 3562 7.8 2840 7.9 2686 7.4 3 1892 3.4 1563 2.8 1942 4.2 1230 2.7 755 2.1 683 1.9 ≥4 1230 2.2 842 1.5 1452 3.2 815 1.8 329 0.9 269 0.7 The ASCVD composite includes myocardial infarction, unstable angina, ischaemic stroke, revascularization procedures (coronary artery bypass grafting/percutaneous coronary intervention), and cardiovascular death. The hard MACE composite includes myocardial infarction, ischaemic stroke, and cardiovascular death. ASCVD, atherosclerotic cardiovascular disease; IS, ischaemic stroke; MACE, major adverse cardiovascular event; MI, myocardial infarction; N, number of patients. Table 5 Distribution of events (%) within the ASCVD composite and the hard MACE composite for multiple events ASCVD prevalent cohort MI incident cohort IS incident cohort ASCVD composite Hard MACE composite ASCVD composite Hard MACE composite ASCVD composite Hard MACE composite MI (%) 30.0 34.0 32.0 46.9 13.7 14.4 IS (%) 20.8 23.5 7.6 11.1 35.2 37.0 CV death (%) 37.4 42.4 28.6 42.0 46.3 48.7 UA (%) 6.9 — 13.8 — 2.4 — CABG/PCI (%) 4.9 — 18.0 — 2.5 — The ASCVD composite includes myocardial infarction, unstable angina, ischaemic stroke, revascularization procedures (coronary artery bypass grafting/percutaneous coronary intervention), and cardiovascular death.

.2 37.0 CV death (%) 37.4 42.4 28.6 42.0 46.3 48.7 UA (%) 6.9 — 13.8 — 2.4 — CABG/PCI (%) 4.9 — 18.0 — 2.5 — The ASCVD composite includes myocardial infarction, unstable angina, ischaemic stroke, revascularization procedures (coronary artery bypass grafting/percutaneous coronary intervention), and cardiovascular death. The hard MACE composite includes myocardial infarction, ischaemic stroke, and cardiovascular death. ASCVD, atherosclerotic cardiovascular disease; CABG/PCI, coronary artery bypass grafting/percutaneous coronary intervention; CV death, cardiovascular death; IS, ischaemic stroke; MACE, major adverse cardiovascular event; MI, myocardial infarction; UA, unstable angina.

The hard MACE composite includes myocardial infarction, ischaemic stroke, and cardiovascular death. ASCVD, atherosclerotic cardiovascular disease; CABG/PCI, coronary artery bypass grafting/percutaneous coronary intervention; CV death, cardiovascular death; IS, ischaemic stroke; MACE, major adverse cardiovascular event; MI, myocardial infarction; UA, unstable angina. The largest proportion of patients experiencing an incident MI during follow-up was seen in the MI incident cohort. Similarly, the largest proportion of patients that had an incident IS during follow-up was seen in the IS incident cohort. The CV death rate per 100 person-years were 4.7, 5.3, and 7.2 in patients with one prior CV event (MI, UA, IS, TIA, coronary revascularization, HF) vs. 6.7, 8.7, and 8.0 in patients with two or more (≥2) prior events in the ASCVD prevalent, the MI incident and the IS incident cohort, respectively (Table 6). Cardiovascular event rates for incident events and multiple events are presented in Table 7. The rates of incident CV events per 100 person-years for the hard MACE composite were 6.3 (ASCVD prevalent cohort), 11.9 (MI incident cohort), and 12.3 (IS incident cohort). Within the ASCVD prevalent cohort, the hard MACE composite rate was found to be higher (7.3 per 100-person years) for the subgroup of patients with a diabetes mellitus (type 1 or 2) diagnosis, compared with the overall study cohort. The rates of incident CV events for the ASCVD composite were 7.0 (ASCVD prevalent cohort), 21.7 (MI incident cohort), and 12.9 (IS incident cohort). The multiple-event MACE composite rates were 8.5, 15.4, and 14.4, respectively, in the ASCVD prevalent, MI incident, and IS incident cohorts. The multiple-event ASCVD composite rate was 9.6, 22.5, and 15.2, in the ASCVD prevalent cohort, the MI incident cohort, and the IS incident cohort, respectively.

ncident cohort). The multiple-event MACE composite rates were 8.5, 15.4, and 14.4, respectively, in the ASCVD prevalent, MI incident, and IS incident cohorts. The multiple-event ASCVD composite rate was 9.6, 22.5, and 15.2, in the ASCVD prevalent cohort, the MI incident cohort, and the IS incident cohort, respectively. Table 6 Cardiovascular death rates by event count ASCVD prevalent cohort MI incident cohort IS incident cohort Event counta N P-Y Rate per 100 P-Y SE per 100 P-Y N P-Y Rate per 100 P-Y SE per 100 P-Y N P-Y Rate per 100 P-Y SE per 100 P-Y 1 4010 85 321 4.7 1.5 2709 51 349 5.3 1.9 2046 28 444 7.2 2.1 ≥2 4721 70 068 6.7 1.4 3938 45 321 8.7 1.5 1574 19 727 8.0 2.4 ASCVD, atherosclerotic cardiovascular disease; IS, ischaemic stroke; MI, myocardial infarction; N, number of patients; P-Y, person-year; SE, standard error. a Event count includes myocardial infarction, ischaemic stroke, unstable angina, transient ischaemic attack, and heart failure. Table 7 Cardiovascular event rates per 100 person-years: incident and multiple-event rates, by outcome

ASCVD prevalent cohort MI incident cohort IS incident cohort Event counta N P-Y Rate per 100 P-Y SE per 100 P-Y N P-Y Rate per 100 P-Y SE per 100 P-Y N P-Y Rate per 100 P-Y SE per 100 P-Y 1 4010 85 321 4.7 1.5 2709 51 349 5.3 1.9 2046 28 444 7.2 2.1 ≥2 4721 70 068 6.7 1.4 3938 45 321 8.7 1.5 1574 19 727 8.0 2.4 ASCVD, atherosclerotic cardiovascular disease; IS, ischaemic stroke; MI, myocardial infarction; N, number of patients; P-Y, person-year; SE, standard error. a Event count includes myocardial infarction, ischaemic stroke, unstable angina, transient ischaemic attack, and heart failure. Table 7 Cardiovascular event rates per 100 person-years: incident and multiple-event rates, by outcome ASCVD prevalent cohort MI incident cohort IS incident cohort Endpoint Events P-Y Rate per 100 P-Y SE per 100 P-Y Events P-Y Rate per 100 P-Y SE per 100 P-Y Events P-Y Rate per 100 P-Y SE per 100 P-Y Hard MACE composite 22 427 356 308 6.3 0.7 18 021 151 317 11.9 0.7 14 039 113 982 12.3 0.8 ASCVD composite 24 100 345 113 7.0 0.6 24 806 114 472 21.7 0.6 14 470 112 351 12.9 0.8 Hard MACE composite— multiple events 33 886 400 685 8.5 0.5 27 255 177 057 15.4 0.6 19 058 132 189 14.4 0.7 ASCVD composite— multiple events 38 406 400 685 9.6 0.4 39 917 177 057 22.5 0.4 20 031 132 189 15.2 0.7 The ASCVD composite includes myocardial infarction, unstable angina, ischaemic stroke, revascularization procedures (coronary artery bypass grafting/percutaneous coronary intervention), and cardiovascular death. The hard MACE composite includes MI, IS, and cardiovascular death.

ASCVD prevalent cohort MI incident cohort IS incident cohort Endpoint Events P-Y Rate per 100 P-Y SE per 100 P-Y Events P-Y Rate per 100 P-Y SE per 100 P-Y Events P-Y Rate per 100 P-Y SE per 100 P-Y Hard MACE composite 22 427 356 308 6.3 0.7 18 021 151 317 11.9 0.7 14 039 113 982 12.3 0.8 ASCVD composite 24 100 345 113 7.0 0.6 24 806 114 472 21.7 0.6 14 470 112 351 12.9 0.8 Hard MACE composite— multiple events 33 886 400 685 8.5 0.5 27 255 177 057 15.4 0.6 19 058 132 189 14.4 0.7 ASCVD composite— multiple events 38 406 400 685 9.6 0.4 39 917 177 057 22.5 0.4 20 031 132 189 15.2 0.7 The ASCVD composite includes myocardial infarction, unstable angina, ischaemic stroke, revascularization procedures (coronary artery bypass grafting/percutaneous coronary intervention), and cardiovascular death. The hard MACE composite includes MI, IS, and cardiovascular death. ASCVD, atherosclerotic cardiovascular disease; CV, cardiovascular; IS, ischaemic stroke; MACE, major adverse cardiovascular events; MI, myocardial infarction; P-Y, person-years; SE, standard error.

ASCVD prevalent cohort MI incident cohort IS incident cohort Endpoint Events P-Y Rate per 100 P-Y SE per 100 P-Y Events P-Y Rate per 100 P-Y SE per 100 P-Y Events P-Y Rate per 100 P-Y SE per 100 P-Y Hard MACE composite 22 427 356 308 6.3 0.7 18 021 151 317 11.9 0.7 14 039 113 982 12.3 0.8 ASCVD composite 24 100 345 113 7.0 0.6 24 806 114 472 21.7 0.6 14 470 112 351 12.9 0.8 Hard MACE composite— multiple events 33 886 400 685 8.5 0.5 27 255 177 057 15.4 0.6 19 058 132 189 14.4 0.7 ASCVD composite— multiple events 38 406 400 685 9.6 0.4 39 917 177 057 22.5 0.4 20 031 132 189 15.2 0.7 The ASCVD composite includes myocardial infarction, unstable angina, ischaemic stroke, revascularization procedures (coronary artery bypass grafting/percutaneous coronary intervention), and cardiovascular death. The hard MACE composite includes MI, IS, and cardiovascular death. ASCVD, atherosclerotic cardiovascular disease; CV, cardiovascular; IS, ischaemic stroke; MACE, major adverse cardiovascular events; MI, myocardial infarction; P-Y, person-years; SE, standard error. Discussion Existing data informing rates of incident CV events are mainly derived from clinical trial populations which may underestimate event rates due to potential patient selection bias from trial recruitment, or to the stringent clinical care received by clinical trial participants. Real-world evidence from this retrospective cohort study demonstrates that CVD burden among high-risk patients with clinically evident ASCVD being treated with moderate or high-intensity statin therapy is substantial, especially in close proximity to an incident CV event.

l care received by clinical trial participants. Real-world evidence from this retrospective cohort study demonstrates that CVD burden among high-risk patients with clinically evident ASCVD being treated with moderate or high-intensity statin therapy is substantial, especially in close proximity to an incident CV event. According to the European Society of Cardiology (ESC) and European Atherosclerosis Society (EAS) guidelines for the management of dyslipidaemias, patients with either documented CVD, very high levels of individual risk factors (such as increased lipid or blood pressure levels), or chronic kidney disease are at very high or high total CV risk.25 The population in this study is similar to the very high-risk population defined in the ESC/EAS guidelines, which advocate for active management of all risk factors. Despite treatment with moderate or high-intensity statins, almost 44% of the patients in the ASCVD prevalent cohort experienced a new event (MI, UA, IS, coronary revascularization, or CV-related death) during a mean follow-up of 7.3 years. In the two incident cohorts CV burden was even higher, with approximately 54% in the MI incident cohort and 40% in the IS incident cohort experiencing at least one new CV event within the ASCVD composite during a mean follow-up of almost 4 years.

scularization, or CV-related death) during a mean follow-up of 7.3 years. In the two incident cohorts CV burden was even higher, with approximately 54% in the MI incident cohort and 40% in the IS incident cohort experiencing at least one new CV event within the ASCVD composite during a mean follow-up of almost 4 years. These findings are in line with previous real-world data showing a high burden of CVD among high-risk patients. Toth et al.26 applied the FOURIER eligibility criteria to identify patients in UK medical records and showed that approximately 33% of patients in the high-risk ASCVD cohort (similar to the ASCVD prevalent cohort) experienced at least one (≥1) new event (MI, IS, UA, coronary revascularization, or CV death) during a mean follow-up of 5.4 years. The corresponding proportions in UK incident cohorts were 27% and 40% over 2.8 years, defined by diagnosis of IS or acute coronary syndrome (ACS) (MI or UA), respectively. The rates of incident CV events per 100 person-years based on UK data were 7.5, 21.1, and 11.9, respectively, for the high-risk ASCVD cohort, the ACS incident cohort, and the IS incident cohort. In addition, Punekar et al.27 showed that 33.9% of high-risk patients experienced at least one new event (MI, IS, coronary revascularization, UA, TIA, or HF) over 2 years follow-up, based on US administrative claims data. The relatively large share of patients (ASCVD prevalent cohort: 16.2%; MI incident cohort: 19.2%; IS incident cohort: 10.9%) who experienced two or more recurrent CV events (≥2 events) in the current study is in line with the results found by Punekar et al. where 12.9% of high-risk patients had three or more (≥3) CV events.

elatively large share of patients (ASCVD prevalent cohort: 16.2%; MI incident cohort: 19.2%; IS incident cohort: 10.9%) who experienced two or more recurrent CV events (≥2 events) in the current study is in line with the results found by Punekar et al. where 12.9% of high-risk patients had three or more (≥3) CV events. For the two incident cohorts, survivors of MI and IS are at immediate risk of having an additional CV event where in most cases, subsequent events are of the same type as previously experienced by the patient. A significant proportion of patients (37.4% in the ASCVD prevalent cohort, 28.6% in the MI incident cohort, and 46.3% in the IS incident cohort) died due to CVD despite moderate or high-intensity statin use. Cardiovascular death rates increased with the number of events, showing that patients who experienced two or more recurrent CV events are at greater risk of dying due to CVD. Most clinical trials examining CV event rates often present results focusing on the first CV event after the index date. However, to fully understand the CV burden and its impact on patients and the healthcare system, it is important to also evaluate subsequent CV events experienced by patients. In addition to the first CV event, this study quantified and captured both the occurrence of multiple and of recurrent events, by also following patients with established CVD past their first outcome event until either death or end of follow-up.

s important to also evaluate subsequent CV events experienced by patients. In addition to the first CV event, this study quantified and captured both the occurrence of multiple and of recurrent events, by also following patients with established CVD past their first outcome event until either death or end of follow-up. This study analysed a large study population combined with long follow-up to study CV outcomes. Mean follow-up was longer in all three study cohorts compared with the 2.2 years of follow-up in the FOURIER clinical trial. The primary outcome in the FOURIER clinical trial, incidence of CV death, MI, IS, hospitalization for UA, or coronary revascularization (corresponding to the ASCVD composite), occurred in 14.6% of participants in the placebo group. The key secondary endpoint, a composite of MI, IS, or CV related death (corresponding to the hard MACE composite), occurred in 7.4% of patients. The multiple-events MACE composite rates in this study were more than two to three times higher compared with the placebo plus standard background therapy arm in the FOURIER trial (4.2 per 100 person-years).28 Several differences may account for the discrepancies between the clinical trial rates and the rates reported from this real-world study. The majority of the high ASCVD risk patients in this study were on moderate-intensity statin treatment at the time of index. As for the FOURIER clinical trial, most participants were high-intensity statin users; the placebo group included 69.1% patients on high-intensity statin therapy, 30.7% on moderate-intensity statin therapy, and 0.2% of patients were classified as low-intensity statin users, unknown intensity, or no data. In addition, the quality of care may differ between the real-world setting and a clinical trial for a variety of reasons including disproportionate recruitment from high performing academic centres, whereas this study comprised patients from nationwide population-based registers with complete coverage. Further, the proportion of women in the current study (ASCVD prevalent cohort: 36.8%; the MI incident cohort: 33.1%; the IS incident cohort: 39.9%) was larger than in the FOURIER outcomes trial (25.0%).

mic centres, whereas this study comprised patients from nationwide population-based registers with complete coverage. Further, the proportion of women in the current study (ASCVD prevalent cohort: 36.8%; the MI incident cohort: 33.1%; the IS incident cohort: 39.9%) was larger than in the FOURIER outcomes trial (25.0%). Strengths of this study include the high degree of validity, completeness and data quality in the Swedish national registers,20,21,23 which enable reliable real-world estimates. The National Patient Register contains more than 99% of hospitalizations, while The Prescription Drug Register covers all prescriptions distributed via pharmacies. Using ICD-10 codes when identifying CV events can be seen as a limitation in the study. Ludvigsson et al.21 have found that positive predictive values of the recorded diagnoses were 85–95% for most diagnoses in the Swedish National Patient Register, when comparing to information in medical records. Further, approximately 90% of patients with a primary diagnosis of MI, had MI as the underlying cause of death according to medical records. Another limitation to the study is that laboratory values, including LDL-C and other lipid values (HDL-C, triglycerides), are not available in the Swedish national registers. The study population’s LDL-C level was thus unknown. The lack of lipid values is anticipated to have limited impact since the patients will be of similar CV risk as the FOURIER trial population based on the inclusion criteria of history of CV events (MI, IS, PAD) as well as statin treatment for elevated LDL-C and other CV risk factors (e.g. diabetes). According to Swedish treatment guidelines, high-risk patients (based on the SCORE-model, ≥10%) and patients with a cholesterol level >4.9 mmol/L, are recommended moderate or high-intensity statin treatment.29,30 Thus, filled prescriptions of moderate and/or high-intensity statins are considered an appropriate proxy for LDL-C. Other risk factors include familial hypercholesterolaemia status and lifestyle conditions such as smoking which were not possible to account for in this real-world data, nor was it the focus of the study to investigate their impact on the risk of CV events. Another limitation is that statin use was not studied during follow-up, meaning that patients may have discontinued their statin treatment several years before experiencing an event.

possible to account for in this real-world data, nor was it the focus of the study to investigate their impact on the risk of CV events. Another limitation is that statin use was not studied during follow-up, meaning that patients may have discontinued their statin treatment several years before experiencing an event. In fact, findings from the EUROASPIRE IV survey, showed that 11.6% of patients discontinued their statin therapy during the one-year period after hospital discharged after a coronary event.31 However, statin adherence is difficult to study in a retrospective dataset based on national registers, compared with a clinical setting such as the FOURIER trial where patients are assumed to receive optimal lipid lowering treatment with statins. The study findings highlight the unmet need and clinical burden among high-risk patients treated with high or moderate-intensity statin therapy and indicate the need for additional and alternative therapeutic options. Evidence from the FOURIER trial indicates that evolocumab, a PCSK9 inhibitor, may provide additional benefit for CV event reduction as add-on compared with moderate or high-intensity statin therapy alone. Furthermore, more recent AHA/ACC recommendations state that in very high-risk patients with multiple high-risk clinical factors, and if LDL-C levels remain ≥70 mg/dL (≥1.8 mmol/L), adding a PCSK9 inhibitor is reasonable if the cost/benefit ratio is favourable.32

red with moderate or high-intensity statin therapy alone. Furthermore, more recent AHA/ACC recommendations state that in very high-risk patients with multiple high-risk clinical factors, and if LDL-C levels remain ≥70 mg/dL (≥1.8 mmol/L), adding a PCSK9 inhibitor is reasonable if the cost/benefit ratio is favourable.32 Conclusions Despite treatment with moderate- or high-intensity statins, the population with clinically evident ASCVD experiences high CV event rates, especially in close proximity to an earlier event. A large proportion of patients experience recurrent CV events and CV death rates show that patients who have recurrent CV events are at greater risk of dying. In this real-world setting, the multiple-event MACE composite rates were more than two to three times higher than in the FOURIER clinical trial, indicating a substantial burden for patients and health care system. Funding This study was funded by Amgen, Inc. Conflict of interest: M.L., J.B., and S.H. are employed by Quantify Research, a contract research organization that provides consultancy services for the pharmaceutical industry. K.M.F. has received consulting fees from Amgen, Inc. M.E. has no conflicts of interest to disclose. M.T., Y.Q. are employed by Amgen, Inc., and own Amgen stock/stock options. G.V. is employed by Amgen (Europe) GmbH, and owns Amgen stock/stock options. M.K.S. is employed by Amgen AB, and owns Amgen stock/stock options.