Browse the corpus

KEY POINTS Question: Does a “smoothing effect” exist between manually recorded and electronically recorded vital sign measurements in postoperative care? Findings: Using a mixed-effects model, we found no relationship between continuous-manual differences and continuous-manual average values for heart rate and respiratory rate, and we found a weak (but clinically insignificant) relationship for oxygen saturation. Meaning: We found that clinical staff in a postoperative ward did not “smooth” vital sign values with a bias toward recording more normal readings because the differences between manual and continuous vital sign measurements were not related to the vital sign values. Manual measurements of the main vital signs—which include respiratory rate, blood pressure, heart rate, temperature, and oxygen saturation (Spo2)—are often inaccurate.1–3 Manual calculations of clinical risk scores are also error prone.4–6 While automated monitoring technology exists, it is mostly confined to high-acuity patients, and manual measurement and documentation of vital signs remain the standard of care in many wards. A potential source of inaccuracy that may exist in the manual vital sign record is data smoothing.7 Clinical staff may be biased toward vital sign values that lie within the assumed limits of normality and record vital sign values that are incorrectly normal— “smoothing” the extremes in the vital sign record. If real, this “smoothing effect”8 may result in lost opportunities for early recognition of physiological deterioration.

aff may be biased toward vital sign values that lie within the assumed limits of normality and record vital sign values that are incorrectly normal— “smoothing” the extremes in the vital sign record. If real, this “smoothing effect”8 may result in lost opportunities for early recognition of physiological deterioration. The smoothing effect has been reported to occur during anesthesia8–13 and in acute ward monitoring.14,15 Most studies use methods based on the comparison of vital sign values from manual and automated measurements sources.9–15 The comparison method makes the cause of “data smoothing” unclear. Vital sign values from monitoring equipment are noisy and may be corrupted with signal artefact, so the smoothing effect may partly result from clinicians correcting spurious values.14 Undersampling may also affect data smoothing based on the magnitude and frequency of extremal values in longitudinal records because sparsely sampled manual observations may not coincide with times of large fluctuations in vital sign values. Analysis compensating for these confounding factors is essential to discover whether the smoothing effect is of clinical relevance.

the magnitude and frequency of extremal values in longitudinal records because sparsely sampled manual observations may not coincide with times of large fluctuations in vital sign values. Analysis compensating for these confounding factors is essential to discover whether the smoothing effect is of clinical relevance. We present a secondary analysis of a large database of postoperative vital sign records to investigate data smoothing of respiratory rate, heart rate, and Spo2. We propose that data smoothing increases the differences between continuous and manual vital sign measurements as the (absolute) value of the vital sign becomes more extreme. We tested whether differences between continuous and manual vital sign recordings are related to the average value of the 2 vital sign recordings. We also assessed agreement between continuous and manual data. Finally, we investigated whether there is an arousal effect caused by the vital sign taker.15 METHODS This article adheres to the Reporting of Studies Conducted Using Observational Routinely-Collected Health Data statement, an extension of the Strengthening the Reporting of Observational Studies in Epidemiology guidelines.16,17

We present a secondary analysis of a large database of postoperative vital sign records to investigate data smoothing of respiratory rate, heart rate, and Spo2. We propose that data smoothing increases the differences between continuous and manual vital sign measurements as the (absolute) value of the vital sign becomes more extreme. We tested whether differences between continuous and manual vital sign recordings are related to the average value of the 2 vital sign recordings. We also assessed agreement between continuous and manual data. Finally, we investigated whether there is an arousal effect caused by the vital sign taker.15 METHODS This article adheres to the Reporting of Studies Conducted Using Observational Routinely-Collected Health Data statement, an extension of the Strengthening the Reporting of Observational Studies in Epidemiology guidelines.16,17 Dataset The database for this retrospective analysis was created during the Computer Alerting Monitoring System 2 (CALMS-2) study, which was granted ethical approval (Mid and South Buckinghamshire ethics committee Research Ethics Committee: 08/H0604/79, December 9, 2008, and Leeds [West] ethics committee Research Ethics Committee: 11/YH/0056, May 20, 2011) and registered in the International Standard Randomised Controlled Trial Number (ISRCTN) database (principal investigator: P.J.W., ISRCTN No: ISRCTN58660550, August 11, 2017). This study assessed whether ambulatory physiological monitoring combined with an alerting system improved recognition and outcomes in patients after major surgery.

he International Standard Randomised Controlled Trial Number (ISRCTN) database (principal investigator: P.J.W., ISRCTN No: ISRCTN58660550, August 11, 2017). This study assessed whether ambulatory physiological monitoring combined with an alerting system improved recognition and outcomes in patients after major surgery. Vital sign data used in the CALMS-2 study were collected in a step-down postoperative ward of the Oxford University Hospitals National Health Service (NHS) Trust, Oxford, between May 2009 and December 2013. Potential participants were screened during preoperative assessment and deemed eligible if they were planned to undergo major upper gastrointestinal surgery. This category was defined as follows: oesophagectomy, oesophagogastrectomy, gastrectomy, Whipple’s operation, liver resection, pancreatectomy, gastric bypass, biliary reconstruction, and splenectomy. Participants were excluded based on the following criteria: participants <16 years of age, pregnant women, participants unable to wear the required monitoring, participants without the capacity to consent, and participants who could not understand written English and for whom no translator could be found. For this secondary analysis, patients who did not receive contemporaneous bedside electronic vital sign monitoring and manual vital sign observations were excluded a priori. Written informed consent was obtained for all subjects.

ipants who could not understand written English and for whom no translator could be found. For this secondary analysis, patients who did not receive contemporaneous bedside electronic vital sign monitoring and manual vital sign observations were excluded a priori. Written informed consent was obtained for all subjects. In the step-down ward used in the CALMS-2 study, high-risk patients are admitted after a period of elective intensive care unit stay, while patients with a lower risk of complication are admitted to the ward immediately after surgery. High-risk patients typically receive an increased level of care for the first 2–48 hours of their ward stay, during which time they undergo conventional bedside monitoring, consisting of continuously measured respiratory rate, heart rate, and Spo2 (Philips M3046A/Intellivue MP50 clinical monitor; Philips Healthcare, Best, the Netherlands). Respiratory rate was measured by impedance pneumography, heart rate was derived from the electrocardiogram, and Spo2 was measured by pulse oximetry. Clinical staff also made manual measurements of blood pressure and temperature, typically at hourly intervals. After this initial period, these patients then join the other patients on the ward in receiving general-ward-level care, which consists of manual vital sign recording typically at 4-hour intervals. The standard of care for measuring respiratory rate on the ward is by counting chest wall movements, and heart rate and Spo2 measurements were likely to be transposed from the monitor screen. In the CALMS-2 study, ambulatory monitoring of heart rate and Spo2 was undertaken. However, for consistency, we restricted this analysis to manually recorded values that could be compared with contemporaneous values from bedside monitoring.

ate and Spo2 measurements were likely to be transposed from the monitor screen. In the CALMS-2 study, ambulatory monitoring of heart rate and Spo2 was undertaken. However, for consistency, we restricted this analysis to manually recorded values that could be compared with contemporaneous values from bedside monitoring. Manual vital sign measurements documented on paper-based bedside charts were double entered into an electronic database. A third researcher reconciled differences with access to the original charts.5 Continuous vital sign data from the bedside monitors were saved directly from the monitor every second. The final dataset obtained for this study consisted of vital sign records of respiratory rate, heart rate, and Spo2 from continuous bedside monitoring equipment and manual vital sign observations. Manual and continuous vital sign values were compared based on the timestamps taken from paper records and the computer-generated timestamps from the patient monitor. Clinical staff manually recorded blood pressure measurements at the time of observation, and these were also logged in the automatically generated data record. We set the manual observation time for all vital signs to the computer-generated timestamps for blood pressure (while checking that data were correctly matched). This calibration method ensured that the data considered contemporaneous from manual and continuous measurement sources were synchronized, and thus could appropriately be used for comparison.

e for all vital signs to the computer-generated timestamps for blood pressure (while checking that data were correctly matched). This calibration method ensured that the data considered contemporaneous from manual and continuous measurement sources were synchronized, and thus could appropriately be used for comparison. Statistical Analysis We summarized the number of vital sign measurements included per patient using the sample median and interquartile range. We sampled the continuous data at the time of each manual observation to create paired measurements of continuous and manual vital signs. We sampled the continuous data by extracting the median value of a 5-minute window centered at the time of manual observation. We used this methodology (also known as “median filtering”) to summarize the continuous data without the effects of measurement noise or short-term variance while retaining long-term vital sign trends. We included all manual observations with contemporaneous periods of continuous data, allowing patients to provide multiple observations in our analysis. We selected a 5-minute window to reflect clinical practice, in line with previous work.13,15 We undertook sensitivity analyses by recomputing the primary assessment method for windows of 1–10 minutes.

ns with contemporaneous periods of continuous data, allowing patients to provide multiple observations in our analysis. We selected a 5-minute window to reflect clinical practice, in line with previous work.13,15 We undertook sensitivity analyses by recomputing the primary assessment method for windows of 1–10 minutes. To obtain the differences between the 2 measurement sources, we subtracted the manual vital sign from the continuous vital sign in each measurement pair. We also calculated the average of the continuous and manual vital signs for each measurement pair. We modeled the relationship between the differences and the averages of the measurement recordings using a linear mixed-effects model, with averages as a fixed effect and subject as a random effect to adjust for repeated measurements. We note that there may be factors other than a smoothing effect that could affect the relationship between the continuous-manual difference and the continuous-manual average, notably the accuracy of the continuous monitoring equipment. However, given our use of gold standard monitoring equipment and median filtering to remove transient outliers, the continuous data were likely to be equally reliable over the range of measurement values. Thus, a linear relationship between the continuous-manual difference and the continuous-manual average would be best explained by a clinical bias in the manual data, allowing the smoothing effect to be tested.

ansient outliers, the continuous data were likely to be equally reliable over the range of measurement values. Thus, a linear relationship between the continuous-manual difference and the continuous-manual average would be best explained by a clinical bias in the manual data, allowing the smoothing effect to be tested. The regression slope of the fixed effect in the mixed-effects model represents the increase in the continuous-manual difference for a 1-unit increase in the continuous-manual average. Statistical significance of the regression slope was calculated with an F test (type III with Kenward–Roger degrees of freedom approximation),18 using a significance of 0.05 to reject the null hypothesis that the regression slope was 0 (no relationship). To assess the clinical significance of the linear relationship, we simulated the mean differences and 95% confidence intervals (CIs) predicted by our model at vital sign values corresponding to important clinical thresholds. We used respiratory rate values of 8 and 24 breaths/min, heart rate values of 40 and 130 beats/min, and an Spo2 value of 91%, which are the maximum National Early Warning Score limits for each vital sign.19

dence intervals (CIs) predicted by our model at vital sign values corresponding to important clinical thresholds. We used respiratory rate values of 8 and 24 breaths/min, heart rate values of 40 and 130 beats/min, and an Spo2 value of 91%, which are the maximum National Early Warning Score limits for each vital sign.19 We generated Bland–Altman plots for the continuous and manual vital sign data by plotting the differences between the 2 measures (y-axis) against the averages of the 2 measures (x-axis). The bias was calculated as the mean difference between continuous and manual measurement pairs. Limits of agreement (LoAs) were calculated using a mixed-effects model, including a subject random effect to adjust for repeated measurements.20 CIs (95%) were calculated for the bias and the LoAs using the method recommended by Bland and Altman.21 Horizontal lines were included on the Bland–Altman plots to show the bias and LoAs. The regression line from the linear mixed-effect model was plotted to visualize the relationship between the differences and the averages.

5%) were calculated for the bias and the LoAs using the method recommended by Bland and Altman.21 Horizontal lines were included on the Bland–Altman plots to show the bias and LoAs. The regression line from the linear mixed-effect model was plotted to visualize the relationship between the differences and the averages. We assessed whether there is an arousal effect caused by the vital sign taker by comparing continuous data before and during the time of manual observation. For the continuous data before the observation, we used the median value of a 15-minute window ending 5 minutes before the time of observation. For the continuous data during the time of observation, we used the median of a 5-minute window, as described in previous paragraphs. These methods replicate those of Taenzer et al,15 to allow comparison. The selection of the continuous data is shown schematically for heart rate in Figure 1. We prepared Bland–Altman plots for the “before” and “during” measurements using the same methodology described in the previous paragraph. Figure 1. An example heart rate observation, for which the “before” window is shown in blue, the “during” window is shown in green, and the manual observation (synchronized to the time of the blood pressure measurement) is shown in red.

We assessed whether there is an arousal effect caused by the vital sign taker by comparing continuous data before and during the time of manual observation. For the continuous data before the observation, we used the median value of a 15-minute window ending 5 minutes before the time of observation. For the continuous data during the time of observation, we used the median of a 5-minute window, as described in previous paragraphs. These methods replicate those of Taenzer et al,15 to allow comparison. The selection of the continuous data is shown schematically for heart rate in Figure 1. We prepared Bland–Altman plots for the “before” and “during” measurements using the same methodology described in the previous paragraph. Figure 1. An example heart rate observation, for which the “before” window is shown in blue, the “during” window is shown in green, and the manual observation (synchronized to the time of the blood pressure measurement) is shown in red. We did not perform sample size calculations for this study because it is a retrospective cohort study of an existing dataset—we used all available data, noting that our dataset was larger than those used in most previous analyses of the smoothing effect.9–13,15,22 We also could not identify the statistical power of the study since previous studies of “the smoothing effect” have not used Bland–Altman analyses (preventing estimation of approximate population regression slopes). We instead provided CIs for all reported outcomes to demonstrate the precision of our results, as recommended by Goodman and Berlin.23

statistical power of the study since previous studies of “the smoothing effect” have not used Bland–Altman analyses (preventing estimation of approximate population regression slopes). We instead provided CIs for all reported outcomes to demonstrate the precision of our results, as recommended by Goodman and Berlin.23 RESULTS Patient inclusion is shown in Figure 2. Concurrent manual and continuous vital sign measurements were available for respiratory rate from 263 patients (3740 paired measurements), heart rate from 267 patients (3844 paired measurements), and Spo2 from 271 patients (3896 paired measurements). The median (interquartile range) number of observations for each patient was 11 (7–18) respiratory rate, 11 (7–18) heart rate, and 11 (7–19) Spo2 measurements. Figure 2. The vital sign data selection process. CALMS-2 indicates Computer Alerting Monitoring System 2; Spo2, oxygen saturation. The mixed-effect model regression slope (95% CI) between the continuous-manual difference and the continuous-manual average was 0.04 (−0.01 to 0.10; P = .11) for respiratory rate, 0.04 (−0.01 to 0.09; P = .11) for heart rate, and 0.10 (0.07–0.14; P < .001) for Spo2 (Figure 3 and Table). The mean differences (95% CI) predicted by the model were −1.14 (−1.57 to −0.71) and −0.44 (−1.57 to 0.04) breaths/min for respiratory rates of 8 and 24 breaths/min, respectively. Likewise, the differences were −2.93 (−4.75 to −1.11) and 0.63 (−4.75 to 3.26) beats/min for heart rates of 40 and 130 beats/min, respectively, and −0.88% (−1.17% to −0.60%) for Spo2 values of 91%.

−1.14 (−1.57 to −0.71) and −0.44 (−1.57 to 0.04) breaths/min for respiratory rates of 8 and 24 breaths/min, respectively. Likewise, the differences were −2.93 (−4.75 to −1.11) and 0.63 (−4.75 to 3.26) beats/min for heart rates of 40 and 130 beats/min, respectively, and −0.88% (−1.17% to −0.60%) for Spo2 values of 91%. Table. Results for the Bland–Altman Analysis Comparing Continuous Vital Sign Data to Manual Observations for Respiratory Rate, Heart Rate, and Oxygen Saturation Figure 3. Bland–Altman plots for RR, HR, and Spo2 showing limits of agreement between the continuous data and the manual observation (continuous data–manual observation). Bias and limits of agreement are shown with blue lines, the regression line is shown in green, and a dashed red line shows the zero y-intercept. HR indicates heart rate; RR, respiratory rate; Spo2, oxygen saturation. The bias (LoA) between the continuous and manual data was −0.74 (−7.80 to 6.32) breaths/min for respiratory rate, −1.13 (−17.4 to 15.1) beats/min for heart rate, and −0.25% (−3.35% to 2.84%) for Spo2 (Figure 3 and Table). The bias (LoA) between continuous data before and during manual recording was −0.57 (−5.63 to 4.48) breaths/min for respiratory rate, −0.71 (−10.2 to 8.73) beats/min for heart rate, and −0.07% (−2.67% to 2.54%) for Spo2 (Figure 4 and Table). CIs (95%) for the bias and LoAs are reported in the Table.

The bias (LoA) between the continuous and manual data was −0.74 (−7.80 to 6.32) breaths/min for respiratory rate, −1.13 (−17.4 to 15.1) beats/min for heart rate, and −0.25% (−3.35% to 2.84%) for Spo2 (Figure 3 and Table). The bias (LoA) between continuous data before and during manual recording was −0.57 (−5.63 to 4.48) breaths/min for respiratory rate, −0.71 (−10.2 to 8.73) beats/min for heart rate, and −0.07% (−2.67% to 2.54%) for Spo2 (Figure 4 and Table). CIs (95%) for the bias and LoAs are reported in the Table. Figure 4. Bland–Altman plots for RR, HR, and Spo2 showing limits of agreement between continuous data sampled before the observation and continuous data sampled during the observation (“before” data – “during” data). Bias and limits of agreement are shown with blue lines, and a dashed red line shows the zero y-intercept. HR indicates heart rate; RR, respiratory rate; Spo2, oxygen saturation. Results for the sensitivity analysis of the primary assessment method for different window sizes (during the manual observation) are given in Supplemental Digital Content 1, Table 1, http://links.lww.com/AA/C522.

Figure 4. Bland–Altman plots for RR, HR, and Spo2 showing limits of agreement between continuous data sampled before the observation and continuous data sampled during the observation (“before” data – “during” data). Bias and limits of agreement are shown with blue lines, and a dashed red line shows the zero y-intercept. HR indicates heart rate; RR, respiratory rate; Spo2, oxygen saturation. Results for the sensitivity analysis of the primary assessment method for different window sizes (during the manual observation) are given in Supplemental Digital Content 1, Table 1, http://links.lww.com/AA/C522. DISCUSSION We found no clinically relevant smoothing effect in postoperative care. If vital sign data were smoothed, then manual measurements would be recorded above abnormally low and below abnormally high continuous measurements, causing continuous-manual differences to be related to vital sign values. However, differences between manual and continuous heart rate and respiratory rate measurements were not related to the average values because the regression slope was not significantly different to 0. Differences between manual and continuous measurements of Spo2 were related to the measurement value. However, this relationship was clinically insignificant (Figure 3), with minor differences for low values of Spo2. For example, for Spo2 measurements of 91%, the model predicted a continuous-manual difference of −0.88% (−1.17% to −0.60%), which is within the measurement error of most pulse oximeters.24 The LoAs (95% CI) between continuously and manually recorded vital signs were large: 14.1 (13.8–14.4) breaths/min, 32.5 (31.9–33.3) heartbeats/min, and 6.2% (6.1%–6.3%) Spo2 between LoAs, suggesting that these recordings cannot be used interchangeably. We found no evidence of an arousal effect from the vital sign taker. The bias between continuous vital sign values recorded before and during manual observation was less than a single breath, heartbeat, or percentage Spo2.

% (6.1%–6.3%) Spo2 between LoAs, suggesting that these recordings cannot be used interchangeably. We found no evidence of an arousal effect from the vital sign taker. The bias between continuous vital sign values recorded before and during manual observation was less than a single breath, heartbeat, or percentage Spo2. Our sensitivity analysis showed that window size affected the relationship between continuous-manual differences and averages but not to a clinically meaningful extent (Supplemental Digital Content 1, Table 1, http://links.lww.com/AA/C522). As window size was reduced, lessening the effect of median filtering, the differences predicted by the model increased, although only by 1–2 heartbeats or breaths. Increasing the window size does not affect our findings, suggesting that the choice of a 5-minute window (also chosen by Reich et al13 and Taenzer et al15) appropriately removes artefact without increasing the sample size beyond what is plausible for bedside measurement. Notably, 4 previous studies found a smoothing effect using automated monitoring signals without temporal averaging, relying on manufacturer settings for artefact removal.8,9,11,12 Sapo et al14 demonstrated that Spo2 values <90% are associated with poor signal quality, suggesting that the effects in these studies may be due to clinicians correctly removing spurious values.

sing automated monitoring signals without temporal averaging, relying on manufacturer settings for artefact removal.8,9,11,12 Sapo et al14 demonstrated that Spo2 values <90% are associated with poor signal quality, suggesting that the effects in these studies may be due to clinicians correctly removing spurious values. Further methodological differences may explain why our results contrast with the previous literature suggesting a smoothing effect.8–10,12,13,22 Four studies compared the magnitude and frequency of extreme values in manual and automated vital sign measurements where the automated measurement had higher measurement rates.9,10,12,22 Frequently sampled signals are more likely to capture transient extreme measurements than sparsely sampled signals, partly explaining the discrepancies found in these articles. Furthermore, 1 study compared manual and automated measures from different patient cohorts,13 while others only presented data from automated measurements8 or used simulated measurements from mannequins.22

t extreme measurements than sparsely sampled signals, partly explaining the discrepancies found in these articles. Furthermore, 1 study compared manual and automated measures from different patient cohorts,13 while others only presented data from automated measurements8 or used simulated measurements from mannequins.22 In contrast to our Spo2 findings, Taenzer et al15 reported a difference of 6.5% between continuous and manual measurements of Spo2 <90% in general medical and postoperative wards. The method used continuous data to group Spo2 measurements >90% or <90%, thus comparing the differences between measurement sources to the values of the continuous source. Bland and Altman25 have shown that this process introduces a false correlation to the data. Hence, we compared the differences to the average values as recommended.25,26 For readers interested in this effect, we have replicated our analyses, comparing the difference against the continuous measure (and finding false correlations) in Supplemental Digital Content 2, Figure 1, http://links.lww.com/AA/C523.25

ata. Hence, we compared the differences to the average values as recommended.25,26 For readers interested in this effect, we have replicated our analyses, comparing the difference against the continuous measure (and finding false correlations) in Supplemental Digital Content 2, Figure 1, http://links.lww.com/AA/C523.25 Manually recorded vital sign measurements varied widely from the continuous measurements. Respiratory rate is difficult to measure clinically,2,3 so the high variance in the differences between continuous and manual measurements is perhaps not unexpected. However, as early warning scores commonly include respiratory rate ranges between 2 scores of 4 breaths/min or less,19,27 these differences would commonly impact clinical care. The LoAs for heart rate and Spo2 were also wide and again would commonly result in different early warning scores. These results are important for those seeking to automate early warning scores,28,29 which have not been designed for use with continuous data, so patients would clearly alert differently.

t clinical care. The LoAs for heart rate and Spo2 were also wide and again would commonly result in different early warning scores. These results are important for those seeking to automate early warning scores,28,29 which have not been designed for use with continuous data, so patients would clearly alert differently. Our study is limited by its single-center design because practices may vary between hospital wards and institutions. There may have been transcription errors in the value or timing of manual vital sign observations. This effect was minimized by double data entry and synchronizing observation times with computer-generated timestamps of blood pressure. One bedside monitoring provider was used in this study, so we could not assess differences between monitors. If we had found a relationship between the continuous-manual differences and averages, then this would have prevented exploration of whether the relationship could be explained by monitor inaccuracy, rather than clinician smoothing. Because there is no clinically significant relationship, this is not a significant issue for our findings. Our results are not influenced by undersampling because the measurement pairs of manual and continuous data sample the same physiology. The strengths of our article are the large dataset used in comparison to previous work and the analysis of 3 different vital signs.

p, this is not a significant issue for our findings. Our results are not influenced by undersampling because the measurement pairs of manual and continuous data sample the same physiology. The strengths of our article are the large dataset used in comparison to previous work and the analysis of 3 different vital signs. We have provided evidence against the existence of a smoothing effect in postoperative care. However, this phenomenon may still exist in the context of anesthesia. It is possible that errors in vital sign documentation increase when clinicians record vital signs from memory after the measurement has been taken.9,10 This may be more pervasive in acute episodes during surgery and may not apply to postoperative wards, where nursing staff are available to measure and document vital signs simultaneously. Further research should investigate the smoothing effect during anesthesia, using the Bland–Altman method of assessing agreement between measurement sources, and comparing manually recorded measures to the median of a continuous window. If confirmed, the effects of the large differences between manually recorded and continuous vital sign measurements on early warning scores require investigation before the measurement and recording of these variables can be safely automated for early warning score computation.

res to the median of a continuous window. If confirmed, the effects of the large differences between manually recorded and continuous vital sign measurements on early warning scores require investigation before the measurement and recording of these variables can be safely automated for early warning score computation. CONCLUSIONS We found no evidence that a clinically significant smoothing effect exists for respiratory rate, heart rate, or Spo2 in postoperative care. We found no evidence of an arousal effect caused by the vital sign taker. Differences between manually recorded and continuous measures of respiratory rate, heart rate, and Spo2 were frequently large, suggesting that the methods cannot be used interchangeably. ACKNOWLEDGMENTS The authors thank several individuals for their work on the Computer Alerting Monitoring System 2 study. Breda Lynch, Theresa Saunders, Sarah Vollam, and Deborah Evans conducted data collection; Jacqueline Birks undertook the statistical analysis and assisted in protocol development; and Julie Darbyshire assisted in proof reading. The authors would also like to thank Dr Lei Clifton for her work in preparing the dataset. Furthermore, the study could not have been undertaken without the support of the clinical staff on the upper gastrointestinal ward. The authors would like to thank the participants of the Computer Alerting Monitoring System 2 study for their commitment to research.

Dr Lei Clifton for her work in preparing the dataset. Furthermore, the study could not have been undertaken without the support of the clinical staff on the upper gastrointestinal ward. The authors would like to thank the participants of the Computer Alerting Monitoring System 2 study for their commitment to research. DETAILS OF ACKNOWLEDGED INDIVIDUALS Breda Lynch, Research Nurse, Oxford University Hospitals NHS Trust, Oxford, United Kingdom; Theresa Saunders, Research Nurse, Oxford University Hospitals NHS Trust, Oxford, United Kingdom; Deborah Evans, Research Nurse, Oxford University Hospitals NHS Trust, Oxford, United Kingdom; Sarah Vollam, Research Nurse, Oxford University Hospitals NHS Trust, Oxford, United Kingdom; Jacqueline Birks, MA, MSc, National Institute for Health Research–Oxford Biomedical Research Centre (NIHR–OXBRC) Senior Medical Statistician, Centre for Statistics in Medicine, Nuffield Department of Orthopaedics, Rheumatology and Musculoskeletal Sciences, University of Oxford, Oxford, United Kingdom; Julie Darbyshire, BA, MA, MSc, Critical Care Research Programme Manager, Oxford University Hospitals NHS Trust, Oxford, United Kingdom; and Dr Lei Clifton, BSc, MSc, PhD, Medical Statistician, Centre for Statistics in Medicine, Botnar Research Centre, Nuffield Department of Orthopaedics, Rheumatology and Musculoskeletal Sciences, University of Oxford, Oxford, United Kingdom. DISCLOSURES Name: Hamish R. Tomlinson, BE (Hons). Contribution: This author helped conduct the analysis and prepare the manuscript. Name: Marco A. F. Pimentel, PhD.

DETAILS OF ACKNOWLEDGED INDIVIDUALS Breda Lynch, Research Nurse, Oxford University Hospitals NHS Trust, Oxford, United Kingdom; Theresa Saunders, Research Nurse, Oxford University Hospitals NHS Trust, Oxford, United Kingdom; Deborah Evans, Research Nurse, Oxford University Hospitals NHS Trust, Oxford, United Kingdom; Sarah Vollam, Research Nurse, Oxford University Hospitals NHS Trust, Oxford, United Kingdom; Jacqueline Birks, MA, MSc, National Institute for Health Research–Oxford Biomedical Research Centre (NIHR–OXBRC) Senior Medical Statistician, Centre for Statistics in Medicine, Nuffield Department of Orthopaedics, Rheumatology and Musculoskeletal Sciences, University of Oxford, Oxford, United Kingdom; Julie Darbyshire, BA, MA, MSc, Critical Care Research Programme Manager, Oxford University Hospitals NHS Trust, Oxford, United Kingdom; and Dr Lei Clifton, BSc, MSc, PhD, Medical Statistician, Centre for Statistics in Medicine, Botnar Research Centre, Nuffield Department of Orthopaedics, Rheumatology and Musculoskeletal Sciences, University of Oxford, Oxford, United Kingdom. DISCLOSURES Name: Hamish R. Tomlinson, BE (Hons). Contribution: This author helped conduct the analysis and prepare the manuscript. Name: Marco A. F. Pimentel, PhD. Contribution: This author helped with conception and design of the analysis and drafting the manuscript. Name: Stephen Gerry, MSc. Contribution: This author helped with conception and design of the statistical methodology, and drafting the manuscript. Name: David A. Clifton, PhD.

Contribution: This author helped conduct the analysis and prepare the manuscript. Name: Marco A. F. Pimentel, PhD. Contribution: This author helped with conception and design of the analysis and drafting the manuscript. Name: Stephen Gerry, MSc. Contribution: This author helped with conception and design of the statistical methodology, and drafting the manuscript. Name: David A. Clifton, PhD. Contribution: This author helped with data acquisition and drafting the manuscript. Name: Lionel Tarassenko, PhD. Contribution: This author helped with conception and design of the analysis, data acquisition, and drafting the manuscript. Name: Peter J. Watkinson, MBChB, MD. Contribution: This author helped with conception and design of the analysis, data acquisition, and drafting the manuscript. This manuscript was handled by: Nancy Borkowski, DBA, CPA, FACHE, FHFMA. Supplementary Material Funding: H.R.T. was funded by the Rhodes Trust (Rhodes Scholarship). M.A.F.P. was funded by the Research Councils United Kingdom (RCUK) Digital Economy Programme grant number EP/G036861/1 (Oxford Centre for Doctoral Training in Healthcare Innovation) and Fundação para a Ciência e a Tecnologia under the grant SFRH/BD/79799/2011. S.G. was funded by a National Institute for Health Research (NIHR) Doctoral Fellowship (DRF-2016-09-073). D.A.C. was funded by the Wellcome Trust and Engineering and Physical Sciences Research Council (EPSRC) under grant number WT 088877/Z/09/Z. L.T. and P.J.W. were supported by the NIHR Biomedical Research Centre Programme, Oxford, which funded the Computer Alerting Monitoring System 2 study.

Fellowship (DRF-2016-09-073). D.A.C. was funded by the Wellcome Trust and Engineering and Physical Sciences Research Council (EPSRC) under grant number WT 088877/Z/09/Z. L.T. and P.J.W. were supported by the NIHR Biomedical Research Centre Programme, Oxford, which funded the Computer Alerting Monitoring System 2 study. The authors declare no conflicts of interest. Supplemental digital content is available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal’s website. Clinical Trial Registration: International Standard Randomised Controlled Trial Number (ISRCTN) database—principal investigator: Dr Peter Watkinson; ISRCTN No: ISRCTN58660550, August 11, 2017 (Computer Alerting Monitoring System 2 study). Reprints will not be available from the authors.

Perioperative neurocognitive disorders, now encompassing acute delirium and longer-lasting postoperative cognitive dysfunction, are major challenges to our rapidly growing aging population that negatively affect cognitive domains such as memory, attention, and concentration after surgery.1 Patients who suffer from perioperative neurocognitive disorder are at risk for significant complications including dementia and even death.2,3 Although postoperative delirium has become the most common complication in older adults,4 the pathophysiology of these conditions remains unknown. Growing evidence suggests a possible role for neuroinflammation in this process because proinflammatory signaling molecules have been identified in both patients and animal models of perioperative neurocognitive disorder. The aim of this review is to discuss recent evidence for the involvement of postoperative neuroinflammation in perioperative neurocognitive disorder, and to highlight possible mechanisms of relevance to perioperative neurocognitive disorder from preclinical and early clinical studies.

Perioperative neurocognitive disorders, now encompassing acute delirium and longer-lasting postoperative cognitive dysfunction, are major challenges to our rapidly growing aging population that negatively affect cognitive domains such as memory, attention, and concentration after surgery.1 Patients who suffer from perioperative neurocognitive disorder are at risk for significant complications including dementia and even death.2,3 Although postoperative delirium has become the most common complication in older adults,4 the pathophysiology of these conditions remains unknown. Growing evidence suggests a possible role for neuroinflammation in this process because proinflammatory signaling molecules have been identified in both patients and animal models of perioperative neurocognitive disorder. The aim of this review is to discuss recent evidence for the involvement of postoperative neuroinflammation in perioperative neurocognitive disorder, and to highlight possible mechanisms of relevance to perioperative neurocognitive disorder from preclinical and early clinical studies. With continuous improvements in surgical technology and anesthesia care, increasingly sicker and older patients are exposed to often life-saving procedures. Unfortunately, many of these frail patients are left with postoperative delirium and longer-lasting cognitive decline, especially after cardiac and even noncardiac surgery. The incidence of delirium is estimated at 26%–53% and postoperative cognitive dysfunction at about 10% at 3 months.5–7 Even though perioperative neurocognitive disorder is observed in patients across different age groups and undergoing different surgical procedures, aging and operations such as cardiac and orthopedic surgery have become well-established risks factors for the development of these neurological complications.8 Recent clinical studies have used different approaches to show that both cardiac and noncardiac surgery trigger neuronal injury, which we briefly summarize below.

ng and operations such as cardiac and orthopedic surgery have become well-established risks factors for the development of these neurological complications.8 Recent clinical studies have used different approaches to show that both cardiac and noncardiac surgery trigger neuronal injury, which we briefly summarize below. BIOMARKERS A recent study by Evered et al9 described a significant increase in plasma neurofilament light and tau, 2 key biomarkers classically associated with neuronal injury, as a result of exposure to general anesthesia and surgery. Several investigators have detected postoperative changes in Alzheimer’s disease biomarkers, including β-amyloid protein and intraneuronal neurofibrillary tangles (tau), in cerebrospinal fluid (CSF). Lower CSF β-amyloid protein/tau ratio has been associated with patients who develop perioperative neurocognitive disorder, suggesting a possible trajectory toward dementia after exposure to anesthesia and surgery.10–12 Changes in Alzheimer’s disease markers and astroglial cell integrity, as well as evidence for blood–brain barrier opening were also found in the CSF of patients after hip arthroplasty,13 confirming some of the earlier findings by Tang et al14 for idiopathic nasal CSF leak correction after surgery. Interestingly, although surgery modifies Alzheimer’s disease biomarkers and potentially accelerates their pathogenesis in some individuals, positron emission tomography imaging of β-amyloid protein plaque deposition has shown limited association with cognitive deficits 6 weeks after cardiac surgery.15 CSF and plasma inflammatory biomarkers, cytokines, and many other immune-soluble factors have been described over the past decade in response to different surgeries.16–20 Higher levels of CSF interleukin-6 were found to predict cognitive decline after coronary artery bypass surgery.21 Other proinflammatory markers such as C-reactive protein and interleukin-1β have also been linked to cognitive decline after cardiac procedures.22 Again, these changes in inflammatory markers are not limited to cardiac surgery and some of its unique aspects, for example, extracorporeal circulation and ischemia-reperfusion injury (reviewed in detail in Ref. 23), but are also detected in noncardiac/nonneurological surgery.

nitive decline after cardiac procedures.22 Again, these changes in inflammatory markers are not limited to cardiac surgery and some of its unique aspects, for example, extracorporeal circulation and ischemia-reperfusion injury (reviewed in detail in Ref. 23), but are also detected in noncardiac/nonneurological surgery. Significant amounts of pro- and anti-inflammatory markers are detectable in plasma and CSF of older adults after knee and hip replacement surgery.24,25 Notably, elevation of inflammatory biomarkers has been noted after general anesthesia and spinal anesthesia.18 Indeed, different anesthetics may modulate immune signaling pathways (reviewed in Ref. 26) and perhaps cognitive outcomes.

ectable in plasma and CSF of older adults after knee and hip replacement surgery.24,25 Notably, elevation of inflammatory biomarkers has been noted after general anesthesia and spinal anesthesia.18 Indeed, different anesthetics may modulate immune signaling pathways (reviewed in Ref. 26) and perhaps cognitive outcomes. NEUROIMAGING Other recent studies have used neuroimaging techniques to visualize changes in brain structure and function after surgery. Kant et al27 recently performed a systematic review of structural magnetic resonance imaging data in perioperative neurocognitive disorder and found a consistent association with neurovascular brain changes. The neurovascular unit is critically involved in neurodegenerative diseases and promoting brain health.28 The role of this interface after surgery is the focus of several ongoing preclinical studies. Initial observations using gadolinium-enhanced magnetic resonance imaging show acute (<24 hours) blood–brain barrier disruption in cardiac surgery patients, and this blood–brain barrier opening seems to correlate with subsequent neurological impairments.29,30 Moreover, in critically ill patients with delirium, endothelial dysfunction and impaired microvascular permeability have also been observed using peripheral artery tonometry and more recently by assessing plasma biomarkers such as S100 calcium-binding protein B, plasminogen activator inhibitor-1, and E-selectin.31,32 Our capacity to image neuroinflammation in humans is limited, but second-generation positron emission tomography tracers directed to the translocator protein have been developed. Although the translocator protein is upregulated by microglia after injury, other cell types and brain vessels express great affinity for the protein (given its location on the mitochondrial membrane), and thus, the proxy of the translocator protein for inflammation may be obscured by other factors.33 Using the ligand [11C]PBR28, Forsberg et al34 conducted the first human imaging study to evaluate neuroinflammation during the perioperative period. Although a strong immunosuppressive response was observed acutely after surgery, microglial activation was detected in a subset of patients with cognitive deficits at 3 months. The interplay between peripheral and central inflammation is a major challenge for clinical perioperative research, and more specific markers are needed to better identify immunocompetent cells using positron emission tomography.

glial activation was detected in a subset of patients with cognitive deficits at 3 months. The interplay between peripheral and central inflammation is a major challenge for clinical perioperative research, and more specific markers are needed to better identify immunocompetent cells using positron emission tomography. As clinical research in this domain intensifies, fundamental questions on “how” surgery and anesthesia affect the central nervous system (CNS) warrant detailed evaluation. Establishing and refining clinically relevant surgical models to study perioperative neurocognitive disorder are contributing to a better understanding of the pathogenesis of cognitive decline and more rigorous evaluation of contributing factors such as different anesthetics, surgical procedures, genetic susceptibilities, and comorbidities. Rodents have been the primary source of preclinical data for perioperative neurocognitive disorder, and rats and mice are most commonly used to evaluate inflammatory changes and cognition after surgery. Cardiac models of surgery, recapitulating cardiopulmonary bypass and deep hypothermic circulatory arrest, have been established to assess neurological complications and perioperative inflammation.35 Notably, cardiac surgery triggers widespread changes in cognition that differ from abdominal surgery, while both generate substantial hippocampal neuroinflammation (ie, microglial activation).36 This suggests that distinct pathways may be involved in the response to injury in different organs, and further studies are needed to dissect these pathways. Abdominal and cardiac surgery were also shown to impair neuronal plasticity, as demonstrated by acute changes in hippocampal neurogenesis and brain-derived neurotrophic factor.36 Importantly, these changes outlasted the neuroinflammatory profile, and remained visible for weeks after surgery. Brain-derived neurotrophic factor in particular has been involved in this response and was found to be dysregulated in several other models.37–39 Yet, the role of inflammation in neuronal deficits and cognitive decline remains undefined, and further studies are needed.

ile, and remained visible for weeks after surgery. Brain-derived neurotrophic factor in particular has been involved in this response and was found to be dysregulated in several other models.37–39 Yet, the role of inflammation in neuronal deficits and cognitive decline remains undefined, and further studies are needed. We have pioneered the development of a clinically relevant orthopedic model consisting of an intramedullary fixation of the mouse tibia.40–42 This has been associated with hippocampal neuroinflammation and synaptic dysfunction due to proinflammatory cytokines such as tumor necrosis factor-α, interleukin-1β, and high-mobility group box 1 protein.43 Models of splenectomy,44 hepatectomy,45,46 abdominal,47,48 and vascular surgery49,50 have reported hippocampal neuroinflammation and behavioral deficits. Notably, even minor surgical procedures, such as skin incisions, trigger neuroinflammation in aged animals, but not younger adults.51 Because aging is a critical risk factor for perioperative neurocognitive disorder, it is paramount that future research systematically evaluates advanced age, as well as sex differences and other common susceptibilities, to ensure successful translation of preclinical data to the bedside.

als, but not younger adults.51 Because aging is a critical risk factor for perioperative neurocognitive disorder, it is paramount that future research systematically evaluates advanced age, as well as sex differences and other common susceptibilities, to ensure successful translation of preclinical data to the bedside. It is well appreciated that the nervous and immune systems bidirectionally communicate with apparent implications for both health and disease.52,53 Indeed, the response to peripheral surgery is able to reach the CNS via multiple pathways. Here we focus on the role of (1) systemic inflammation, (2) the neurovascular unit/blood–brain barrier, and (3) neuroinflammation after surgery.

rectionally communicate with apparent implications for both health and disease.52,53 Indeed, the response to peripheral surgery is able to reach the CNS via multiple pathways. Here we focus on the role of (1) systemic inflammation, (2) the neurovascular unit/blood–brain barrier, and (3) neuroinflammation after surgery. SYSTEMIC INFLAMMATORY RESPONSE Systemic inflammation produces physiological and behavioral changes in humans and animals that are characterized by a decline in cognitive function, fever, decreased food intake, somnolence, hyperalgesia, and general fatigue—commonly referred to as “sickness behavior.”54 Sterile inflammation activates similar innate immune pathways to other stressors, such as lipopolysaccharide, by releasing damage-associated molecular patterns, such as high-mobility group box 1 protein, and cytokines.55 These soluble mediators can trigger a systemic inflammatory response via activation of pattern recognition receptors, including toll-like receptors, cytokines such as interleukin-1, interleukin-6, and tumor necrosis factor-α, as well as S100 Ca2+ binding proteins and oxidative stress pathways. We and others have shown a pivotal role for systemic alarmins and cytokines such as interleukin-1β,40 tumor necrosis factor-α,41,56 interleukin-6,57,58 and high-mobility group box 1 protein43,46 in triggering neuroinflammation after peripheral surgery in rodent models. Similar changes in biomarkers have been described in clinical samples and provide critical insights into the temporal profile of different cytokines after surgery. This should be taken into consideration as larger datasets become available for perioperative neurocognitive disorder diagnosis and treatment.

odent models. Similar changes in biomarkers have been described in clinical samples and provide critical insights into the temporal profile of different cytokines after surgery. This should be taken into consideration as larger datasets become available for perioperative neurocognitive disorder diagnosis and treatment. It is important to note that inflammation is overall a protective response to injury, but the improper control of its normal resolution can become harmful and contribute to pathological hallmarks including neuroinflammation.59–61 The impact of systemic inflammation on the brain can be profound. Mounting evidence indicates that blood-borne factors as well as the proinflammatory systemic milieu can negatively impact CNS function, directly affecting synaptic plasticity and cognitive function during normal aging.62,63 Uniform immune signaling responses, including monocyte activation, have also been linked to surgical trauma and may serve as predictors for different recovery profiles in at-risk patients.64 Peripheral cells have become attractive targets for perioperative neurocognitive disorder research because they can be easily accessed in patients and may inform subsequent changes in the CNS without accessing CSF or performing neuroimaging. Monocyte-derived macrophages migrate into the brain parenchyma after surgical trauma, and this plays a role in the pathophysiology of neurological complications including postoperative cognitive decline.43,58,65 In particular, it has been proposed that elevated cerebral monocyte chemoattractant protein-1 contributes to the recruitment of monocytes to the CNS and the ensuing neuroinflammatory response.58,66 Both tumor necrosis factor-α and high-mobility group box 1 protein have been implicated in the regulation of monocyte chemoattractant protein-1 after surgery,41,66 and these may serve as targets for clinical studies. Circulating monocytes, neutrophils, and other peripheral systemic factors can contribute to changes in neuronal function, synaptic plasticity, and glial homeostasis67; however, they are also critically involved in the release of neuroprotective factors, which is crucial in the context of surgical recovery.68 Overall, the contribution of other cellular factors (including T-cells and components of adaptive immunity) to perioperative neurocognitive disorder is largely understudied.

s67; however, they are also critically involved in the release of neuroprotective factors, which is crucial in the context of surgical recovery.68 Overall, the contribution of other cellular factors (including T-cells and components of adaptive immunity) to perioperative neurocognitive disorder is largely understudied. ENDOTHELIAL DYSFUNCTION AND NEUROVASCULAR UNIT/BLOOD–BRAIN BARRIER OPENING Endothelial cells, pericytes, and astrocytic end-feet are core components of the neurovascular unit.69 Together with tight junction and adherent proteins of the endothelial cell layer, they ensure proper barrier formation and protection against potentially harmful peripheral molecules.70 Under pathological conditions, the blood–brain barrier allows extravasation of various immune cells and systemic markers including plasma proteins, prostaglandins, cytokines, and chemokines into brain parenchyma.71 Surgery triggers inflammation and pattern recognition receptors expressed at the blood–brain barrier surface can lead to endothelial inflammation and subsequent neuroinflammation.65,72–75 This disruption is also found in several neurological disorders such as traumatic injury, stroke, and neurodegenerative diseases.76,77 Cytokines and migration of peripheral immunocompetent cells across the blood–brain barrier have been associated with perioperative neurocognitive disorder in animal models.65 After orthopedic surgery, we found opening of the blood–brain barrier with parenchymal fibrinogen deposition in the hippocampus.65 Using Cx3cr1GFP/+Ccr2RFP/+ transgenic mice, we described acute infiltration of monocytes C-C chemokine receptor type 2 into the brain parenchyma via processes partly mediated by tumor necrosis factor-α/nuclear factor kappa-light-chain-enhancer of activated B cells signaling in monocytes.58,65 Macrophage-specific deletion of IκB kinase, a central coordinator of tumor necrosis factor-α activation of nuclear factor kappa-light-chain-enhancer of activated B cells, prevented subsequent infiltration into the hippocampus after surgery. D’Mello et al66 described a similar immune-to-brain communication pathway after hepatic inflammation, and also demonstrated that tumor necrosis factor-α-stimulated microglia produce monocyte chemoattractant protein-1, which subsequently causes monocyte infiltration into the brain.

on into the hippocampus after surgery. D’Mello et al66 described a similar immune-to-brain communication pathway after hepatic inflammation, and also demonstrated that tumor necrosis factor-α-stimulated microglia produce monocyte chemoattractant protein-1, which subsequently causes monocyte infiltration into the brain. Monocyte chemoattractant protein-1 is elevated in the CSF of a limited subset of patients with delirium after orthopedic surgery,24 suggesting that similar mechanisms may occur in humans. Other preclinical surgical models have found similar changes in blood–brain barrier ultrastructure, with infiltration of exogenous tracers into the brain parenchyma, as well as astrocyte pathology.72,78 Cardiac surgery was shown to impair expression of tight junctions in rats.79 Laparotomy, especially in aged mice, triggers changes in several markers including claudins, occludins, and adhesion molecules, leading to blood–brain barrier opening and cognitive decline in a process dependent on interleukin-6 signaling.74 Notably, other studies have shown that administration of interleukin-6 monoclonal antibody and targeting of tumor necrosis factor-α (upstream from interleukin-1 and interleukin-6 signaling) prevent perioperative neurocognitive disorder.41,73 Surgery was shown to upregulate enzymes that break down extracellular matrix, such as matrix metallopeptidase 9, and lead to blood–brain barrier opening and neuroinflammation.75 Importantly, different concentrations of sevoflurane anesthesia differentially regulate matrix metallopeptidase 9 and 2,80 suggesting that anesthesia per se contributes to these changes in the aging brain. Significantly more work is needed to define the role of different anesthetics in blood–brain barrier/neurovascular unit perioperative changes. Many of these pathological features, including blood–brain barrier opening, neurovascular unit dysfunction, and cell infiltration into the CNS, have been implicated in many neurological disorders.81–83 Yet in some cases, the infiltration of blood-derived cells, such as macrophages, is necessary to boost tissue recovery in unique ways that resident microglia, astrocytes, and oligodendrocytes cannot.84,85 Therefore, the role and timing of blood–brain barrier/neurovascular unit opening after surgery require further investigation to possibly develop strategies to effectively limit neuroinflammation in the perioperative period.

e recovery in unique ways that resident microglia, astrocytes, and oligodendrocytes cannot.84,85 Therefore, the role and timing of blood–brain barrier/neurovascular unit opening after surgery require further investigation to possibly develop strategies to effectively limit neuroinflammation in the perioperative period. NEUROINFLAMMATION Neuroinflammation has become a key feature of virtually every neurological complication.86,87 Microglial activation plays a critical role in CNS dyshomeostasis.88 Microglia, the resident immune cells of the CNS, are highly motile cells that continuously survey the brain microenvironment, facilitating synaptic activity, pruning, and remodeling.89,90 Under normal conditions, microglia display highly complex and morphologies with small nuclei and slender processes.91 Upon injury, these cells can shift to a “reactive” phenotype, losing their ramified morphology to become enlarged and stumpy. Activated microglia have been implicated as the primary source of the CNS pro- and anti-inflammatory milieu.92 Activated microglia secrete proinflammatory factors such as cytokines, eicosanoids, complement factors, excitatory amino acids, reactive oxygen radicals, and nitric oxide.93 While dysregulation of these factors can lead to pathology, microglial activation is also responsible for jumpstarting reparative processes and releasing neuroprotective factors. For example, in Alzheimer’s disease, microglia contribute to the clearing of amyloid deposits, and support synaptic remodeling by releasing growth factors.94 However, microglia also contribute to pathological features in the Alzheimer’s disease brain, including hyperphosphorylation of tau and neuronal loss via cytokine release.95 Thus, microglia display both defensive and protective functions, making their role in neurological conditions paradoxical and poorly understood. Microglial activation has been described in several rodent models after peripheral surgery and recently, in human subjects, and is associated with longer-lasting cognitive impairments.34 Conventional histology is still the most common strategy for evaluating microglia in the CNS. Ionized calcium-binding adaptor molecule 1, a protein found on microglia and macrophages,96 is a classic marker for morphological evaluation that has often been used in perioperative neurocognitive disorder models. However, recent advancements in technology are revolutionizing our understanding of neuroinflammation in health and disease.

ding adaptor molecule 1, a protein found on microglia and macrophages,96 is a classic marker for morphological evaluation that has often been used in perioperative neurocognitive disorder models. However, recent advancements in technology are revolutionizing our understanding of neuroinflammation in health and disease. Cell sequencing and multiomics approaches are now revealing unique phenotypes and functions of these cells during normal aging and neurodegeneration that go beyond morphological changes and immunostaining.97–99 Evaluation of microglia across discrete brain regions in mice has revealed selective regional sensitivity to neuroinflammation.100 Further, microglia isolated from different brain regions respond differently to challenges, possibly the reason that neurological disorders affect specific areas and cell populations.101,102 To date, a large number of studies interrogating perioperative neurocognitive disorder in rodent models have focused on the hippocampus, given its essential role in learning and memory. Within this larger framework, studies that evaluate multiple brain regions may reveal different response profiles for microglia, and possibly other cell types, related to surgery and anesthesia. Other technologies have been implemented with direct implications to neuroimmunology. Tissue clarification allows investigators to evaluate complex structures in intact specimens, including the brain.103,104 CLARITY is offering novel insights into 3D imaging.105–107 This pioneering technique preserves cellular integrity while rendering tissues visually transparent for deeper optical imaging. Our own work using CLARITY is providing ways to evaluate changes in microglial morphology after surgery and to further evaluate their relationship with other cell types including endothelial cells, neurons, and astrocytes. Indeed, microglia can induce astrocyte activation, which leads to neuronal death and toxicity.108 Astrocytes are also activated postoperatively in perioperative neurocognitive disorder models of major surgery (eg, liver surgery).109 Tibial fracture induces morphological and functional changes in astrocytes110,111 that contribute to the disruption of neuroglial metabolic coupling and subsequent neuronal dysfunction.

Astrocytes are also activated postoperatively in perioperative neurocognitive disorder models of major surgery (eg, liver surgery).109 Tibial fracture induces morphological and functional changes in astrocytes110,111 that contribute to the disruption of neuroglial metabolic coupling and subsequent neuronal dysfunction. Complement activation, in particular C3 and C3R, is increased in microglia and astrocytes after orthopedic surgery thereby contributing to synaptic loss and hippocampal inflammation.112 Similar mechanisms involved in synaptic pruning during development were described by Stevens et al,113 and were found to be hijacked in Alzheimer’s disease.114 Defining the role of complement signaling and mechanisms of communication between glia and neurons in perioperative neurocognitive disorder will require extensive studies.

hanisms involved in synaptic pruning during development were described by Stevens et al,113 and were found to be hijacked in Alzheimer’s disease.114 Defining the role of complement signaling and mechanisms of communication between glia and neurons in perioperative neurocognitive disorder will require extensive studies. The role of inflammation in perioperative brain function is becoming apparent (Figure). Although this is a necessary response to tissue trauma, defective resolution and nonresolving inflammation are now appreciated as key contributors to chronic and maladaptive states.61 In perioperative neurocognitive disorder, we are beginning to understand that “fine-tuning” of immune signaling may be a way forward to limit secondary CNS damage. Broad approaches that block inflammation, for example, treating with dexamethasone or statins, yield limited results in clinical trials.115,116 Harnessing endogenous pathways and mediators, such as cholinergic signaling and lipids biosynthesized from omega-3 fatty acids, may provide unique opportunities to curtail inflammation after surgery without causing unwanted side effects.117 In particular, omega-3 fatty acids are important catalysts in the synthesis of potent specialized proresolving mediators,61 which can exert proresolving and anti-inflammatory actions after surgery and several other conditions (reviewed in Ref. 118).110,119 Alternative approaches to regulate immunity at a neuronal level are also under development. The cholinergic anti-inflammatory reflex is one of the exemplary circuits that can regulate inflammation by stimulating the vagus nerve.53,120,121 The establishment of bioelectronic approaches is already able to reduce inflammation in rheumatoid arthritis patients,122 and may be effective in treating neuroinflammation although further research is needed.

flex is one of the exemplary circuits that can regulate inflammation by stimulating the vagus nerve.53,120,121 The establishment of bioelectronic approaches is already able to reduce inflammation in rheumatoid arthritis patients,122 and may be effective in treating neuroinflammation although further research is needed. Figure. Schematic overview of mechanisms involved in postoperative neuroinflammation. Surgery activates the innate immune system resulting in release of proinflammatory mediators (cytokines, chemokines, alarmins, eicosanoids, etc) and activation of systemic immunocompetent cells. These processes negatively affect the blood–brain barrier, resulting in endothelial dysfunction and infiltration of peripheral cells/factors into the brain parenchyma. Within the central nervous system, astrocytes and microglia shift from their resting state and contribute to overall neuronal dysfunction and memory function. Harnessing resolution programs as early as the inflammatory response begins may translate into neuroprotective strategies for PND. An example of microglial activation after surgery using CLARITY is provided on the right panel. Scale bar: 150 µm. BDNF indicates brain-derived neurotrophic factor; Ca+, calcium; DAMP, damage-associated molecular pattern; HMGB1, high-mobility group box 1 protein; IL, interleukin; Mω, macrophage; MCP, monocyte chemoattractant protein; PND, perioperative neurocognitive disorder; TNF, tumor necrosis factor.

t panel. Scale bar: 150 µm. BDNF indicates brain-derived neurotrophic factor; Ca+, calcium; DAMP, damage-associated molecular pattern; HMGB1, high-mobility group box 1 protein; IL, interleukin; Mω, macrophage; MCP, monocyte chemoattractant protein; PND, perioperative neurocognitive disorder; TNF, tumor necrosis factor. The human brain is the source of all human thought, and also the target of many neurological disorders. These disorders can cause disturbances in behavior, cognition, and emotions that can sometimes even interfere with the essence of who we are as human beings. We have an opportunity to protect our brain, at least within the perioperative space, and preserve its fundamental functions including our capacity to reason. René Descartes’ classic line from 1637 could not be more relevant for today’s research in neuroprotection: Cogito, ergo sum. ACKNOWLEDGMENTS We thank Kathy Gage, BS (Department of Anesthesiology, Duke University Medical Center, Durham, NC) for editorial assistance. We apologize to all colleagues whose work was not cited owing to space constraints. DISCLOSURES Name: Saraswathi Subramaniyan, PhD. Contribution: This author helped write the manuscript. Name: Niccolò Terrando, PhD. Contribution: This author helped write and edit the manuscript. This manuscript was handled by: Gregory J. Crosby, MD. Published ahead of print 31 December 2018. Funding: N.T. is supported by the National Institutes of Health R01 AG057525, R21 AG055877-01A1, the Duke Institute for Brain Sciences Incubator Award, and Duke Anesthesiology. The authors declare no conflicts of interest.

Contribution: This author helped write and edit the manuscript. This manuscript was handled by: Gregory J. Crosby, MD. Published ahead of print 31 December 2018. Funding: N.T. is supported by the National Institutes of Health R01 AG057525, R21 AG055877-01A1, the Duke Institute for Brain Sciences Incubator Award, and Duke Anesthesiology. The authors declare no conflicts of interest. Reprints will not be available from the authors.

Severe sepsis and septic shock are major causes of death in critically ill patients,1 with a mortality rate of 14% to 40%.2 Increased compliance with sepsis performance bundles is associated with a reduction in mortality.3,4 As fundamental principles for sepsis management, early recognition, control of infection, early and appropriate administration of antibiotics, and resuscitation with IV fluids and vasoactive drugs are widely accepted by intensivists.5 Early adequate hemodynamic resuscitation is emphasized, as the key elements should be focused on saving lives. A proof-of-concept trial6 found that, when compared with a control protocol, early hemodynamic resuscitation with a specific protocol termed early goal-directed therapy (EGDT) improved outcomes in patients with severe sepsis. Several additional studies using a similar protocol involving central venous pressure, mean arterial pressure, and central venous oxygen saturation (Scvo2) to guide hemodynamic resuscitation also found a survival benefit with EGDT.7–9 As a result, EGDT principles are subsequently incorporated into the early hemodynamic resuscitation bundle of the Surviving Sepsis Campaign guidelines.5 However, in 3 multicenter randomized trials10–12 (Protocolized Care for Early Septic Shock [ProCESS] study, the Australasian Resuscitation in Sepsis Evaluation [ARISE] study, and Protocolized Management in Sepsis [ProMISe] trial) published recently, EGDT did not decrease mortality in severe sepsis and septic shock when compared with control care.

ndomized trials10–12 (Protocolized Care for Early Septic Shock [ProCESS] study, the Australasian Resuscitation in Sepsis Evaluation [ARISE] study, and Protocolized Management in Sepsis [ProMISe] trial) published recently, EGDT did not decrease mortality in severe sepsis and septic shock when compared with control care. The controversial question of whether EGDT improves outcome of severe sepsis and septic shock thus remains relevant. Moreover, whether specific patient characteristics are associated with a potential benefit of EGDT is also unclear. Finally, whether all elements of the protocol are necessary in the hemodynamic resuscitation of severe sepsis and septic shock is unknown. Our goal was to perform a meta-analysis to examine whether EGDT improved outcome when employed in the resuscitation of adult patients with severe sepsis and septic shock compared with control therapy. METHODS Approval Our IRB does not require ethics approval for systematic reviews, including network meta-analyses, because there are no data being collected from patients. We evaluated and synthesized only data in published trials.

The controversial question of whether EGDT improves outcome of severe sepsis and septic shock thus remains relevant. Moreover, whether specific patient characteristics are associated with a potential benefit of EGDT is also unclear. Finally, whether all elements of the protocol are necessary in the hemodynamic resuscitation of severe sepsis and septic shock is unknown. Our goal was to perform a meta-analysis to examine whether EGDT improved outcome when employed in the resuscitation of adult patients with severe sepsis and septic shock compared with control therapy. METHODS Approval Our IRB does not require ethics approval for systematic reviews, including network meta-analyses, because there are no data being collected from patients. We evaluated and synthesized only data in published trials. Search Strategy for Identification of Relevant Studies We searched the following databases: Medline, Elsevier, Cochrane Central Register of Controlled Trials, and Web of Science databases. The following keywords were used as searching terms: “goal-directed” or “goal-directed resuscitation” or “early goal-directed therapy” or “EGDT” or “bundle” or “sepsis bundle” and “sepsis” or “severe sepsis” or “septic shock” or “shock” or “critical ill” or “critical illness” or “intensive care units” or “intensive care” or “critical care” or “ICU.” No language restrictions were placed on the search. All databases were searched for articles published from inception until March 17, 2015. Additional files and supplementary appendices of the relevant articles were also reviewed. Detailed search strategies are shown in Supplemental Digital Content 1 (http://links.lww.com/AA/B409).

estrictions were placed on the search. All databases were searched for articles published from inception until March 17, 2015. Additional files and supplementary appendices of the relevant articles were also reviewed. Detailed search strategies are shown in Supplemental Digital Content 1 (http://links.lww.com/AA/B409). Study Selection One reviewer screened the search results, and the full-text manuscripts of all potentially eligible studies were acquired. All the articles were then reviewed by 2 reviewers independently in accordance with the inclusion and exclusion criteria. Twelve disagreements between the 2 reviewers were resolved by consensus and discussion including a third reviewer. Any inconsistency in study inclusion and exclusion and their reason lead to the discussion. Details of the consensus are shown in Supplemental Digital Content 2 (http://links.lww.com/AA/B410). Inclusion and Exclusion Criteria We included trials with the following features: Type of trials: randomized controlled clinical trials. Population: trials including adult population with severe sepsis and septic shock. Severe sepsis was defined as sepsis plus sepsis-induced organ dysfunction or tissue hypoperfusion.5 Intervention: patients submitted to EGDT, which used the protocol involving central venous pressure, mean arterial pressure, and Scvo2 to guide hemodynamic resuscitation. Comparison: control care, including usual care, protocol-based therapy, etc. Outcome: the primary outcome was all-cause mortality, including 28-day mortality, 90-day mortality, or mortality at other time points.

Intervention: patients submitted to EGDT, which used the protocol involving central venous pressure, mean arterial pressure, and Scvo2 to guide hemodynamic resuscitation. Comparison: control care, including usual care, protocol-based therapy, etc. Outcome: the primary outcome was all-cause mortality, including 28-day mortality, 90-day mortality, or mortality at other time points. Trials were excluded because of following reason: If they were not published in English or Chinese. If they were not published as original articles. If they did not enroll adult patients. If they did not compare EGDT with control care. If they included no data on mortality in patients with severe sepsis and septic shock. If full-text articles were not available. Quality Assessment The quality of each article was assessed by 2 reviewers independently. Disagreement that occurred once was resolved by consulting a third reviewer. The 5-point Jadad scale13 was calculated to assess the quality of the trial. This scale includes the method of randomization, blinding, and loss to follow-up. In addition, sequences generation, allocation concealment, incomplete outcome data, selective reporting, and other bias were also inspected to assess the risk of bias. The latter was reported as low risk, unclear risk, or high risk for each trial. Low risk was defined as low risk of bias in all domains. Unclear risk was defined as unclear risk of bias in at least 1 domain with no high risk of bias domains. High risk was defined as high risk of bias in 1 or more domains.

risk of bias. The latter was reported as low risk, unclear risk, or high risk for each trial. Low risk was defined as low risk of bias in all domains. Unclear risk was defined as unclear risk of bias in at least 1 domain with no high risk of bias domains. High risk was defined as high risk of bias in 1 or more domains. Data Extraction and Management Using a data extraction table, 2 reviewers independently extracted data. Disagreements that occurred twice were resolved by discussion with another reviewer until a consensus was achieved. Then, data were proofread by another reviewer.

risk of bias. The latter was reported as low risk, unclear risk, or high risk for each trial. Low risk was defined as low risk of bias in all domains. Unclear risk was defined as unclear risk of bias in at least 1 domain with no high risk of bias domains. High risk was defined as high risk of bias in 1 or more domains. Data Extraction and Management Using a data extraction table, 2 reviewers independently extracted data. Disagreements that occurred twice were resolved by discussion with another reviewer until a consensus was achieved. Then, data were proofread by another reviewer. Mortality data were recorded as primary predefined outcome parameters during the data extraction. When >1 value for mortality was provided by the article, the mortality for the longest complete follow-up was preferentially used in the meta-analysis. If no 28-day or 90-day mortality values were presented, intensive care unit (ICU) or hospital mortality, or mortality at other time points, was recorded. Ninety-day mortality was reported by the ProCESS study,10 the ARISE study,11 and the ProMISe study.12 Sixty-day mortality was reported by the study of Rivers et al.6 Twenty-eight-day mortality was reported by the study of Yan et al.14 Fourteen-day mortality was reported by the study of Wang et al.15 In-hospital mortality was reported by the study of Rivers et al,6 the ProCESS study,10 the ARISE study,11 the ProMISe study,12 the study of Jones et al,16 and the study of Lu et al.17 ICU mortality was reported by the ARISE study,11 the study of Yan et al,14 the study of Wang et al,15 and the study of Chen et al.18

.15 In-hospital mortality was reported by the study of Rivers et al,6 the ProCESS study,10 the ARISE study,11 the ProMISe study,12 the study of Jones et al,16 and the study of Lu et al.17 ICU mortality was reported by the ARISE study,11 the study of Yan et al,14 the study of Wang et al,15 and the study of Chen et al.18 Other data, including ICU length of stay, study characteristics, inclusion and exclusion criteria, sample size of the trial, resuscitation end points, and detailed information, were extracted as secondary predefined outcome parameters. If there was insufficient information in the publications, the authors were contacted to obtain missing information. The effects of EGDT on all-cause mortality, hospital mortality, ICU mortality, and ICU length of stay in severe sepsis and septic shock patients were observed in the meta-analysis. Then effects of EGDT on mortality in severe sepsis and septic shock patients with different severity of illness and with/without venous oxygen saturation were evaluated.

se mortality, hospital mortality, ICU mortality, and ICU length of stay in severe sepsis and septic shock patients were observed in the meta-analysis. Then effects of EGDT on mortality in severe sepsis and septic shock patients with different severity of illness and with/without venous oxygen saturation were evaluated. Statistical Analysis Data were analyzed by Review Manager 4.2 (The Nordic Cochrane Center, Rigshospitalet, Copenhagen, Denmark). The relative risk for dichotomous data and mean differences for continuous data with 99% confidence intervals (CIs) were calculated. The statistical heterogeneity of the data was explored and quantified using the I2 test. Heterogeneity was predefined as P < 0.05. I2value of 0% to 24.9%, 25% to 49.9%, 50% to 74.9%, and 75% to 100% were considered as none, low, moderate, and high thresholds, respectively.19,20 The randomized-effects model was used if heterogeneity was observed21; otherwise, the fixed-effects model was used. To explore the significant heterogeneity, sensitivity analyses were performed. Results were considered statistically significant at 2-sided P < 0.01. Most continuous data were displayed by mean ± SD. However, the ICU length of stay was reported by median and interquartile ranges in ARISE11 and ProMISe trial,12 with large sample sizes. As a result, we used median instead of mean, and SD calculated by interquartile range divided by 1.35.

Statistical Analysis Data were analyzed by Review Manager 4.2 (The Nordic Cochrane Center, Rigshospitalet, Copenhagen, Denmark). The relative risk for dichotomous data and mean differences for continuous data with 99% confidence intervals (CIs) were calculated. The statistical heterogeneity of the data was explored and quantified using the I2 test. Heterogeneity was predefined as P < 0.05. I2value of 0% to 24.9%, 25% to 49.9%, 50% to 74.9%, and 75% to 100% were considered as none, low, moderate, and high thresholds, respectively.19,20 The randomized-effects model was used if heterogeneity was observed21; otherwise, the fixed-effects model was used. To explore the significant heterogeneity, sensitivity analyses were performed. Results were considered statistically significant at 2-sided P < 0.01. Most continuous data were displayed by mean ± SD. However, the ICU length of stay was reported by median and interquartile ranges in ARISE11 and ProMISe trial,12 with large sample sizes. As a result, we used median instead of mean, and SD calculated by interquartile range divided by 1.35. Predefined subgroup analysis was conducted comparing EGDT protocols with and without Scvo2. In addition, we performed post hoc analyses according to setting (emergency department versus ICU), timing (within 6 hours versus unclear), and fluids (fluid resuscitation volume in 6 hours >4 L versus fluid volume in 6 hours <4 L).

efined subgroup analysis was conducted comparing EGDT protocols with and without Scvo2. In addition, we performed post hoc analyses according to setting (emergency department versus ICU), timing (within 6 hours versus unclear), and fluids (fluid resuscitation volume in 6 hours >4 L versus fluid volume in 6 hours <4 L). With the type I errors resulting from an increased risk of error and repeated significance testing, trial sequential analysis (TSA; TSA software version 0.9 Beta; Copenhagen Trial Unit, Copenhagen, Denmark) was performed to combine information size estimation with an adjusted threshold for statistical significance in the cumulative meta-analysis. Information size was calculated as diversity-adjusted information size, suggested by the relative risk reduction of the intervention in the included studies. RESULTS Study Location and Selection Figure 1. Flow diagram of the search process and study selection. A total of 1135 records were identified through the initial search, and 149 were removed as duplicates. The remainder of the 986 records was screened. After assessment of the titles and abstracts, 943 articles were excluded as irrelevant; 2 full-text articles were unavailable. The flow diagram is presented in Figure 1. In total, 41 potentially eligible studies were identified and 32 of these were excluded, leaving 9 studies6,10–12,14–18 that met inclusion criteria and compared EGDT with control care. Detailed excluded articles are listed in Supplemental Digital Content 2 (http://links.lww.com/AA/B410).

agram is presented in Figure 1. In total, 41 potentially eligible studies were identified and 32 of these were excluded, leaving 9 studies6,10–12,14–18 that met inclusion criteria and compared EGDT with control care. Detailed excluded articles are listed in Supplemental Digital Content 2 (http://links.lww.com/AA/B410). Study Characteristics The characteristics of the included studies are shown in Supplemental Digital Content 3 (http://links.lww.com/AA/B411). The meta-analysis included 5202 severe sepsis and septic shock patients, 2382 in the EGDT group and 2820 in the control group. Among the 9 trials, 510–12,14,16 were multiple-center studies, and the remaining 46,15,17,18 were single-center studies. Four trials14,15,17,18 were conducted in the ICU, and the remaining 56,10–12,16 were conducted in the emergency department. Seven trials6,10–12,14,17,18 enrolled and resuscitated patients within 6 hours, and 2 trials15,16 reported unclear timing of enrollment. The severity of disease was reported in each study by the acute physiology and chronic health evaluation II score (APACHE II score), and data were presented as mean ± SD or medians (interquartile ranges). The distinction between higher severity of disease and less severity was differentiated by us according to the means of APACHE II scores in each article. The APACHE II scores of the included patients in 5 trials6,10,14,15,17 were >20; 311,12,18 of the remaining were <20 or unclear. The fluid volume for resuscitation of the included patients in 2 trials6,15 was >4 L, in 5 trials10–12,16,17 was <4 L, and in the other 2 trials14,18 was unclear.

E II scores in each article. The APACHE II scores of the included patients in 5 trials6,10,14,15,17 were >20; 311,12,18 of the remaining were <20 or unclear. The fluid volume for resuscitation of the included patients in 2 trials6,15 was >4 L, in 5 trials10–12,16,17 was <4 L, and in the other 2 trials14,18 was unclear. Details of intervention used in included studies are shown in Table 1. Intervention goals of the EGDT group included CVP ≥8 to 12 mm Hg, MAP ≥65 mm Hg, and Scvo2 ≥70% in all included studies. Four trials6,14,16,18 used the protocol, including identical goal of CVP and MAP but without Scvo2 as control care, the other 5 trials10–12,15,17 used control care with different protocols. Table 1. Details of Interventions Used in Included Studies Table 2. Quality Assessment of the Included Studies Quality assessment of the included studies is shown in Table 2. Three trials6,10,16 were judged to be at low risk of bias or unclear risk of bias, and the remaining 6 trials11,12,14,15,17,18 were judged to be at high risk of bias. None of the 9 studies was double-blinded because of the extreme difficulty in blinding required to evaluate the complex intervention such as EGDT. However, we judged that mortality and ICU length of stay were not likely to be influenced by lack of blinding.

rials11,12,14,15,17,18 were judged to be at high risk of bias. None of the 9 studies was double-blinded because of the extreme difficulty in blinding required to evaluate the complex intervention such as EGDT. However, we judged that mortality and ICU length of stay were not likely to be influenced by lack of blinding. The Effects of EGDT on Mortality in Severe Sepsis and Septic Shock Patients Figure 2. The effects of early goal-directed therapy (EGDT) on mortality in patients with severe sepsis and septic shock. Mortality data were recorded during the data extraction. When there is >1 value about mortality published in the article, the longest complete follow-up was preferentially used for evaluation of all-cause mortality. However, when 28-day or 90-day mortality values were not presented, intensive care unit (ICU) or hospital mortality or mortality at other time points were recorded. Ninety-day mortality was reported by the ARISE study,11 the ProCESS study,10 and the ProMISe study.12 Sixty-day mortality was reported by the study of Rivers et al.6 Twenty-eight-day mortality was reported by the Yan study.14 Fourteen-day mortality was reported by the study of Wang et al.15 In-hospital mortality was reported by the study of Rivers et al,6 the study of Jones et al,16 Lu study,17 ARISE study,11 ProCESS study,10 and ProMISe study.12 ICU mortality was reported by the study of Chen et al,18 the study of Wang,15 Yan study,14 and ARISE study.12 CI indicates confidence interval; RR, relative risk.

n-hospital mortality was reported by the study of Rivers et al,6 the study of Jones et al,16 Lu study,17 ARISE study,11 ProCESS study,10 and ProMISe study.12 ICU mortality was reported by the study of Chen et al,18 the study of Wang,15 Yan study,14 and ARISE study.12 CI indicates confidence interval; RR, relative risk. The effects of EGDT on mortality in patients with severe sepsis were estimated from 9 trials (Figure 2), and the heterogeneity was observed (P = 0.04, I2 = 51.6%). The longest complete follow-up mortality rates of all the trials were evaluated in the analysis of all-cause mortality. The overall mortality in the EGDT and control group was 636 of 2382 (26.7%) and 831 of 2820 (29.5%), respectively. No significantly reduced all-cause mortality was observed in EGDT group compared with control care (relative risk [RR], 0.89; 99% CI, 0.74–1.07; P = 0.10). The Effects of EGDT on Hospital Mortality in Severe Sepsis and Septic Shock Patients Six trials6,10–12,16,17 reported hospital mortality in patients with severe sepsis and septic shock. Hospital mortality was not significantly different between EGDT and control care (RR, 0.98; 99% CI, 0.78–1.24; P = 0.86). The Effects of EGDT on ICU Mortality in Severe Sepsis and Septic Shock Patients Four trials11,14,15,18 reported ICU mortality in patients with severe sepsis and septic shock. EGDT significantly reduced ICU mortality in severe sepsis and septic shock patients (RR, 0.72; 99% CI, 0.57–0.90; P = 0.0002).

The Effects of EGDT on Hospital Mortality in Severe Sepsis and Septic Shock Patients Six trials6,10–12,16,17 reported hospital mortality in patients with severe sepsis and septic shock. Hospital mortality was not significantly different between EGDT and control care (RR, 0.98; 99% CI, 0.78–1.24; P = 0.86). The Effects of EGDT on ICU Mortality in Severe Sepsis and Septic Shock Patients Four trials11,14,15,18 reported ICU mortality in patients with severe sepsis and septic shock. EGDT significantly reduced ICU mortality in severe sepsis and septic shock patients (RR, 0.72; 99% CI, 0.57–0.90; P = 0.0002). The Effects of EGDT on Mortality in Severe Sepsis and Septic Shock Patients With Different Severity of Illness Figure 3. The effects of early goal-directed therapy (EGDT) on mortality in severe sepsis and septic shock patients with different severity of illness. The severity of illness was reported in each study by the acute physiology and chronic health evaluation II score (APACHE II score), and the data were presented as mean ± SD or medians (interquartile ranges). The distinction between higher severity of illness and less severity was differentiated by us according to the means of APACHE II scores in each article. The APACHE II score of the included patients in 5 trials6,10,14,15,17 were >20, the remaining 311,12,18 were <20 or unclear. CI indicates confidence interval; RR, relative risk. For patients with a higher severity of disease (APACHE II score ≥ 20), mortality benefit trended toward EGDT (RR, 0.87; 99% CI, 0.74–1.03; P = 0.03) when compared with control care (Figure 3).

The Effects of EGDT on Mortality in Severe Sepsis and Septic Shock Patients With Different Severity of Illness Figure 3. The effects of early goal-directed therapy (EGDT) on mortality in severe sepsis and septic shock patients with different severity of illness. The severity of illness was reported in each study by the acute physiology and chronic health evaluation II score (APACHE II score), and the data were presented as mean ± SD or medians (interquartile ranges). The distinction between higher severity of illness and less severity was differentiated by us according to the means of APACHE II scores in each article. The APACHE II score of the included patients in 5 trials6,10,14,15,17 were >20, the remaining 311,12,18 were <20 or unclear. CI indicates confidence interval; RR, relative risk. For patients with a higher severity of disease (APACHE II score ≥ 20), mortality benefit trended toward EGDT (RR, 0.87; 99% CI, 0.74–1.03; P = 0.03) when compared with control care (Figure 3). The Effects of Goal-Directed Therapy on Mortality in Severe Sepsis and Septic Shock Patients With and Without Central Venous Oxygen Saturation Figure 4. The effects of goal-directed therapy on mortality in severe sepsis and septic shock patients with and without central venous oxygen saturation. Mortality data were recorded during the data extraction. When there is >1 value about mortality published in the article, the longest complete follow-up was preferentially used for evaluation of all-cause mortality. However, when 28-day or 90-day mortality values were not presented, intensive care unit (ICU) or hospital mortality or mortality at other time points were recorded. Ninety-day mortality was reported by the ARISE study,11 the ProCESS study,10 and the ProMISe study.12 Sixty-day mortality was reported by the study of Rivers et al.6 Twenty-eight-day mortality was reported by Yan study.14 Fourteen-day mortality was reported by the study of Wang et al.15 In-hospital mortality was reported by the study of Rivers et al,6 the study of Jones et al,16 Lu study,17 ARISE study,11 ProCESS study,10 and ProMISe study.12 ICU mortality was reported by the study of Chen et al,18 study of Wang et al,15 Yan study,14 and ARISE study.12 CI indicates confidence interval; RR, relative risk.

hospital mortality was reported by the study of Rivers et al,6 the study of Jones et al,16 Lu study,17 ARISE study,11 ProCESS study,10 and ProMISe study.12 ICU mortality was reported by the study of Chen et al,18 study of Wang et al,15 Yan study,14 and ARISE study.12 CI indicates confidence interval; RR, relative risk. Predefined subgroup analysis according to protocol with versus without Scvo2 suggested no significant difference of mortality between the 2 protocols; however, mortality benefit trended toward Scvo2 monitoring (RR, 0.88; 99% CI, 0.73–1.06; P = 0.07) when compared with protocol including identical remaining intervention goals (Figure 4). Post Hoc Subgroup Analyses for Overall Mortality Table 3. Subgroup Analyses for Overall Mortality Table 4. Sensitivity Analyses of the Studies

Predefined subgroup analysis according to protocol with versus without Scvo2 suggested no significant difference of mortality between the 2 protocols; however, mortality benefit trended toward Scvo2 monitoring (RR, 0.88; 99% CI, 0.73–1.06; P = 0.07) when compared with protocol including identical remaining intervention goals (Figure 4). Post Hoc Subgroup Analyses for Overall Mortality Table 3. Subgroup Analyses for Overall Mortality Table 4. Sensitivity Analyses of the Studies Post hoc subgroup analyses (Table 3) according to the setting of EGDT for resuscitation suggested that mortality benefit was only seen in the subgroup when EGDT was conducted in the ICU (RR, 0.67; 99% CI, 0.51–0.88; P = 0.0002) but not in the subgroup conducted in the emergency department. Moreover, although there was a lack of statistical significance, more trending toward reduced mortality was found when EGDT was performed within 6 hours than when timing of EGDT was unclear (RR, 0.87; 99% CI, 0.72–1.05; P = 0.05) or when there was EGDT with more fluid volume in patients with severe sepsis and septic shock (RR, 0.77; 99% CI, 0.55–1.06; P = 0.04). Sensitivity analyses for the effects of EGDT on mortality in severe sepsis and septic shock patients with higher severity of illness and timing were performed (Table 4). The Effects of EGDT on ICU Length of Stay in Severe Sepsis and Septic Shock Patients Figure 5. The effects of early goal-directed therapy (EGDT) on intensive care unit (ICU) length of stay in severe sepsis and septic shock patients. CI indicates confidence interval.

Post hoc subgroup analyses (Table 3) according to the setting of EGDT for resuscitation suggested that mortality benefit was only seen in the subgroup when EGDT was conducted in the ICU (RR, 0.67; 99% CI, 0.51–0.88; P = 0.0002) but not in the subgroup conducted in the emergency department. Moreover, although there was a lack of statistical significance, more trending toward reduced mortality was found when EGDT was performed within 6 hours than when timing of EGDT was unclear (RR, 0.87; 99% CI, 0.72–1.05; P = 0.05) or when there was EGDT with more fluid volume in patients with severe sepsis and septic shock (RR, 0.77; 99% CI, 0.55–1.06; P = 0.04). Sensitivity analyses for the effects of EGDT on mortality in severe sepsis and septic shock patients with higher severity of illness and timing were performed (Table 4). The Effects of EGDT on ICU Length of Stay in Severe Sepsis and Septic Shock Patients Figure 5. The effects of early goal-directed therapy (EGDT) on intensive care unit (ICU) length of stay in severe sepsis and septic shock patients. CI indicates confidence interval. The ICU length of stay was also evaluated (Figure 5). Five studies10,14–17 reported ICU length of stay by mean ± SD. However, the ICU length of stay was reported by median and interquartile ranges in the ARISE11 and the ProMISe trial12 with a large sample size. As a result, we used median instead of mean, and SD was calculated by interquartile range divided by 1.35. Compared with control care, EGDT displayed no beneficial effect on ICU length of stay in severe sepsis and septic shock patients.

terquartile ranges in the ARISE11 and the ProMISe trial12 with a large sample size. As a result, we used median instead of mean, and SD was calculated by interquartile range divided by 1.35. Compared with control care, EGDT displayed no beneficial effect on ICU length of stay in severe sepsis and septic shock patients. Random Errors Figure 6. Trial sequential analysis for a relative risk reduction of all-cause mortality of 9.5% in control group in 9 trials. A required diversity-adjusted information size of 21,342 patients was calculated based on a control event proportion of 9.5%, early goal-directed therapy (EGDT) induced relative risk reduction of mortality of 9.5% suggested by all trials, α = 0.05 two-sided, β = 0.20 (power = 80%). The cumulated Z-curve (blue) crosses the traditional boundary (P = 0.05) but not the trial sequential monitoring boundary, indicating lack of firm evidence for a beneficial effect of EGDT. To correct for random error and repetitive testing of sparse data, TSA was calculated with α = 0.05 and β = 0.20 (power 80%). The required diversity-adjusted information size based on the intervention effect was suggested by the included trials using a random-effects model (with the relative risk reduction of 9.5% regarding mortality and 21,342 patients). TSA indicated lack of reliable and conclusive evidence for a beneficial effect of EGDT for the longest complete follow-up mortality in severe sepsis and septic shock patients (Figure 6) because the monitoring boundary was not surpassed and the required information size was not reached.

g mortality and 21,342 patients). TSA indicated lack of reliable and conclusive evidence for a beneficial effect of EGDT for the longest complete follow-up mortality in severe sepsis and septic shock patients (Figure 6) because the monitoring boundary was not surpassed and the required information size was not reached. DISCUSSION Our meta-analysis showed no significantly reduced all-cause mortality in patients resuscitated with EGDT when compared with control therapy. This meta-analysis differs considerably from those of recently published meta-analyses, but we obtained similar results.22–25 In our meta-analysis, TSA indicated lack of firm evidence for our results because of considerable heterogeneity between groups. Our results suggest that although 3 recent multicenter randomized controlled studies reported negative results, conclusive evidence regarding the benefit of EGDT is not possible, and more randomized controlled trials are needed. Our results, however, suggest that some patient subgroups may benefit from EGDT. In our meta-analysis, for example, we found a trend toward the longest complete follow-up mortality benefit with EGDT in patients with a higher severity of disease (APACHE II score ≥ 20). In addition, EGDT was associated with decreased ICU mortality when compared with control care. These results suggest that EGDT may have some benefit in more critically ill patients.

nd toward the longest complete follow-up mortality benefit with EGDT in patients with a higher severity of disease (APACHE II score ≥ 20). In addition, EGDT was associated with decreased ICU mortality when compared with control care. These results suggest that EGDT may have some benefit in more critically ill patients. As the cornerstone for resuscitation in patients with severe sepsis and septic shock, EGDT has been recommended by the Surviving Sepsis Guidelines,5 and studies have suggested3 that every 10% increase in compliance is associated with a significant decrease in the odds ratio for mortality. However, the role of EGDT in the treatment of sepsis remains controversial. Different study time periods may influence the value of EGDT. Earlier trials supported EGDT for severe sepsis and septic shock patients; however, after 15 years, the most recent 3 trials all failed to show any benefit of EGDT. When comparing these studies with the original study of Rivers et al, there were concerns, including the amount of fluids before randomization and the lower overall mortality rate. These concerns could be explained by the broad implementation of, and compliance with, the Guidelines, suggesting that patients with severe sepsis and septic shock need early attention and resuscitation. The earlier we treat sepsis, the better outcome is shown.

before randomization and the lower overall mortality rate. These concerns could be explained by the broad implementation of, and compliance with, the Guidelines, suggesting that patients with severe sepsis and septic shock need early attention and resuscitation. The earlier we treat sepsis, the better outcome is shown. Three recent large multicenter trials10–12 with lower mortality (ProCESS: 21.0% vs 18.9% vs 18.2%; ARISE: 18.6% vs 18.8%; ProMISe: 29.5% vs 29.2%) compared with the study of Rivers et al (30.5% vs 46.5%) have found no difference in mortality between EGDT and control care, leading to lively discussions.26–29 Possible explanations for the difference between trial results include the study population, intervention methods, and goals. The timing and volume of fluids, for example, may play an important role.30 As a result, the focus during the golden time of resuscitation for septic shock should be fluid administration. However, in the recent trials, the volume of fluid administration is less (ProCESS: 2805 ± 1957 vs 3285 ± 1743 vs 2279 ± 1881 mL; ARISE: 1964 ± 1415 vs 1713 ± 1401 mL; ProMISe: 2226 ± 1443 vs 2022 ± 1271 mL) than that in the study of Rivers et al (4981 ± 2984 vs 3499 ± 2438 mL).

ic shock should be fluid administration. However, in the recent trials, the volume of fluid administration is less (ProCESS: 2805 ± 1957 vs 3285 ± 1743 vs 2279 ± 1881 mL; ARISE: 1964 ± 1415 vs 1713 ± 1401 mL; ProMISe: 2226 ± 1443 vs 2022 ± 1271 mL) than that in the study of Rivers et al (4981 ± 2984 vs 3499 ± 2438 mL). Many EGDT protocols specify using CVP and Scvo2 monitoring to guide management of fluids, vasopressors, packed red-blood-cell transfusions, and dobutamine. Some negative studies16,31 have challenged the role of Scvo2 in EGDT protocols and have used lactate clearance instead. Another multicenter study32 reported that abnormal Scvo2 values (90%–100%) observed in the emergency department were associated with increased mortality, indicating that an Scvo2 target should be achieved in the resuscitation of sepsis. Our analysis has several limitations. First, only 9 trials were included in our meta-analysis, and some were at high risk of bias. Second, the protocols of control care were different, which may have affected results. Third, different end points were used for mortality evaluation, which may influence the overall results; and ICU length of stay might not be a reliable marker of success because criteria for discharge from the ICU are not uniform. CONCLUSIONS The results of this meta-analysis suggested a nonsignificant trend toward reduction in the longest all-cause mortality in patients resuscitated with EGDT. However, TSA indicated lack of firm evidence for the results. High powered, randomized controlled trials are needed to determine the effects.

Our analysis has several limitations. First, only 9 trials were included in our meta-analysis, and some were at high risk of bias. Second, the protocols of control care were different, which may have affected results. Third, different end points were used for mortality evaluation, which may influence the overall results; and ICU length of stay might not be a reliable marker of success because criteria for discharge from the ICU are not uniform. CONCLUSIONS The results of this meta-analysis suggested a nonsignificant trend toward reduction in the longest all-cause mortality in patients resuscitated with EGDT. However, TSA indicated lack of firm evidence for the results. High powered, randomized controlled trials are needed to determine the effects. Supplementary Material DISCLOSURES Name: Jing-Yuan Xu, MD. Contribution: This author performed the analysis and interpretation of data and participated in drafting, editing, and submitting the manuscript. Name: Qi-Hong Chen, MD. Contribution: This author reviewed all the articles independently in accordance with the inclusion criteria. Name: Song-Qiao Liu, MD. Contribution: This author reviewed all the articles independently in accordance with the inclusion criteria. Name: Chun Pan, MD. Contribution: This author was responsible for identifying the clinical question and performing the literature search. Name: Xiu-Ping Xu, MD. Contribution: This author was responsible for identifying the clinical question and performing the literature search. Name: Ji-Bin Han, MD.

Contribution: This author reviewed all the articles independently in accordance with the inclusion criteria. Name: Chun Pan, MD. Contribution: This author was responsible for identifying the clinical question and performing the literature search. Name: Xiu-Ping Xu, MD. Contribution: This author was responsible for identifying the clinical question and performing the literature search. Name: Ji-Bin Han, MD. Contribution: This author was responsible for extracting methodological data and identifying the approach for data extraction from graphical sources. Name: Jian-Feng Xie, MD. Contribution: This author was responsible for extracting methodological data and identifying the approach for data extraction from graphical sources. Name: Ying-Zi Huang, MD. Contribution: This author was responsible for extracting numerical data, performing the data analysis, and interpreting the results. Name: Feng-Mei Guo, MD. Contribution: This author contributed to study analysis and interpretation of data. Name: Yi Yang, MD, PhD. Contribution: This author was responsible for revising the manuscript for important intellectual content. Name: Hai-Bo Qiu, MD, PhD. Contribution: This author was responsible for conception and design and revising the manuscript for important intellectual content. This manuscript was handled by: Avery Tung, MD. ACKNOWLEDGMENTS The authors thank Prof. Nian-Fang Lu for providing data from the study of Lu et al and Prof. Xiao-Zhi Wang for providing data from the study of Wang et al. Published ahead of print April 5, 2016

Contribution: This author was responsible for conception and design and revising the manuscript for important intellectual content. This manuscript was handled by: Avery Tung, MD. ACKNOWLEDGMENTS The authors thank Prof. Nian-Fang Lu for providing data from the study of Lu et al and Prof. Xiao-Zhi Wang for providing data from the study of Wang et al. Published ahead of print April 5, 2016 Funding: This work is partially supported by grants from the National Natural Science Foundations of China (81170057, 81201489, 81300043, 81300060, 81372093, and 81501705), grants from the Clinical Medicine Science and Technology program of Jiangsu Province (BL2013030). The authors declare no conflicts of interest. Supplemental digital content is available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal’s website. Reprints will not be available from the authors.

Endotracheal intubation is a common procedure performed in both the operating room and the intensive care unit. The inadvertent movement of the endotracheal tube (ETT) is potentially life-threatening. For this reason, the American Heart Association’s 2005 Advanced Cardiac Life Support guidelines recommend securing the ETT with either a tape or a commercial device.1 In current clinical practice, there is wide variability of methods used to secure the ETT including (1) adhesive tape, (2) cloth tape ties, and (3) commercial ETT holders. Prior studies have demonstrated diverging expert opinions on the optimum method for securing an ETT.2–8 The ideal method would minimize movement of the ETT, be quick and easy to apply, and have a low risk of injury to the oropharyngeal structures. The Haider Tube-Guard® is a new, commercially available device designed to optimally secure an ETT. The device is composed of a “clamshell”-like silicone body that wraps around and attaches onto the ETT via an adjustable, one-size-fits-all plastic clamp (Fig. 1A). The device plus ETT combination rests between the upper and the lower teeth, using the immobility of the maxilla for rigid fixation of the ETT to the patient (Fig. 1B). A Velcro neck strap (Posey foam/Velcro tie, Arcadia CA) is then attached for extra security. The Haider Tube-Guard device also incorporates a built-in soft bite block and, according to the company, is engineered to be gentle on the oropharyngeal and facial soft tissues during clinical use.

ETT to the patient (Fig. 1B). A Velcro neck strap (Posey foam/Velcro tie, Arcadia CA) is then attached for extra security. The Haider Tube-Guard device also incorporates a built-in soft bite block and, according to the company, is engineered to be gentle on the oropharyngeal and facial soft tissues during clinical use. Figure 1. Haider Tube-Guard device. A, Photograph of the Haider Tube-Guard device. The blue silicone portion wraps around and grips the endotracheal tube while also acting as a bite block. The white adjustable clamp maximizes grip and enables fixation to a Velcro neck strap. The device fits all adult-sized endotracheal tubes (size 6.0–8.5 mm ID). B, Photograph of the Haider Tube-Guard shown in situ with the Velcro neck strap being applied. The goal of this study was to compare ETT mobility when securing the ETT with the Haider Tube-Guard versus adhesive tapes, our current standard of practice. We hypothesized that the Haider Tube-Guard would provide more secure ETT attachment in adults undergoing general endotracheal anesthesia when compared with tapes and adhesives. The primary objective of the study was to quantify the movement of the ETT caused by a standardized force of dislodgement, comparing the Haider Tube-Guard with adhesive tape. The secondary objective of the study was to determine whether the Haider Tube-Guard device was well tolerated during routine clinical use.

sives. The primary objective of the study was to quantify the movement of the ETT caused by a standardized force of dislodgement, comparing the Haider Tube-Guard with adhesive tape. The secondary objective of the study was to determine whether the Haider Tube-Guard device was well tolerated during routine clinical use. METHODS After approval by our institutional review board and obtaining written informed consent, 30 adult patients undergoing general anesthesia with endotracheal intubation and neuromuscular blockade were prospectively enrolled in this study. The trial was registered with the clinicatrials.gov registry (NCT02347488; PI: Nir N. Hoftman, January 8, 2015). A careful history and physical examination were performed preoperatively; patients with loose or missing teeth, temporomandibular joint disease, severe asthma, and immunosuppression and patients undergoing surgery of the oropharynx or surgery who required the prone position were excluded from the trial.

nuary 8, 2015). A careful history and physical examination were performed preoperatively; patients with loose or missing teeth, temporomandibular joint disease, severe asthma, and immunosuppression and patients undergoing surgery of the oropharynx or surgery who required the prone position were excluded from the trial. Immediately after endotracheal intubation with a standard adult (6.5–8.0 mm ID) ETT (Mallinckrodt™; Covidien, Mansfield, OH), a single-study author with >10 years’ experience in thoracic anesthesia positioned the ETT tip in the distal one-third of the trachea using a bronchoscope. The exact distance from the carina to the ETT tip was measured and recorded, and the ETT was then secured at the discretion of the anesthesiologist caring for the patient. A force transducer (IWS V2-30, Jennings, Vancouver, BC, Canada) accurate to within 0.1 N was then attached to the ETT assembly and linear force applied parallel to the ETT such that it would pull the tube from the trachea if the tube was not adequately secured. While securing the head in a neutral position to prevent unwanted movement, we increased the force over approximately 5 seconds until the target of 15 N was reached or until the principal investigator deemed that the force be aborted to prevent possible tracheal extubation. Based on our previous mannequin testing, we selected a 15-N force because it could reliably displace inadequately secured ETTs. This force was at least an order of magnitude less than that generated during routine mastication and less than half the force routinely used during the application of cricoid pressure.9,10 We thus thought that this force would be safe in all patients, but as a precaution, we excluded any patients with actual or perceived loose teeth or any temporomandibular joint pain or disease. The change in ETT tip position caused by this force was then measured with the indwelling bronchoscope, rounded up to the nearest 5 mm, and recorded. After removal of the adhesive tape, the ETT was repositioned in the distal trachea, and the Haider Tube-Guard device was inserted and used to secure the ETT as recommended by the manufacturer. The linear force was reapplied as before, and the change in ETT tip position caused by the force was recorded. Linear force was aborted at the discretion of the investigators if tracheal extubation was deemed likely to occur without intervention.

s inserted and used to secure the ETT as recommended by the manufacturer. The linear force was reapplied as before, and the change in ETT tip position caused by the force was recorded. Linear force was aborted at the discretion of the investigators if tracheal extubation was deemed likely to occur without intervention. In such cases, the greatest achievable force was recorded along with the ETT displacement distance at the time of force abort. The bronchoscope was then removed and the Haider Tube-Guard device left in place for the duration of the case. Any conditions that would interfere with tape adhesion such as facial hair, sweating, and oily or flaky skin were documented. During the intraoperative portion as well as at admission and discharge from the postanesthesia care unit, an anesthesiologist who was part of the study team thoroughly examined the patient’s face and oropharynx for any evidence of minor tissue trauma. On discharge from the postanesthesia care unit, the patient also answered a brief survey assessing any subjective evidence of minor facial or oropharyngeal tissue trauma.

it, an anesthesiologist who was part of the study team thoroughly examined the patient’s face and oropharynx for any evidence of minor tissue trauma. On discharge from the postanesthesia care unit, the patient also answered a brief survey assessing any subjective evidence of minor facial or oropharyngeal tissue trauma. ETT movements >1 cm were defined as clinically significant because they could lead to patients’ coughing or moving. Furthermore, specialty ETTs such as double-lumen endobronchial tubes or electromyography tubes could become dislodged and malfunction with such movements. ETT movements >4 cm were defined as “potentially high risk for extubation” based on the dimensions of standard ETT cuffs and presumed positioning of the tube cuff in the proximal third of the trachea. Given that the average human trachea is 10 to 12 cm in length, movements >4 cm in such a positioned ETT would be highly likely to displace the majority of the balloon cuff beyond the vocal cords, risking imminent tracheal extubation. Our protocol was designed with safety nets in place to eliminate any chance of extubation during the study period. First, the ETT was positioned in the distal trachea to increase the margin of safety. Second, the principal investigator continuously ensured in real time that the ETT was well within the trachea during the data collection period using the indwelling bronchoscope. Third, a second study anesthesiologist and the clinical provider anesthesiologist not participating in the study were both constantly observing the patient to ensure proper ETT placement. Any of those providers had overriding power to request that the force be aborted should safety concerns arise. Finally, the bronchoscope was left in the trachea during the entire data collection period to act as a guide for reintubation in the extremely unlikely event that the ETT was pulled out of the trachea.

cement. Any of those providers had overriding power to request that the force be aborted should safety concerns arise. Finally, the bronchoscope was left in the trachea during the entire data collection period to act as a guide for reintubation in the extremely unlikely event that the ETT was pulled out of the trachea. A power analysis performed before patient enrollment determined that 17 patients would need to be enrolled to show a difference of 1 SD from the mean between the 2 fixation techniques at 80% power. We chose to enroll 30 patients to increase the power of our results and include a larger variety of patients undergoing different surgical procedures. The comparison of average ETT movement during maximal force was first performed using the paired t test. Given that the order of securing the ETT (tape versus Haider-Tube Guard) was not randomized, we also ran an analysis of covariance (ANCOVA) model to confirm our findings from the paired t test. The ANCOVA model was constructed with the pairwise difference between methods of securement as the dependent variable and tape movement (centered) for each patient as the covariate. The intercept from this model represents the average difference between methods, and the slope represents the expected difference for each 1-cm increase in tape movement. The slope from this model was estimated to be 0.98 (95% confidence interval [CI], 0.86–1.11; P <0.001), indicating that for each 1-cm increase in tape movement, the expected difference between methods of securement increased by 0.98 cm.

e slope represents the expected difference for each 1-cm increase in tape movement. The slope from this model was estimated to be 0.98 (95% confidence interval [CI], 0.86–1.11; P <0.001), indicating that for each 1-cm increase in tape movement, the expected difference between methods of securement increased by 0.98 cm. The absolute movement values were then dichotomized into clinically relevant thresholds (>1 cm and >4 cm). The rate of patients exceeding these thresholds was compared between the 2 ETT securing methods using the McNemar test for paired proportions. P values <0.05 were considered statistically significant. All statistical analyses were performed using SAS 9.4 (Cary, NC). RESULTS Thirty patients successfully completed the protocol, and no patients had to be excluded once they were recruited. Surgical case mix included neurological, vascular, gynecological, colorectal, thoracic, and oncological surgery. The mean surgical time and, therefore, dwell time of the Haider Tube-Guard was 189 minutes (range 65–410 minutes). All patients were positioned supine for the duration of surgery.

d once they were recruited. Surgical case mix included neurological, vascular, gynecological, colorectal, thoracic, and oncological surgery. The mean surgical time and, therefore, dwell time of the Haider Tube-Guard was 189 minutes (range 65–410 minutes). All patients were positioned supine for the duration of surgery. The Haider Tube-Guard significantly reduced the mobility of the ETT when compared with adhesive tape. Under standardized traction, the ETT withdrew a mean distance of 3.4 cm (SD ±1.8) when secured with adhesive tape versus 0.3 cm (SD ±0.6) when secured with the Haider Tube-Guard (average difference = 3.1 [SD ±1.8; 95% CI, 2.88–3.32; P <0.001; ANCOVA]). The paired t test results were similar to the ANCOVA with the average difference of 3.1 (95% CI, 2.41–3.79; P <0.001). The ETT displacement distances measured with each of the securing methods for every patient are shown graphically in Figure 2. Figure 2. Results summary. This graph highlights the study results. Every patient is represented by a number and plotted on the x-axis. Endotracheal tube (ETT) movement during applied traction is plotted for each patient on the y-axis in centimeters. Movement distance is denoted by blue diamonds during fixation with adhesive tape and red squares during fixation with the Haider Tube-Guard. Individual patients have both symbols shown because each patient served as his or her own control. Horizontal green lines highlight 2 clinically relevant categories: (1) >1-cm ETT movement and (2) >4 cm of ETT movement.

monds during fixation with adhesive tape and red squares during fixation with the Haider Tube-Guard. Individual patients have both symbols shown because each patient served as his or her own control. Horizontal green lines highlight 2 clinically relevant categories: (1) >1-cm ETT movement and (2) >4 cm of ETT movement. Ninety-seven percent of patients (29/30) experienced clinically significant ETT movement (>1 cm) when adhesive tape was used to secure the tube versus 3% (1/30) when the Haider Tube-Guard was used (P < 0.001). Thirty percent of patients (9/30) experienced ETT movement >4 cm (potential high extubation risk) when the ETT was secured with tape versus 0% (0/30) when secured with the Haider Tube-Guard (P = 0.004). Six patients required that traction be aborted before 15 N of force was achieved to prevent extubation because the tape either completely separated from the face or stretched enough to allow for significant ETT excursion. Three of these 6 patients had short facial hair, and despite significant efforts to tape and glue the ETT, in 2 cases, the tape disconnected under load. Five of the 6 patients with tape disconnects experienced ETT movement >4 cm and were categorized into the high extubation risk group, whereas the sixth (who had a short mustache) was not because the ETT excursion was only 4.0 cm at the time the force was aborted. The details of the patients deemed potentially high extubation risk are summarized in Table 1. Table 1. Characteristics and Details of Patients with Endotracheal Tube (ETT) Movements >4.0 cm

Ninety-seven percent of patients (29/30) experienced clinically significant ETT movement (>1 cm) when adhesive tape was used to secure the tube versus 3% (1/30) when the Haider Tube-Guard was used (P < 0.001). Thirty percent of patients (9/30) experienced ETT movement >4 cm (potential high extubation risk) when the ETT was secured with tape versus 0% (0/30) when secured with the Haider Tube-Guard (P = 0.004). Six patients required that traction be aborted before 15 N of force was achieved to prevent extubation because the tape either completely separated from the face or stretched enough to allow for significant ETT excursion. Three of these 6 patients had short facial hair, and despite significant efforts to tape and glue the ETT, in 2 cases, the tape disconnected under load. Five of the 6 patients with tape disconnects experienced ETT movement >4 cm and were categorized into the high extubation risk group, whereas the sixth (who had a short mustache) was not because the ETT excursion was only 4.0 cm at the time the force was aborted. The details of the patients deemed potentially high extubation risk are summarized in Table 1. Table 1. Characteristics and Details of Patients with Endotracheal Tube (ETT) Movements >4.0 cm The study team examined the oropharynx of all 30 patients postoperatively, and all but one patient completed the study questionnaire (one patient was aphasic postoperatively). None of the patients suffered any significant injury from the Haider Tube-Guard device. Sixty-five percent of patients reported a mild sore throat deemed to be secondary to tracheal intubation. One patient reported mild discomfort of the gum near a rear molar thought to be unrelated to the Haider Tube-Guard. Another patient had a minor abrasion of the tongue frenulum possibly because of a difficult nasogastric and orogastric tube insertion, although the Haider Tube-Guard could not be excluded as a causative agent.

atient reported mild discomfort of the gum near a rear molar thought to be unrelated to the Haider Tube-Guard. Another patient had a minor abrasion of the tongue frenulum possibly because of a difficult nasogastric and orogastric tube insertion, although the Haider Tube-Guard could not be excluded as a causative agent. DISCUSSION The Haider Tube-Guard significantly outperformed all types and combinations of adhesive tape because it affixes to the patient in a far more secure fashion. Whereas adhesive tape attaches to the facial surface, the Haider Tube-Guard anchors to the maxilla and mandible. Forces applied to a taped ETT distort and pull on mobile facial tissues, allowing for significant ETT movements to occur even in the absence of an adhesive failure. An ETT secured with the Haider Tube-Guard on the contrary is pinned between rigid bony structures that do not move under such loads. This lack of reliance on a tape–skin interface explains why the Haider Tube-Guard functioned well in notoriously difficult settings such as in patients with facial hair.

adhesive failure. An ETT secured with the Haider Tube-Guard on the contrary is pinned between rigid bony structures that do not move under such loads. This lack of reliance on a tape–skin interface explains why the Haider Tube-Guard functioned well in notoriously difficult settings such as in patients with facial hair. The Haider Tube-Guard all but eliminated ETT movements >1 cm under maximal load, in contrast to adhesive tapes, which failed to prevent such movements in all but 1 patient. The Haider Tube-Guard may prove especially useful in cases using specialty ETTs such as double-lumen endobronchial tubes or electromyography tubes, where small movements can dislodge the bronchial balloon or electrode interface, respectively. Further studies should be conducted in such scenarios as well as in patients not undergoing neuromuscular blockade to determine whether in fact the Haider Tube-Guard reduces unwanted intraoperative coughing by eliminating such small ETT movements. Of note, in one patient with a very large mouth opening, the Haider Tube-Guard separated from the teeth under load, but the Velcro neck strap served as a backup and kept the ETT from moving beyond 3 cm. The control method of taping in this patient, which included 5 different tapes and adhesives, faired no better by allowing 3 cm of movement under load.

y large mouth opening, the Haider Tube-Guard separated from the teeth under load, but the Velcro neck strap served as a backup and kept the ETT from moving beyond 3 cm. The control method of taping in this patient, which included 5 different tapes and adhesives, faired no better by allowing 3 cm of movement under load. Nearly one-third (n = 9) of the study population experienced ETT movement >4 cm when adhesive tape was used to affix the ETT and could have been highly vulnerable to extubation under normal, uncontrolled circumstances. In a tenth patient (with a mustache), whose force was aborted at 4 cm, ETT movement because of tape disconnect could have been included in the high extubation risk category; we chose not to include this patient to stay true to our original definition in the protocol. In all 10 of these patients, securing the ETT with the Haider Tube-Guard mitigated this extubation vulnerability by eliminating excessive ETT movement. Our definition of high extubation risk (movement >4 cm) was quite conservative, and the actual population at risk in this study may have been higher.2–8 Reported rates of unintentional extubation range from 0.03% to 2.5% depending on the clinical setting.11–14 Undeniably, unintentional extubation can lead to increased patient morbidity and even mortality.15–17 In fact, 12% of all airway cases in the anesthesia closed claims analysis were caused by unintentional extubation.18 Several commercial devices have been introduced into clinical practice over the years with the aim of mitigating this risk. Clinical trials comparing various commercial ETT holders to traditional adhesive tape have shown mixed results; traditional tape was found to be superior to these devices in some but not all trials.2–8 These findings, combined with the perceived extra cost of the commercial devices, have slowed their adoption into clinical practice. The Haider Tube-Guard’s superior and unequivocal performance versus traditional tape may indicate improved design compared with other available commercial devices, although head-to-head testing should be performed to determine any such benefit. The Haider Tube-Guard may prove beneficial in clinical settings prone to ETT motion including (1) emergency transport (civilian and military) of trauma patients, (2) prone and lateral positioning, and (3) extensive head and neck surgery.

, although head-to-head testing should be performed to determine any such benefit. The Haider Tube-Guard may prove beneficial in clinical settings prone to ETT motion including (1) emergency transport (civilian and military) of trauma patients, (2) prone and lateral positioning, and (3) extensive head and neck surgery. Blood, sweating, oral and facial secretions, and burns can make securing the ETT in these cases even more challenging, and further studies using the Haider Tube-Guard in such scenarios are warranted. This clinical study had several methodological advantages when compared with previous ETT-holding device trials. We conducted the trial in the operating room on patients undergoing surgery with each patient serving as his or her own control for a true comparison of the Haider Tube-Guard versus adhesive tape. Many previous studies were performed on human cadavers or plastic simulation mannequins, limiting the generalizability of the results.2–5 Also, our protocol used bronchoscopy to measure the distance between the ETT tip and the carina, thus defining the actual ETT movement within the airway. Other studies used only indirect surrogates of ETT airway movement such as distance from the incisor or lip.3,5 Finally, all patients in the study received neuromuscular blockade during data collection to eliminate any baseline intrinsic jaw muscle tone, ensuring that only the device’s inherent grip was being measured.

Other studies used only indirect surrogates of ETT airway movement such as distance from the incisor or lip.3,5 Finally, all patients in the study received neuromuscular blockade during data collection to eliminate any baseline intrinsic jaw muscle tone, ensuring that only the device’s inherent grip was being measured. Our study did have several limitations that should be noted. First, we introduced the force in a crescendo pattern, building to maximal force over about 5 seconds, and then held the force during bronchoscopy measurements. This was necessary to allow for accurate distance measurements and also to ensure patient safety by carefully controlling the force. Such prolonged forces may not represent all forces encountered clinically, some of which may be fast, jerking motions. Second, we applied the force in the same linear direction in all patients, whereas real-life forces encountered in clinical medicine may be multidirectional and rapidly changing. Third, we excluded edentulous patients or patients with loose teeth from the study. Further studies should be performed to assess the Haider Tube-Guard device in patients with poor dentition because this could possibly lead to reduced device grip. Fourth, our study was not blinded to the treatment group, but given the nature of the device, we thought that blinding would not be possible. Furthermore, constant observation of the ETT–face interface was required during application of force to ensure safety, making blinding impossible. Finally, although our study demonstrated that the Haider Tube-Guard was well tolerated, it was not statistically powered to determine device safety.

g would not be possible. Furthermore, constant observation of the ETT–face interface was required during application of force to ensure safety, making blinding impossible. Finally, although our study demonstrated that the Haider Tube-Guard was well tolerated, it was not statistically powered to determine device safety. In conclusion, the Haider Tube-Guard was superior to adhesive tape in securing the ETT. The device significantly reduced ETT movement and may prevent unplanned patient extubation. Larger studies would need to be conducted to demonstrate utility in diverse clinical settings and to establish a patient safety record. DISCLOSURES Name: Jack C. Buckley, MD. Contribution: This author helped design the study, conduct the study, analyze the data, and write the manuscript. Attestation: Jack C. Buckley has seen the original study data, reviewed the analysis of the data, and approved the final manuscript. Name: Adam P. Brown, MD. Contribution: This author helped conduct the study and write the manuscript. Attestation: Adam P. Brown approved the final manuscript. Name: John S. Shin, MD. Contribution: This author helped conduct the study and write the manuscript. Attestation: John S. Shin approved the final manuscript. Name: Kirsten M. Rogers, BA. Contribution: This author helped conduct the study. Attestation: Kirsten M. Rogers approved the final manuscript. Name: Nir N. Hoftman, MD. Contribution: This author helped design the study, conduct the study, analyze the data, and write the manuscript.

Attestation: John S. Shin approved the final manuscript. Name: Kirsten M. Rogers, BA. Contribution: This author helped conduct the study. Attestation: Kirsten M. Rogers approved the final manuscript. Name: Nir N. Hoftman, MD. Contribution: This author helped design the study, conduct the study, analyze the data, and write the manuscript. Attestation: Nir N. Hoftman has seen the original study data, reviewed the analysis of the data, approved the final manuscript, and is the author responsible for archiving the study files. This manuscript was handled by: Sorin J. Brull, MD. ACKNOWLEDGMENT The authors thank Tristan Grogan, MS, for his expertise and assistance in statistical analysis. Funding: All funding came from departmental research funds, with the exception of the Haider Tube-Guard devices, which were supplied by Haider Biologics free of charge. The authors declare no conflicts of interest. Reprints will not be available from the authors.

Since their invention several decades ago,1,2 pulse oximeters have successfully used the principle of photoplethysmography (PPG). In PPG, the intensity of light that travels through tissue is modulated by the absorption of pulsatile blood volume.3 In pulse oximetry, the relative PPG amplitudes at 2 or more wavelengths provide a noninvasive estimate, termed Spo2, of the arterial oxygen saturation Sao2. Many technical improvements over the past decades have led to Spo2 as one of the common vital signs to be monitored clinically, but it has always remained a contact measurement. However, development of inexpensive and sensitive digital imaging sensors has enabled several investigators4–8 to change the conventional geometry and measure PPG in a fully contactless manner, with both light source and detector away from the skin. The potential advantages of contactless PPG include avoiding skin damage in fragile patients and a freedom to select a more physiologically central location with a possibly faster response.9,10

e conventional geometry and measure PPG in a fully contactless manner, with both light source and detector away from the skin. The potential advantages of contactless PPG include avoiding skin damage in fragile patients and a freedom to select a more physiologically central location with a possibly faster response.9,10 In the past decade, a rapidly increasing number of articles11 has been published on contactless PPG addressing pulse or respiration rate, while only a few4,5,12–16 explore Spo2. Humphreys et al5 sought to emulate the conventional source-detector geometry by illuminating the skin at one location and measuring PPG (with a camera) right next to it. Wieringa et al4 used wide-field illumination but focused on imaging vasculature. However, none of these has convincingly demonstrated that camera-based Spo2 is fundamentally calibratable. In this article, we address the question “Can a single calibration curve provide Spo2 estimates for a population of individuals with acceptable accuracy?” This is not a trivial question because there is a fundamental difference between the conventional contact source–detector (Figure 1A), and the contactless, wide-field illumination-detection geometries (Figure 1B). Although the former geometry collects light that has travelled through relatively deep vasculature (both in transmissive and reflective, or “surface” mode),17 the latter predominantly collects light that has travelled through much shallower tissue depths15,18 over much smaller distances.

tection geometries (Figure 1B). Although the former geometry collects light that has travelled through relatively deep vasculature (both in transmissive and reflective, or “surface” mode),17 the latter predominantly collects light that has travelled through much shallower tissue depths15,18 over much smaller distances. Figure 1. In conventional contact probes (A), albeit transmissive or reflective, the skin interrogation depth is significantly deeper than in a wide-field illumination camera geometry (B) which has consequences for the signal strength and potentially also for calibration of Spo2. Proportions not to scale.

tection geometries (Figure 1B). Although the former geometry collects light that has travelled through relatively deep vasculature (both in transmissive and reflective, or “surface” mode),17 the latter predominantly collects light that has travelled through much shallower tissue depths15,18 over much smaller distances. Figure 1. In conventional contact probes (A), albeit transmissive or reflective, the skin interrogation depth is significantly deeper than in a wide-field illumination camera geometry (B) which has consequences for the signal strength and potentially also for calibration of Spo2. Proportions not to scale. Currently, it is not known whether the PPG signal measured in the camera geometry stems mostly from the deeper arterioles or also (partly) from the shallow capillaries. In fact, there is some controversy on whether the PPG signal stems directly from blood volume changes at all.19–21 We can only hypothesize on PPG origins and the consequences for Spo2 calibratability. Due to their shallow location, superficial capillaries may contribute to PPG amplitude or even be the predominant source (Figure 1B), even if they are only slightly pulsatile compared with arterioles. If their oxygen saturation is not representative of arterial blood, calibratability becomes questionable. Another potential risk is shunt light, light that has travelled through the bloodless epidermis only, or specularly reflected light. The shallow skin layers involved in camera-based PPG are essentially “uncharted territory” in terms of pulse oximetry. To address this fundamental question of calibratability, we have taken an experimental approach. However, we did not address technical challenges and used only artifact-free data.

arly reflected light. The shallow skin layers involved in camera-based PPG are essentially “uncharted territory” in terms of pulse oximetry. To address this fundamental question of calibratability, we have taken an experimental approach. However, we did not address technical challenges and used only artifact-free data. We have found that some individuals can induce significant changes in Spo2 by holding their breath. Trending of contactless Spo2 can thus be demonstrated without the aid of hypoxic environments (Figure 2). However, to calibrate and study accuracy on a broader population, stable desaturation levels are needed to exclude discrepancies related to the temporal filtering in contact references and synchronization between reference and camera traces. Guazzi et al16 were the first to report contactless Spo2 in a more controlled manner in a hypoxic chamber but their analysis aimed at trending and resulted in different calibration constants for each of the 5 individuals studied. Figure 2. Normalized raw PPG signals at a normoxic (A) and hypoxic (B) segments. Breath-holds are shown along with Spo2 traces from finger probe and camera in (C). Time segments in (A) and (B) are indicated in (C) by “n” and “h.” The camera Spo2 trace is delayed by 21 seconds for a good visual match. Tentative calibration constants were used.

PPG signals at a normoxic (A) and hypoxic (B) segments. Breath-holds are shown along with Spo2 traces from finger probe and camera in (C). Time segments in (A) and (B) are indicated in (C) by “n” and “h.” The camera Spo2 trace is delayed by 21 seconds for a good visual match. Tentative calibration constants were used. In the present study, we use a dedicated hypoxic chamber with the goal of investigating whether a single calibration curve can describe a population of individuals with acceptable accuracy. We realize that calibratibility may depend on many variables such as physiological status, age, and skin condition. We only included healthy adults as subjects under normoxic and hypoxic conditions. Additionally, we explored the impact of simulated centralization on the calibratibility of contactless pulse oximetry. To this goal, we exposed subjects to low temperatures in a climate room study under normoxic conditions. METHODS IRB/Consent All experiments were approved by the Philips IRB (Internal Committee on Biomedical Experiments) before experimentation. Written informed consent was obtained from all subjects. Three different experimental protocols (studies N, H, and C) were submitted to and approved by the Philips IRB. The methods used for the 3 experiments are described in separate sections below. IRB contact: PHM Keizer, Senior Ethical & Biomedical Officer, High Tech Campus 34, 5656 AE Eindhoven, The Netherlands.

ubjects. Three different experimental protocols (studies N, H, and C) were submitted to and approved by the Philips IRB. The methods used for the 3 experiments are described in separate sections below. IRB contact: PHM Keizer, Senior Ethical & Biomedical Officer, High Tech Campus 34, 5656 AE Eindhoven, The Netherlands. Clinical Trial Registration The study was not registered before patient enrollment because it was not a clinical trial; it was a basic research study on healthy volunteers. It did not include treatment or control groups. The Basic Set-up (Study N) During data collection, the subjects were seated, and head motion was minimized using a head support mounted on the chair. Two tripod-mounted monochrome Stingray F046-B FireWire cameras (Allied Vision Technologies, Stadtroda, Germany) with Computar 50 mm 1:1.3 lenses (CBC, Commack, NY) were each equipped with spectral bandpass filters with red and infrared (IR) center wavelengths of 675 and 842 nm, respectively (Semrock FF02-675/67-24-D and FF01-842/56-24D; IDEX Corp., Lake Forrest, IL). The cameras recorded movies of the subject’s face from a distance of about 1.6 meters at 15 frames per second. Illumination was provided by 2 armatures (Falcon Eyes, Hong Kong, China), each equipped with 9 incandescent lamps (Philips 40 W) at a distance of about 1 m from the subjects. Current limited DC power supplies set to 150 V, 1.25 A (SM3004-D, Delta Electronica, Zierikzee, The Netherlands) powered the lamps.

The Basic Set-up (Study N) During data collection, the subjects were seated, and head motion was minimized using a head support mounted on the chair. Two tripod-mounted monochrome Stingray F046-B FireWire cameras (Allied Vision Technologies, Stadtroda, Germany) with Computar 50 mm 1:1.3 lenses (CBC, Commack, NY) were each equipped with spectral bandpass filters with red and infrared (IR) center wavelengths of 675 and 842 nm, respectively (Semrock FF02-675/67-24-D and FF01-842/56-24D; IDEX Corp., Lake Forrest, IL). The cameras recorded movies of the subject’s face from a distance of about 1.6 meters at 15 frames per second. Illumination was provided by 2 armatures (Falcon Eyes, Hong Kong, China), each equipped with 9 incandescent lamps (Philips 40 W) at a distance of about 1 m from the subjects. Current limited DC power supplies set to 150 V, 1.25 A (SM3004-D, Delta Electronica, Zierikzee, The Netherlands) powered the lamps. For Spo2 reference, rather than blood gas analysis, we used 4 conventional Spo2 probes coupled to Philips MP2 patient monitors: a Philips finger sensor, a Philips ear sensor (M1191B and M1194A, Philips Medizin-Systeme, Böblingen, Germany), a Masimo finger sensor (LNCS DC-I, Masimo Corporation, Irvine), and a Nellcor finger sensor (DS-100A Medtronic, Dublin, Ireland). A sample-wise (1 Hz) median of all 4 probes was defined as the reference signal Spo2,ref(t). Twenty-five individuals participated in study N (Supplemental Digital Content, Supplemental Table 1, http://links.lww.com/AA/B428).

Masimo Corporation, Irvine), and a Nellcor finger sensor (DS-100A Medtronic, Dublin, Ireland). A sample-wise (1 Hz) median of all 4 probes was defined as the reference signal Spo2,ref(t). Twenty-five individuals participated in study N (Supplemental Digital Content, Supplemental Table 1, http://links.lww.com/AA/B428). The Hypoxic Tent (Study H) For the second study, the setup described above was used in a normobaric hypoxic tent (At Home Cubicle, Hypoxico Inc, New York, NY). Oxygen concentration and temperature inside the tent were continuously monitored (Model AD300, Teledyne Analytical Instruments, City of Industry, CA and Model 54II B, Fluke Corp, Everett, WA, respectively). After about 4 hours of pumping with the hypoxic generator (Everest Summit 2 Hypoxic Generator, Hypoxico Inc), the oxygen concentration stabilized at approximately 15% at which point the subject entered the tent, carefully and quickly to minimize air exchange. An electric fan (Cool Air System 40, Philips) homogenized the air inside the tent while the outlet of the oxygen pump was relatively close (eg, 20 cm) to the mouth of the subject and the probe of the oxygen meter at about 1 m away, probing homogenized air. During data collection, oxygen concentration was measured to be 15.0% ± 0.5% (average and standard deviation over all recordings) and air temperature typically increased to approximately 25°C. While results varied strongly among subjects, an Spo2 reduction of about 7% on average was induced by the hypoxic conditions: median Spo2 values were 98.2% and 90.6% for studies N and H, respectively (Supplemental Digital Content, Supplemental Table 1, http://links.lww.com/AA/B428). Twenty-one individuals participated in study H (Supplemental Digital Content, Supplemental Table 2, http://links.lww.com/AA/B428), three were also in study N.

ns: median Spo2 values were 98.2% and 90.6% for studies N and H, respectively (Supplemental Digital Content, Supplemental Table 1, http://links.lww.com/AA/B428). Twenty-one individuals participated in study H (Supplemental Digital Content, Supplemental Table 2, http://links.lww.com/AA/B428), three were also in study N. The Climate Chamber (Study C) In the third study, the subject was seated comfortably in a chair in a climate-controlled room (Environmental Stress Chamber GMAPX-15CW, Hielkema Testequipment, Uden, NL), with an initial temperature of approximately 22°C. The chamber temperature was then gradually reduced to about 5°C. This process took approximately 30 minutes. The room temperature was continuously monitored using the temperature probe of a patient monitor (MP5, Philips Medizin-Systeme). Video recordings of the subject’s head were made before the cooling phase at about 22°C, in the middle of the cooling phase at about 11°C, and at the end at about 5°C. The climate chamber’s cooling fan was turned off during the recordings to avoid any air turbulence moving the cameras or illumination. Due to limited space in the climate room, only 1 illumination armature was used instead of 2 in studies N and H. Other than this difference, the temperature was the only variable that changed.

mate chamber’s cooling fan was turned off during the recordings to avoid any air turbulence moving the cameras or illumination. Due to limited space in the climate room, only 1 illumination armature was used instead of 2 in studies N and H. Other than this difference, the temperature was the only variable that changed. To obtain feedback on cold-induced skin physiological changes, skin reflectance spectra (380–740 nm) were measured on the forehead with a spectrophotometer (CM-2600d, Konica-Minolta, Osaka, Japan). Immediately after each video recording at each temperature regime, 5 measurements were made and used as input for a median spectrum. Five individuals participated in study C (Supplemental Digital Content, Supplemental Table 2, http://links.lww.com/AA/B428). Three of these individuals also participated in studies N and/or H.

iately after each video recording at each temperature regime, 5 measurements were made and used as input for a median spectrum. Five individuals participated in study C (Supplemental Digital Content, Supplemental Table 2, http://links.lww.com/AA/B428). Three of these individuals also participated in studies N and/or H. Signal Processing for All Experiments The camera and reference probe signals were acquired and processed by LabVIEW (National Instruments, Austin, TX). The raw PPG signals R(t) and IR(t) (Appendix) are the spatially averaged pixel intensity in a manually selected region of interest on the forehead. Next, using a window size of 10 seconds and step size of 1 second, R(t) and IR(t) were normalized (by the respective temporal mean values over the window) and temporally filtered (second-order Butterworth bandpass with the cutoff frequencies 0.5 and 1.5 Hz), giving signals Rn(t) and IRn(t). For each channel, the difference between median values of peaks and valleys was defined as the pulsatility amplitude, now at a temporal resolution of the window step size. The ratio-of-ratio signal RR(t) was finally computed as the sample-wise quotient of red and IR pulsatility amplitudes, which is subsequently linearly transformed into an Spo2 trace with calibration constants C1 and C2:

lleys was defined as the pulsatility amplitude, now at a temporal resolution of the window step size. The ratio-of-ratio signal RR(t) was finally computed as the sample-wise quotient of red and IR pulsatility amplitudes, which is subsequently linearly transformed into an Spo2 trace with calibration constants C1 and C2: (1) Camera Spo2 traces were typically about 20 seconds faster than the Spo2,ref., due to a limited signal processing of the camera signals compared with the reference Spo2 probes and likely also a physiological delay at the periphery with respect to the forehead. Most traces have insufficient features to individually determine a delay. To avoid fitting the camera data to the reference probe, we tried to determine an average delay for the whole data set. We identified 12 traces with distinct temporal features in the Spo2 traces. For each of these traces, we determined the optimal delay τ for temporal alignment of the reference and camera Spo2 traces, using Equation 2: (2) Four of the 12 traces are shown in Figure 3, A–D. For the 12 traces, an average delay of 21.1 seconds was found. An SD of 2.1 seconds partly represents physiological variation but also expresses accuracy of determination of τ. Figure 3. Example Spo2 traces for study H. Values for (tentative) calibration constants C1 and C2 were the same as those use for the breath hold experiment (Figure 2). For a good temporal match with the Spo2 ref, delays τ were determined as 20, 22, 24, and 24 seconds in (A–D), respectively.

(2) Four of the 12 traces are shown in Figure 3, A–D. For the 12 traces, an average delay of 21.1 seconds was found. An SD of 2.1 seconds partly represents physiological variation but also expresses accuracy of determination of τ. Figure 3. Example Spo2 traces for study H. Values for (tentative) calibration constants C1 and C2 were the same as those use for the breath hold experiment (Figure 2). For a good temporal match with the Spo2 ref, delays τ were determined as 20, 22, 24, and 24 seconds in (A–D), respectively. Next, we computed the temporal median RR of the signal RR(t). This procedure was justified since we considered individuals in steady-state conditions and we were primarily interested in long-term calibratability of camera-based Spo2 measurements. Further advantages are that each individual entered the calibration process with the same weight and that recording length did not introduce a bias. It also further minimized synchronization issues between reference and camera-based traces and among the reference probes themselves. Similarly, for each recording, 1 median Spo2,ref value was computed for each Spo2,ref(t) trace. With this approach, each recording in studies N and H provided 1 data point (median RR, median Spo2,ref) to populate a calibration data set.

etween reference and camera-based traces and among the reference probes themselves. Similarly, for each recording, 1 median Spo2,ref value was computed for each Spo2,ref(t) trace. With this approach, each recording in studies N and H provided 1 data point (median RR, median Spo2,ref) to populate a calibration data set. Despite head fixation, some recordings were frequently disrupted due to motion. In other recordings, the signal amplitude of the red PPG signal was so small that noise from illumination or camera disrupted proper measurement of the pulsatility. Because we sought a fundamental assessment of calibratability, we excluded such recordings from the data set in an objective manner. To this goal, a signal quality Q(t), essentially a signal-to-noise ratio, was derived from the raw red PPG signal to allow the deselection of data of low signal quality (see Results, Figure 4B). Hereby, Rn(t), the weakest signal, was processed in 20-s sliding windows, in steps of 1 second. Within each window, the Fourier transform of R(t) was computed. The magnitude of the Fourier spectrum at the heart rate was used as a measure of signal strength, whereas the average magnitude at frequencies of heart rate ±0.3 Hz was used as a measure for noise strength. The Q-value for the window was finally computed as log10(signal amplitude/noise amplitude). Example signals of strong, weak, and motion-distorted signals are shown in Figure 4, A–C, respectively.

ignal strength, whereas the average magnitude at frequencies of heart rate ±0.3 Hz was used as a measure for noise strength. The Q-value for the window was finally computed as log10(signal amplitude/noise amplitude). Example signals of strong, weak, and motion-distorted signals are shown in Figure 4, A–C, respectively. Figure 4. Normalized PPG signal segments with Q values of 2.2, 1.0, and 0.6 in (A), (B), and (C), respectively. The PPG signals may be described as “strong,” “weak,” and “motion corrupted,” respectively. Figure 5. All candidate data points for calibration are shown in (A) with Q values indicated and regression lines for different values for Qthr. Using data with Q > 0.5 shows a regression line (dashed) that is biased by outlier points. Too strict filtering with Qthr = 2.3 gives a regression line (dashed) determined by too few data points and low value for R2. Regression has high values for R2 and, more importantly, is stable for 1.4 < Qthr < 1.9, (A and B). Fourteen regression lines (solid) are shown for this range of Qthr. Analogous to the temporal medians for Spo2,ref, and RR, we also defined a single Q value for each recording, as the temporal median of Q(t). As shown in the Results section, a threshold value Qthr was determined as an objective measure to only use reliable data in the calibration procedure.

Figure 5. All candidate data points for calibration are shown in (A) with Q values indicated and regression lines for different values for Qthr. Using data with Q > 0.5 shows a regression line (dashed) that is biased by outlier points. Too strict filtering with Qthr = 2.3 gives a regression line (dashed) determined by too few data points and low value for R2. Regression has high values for R2 and, more importantly, is stable for 1.4 < Qthr < 1.9, (A and B). Fourteen regression lines (solid) are shown for this range of Qthr. Analogous to the temporal medians for Spo2,ref, and RR, we also defined a single Q value for each recording, as the temporal median of Q(t). As shown in the Results section, a threshold value Qthr was determined as an objective measure to only use reliable data in the calibration procedure. Statistical Approach and Analysis Before our measurements, we faced many unknowns such as variations in PPG signal strength among individuals, subject motion, and the extent to which hypoxia could be induced in the normobaric hypoxic tent. Rather than conducting a pilot study to investigate each of these unknowns, we recruited relatively large numbers of subjects for studies N and H (25 each), in the anticipation that this would provide a sufficiently large and clean data set to determine a stable linear regression between RR and Spo2,ref. Thus, we enrolled many subjects in the study and determined a posteriori if we succeeded in establishing a stable linear regression. This process is described in the Results section.

cipation that this would provide a sufficiently large and clean data set to determine a stable linear regression between RR and Spo2,ref. Thus, we enrolled many subjects in the study and determined a posteriori if we succeeded in establishing a stable linear regression. This process is described in the Results section. All measurements for studies N and H were performed consecutively without intermediate analysis to adjust the study design. Inclusion and exclusion criteria for studies N, H, and C are in Supplemental Digital Content, Supplemental Table 3 (http://links.lww.com/AA/B428). For study N, all 25 recruited individuals were included while for study H, 21 individuals were included after checking for inclusion and exclusion criteria. For study C, we were somewhat limited in inclusion because the climate room had limited availability. The purpose of study C was exploratory.

nks.lww.com/AA/B428). For study N, all 25 recruited individuals were included while for study H, 21 individuals were included after checking for inclusion and exclusion criteria. For study C, we were somewhat limited in inclusion because the climate room had limited availability. The purpose of study C was exploratory. For studies N and H, an objective measure was used to discard recordings with unreliable RR. As shown in the Results section, indeed the 17 cleanest data points resulted in a regression line that was essentially the same as the regression for the 31 cleanest data points by which we mean that any error due to variation in regression was irrelevant (much smaller) compared with inaccuracies caused by variations among subjects. The determination of C1 and C2 was eventually based on the 31 cleanest data points. Next, with these calibration constants we proceeded to answer the overall study goal (Can a single calibration curve provide Spo2 estimates for a population of individuals with acceptable accuracy?). For this, we adopted Arms, a metric used for pulse oximeters to describe accuracy (ISO 80601-2-61, 2011, section 201.12.1.101.2.2). An * superscript is used to indicate at least 1 major difference with the International Organization for Standardization (ISO) standard, which is that in our formulation (Equation 3) we used median values of the traces, thus discarding short-term errors. (3) Because, in our calibration approach, this expression was mathematically similar to that of a standard deviation, we used the χ2 approach to calculate an error estimate for A*rms.

For studies N and H, an objective measure was used to discard recordings with unreliable RR. As shown in the Results section, indeed the 17 cleanest data points resulted in a regression line that was essentially the same as the regression for the 31 cleanest data points by which we mean that any error due to variation in regression was irrelevant (much smaller) compared with inaccuracies caused by variations among subjects. The determination of C1 and C2 was eventually based on the 31 cleanest data points. Next, with these calibration constants we proceeded to answer the overall study goal (Can a single calibration curve provide Spo2 estimates for a population of individuals with acceptable accuracy?). For this, we adopted Arms, a metric used for pulse oximeters to describe accuracy (ISO 80601-2-61, 2011, section 201.12.1.101.2.2). An * superscript is used to indicate at least 1 major difference with the International Organization for Standardization (ISO) standard, which is that in our formulation (Equation 3) we used median values of the traces, thus discarding short-term errors. (3) Because, in our calibration approach, this expression was mathematically similar to that of a standard deviation, we used the χ2 approach to calculate an error estimate for A*rms. A mean difference or “bias” was according to Equation 4, also following the ISO standard.

For studies N and H, an objective measure was used to discard recordings with unreliable RR. As shown in the Results section, indeed the 17 cleanest data points resulted in a regression line that was essentially the same as the regression for the 31 cleanest data points by which we mean that any error due to variation in regression was irrelevant (much smaller) compared with inaccuracies caused by variations among subjects. The determination of C1 and C2 was eventually based on the 31 cleanest data points. Next, with these calibration constants we proceeded to answer the overall study goal (Can a single calibration curve provide Spo2 estimates for a population of individuals with acceptable accuracy?). For this, we adopted Arms, a metric used for pulse oximeters to describe accuracy (ISO 80601-2-61, 2011, section 201.12.1.101.2.2). An * superscript is used to indicate at least 1 major difference with the International Organization for Standardization (ISO) standard, which is that in our formulation (Equation 3) we used median values of the traces, thus discarding short-term errors. (3) Because, in our calibration approach, this expression was mathematically similar to that of a standard deviation, we used the χ2 approach to calculate an error estimate for A*rms. A mean difference or “bias” was according to Equation 4, also following the ISO standard. (4) In study C, the protocol was repeated once for each of the 5 individuals. To assess statistical significance among parameters (eg, DC reflectance, PPG signal strength or Spo2,cam–Spo2,ref), caused by exposure to different temperatures, we averaged the values from the 2 sessions and then treated these averages as 5 independent samples (from 5 individuals).

once for each of the 5 individuals. To assess statistical significance among parameters (eg, DC reflectance, PPG signal strength or Spo2,cam–Spo2,ref), caused by exposure to different temperatures, we averaged the values from the 2 sessions and then treated these averages as 5 independent samples (from 5 individuals). Throughout this study, we reported 99% confidence intervals (CIs) because no prior data were used to choose our sample sizes. When Student t tests were used, we reported the P values.

once for each of the 5 individuals. To assess statistical significance among parameters (eg, DC reflectance, PPG signal strength or Spo2,cam–Spo2,ref), caused by exposure to different temperatures, we averaged the values from the 2 sessions and then treated these averages as 5 independent samples (from 5 individuals). Throughout this study, we reported 99% confidence intervals (CIs) because no prior data were used to choose our sample sizes. When Student t tests were used, we reported the P values. RESULTS Filtering the Data Set and Determination of Qthr and Calibration Constants C1, C2 A total of 46 data points from 39 individuals for studies N and H are shown in Figure 5A and listed in Supplemental Table 1 (Supplemental Digital Content, http://links.lww.com/AA/B428). As expected, similar to calibration curves for conventional pulse oximetry, a negative correlation between median RR and median Spo2,ref is seen. Some outlier data points are from signals similar to those in Figure 4B or C, that are obviously corrupted. As expected, similar to calibration curves for conventional pulse oximetry, a negative correlation between median RR and median Spo2,ref. is seen. Some outlier data points are from signals similar to those in Figure 4B or C that are obviously corrupted. As mentioned, we wished to discard such data in an objective manner. We defined an objective quality metric Q and indeed, the signals in Figure 4B and C, had low Q values. However, the question remains which threshold value Qthr to use for this metric to discard data for which the signal corruption is less obvious. In the following paragraph, we describe the process of determining Qthr in an objective manner.

quality metric Q and indeed, the signals in Figure 4B and C, had low Q values. However, the question remains which threshold value Qthr to use for this metric to discard data for which the signal corruption is less obvious. In the following paragraph, we describe the process of determining Qthr in an objective manner. In Figure 5B, we used Qthr as a parameter to discard data with Q is less than Qthr and plot the coefficient of determination (R2) for regression on the remaining data. Three domains can be identified: insufficient filtering, appropriate filtering, and too aggressive filtering. At Qthr <1.4, recordings with corrupted data had a large impact on the regression; only moderate R2 values were found. For 1.4 < Qthr <1.9, R2 was high but more importantly, regression lines were very similar to each other (Figure 5A). For Qthr >1.9, both the R2 values and regression slopes were unstable. Here, the filtering was too aggressive and interindividual variations combined with poor statistics (too few data points) dominated the regression slope. We chose Qthr = 1.4 as the filtering value to obtain a clean set of data for the calibration although any value between 1.4 and 1.9 gave essentially equal calibration results. In other words, the 17 cleanest recordings gave the same regression as the 31 cleanest recordings and those in between (14 regression lines shown in Figure 5A), indicating that the data set was sufficiently large to determine a population calibration.

alue between 1.4 and 1.9 gave essentially equal calibration results. In other words, the 17 cleanest recordings gave the same regression as the 31 cleanest recordings and those in between (14 regression lines shown in Figure 5A), indicating that the data set was sufficiently large to determine a population calibration. Figure 6. The calibration curve (regression line) based on filtered data (Q >1.4) along with its corresponding 99% confidence interval. Discarded data (Q <1.4) not used in calibration are also shown. The regression line for Qthr = 1.4 defined the calibration curve with constants C1 =118.0 and C2 =45.9 (Figure 6). With calibration defined, the next step was to analyze the A*rms as this determined an estimate of the system accuracy. After linear regression on 31 data points, the point estimate of the A*rms was 1.15%. We applied a 1-sided, 99% CI based on the χ2 distribution with 29 degrees of freedom for the root mean square error after a linear regression of our data. This gave an upper limit of 1.65% for A*rms.

n estimate of the system accuracy. After linear regression on 31 data points, the point estimate of the A*rms was 1.15%. We applied a 1-sided, 99% CI based on the χ2 distribution with 29 degrees of freedom for the root mean square error after a linear regression of our data. This gave an upper limit of 1.65% for A*rms. Perturbation of Physiological Conditions and Validation of Calibration Under Normoxic Conditions One source of feedback on the effect of cooling was obtained with reflectance spectra on the forehead. Typically, the reflectance spectra at low temperatures were slightly higher than that at room temperature for wavelengths between 450 and 600 nm that was likely due to a lower blood volume caused by vasoconstriction. Figure 7A shows reflectance spectra for room and cold temperatures for one individual, as well as the average difference for 5 individuals. The difference just reached statistical significance (at 99% CI) for a few yellow and green wavelengths. An increased reflectance in this wavelength region only was consistent with a reduced blood volume at skin depths of about 0.1 to 0.3 mm (Figure 7B).22 These skin layers were dominant in camera-based PPG measurements (Figure 1B). While barely significant, we considered this a first sign of successful perturbation relevant for investigating calibration of camera-based pulse oximetry.

stent with a reduced blood volume at skin depths of about 0.1 to 0.3 mm (Figure 7B).22 These skin layers were dominant in camera-based PPG measurements (Figure 1B). While barely significant, we considered this a first sign of successful perturbation relevant for investigating calibration of camera-based pulse oximetry. Figure 7. Reflectance spectra for one individual (A, left axis) and the average difference for five individuals (B, right axis) shown along with the 99% confidence interval. Typically, reflectance for green and yellow light increases slightly at low temperature although the difference reaches significance at a few wavelengths only. Such a difference is consistent with reduced blood volume in superficial skin (B). B, The relative penetration depths of the various wavelengths indicated. The PPG signal strength strongly reduces with temperature (C) for both finger contact probe and camera.

e difference reaches significance at a few wavelengths only. Such a difference is consistent with reduced blood volume in superficial skin (B). B, The relative penetration depths of the various wavelengths indicated. The PPG signal strength strongly reduces with temperature (C) for both finger contact probe and camera. A stronger manifestation of physiological perturbation was the dependency of PPG signal strength on temperature, for both contact and contactless PPG (Figure 7C). Pulsatile strengths varied considerably at any temperature, which was due to a natural variation among individuals (Supplemental Digital Content, Supplemental Table 3, http://links.lww.com/AA/B428) and the fact that temperature control was not perfect. A strong wind from the cooling ventilators caused us to only record when they were off. Thus, actual temperatures increased quite rapidly during recordings and indicated temperatures per data point were only averages. Nevertheless, the impact of ambient temperature on pulsatility in the finger was significant. A paired, 1-sided t test analysis of the changes of pulsatile strengths with temperature shows that pulsatility at room temperature was larger than that at medium temperatures (P = 0.0096) and similarly for medium and cold temperatures (P = 0.006) for the finger probe. For the forehead, this was similar for room and medium temperature (P = 0.002). For medium and cold temperatures, the difference did not reach a significant level (P = 0.034); here, only 3 data were available with Q is greater than Qthr. Overall, the amplitude reduction at the finger was stronger than on the forehead by approximate factors of 30 and 4, respectively, for the temperature range of 23 to 7°C. More interestingly, however, we observed that the regression lines for red and IR were more or less parallel, which illustrates that the RR, critical for Spo2 calibratability, did not appear to be negatively affected. Please note that the wavelengths used in contact probes were different than those used for the camera, which explains the different red/IR ratios. To investigate calibration robustness with temperature, we did not further analyze the red and IR regression lines because this would have implied that we assumed the Spo2 did not change with temperature. In fact, on average, Spo2,ref increased slightly by 0.21% (±0.08%, 99% CI) per degree Celsius cooling. Instead, we computed Spo2,cam values from the RR and the earlier established calibration constants.

red and IR regression lines because this would have implied that we assumed the Spo2 did not change with temperature. In fact, on average, Spo2,ref increased slightly by 0.21% (±0.08%, 99% CI) per degree Celsius cooling. Instead, we computed Spo2,cam values from the RR and the earlier established calibration constants. The results are shown in Figure 8 combined with the main results of studies N and H to allow an easy visual comparison. The data used for calibration are shown along with the computed upper limit for A*rms (99% CI), indicated as dashed lines. The ISO requirement of Arms <4% is also indicated for reference. Figure 8. Calibration data from studies N and H, shown along with data from study C (normoxic conditions), all computed with one set of calibration constants. The labels “Low,” “Medium,” and “Room” refer to temperatures of approximately 7, 14, and 22°C, respectively.

The results are shown in Figure 8 combined with the main results of studies N and H to allow an easy visual comparison. The data used for calibration are shown along with the computed upper limit for A*rms (99% CI), indicated as dashed lines. The ISO requirement of Arms <4% is also indicated for reference. Figure 8. Calibration data from studies N and H, shown along with data from study C (normoxic conditions), all computed with one set of calibration constants. The labels “Low,” “Medium,” and “Room” refer to temperatures of approximately 7, 14, and 22°C, respectively. It is encouraging that with a different experimental setting, the Spo2,cam values from study C were estimated quite well with the calibration derived from studies N and H. Even though study C explored the impact of ambient temperature on calibratibility rather than validation of the calibration, the data for 2 individuals (40 and 41) that were not in the calibration studies at room temperature can be considered a first validation of the calibration at normoxic conditions. While the mean difference values for room, medium, and cold temperature were small (−0.7%, −0.5%, and −0.7%), the 99% CI intervals were relatively large (±0.7%, ±2.1%, and ±2.4%, respectively), expressing the poor statistics in study C. An indication of whether calibration is robust for cold perturbation is to analyze the pairwise changes in Spo2 discrepancies (Spo2,cam – Spo2,ref) per individual for a change in temperature because such an analysis discards any potential systematic errors resulting from the different experimental setting. With Q is greater than Qthr, we had 5 Spo2,cam – Spo2,ref values at room and medium temperature but only 3 at cold temperature, which gave 5 pairs for room-medium and 3 pairs for medium-cold. An additional comparison was made by comparing the values for room and cold temperature. The average changes in discrepancies for these 3 temperature steps were small: 0.3%, 0.5%, and −0.1% for room-medium, medium-cold, and room-cold, respectively. However, the 99% CI were relatively large: ±2.0%, ±0.7%, and ±1.3%, respectively.

nal comparison was made by comparing the values for room and cold temperature. The average changes in discrepancies for these 3 temperature steps were small: 0.3%, 0.5%, and −0.1% for room-medium, medium-cold, and room-cold, respectively. However, the 99% CI were relatively large: ±2.0%, ±0.7%, and ±1.3%, respectively. DISCUSSION The presented data provide strong evidence that a single calibration curve for a population of healthy adult individuals can be used to estimate Spo2 contactlessly with an acceptable accuracy of A*rms <1.65% (upper 99%, one-sided confidence limit). While noting that we discarded short-term errors, we consider this accuracy acceptable in view of typical Arms values of 2% or 3% for commercial pulse oximeters and the maximum of 4% as specified by the ISO standard (80601-2-61, 2011). In our view, this finding was not trivial because the contactless nature of camera-based pulse oximetry implies interrogation of much shallower skin layers than conventional pulse oximetry.

Arms values of 2% or 3% for commercial pulse oximeters and the maximum of 4% as specified by the ISO standard (80601-2-61, 2011). In our view, this finding was not trivial because the contactless nature of camera-based pulse oximetry implies interrogation of much shallower skin layers than conventional pulse oximetry. The calibration was performed on the basis of data for 26 individuals from a data pool of 39. Using an objective measure, the data for 15 individuals were discarded because subject motion and/or low pulsatile strengths caused the signals to be unreliable for accurate assessment of pulsatile strength. Overall pulsatile strength varied significantly among individuals: the lowest and highest median IR pulsatile strengths in the data with Q > 1.4 were 0.9 × 10−3 and 4.6 × 10−3, respectively. The discarded data tended to have lower average pulsatility than the nondiscarded data (IR pulsatility: 0.9 × 10−3 and 1.2 × 10−3, respectively). Even though we found that 14 regression lines were stable for 1.4 < Q < 1.9, we wanted to verify that the corresponding measured pulsatile strengths were clean, that is, the signal was due to true PPG and not due to motion or noise (camera or lamp). In our approach of assessing pulsatile strength, noise artificially increased pulsatile strength in both channels (red and IR) to the same extent, pushing the RR value closer toward one. If this were the case, the discrepancy (Spo2,cam − Spo2,ref) would then tend to become negative for traces with relatively low pulsatile strength. Analysis of our data (Supplemental Digital Content, Supplemental Table 1, http://links.lww.com/AA/B428) showed that (Spo2,cam − Spo2,ref) in fact slightly trended in the opposite direction (not significant) illustrating that the measured pulsatilities were caused by true PPG (skin color changes) and did not have a significant noise component. Apparently, the approach of discarding unreliable data based on Q provided satisfactory filtering of the data. Moreover, in study C, where PPG amplitudes were reduced significantly, we measured traces with Q >1.4 at a pulsatile strength as low as 0.7 × 10−3.

olor changes) and did not have a significant noise component. Apparently, the approach of discarding unreliable data based on Q provided satisfactory filtering of the data. Moreover, in study C, where PPG amplitudes were reduced significantly, we measured traces with Q >1.4 at a pulsatile strength as low as 0.7 × 10−3. PPG pulse amplitude on the forehead decreased to a much smaller extent than at the finger. This was qualitatively consistent with observations of Bebout and Mannheimer9 who also compared pulsatile strengths on the forehead with the finger. It should be noted, however, that the contact forehead sensor interrogated deeper skin layers than the camera (Figure 1 and comments by Mannheimer17 on referring to such probes as “surface” versus “reflective” probes). Other differences in the present study were that the temperature reduction was slightly larger and acclimatization was shorter. Unfortunately, the discarded group included 5 of 7 people with dark skin, leaving only 2 remaining subjects in the calibration set (subjects 7 and 29, Supplemental Digital Content, Supplemental Table 3, http://links.lww.com/AA/B428). Although not enough data were available for statistical proof, the 2 data points had relatively small errors (0.5% and 1.5%), suggesting that there was not necessarily relevant bias due to melanin concentration. While melanin reduces the signal to noise ratio, we did not see evidence, experimental nor theoretical, of a potential calibration problem in contactless pulse oximetry caused by melanin.

s had relatively small errors (0.5% and 1.5%), suggesting that there was not necessarily relevant bias due to melanin concentration. While melanin reduces the signal to noise ratio, we did not see evidence, experimental nor theoretical, of a potential calibration problem in contactless pulse oximetry caused by melanin. Figure 9. Differences between Spo2,cam and Spo2,ref for 10 second segments with Q >1.4 from all recordings of studies N, H, and C. The calibration constants are validated for a population of 26 individuals that were in the calibration (“+” symbols), and 16 new individuals (“−“ symbols) that were not in the calibration set. Six individuals with fotoype IV or higher are now represented in total and are indicated by “*” in the legend and emphasized by larger symbols in the chart. The A**rms is 2.54% where the asterisks serve to indicate that the individuals are not equally weighed for the calculation.

) that were not in the calibration set. Six individuals with fotoype IV or higher are now represented in total and are indicated by “*” in the legend and emphasized by larger symbols in the chart. The A**rms is 2.54% where the asterisks serve to indicate that the individuals are not equally weighed for the calculation. In this study, we focused on fundamental calibratability based on median values of traces, discarding short-term errors. However, to obtain an impression of general accuracy, we considered all recordings of all individuals and considered segments of 10 seconds. In Figure 9, each point represents the average value for a 10-second segment. We filtered with Q >1.4, per segment, resulting in 2500 segments (59% of all segments). Please note that the data in Figure 9 now include 6 individuals with skin fototype IV or higher (indicated by asterisks in the legend). For data that were in the calibration set, “+” symbols are used, while for the additional data we use “−.” Data for dark-skinned individuals are emphasized by using larger symbols. Some data points feature a relatively large error. It should be noted that the Qthr of 1.4 was derived for traces and is likely not strict enough for the 10-second shorter segments. Also, the data that were not used in the calibration stem from traces with lower Q values; thus, it is not surprising that these data feature larger errors than the segments stemming from the traces with higher Q values. We include this figure to give the reader an impression of data that were discarded in the calibration. Also, it serves to remind that the calibration error A*rms is an underestimate because it discards short-term errors. While noting that some individuals contribute much more than others, the overall A**rms for Figure 9 is 2.54%. Short-term errors can be mitigated with various methods (eg, filtering and error concealment) and of course algorithms that address motion, unlike the approach used in this study. Many factors therefore will change (reduce) the eventual Arms. The value of 2.54% is therefore merely illustrative which is why we did not compute a confidence interval. Nevertheless, we believe that the finding of an A**rms of 2.54%, together with the A*rms <1.65% (upper 99% 1-sided confidence limit) for long-term calibratibility, provides strong evidence for the fundamental feasibility of contactless pulse oximetry.

illustrative which is why we did not compute a confidence interval. Nevertheless, we believe that the finding of an A**rms of 2.54%, together with the A*rms <1.65% (upper 99% 1-sided confidence limit) for long-term calibratibility, provides strong evidence for the fundamental feasibility of contactless pulse oximetry. APPENDIX CONCLUSIONS Calibration of contactless, camera-based pulse oximetry was performed by robust linear regression on 31 data points measured on a population of 26 healthy individuals under normoxic and hypoxic conditions (Spo2 83% – 100%). Discarding short-term errors, an accuracy of A*rms <1.65% (99% one-sided, upper confidence limit) was found which compared well with Arms values for conventional pulse oximeters (typical values are 2% – 3%) and the maximum (4%) allowed by ISO 80601-2-61, 2011. By exposing subjects to temperature changes from room temperature down to cold (about 5°C), discrepancies between Spo2,ref and Spo2,cam were 0.3%, 0.5%, and −0.1% for room-medium, medium-cold, and room-cold, respectively. With 99% CI intervals of ±2.0%, ±0.7%, and ±1.3%, respectively, these accuracies do not necessarily violate the ISO standard. Although further research on more than just 5 individuals is needed to narrow the intervals, these results are encouraging. Challenges such as subject motion and low pulsatile strength have to be addressed to make this new measurement practical and successful. DISCLOSURES Name: Wim Verkruysse, PhD. Contribution: This author was involved in study design, conduct of the study, and manuscript preparation. Name: Marek Bartula, MSc.

By exposing subjects to temperature changes from room temperature down to cold (about 5°C), discrepancies between Spo2,ref and Spo2,cam were 0.3%, 0.5%, and −0.1% for room-medium, medium-cold, and room-cold, respectively. With 99% CI intervals of ±2.0%, ±0.7%, and ±1.3%, respectively, these accuracies do not necessarily violate the ISO standard. Although further research on more than just 5 individuals is needed to narrow the intervals, these results are encouraging. Challenges such as subject motion and low pulsatile strength have to be addressed to make this new measurement practical and successful. DISCLOSURES Name: Wim Verkruysse, PhD. Contribution: This author was involved in study design, conduct of the study, and manuscript preparation. Name: Marek Bartula, MSc. Contribution: This author was involved in study design and manuscript preparation. Name: Erik Bresch, PhD. Contribution: This author was involved in conduct of the study, data analysis, and manuscript preparation. Name: Mukul Rocque, MSc. Contribution: This author was involved in conduct of the study, data analysis, and manuscript preparation. Name: Mohammed Meftah, BSc. Contribution: This author was involved in study design, conduct of the study, and manuscript preparation. Name: Ihor Kirenko, PhD. Contribution: This author was involved in conduct of the study and critical reading. This manuscript was handled by: Maxime Cannesson, MD, PhD.

Contribution: This author was involved in conduct of the study, data analysis, and manuscript preparation. Name: Mohammed Meftah, BSc. Contribution: This author was involved in study design, conduct of the study, and manuscript preparation. Name: Ihor Kirenko, PhD. Contribution: This author was involved in conduct of the study and critical reading. This manuscript was handled by: Maxime Cannesson, MD, PhD. ACKNOWLEDGMENTS We greatly appreciate the invaluable suggestions from Siegfried Kaestle, Andreas Schlack, Alexander Dubielczyk, and Rolf Neumann from Philips Medizin-Systeme, Böblingen, Germany. Furthermore, we also thank our software expert Patriek Bruins from Philips Innovation Services, Eindhoven, The Netherlands. Supplementary Material Published ahead of print May 31, 2016. Funding: None. The authors declare no conflicts of interest. Supplemental digital content is available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal’s website. Reprints will not be available from the authors.

Levobupivacaine and ropivacaine are 2 long-acting local anesthetics that have been developed since the recognition of the severe toxicity caused by bupivacaine. These 2 drugs are pure left-handed enantiomers, and are associated with less potential for both central nervous system and cardiovascular system toxicity than bupivacaine.1,2 However, there have been several clinical reports of cardiovascular collapse induced by both ropivacaine and levobupivacaine.3–10

by bupivacaine. These 2 drugs are pure left-handed enantiomers, and are associated with less potential for both central nervous system and cardiovascular system toxicity than bupivacaine.1,2 However, there have been several clinical reports of cardiovascular collapse induced by both ropivacaine and levobupivacaine.3–10 Comparative studies investigating the systemic toxicity of levobupivacaine and ropivacaine have been published. The cumulative dose of ropivacaine to induce circulatory collapse was greater than that of levobupivacaine in anesthetized ewes and dogs11,12 Additionally, the cumulative doses of ropivacaine that produced dysrhythmias and asystole were larger than corresponding doses of levobupivacaine immediately after the discontinuation of anesthesia in rat models.13 Thus, ropivacaine is required more to produce cardiac collapse than levobupivacaine in ewe, dog, and rat models. In contrast, it has been reported that levobupivacaine is less toxic than bupivacaine, but is no different from ropivacaine for lethality in anesthetized swine.14 In rats, significantly less epinephrine was needed to treat ropivacaine-induced cardiac arrest than to treat levobupivacaine-induced or bupivacaine-induced cardiac arrest under ventilation with 100% oxygen and external cardiac compressions.13 Ropivacaine-induced cardiac arrest also appeared to respond better to epinephrine than cardiac arrest induced by bupivacaine or levobupivacaine.12,13 As noted above, reports of the systemic toxic effects of these 2 local anesthetics have been variable.

r ventilation with 100% oxygen and external cardiac compressions.13 Ropivacaine-induced cardiac arrest also appeared to respond better to epinephrine than cardiac arrest induced by bupivacaine or levobupivacaine.12,13 As noted above, reports of the systemic toxic effects of these 2 local anesthetics have been variable. Use of intravenous (IV) lipid emulsion has been reported as a rescue therapy for local anesthetic toxicity, especially in the situations of local anesthetic–induced cardiovascular collapse.15,16 The mechanism of lipid emulsion therapy is not completely understood, and several new important mechanisms have been proposed; however, its effect may, in part, appear related to its ability to extract bupivacaine (or other lipophilic drugs) from plasma or tissue. This theorized mechanism of action, known as the “lipid sink” effect, may relate to the lipophilicity of local anesthetics.15,16 To the best of our knowledge, there have been no comparisons of lipid emulsion therapy between levobupivacaine-induced and ropivacaine-induced cardiac arrest. To test the hypothesis that lipid therapy is more effective for resuscitation of levobupivacaine-induced (high lipophilicity) cardiac arrest than of ropivacaine-induced (lower lipophilicity) induced cardiac arrest, the difference in efficacy of lipid infusion to treat cardiac arrest induced by these drugs was examined in awake rats.

pothesis that lipid therapy is more effective for resuscitation of levobupivacaine-induced (high lipophilicity) cardiac arrest than of ropivacaine-induced (lower lipophilicity) induced cardiac arrest, the difference in efficacy of lipid infusion to treat cardiac arrest induced by these drugs was examined in awake rats. METHODS Approval was obtained from the institutional animal care committee of Akita University Graduate School of Medicine (a-1-2652) before initiation of the study. This article adheres to the applicable ARRIVE guidelines of the EQUATOR network. Twenty-eight healthy, nonpregnant female Sprague-Dawley rats weighing between 200 and 284 g were studied. The rats were fasted for 12 hours before the experiments, with free access to water.

icine (a-1-2652) before initiation of the study. This article adheres to the applicable ARRIVE guidelines of the EQUATOR network. Twenty-eight healthy, nonpregnant female Sprague-Dawley rats weighing between 200 and 284 g were studied. The rats were fasted for 12 hours before the experiments, with free access to water. Anesthesia was induced in an acryl box, using 5% sevoflurane in a mixture of 33% oxygen in nitrogen. After sedation, the rats were placed on a surgical board, and anesthesia was maintained via mask using 3% sevoflurane in a 33% oxygen in nitrogen mixture, breathing spontaneously. A tracheostomy was performed, followed by tracheal intubation using a 14-gauge plastic cannula. Catheters were inserted into the left femoral artery to monitor blood pressure (BP) and measure arterial blood gases, as well as the right femoral vein for infusion of drugs. The catheters were directed to run subcutaneously in the animals, were pierced over the posterior midthorax, and were fixed by a sutured swivel, allowing the rats to move spontaneously in their cages after they had awakened from anesthesia. Rectal temperature was maintained at 37.5°C ± 0.5°C throughout the procedure by placing the rats on a temperature-controlled heating pad (CMA/150, Stockholm, Sweden) and under a heating lamp.

were fixed by a sutured swivel, allowing the rats to move spontaneously in their cages after they had awakened from anesthesia. Rectal temperature was maintained at 37.5°C ± 0.5°C throughout the procedure by placing the rats on a temperature-controlled heating pad (CMA/150, Stockholm, Sweden) and under a heating lamp. After the procedure, sevoflurane was discontinued, and the rats were left in their cages for 2 hours and allowed to move freely to exclude the effect of residual sevoflurane before commencing the study. Before the experiment, rats were blindly randomized by number drawing into 1 of 2 groups: the levobupivacaine group (n = 14) and the ropivacaine group (n = 14). Each group was then divided into a lipid emulsion group (n = 7) and a control group (n = 7). Before the local anesthetic challenge, BP and heart rate (HR) values were measured, and arterial blood was sampled to analyze arterial blood gases (ABL510; Radiometer Co, Copenhagen, Denmark) to confirm an arterial carbon dioxide tension between 30 and 46 mm Hg, and a pH between 7.35 and 7.46 (breathing spontaneously on room air). An anesthesiologist who was not related to the study prepared the local anesthetic solution, and the researchers who performed this experiment were blinded to the type of local anesthetic that was administered. Throughout the experiment, an electrocardiogram was recorded using 2 subcutaneous needle electrodes, and femoral arterial pressure was also recorded (CM−615G; Nihon Koden, Tokyo, Japan).

hetic solution, and the researchers who performed this experiment were blinded to the type of local anesthetic that was administered. Throughout the experiment, an electrocardiogram was recorded using 2 subcutaneous needle electrodes, and femoral arterial pressure was also recorded (CM−615G; Nihon Koden, Tokyo, Japan). Rats in the levobupivacaine group received levobupivacaine 0.2%, while those in the ropivacaine group received ropivacaine 0.2% at a rate of 2 mg/kg/min. The cumulative doses of local anesthetic required to induce the first seizure activity (a tonic-clonic seizure) and a pulse pressure of 0 mm Hg were calculated.

hetic solution, and the researchers who performed this experiment were blinded to the type of local anesthetic that was administered. Throughout the experiment, an electrocardiogram was recorded using 2 subcutaneous needle electrodes, and femoral arterial pressure was also recorded (CM−615G; Nihon Koden, Tokyo, Japan). Rats in the levobupivacaine group received levobupivacaine 0.2%, while those in the ropivacaine group received ropivacaine 0.2% at a rate of 2 mg/kg/min. The cumulative doses of local anesthetic required to induce the first seizure activity (a tonic-clonic seizure) and a pulse pressure of 0 mm Hg were calculated. When pulse pressure decreased to 0, continuous infusion of local anesthetic was discontinued, and this point was defined as time 0. Mechanical ventilation with 100% oxygen through the tracheostomy tube was started immediately using a rodent ventilator (Ugo Basile Cat. No. 7025; Muromachi Kikai Co, Ltd, Tokyo, Japan) to deliver a tidal volume of 2.5 mL at a rate of 60 breaths/min. At the same time, external manual chest compressions were commenced at a rate of 240/min to obtain a systolic arterial pressure >40 mm Hg. The lipid emulsion group received an IV infusion of 20% lipid emulsion as a 5 mL/kg bolus immediately after commencing mechanical ventilation and chest compressions, followed by a continuous infusion of 0.5 mL/kg/min for 10 minutes (Intralipid 20%; Fresenius Kabi AB, Uppsala, Sweden). Chest compressions were continued to achieve a rate-pressure product (RPP; RPP = systolic pressure × HR) of at least 20% of baseline, which was defined as the criterion for the return of spontaneous circulation. Chest compressions were interrupted for 5 seconds every minute to assess whether the native RPP had increased by >20% of baseline. The operator who was assigned to the role of chest compressions commenced the training 1 month in advance. Competence in the application of chest compressions was assured by the demonstration of 20 consecutive successful applications at a rate of 240/min and at a pressure of 45 mm Hg, using a metronome and weighing scale. The selected operator performed chest compression in all experiments and was blinded to the type of local anesthetic infused. Mechanical ventilation with 100% oxygen was continued from time 0 (when infusion of local anesthetic was discontinued) until the end of the experiment (10-minute time point). Chest compressions and mechanical ventilation with 100% oxygen were stopped at the 10-minute time point, even if native RPP had not increased by >20% of baseline. At the 10-minute time point, arterial blood gas analysis was performed. The saline control group received an infusion of the same volume of saline, similar to the rats receiving lipid emulsion. Other procedures were the same as in the lipid emulsion group.

oint, even if native RPP had not increased by >20% of baseline. At the 10-minute time point, arterial blood gas analysis was performed. The saline control group received an infusion of the same volume of saline, similar to the rats receiving lipid emulsion. Other procedures were the same as in the lipid emulsion group. Power analysis was based on the results of preliminary experiments comparing RPP at 10 minutes between levobupivacaine and ropivacaine groups and yielding a sample size of n = 5 for each group; power was set at 0.8; significance was set at 0.05, and effect size was estimated as 2, with sigma at 0.9. Data from blood gas analysis are expressed as medians and ranges. Values were compared among groups using the Mann-Whitney U test. Mean arterial blood pressure (MAP) and HR values are expressed as mean ± standard deviation. The unpaired Student t test was used to compare the MAP and HR values between the levobupivacaine and ropivacaine groups receiving lipid emulsion, as well as between the lipid emulsion and control groups in each levobupivacaine or ropivacaine group, respectively. The MAP and HR variables of the 2 groups measured across time were analyzed using 1-way repeated-measures analysis of variance and post hoc Student-Newman-Keuls testing (the results are applied to Figure 2). The χ2 test was used to compare the number of rats attaining return of spontaneous circulation at 10 minutes in the levobupivacaine and ropivacaine group treated with lipid emulsion. For all comparisons, P < .05 was considered to be statistically significant.

-Newman-Keuls testing (the results are applied to Figure 2). The χ2 test was used to compare the number of rats attaining return of spontaneous circulation at 10 minutes in the levobupivacaine and ropivacaine group treated with lipid emulsion. For all comparisons, P < .05 was considered to be statistically significant. RESULTS Arterial blood gas values, MAP, and HR before infusion of local anesthetics did not differ among the 4 groups (Table 1). The levobupivacaine and ropivacaine groups showed no significant difference in term of the total amount that induced the first seizure activity (11.5 ± 3.5 vs 12.2 ± 4.3 mg/kg, respectively) and a pulse pressure of 0 mm Hg (18.0 ± 4.3 vs 19.8 ± 4.7 mg/kg, respectively) (Figure 1). MAP and HR did not differ among the 4 groups when pulse pressure decreased to 0 mm Hg (Table 2). Table 1. Baseline (Before Infusion of Local Anesthetics) Physiologic Values Table 2. Mean Arterial Blood Pressure and Heart Rate at a Time of Pulse Pressure = 0 Figure 1. Cumulative doses of levobupivacaine (n = 14) and ropivacaine (n = 14) for inducing convulsions and cardiac arrest (pulse pressure = 0). Data are expressed as mean ± standard deviation. MAP values were higher in the levobupivacaine group than in the ropivacaine group at 2, 3, 4, 5, and 10 minutes after the start of resuscitation with lipid infusion (P < .05). The HR values were higher in the levobupivacaine group than in the ropivacaine group at 5 minutes after the start of resuscitation with lipid infusion (P < .05) (Figure 2, A-1, A-2).

vacaine group than in the ropivacaine group at 2, 3, 4, 5, and 10 minutes after the start of resuscitation with lipid infusion (P < .05). The HR values were higher in the levobupivacaine group than in the ropivacaine group at 5 minutes after the start of resuscitation with lipid infusion (P < .05) (Figure 2, A-1, A-2). Figure 2. A, Changes in mean arterial blood pressure (A-1) and heart rate (A-2) after the start of resuscitation with lipid infusion in the levobupivacaine group (n = 7) and the ropivacaine group (n = 7). Data are expressed as mean ± SD. *P < .05 compared with the ropivacaine group. B, Changes in MAP in the lipid emulsion group and the control group after the start of resuscitation from levobupivacaine-induced cardiac arrest (B-1) and ropivacaine-induced cardiac arrest (B-2) (n = 7 in each group). The same MAP values are presented in the lipid emulsion group of A-1 and B-1. Data are expressed as mean ± SD. *P < .05 compared with the control group. C, Changes in HR in the lipid group and the control group after the start of resuscitation from levobupivacaine-induced cardiac arrest (C-1) and ropivacaine-induced cardiac arrest (C-2) (n = 7 in each group). The same HR values are presented in the lipid emulsion group of A-2 and C-2. Data are expressed as mean ± SD. *P < .05 compared with the control group. HR indicates heart rate; MAP, mean arterial blood pressure; SD, standard deviation.

iac arrest (C-1) and ropivacaine-induced cardiac arrest (C-2) (n = 7 in each group). The same HR values are presented in the lipid emulsion group of A-2 and C-2. Data are expressed as mean ± SD. *P < .05 compared with the control group. HR indicates heart rate; MAP, mean arterial blood pressure; SD, standard deviation. MAP values were higher in the lipid group than in the control group at 2, 3, 4, 5, and 10 minutes after the start of resuscitation from levobupivacaine-induced cardiac arrest (P < .05). However, there was no significant difference in MAP values between the lipid and the control groups in ropivacaine-induced cardiac arrest (Figure 2, B-1, B-2). HRs of the lipid group were higher than that of the control group from 4 to 10 minutes after the initiation of resuscitation from levobupivacaine-induced cardiac arrest; however, in ropivacaine-induced cardiac arrest, significant differences between the lipid and control groups were not found (P < .05) (Figure 2, C-1, C-2). There were no significant differences in MAP and HR values between ropivacaine and levobupivacaine in the control groups. Data from all animals for which native RPP is ≤20% of baseline value were included (Figure 2).

ificant differences between the lipid and control groups were not found (P < .05) (Figure 2, C-1, C-2). There were no significant differences in MAP and HR values between ropivacaine and levobupivacaine in the control groups. Data from all animals for which native RPP is ≤20% of baseline value were included (Figure 2). Table 3 summarizes the number of animals attaining return of spontaneous circulation (native RPP >20% of baseline value) in each group over time. In the lipid-infusion group, all rats were resuscitated successfully at 10 minutes after levobupivacaine-induced cardiac arrest, while 1 of 7 rats had not achieved successful resuscitation 10 minutes after ropivacaine-induced cardiac arrest. In contrast, in the control group, 5 of 7 rats were not successfully resuscitated at 10 minutes after both levobupivacaine-induced and ropivacaine-induced cardiac arrest. There was no significant difference between the numbers of rats attaining return of spontaneous circulation at 10 minutes in the levobupivacaine and ropivacaine group treated with lipid emulsion. Table 3. Animals Attaining Return of Spontaneous Circulation (Native RPP >20% of Baseline Value) for Each Group and Time Table 4. Metabolic Values 10 Minutes After Resuscitation

Table 3 summarizes the number of animals attaining return of spontaneous circulation (native RPP >20% of baseline value) in each group over time. In the lipid-infusion group, all rats were resuscitated successfully at 10 minutes after levobupivacaine-induced cardiac arrest, while 1 of 7 rats had not achieved successful resuscitation 10 minutes after ropivacaine-induced cardiac arrest. In contrast, in the control group, 5 of 7 rats were not successfully resuscitated at 10 minutes after both levobupivacaine-induced and ropivacaine-induced cardiac arrest. There was no significant difference between the numbers of rats attaining return of spontaneous circulation at 10 minutes in the levobupivacaine and ropivacaine group treated with lipid emulsion. Table 3. Animals Attaining Return of Spontaneous Circulation (Native RPP >20% of Baseline Value) for Each Group and Time Table 4. Metabolic Values 10 Minutes After Resuscitation Metabolic values at 10 minutes after resuscitation are shown in Table 4. In the saline group (for both levobupivacaine and ropivacaine), the sample size was reduced to 4 because arterial blood could not be drawn due to the low BP. Base excess and pH values were higher in the levobupivacaine group than in the ropivacaine group when lipid emulsion was infused (P < .05). There were no significant differences in the other values because of the sample size reduction in the saline groups.

4 because arterial blood could not be drawn due to the low BP. Base excess and pH values were higher in the levobupivacaine group than in the ropivacaine group when lipid emulsion was infused (P < .05). There were no significant differences in the other values because of the sample size reduction in the saline groups. DISCUSSION Lipid emulsion treatment was more effective in treating cardiac arrest induced by levobupivacaine than by ropivacaine, although there were no significant differences between the total amount of levobupivacaine and ropivacaine required to induce seizures and cardiac arrest. Lipid therapy was superior to saline in achieving resuscitation in both levobupivacaine-induced and ropivacaine-induced cardiac arrest; however, successful resuscitation from ropivacaine-induced cardiac arrest took more time when compared with levobupivacaine. There have been many reports investigating lipid rescue in bupivacaine-induced cardiovascular collapse.17 However, to the best of our knowledge, no studies have compared the effects of lipid emulsion on the recovery from levobupivacaine-induced versus ropivacaine-induced cardiac arrest. Ohmura et al13 compared the number of successful resuscitations between levobupivacaine-induced and ropivacaine-induced cardiac arrest and concluded that ropivacaine-induced cardiac arrest may be more responsive to treatment than cardiac arrest induced by levobupivacaine. However, the drug used for treating cardiac arrest in their study was epinephrine, not a lipid emulsion.

ations between levobupivacaine-induced and ropivacaine-induced cardiac arrest and concluded that ropivacaine-induced cardiac arrest may be more responsive to treatment than cardiac arrest induced by levobupivacaine. However, the drug used for treating cardiac arrest in their study was epinephrine, not a lipid emulsion. Lipid emulsion therapy has become an additional option of treatment for local anesthetic–induced cardiac arrest.18 However, the effects of lipid infusion on local anesthetic-induced cardiac arrest are not likely uniform and depend on the chemical properties of the local anesthetic concerned. In fact, it has been reported that lipid emulsion improves the recovery from cardiac arrest induced by bupivacaine, but not from cardiac arrest induced by ropivacaine or mepivacaine in the isolated rat heart.19 The lipophilicity of local anesthetics might have an impact on the efficiency of lipid infusions to treat cardiac arrest.20 Because ropivacaine and levobupivacaine have very similar characteristics in terms of pKa and percentage of protein binding,1,2 the key difference is that levobupivacaine is more lipophilic than ropivacaine.21 However, the difference between the lipophilicity of levobupivacaine and ropivacaine may not be sufficient to support our results. In fact, some new mechanisms have been advocated. Large lipid dose could reverse the inhibition of fatty acid metabolism in cardiac mitochondria because lipids are the energy matrix of the heart. These metabolic or other direct cardiac effects may be even more important than the lipid sink because, in more recent studies, very high doses of lipid emulsion have caused a rise in arterial BP, HR, and cardiac blood flow in rats, possibly through inotropic and lusitropic mechanisms.22,23 Moreover, a microemulsion both improves cardiac output and rapidly transports the drug away from organs subject to toxicity.22 Other recent experiments also support these new mechanisms.23 However, we have not proved the relationship between these mechanisms and the difference of the lipid emulsion effects on levobupivacaine and ropivacaine toxicity.

mproves cardiac output and rapidly transports the drug away from organs subject to toxicity.22 Other recent experiments also support these new mechanisms.23 However, we have not proved the relationship between these mechanisms and the difference of the lipid emulsion effects on levobupivacaine and ropivacaine toxicity. One of the characteristic observations in the present study is that elevated MAP was seen from 2 minutes after lipid resuscitation in the levobupivacaine group, whereas MAP elevation did not occur until 5 minutes in the ropivacaine group. However, it would be inappropriate to conclude that lipid emulsion is not effective for ropivacaine-induced cardiac arrest. As the result shows, 6 of 7 rats that received lipid emulsion attained return of spontaneous circulation 10 minutes after ropivacaine-induced cardiac arrest. Meanwhile, only 2 of 7 rats receiving saline recovered after ropivacaine-induced cardiac arrest. This suggests that lipid emulsion therapy for ropivacaine toxicity is not ineffective, but does takes a longer time to achieve its effect. Although the lipophilicity of the local anesthetics concerned may be 1 factor explaining the difference between the 2 groups studied, there may be other important mechanisms at play. Our results suggest that early treatment with lipid infusion is an effective strategy, as recommended in the American Society of Regional Anesthesia and Pain Medicine practice advisory on local anesthetic systemic toxicity,18 especially in ropivacaine-induced cardiac arrest.

e may be other important mechanisms at play. Our results suggest that early treatment with lipid infusion is an effective strategy, as recommended in the American Society of Regional Anesthesia and Pain Medicine practice advisory on local anesthetic systemic toxicity,18 especially in ropivacaine-induced cardiac arrest. It has been reported that lipid emulsion therapy was successful in treating ropivacaine-induced cardiovascular collapse.8,10 However, in previous clinical reports, several different drugs were used during resuscitation efforts. For example, both epinephrine and lipid emulsion were infused to treat a case of ropivacaine-induced cardiovascular collapse.8 Moreover, there have been some clinical reports describing failed reversal of ropivacaine-induced neurotoxicity.24 In these clinical situations, it is uncertain whether lipid emulsion therapy was solely effective in treating ropivacaine-induced cardiovascular collapse.

e of ropivacaine-induced cardiovascular collapse.8 Moreover, there have been some clinical reports describing failed reversal of ropivacaine-induced neurotoxicity.24 In these clinical situations, it is uncertain whether lipid emulsion therapy was solely effective in treating ropivacaine-induced cardiovascular collapse. In the present study, there were no significant differences between the total doses of levobupivacaine and ropivacaine required to induce cardiac arrest. This finding is consistent with some previous reports14 but not with others.11–13 The cumulative dose of ropivacaine to induce circulatory collapse was greater than that of levobupivacaine in anesthetized ewes, dogs and immediately discontinued anesthesia in rat model.11–13 On the other hand, levobupivacaine is less toxic than bupivacaine but was no different from ropivacaine for lethality in anesthetized swine.14 Our result indicates that there was no statistical difference of lethal doses between levobupivacaine and ropivacaine in the awake rat model. One of the causes for this difference may be the definition of “lethal.” We defined lethal as a pulse pressure of 0, while previous reports defined it as a MAP ≤45 mm Hg,12 or the lack of recognizable beat on the ECG for 1 minute after the appearance of the last systole.13 Another cause of this difference may be attributable to whether anesthetized or awake animals were used in the experiment. In our study, general anesthesia was used only in the preparation of the animals. Experiments commenced 2 hours after discontinuation of sevoflurane to exclude the effects of this agent because volatile anesthetic agents may increase the convulsive thresholds of local anesthetics. Sevoflurane is known to attenuate bupivacaine-induced arrhythmias and seizures in rats.25 The doses of local anesthetic required to depress cardiac index and cause asystole were higher in the group receiving volatile anesthesia.26 These 2 factors may account for the differences between the results of previous studies and our study. Therefore, we do not suppose that our findings are necessarily rat specific.

The doses of local anesthetic required to depress cardiac index and cause asystole were higher in the group receiving volatile anesthesia.26 These 2 factors may account for the differences between the results of previous studies and our study. Therefore, we do not suppose that our findings are necessarily rat specific. We used the long-chain fatty acid Intralipid for resuscitation. According to previous studies, long-chain triglyceride is more effective in vivo.27 On the other hand, long-chain and medium-chain triglyceride is more effective in vitro, which may have been predicted based on the partition constants (log P) of these drugs.28 The log P is a measure of the differential solubility of a compound in octanol and water, and thus is a measure of how hydrophilic a substance is, with higher log P values indicating greater hydrophobicity. These 2 studies draw opposite conclusions, and further study is needed to resolve which composition of lipid is effective.

is a measure of the differential solubility of a compound in octanol and water, and thus is a measure of how hydrophilic a substance is, with higher log P values indicating greater hydrophobicity. These 2 studies draw opposite conclusions, and further study is needed to resolve which composition of lipid is effective. There were several limitations to this study. First, serum concentrations of local anesthetics were not measured because it was believed that excessive blood sampling would have affected the results. However, had this been performed, a correlation could have been found between the efficacy of resuscitation measures and serum concentrations of local anesthetics. Second, clinical cardiac collapse induced by local anesthetic agents is usually the result of accidental IV injection of local anesthetic, whereas our protocol used an incremental increase of local anesthetic dosage. A slower rate of infusion, compared with bolus injection, requires larger doses of local anesthetic before the onset of toxicity.29 Third, female rats were used, despite the possible effects of the sexual cycle on local anesthetic toxicity. One of the reasons we used female rats was to obtain fundamental data from females for our future experiments, which we are planning to conduct using pregnant rats. Although pregnant females may have a lower threshold for local anesthetic toxicity,30 the effect of lipid therapy remains unknown. In fact, female humans also encounter local anesthetic toxicity; therefore, study involving female rats will be needed.

re experiments, which we are planning to conduct using pregnant rats. Although pregnant females may have a lower threshold for local anesthetic toxicity,30 the effect of lipid therapy remains unknown. In fact, female humans also encounter local anesthetic toxicity; therefore, study involving female rats will be needed. In conclusion, lipid emulsion therapy was more effective for resuscitation of levobupivacaine-induced cardiac arrest than that induced by ropivacaine. Lipid therapy was effective in our model of ropivacaine-induced cardiac arrest, but has a quicker effect for levobupivacaine-induced cardiac arrest. Our results suggest that the lipophilicity of local anesthetics can influence the efficacy of lipid infusion for treating cardiac arrest induced by these drugs. ACKNOWLEDGMENTS The authors thank Yoshitsugu Tobe, BS, for assistance with preparing the experiment. DISCLOSURES Name: Masashi Yoshimoto, DMD. Contribution: This author helped collect the data and prepare the manuscript. Name: Takashi Horiguchi, MD. Contribution: This author helped design and conduct the study, perform the analysis, and write the manuscript. Name: Tetsu Kimura, MD. Contribution: This author helped review the original study data and revise the manuscript. Name: Toshiaki Nishikawa, MD. Contribution: This author helped review the original study data and revise the manuscript. This manuscript was handled by: Markus W. Hollmann, MD, PhD. Published ahead of print September 2, 2017. Funding: This study was supported by a grant in aid for scientific research from the Japan Society for the Promotion of Science (No. 26861221).

Contribution: This author helped review the original study data and revise the manuscript. This manuscript was handled by: Markus W. Hollmann, MD, PhD. Published ahead of print September 2, 2017. Funding: This study was supported by a grant in aid for scientific research from the Japan Society for the Promotion of Science (No. 26861221). The authors declare no conflicts of interest. Reprints will not be available from the authors.

KEY POINTS Question: Does perioperative management affect long-term outcomes in patients after lung cancer surgery? Findings: Certain factors in particular perioperative dexamethasone and flurbiprofen axetil therapy may improve patients’ long-term survival after surgery for non–small-cell lung cancer. Meaning: Further studies to determine whether combined use of perioperative dexamethasone and nonsteroid anti-inflammatory drugs improves patients’ long-term survival after lung cancer surgery are urgently needed. Cancer is the leading cause of death worldwide. Global cancer statistics1 showed that about 14.1 million new cancer cases were diagnosed in 2012; among them 1.8 million were lung cancer cases, accounting for 13% of the total cancer diagnosis. Lung cancer is the primary cause of cancer deaths among men globally and among women in the developed countries. According to China’s cancer statistics,2 733,300 new lung cancer cases (509,300 men and 224,000 women) were diagnosed in 2015, accounting for 17.1% of all new cancer diagnosis; 610,200 lung cancer patients (432,400 men and 177,800 women) died during the same period, accounting for 21.7% of all cancer deaths. The incidence and mortality of lung cancer are among the highest of all malignant tumors; it has the highest incidence and mortality in men and the second incidence (lower than breast cancer) but the highest mortality in women.2 The 5-year survival rate after lung cancer surgery remains low.3

1.7% of all cancer deaths. The incidence and mortality of lung cancer are among the highest of all malignant tumors; it has the highest incidence and mortality in men and the second incidence (lower than breast cancer) but the highest mortality in women.2 The 5-year survival rate after lung cancer surgery remains low.3 Surgical resection is the first-line treatment for non–small-cell lung cancer (NSCLC). However, it is unavoidable that some tumor cells will be disseminated into the blood or the lymphatic systems during surgery. The outcome depends on the balance between tumor-promoting factors and immune function of the body during the perioperative period. Studies showed that stress response induced by surgery attenuates the cytotoxic effect of natural killer cells and the reaction of T cells, and thus inhibits the cell-mediated immunity.4 Indeed, immunosuppression occurs within hours after surgery and lasts for several days, depending on the severity of surgical trauma.5 In addition to surgery per se, the anesthesia management during perioperative period, including type of anesthesia, anesthetic drugs, blood transfusion, and hypothermia, can all affect the immune function of patients; for example, volatile anesthetics and opioids might aggravate the immunosuppression and potentially worsen long-term outcome, whereas regional anesthesia and nonsteroidal anti-inflammatory drugs (NSAIDs) might attenuate the immunosuppression and exert protective effects.6,7 These indicate that perioperative management may contribute to the long-term outcome of patients after lung cancer surgery. The purpose of this study was, therefore, to identify factors that were closely related to perioperative management and beyond in affecting patients’ long-term survival after surgery for NSCLC.

,7 These indicate that perioperative management may contribute to the long-term outcome of patients after lung cancer surgery. The purpose of this study was, therefore, to identify factors that were closely related to perioperative management and beyond in affecting patients’ long-term survival after surgery for NSCLC. METHODS This was a retrospective cohort study of prospectively collected data. The study protocol was approved by the Clinical Research Ethics Committee of Beijing University Cancer Hospital (ethics approval number 2014[074]). Considering that the study was observational and that patients who would be enrolled in this study underwent surgery years ago and lived in different regions nationwide, the Ethics Committee agreed to exempt the written informed consent, but all enrolled patients had verbally agreed to participate for long-term outcome follow-up. This manuscript adheres to the applicable Enhancing the QUAlity and Transparency Of health Research (EQUATOR) guidelines. Patients Patients who underwent lung cancer surgery from January 1, 2006, to December 31, 2009, in Beijing University Cancer Hospital were screened using the electronic medical records system. The inclusion criteria were as follows: (1) age ≥18 years; and (2) the diagnosis of NSCLC was confirmed by postoperative pathological examination. Patients were excluded if they met any of the following criteria: (1) complicated with primary malignant tumor in other place; (2) recurrent metastatic lung tumor; (3) long-term steroid exposure; or (4) impossible follow-up due to incomplete information.

of NSCLC was confirmed by postoperative pathological examination. Patients were excluded if they met any of the following criteria: (1) complicated with primary malignant tumor in other place; (2) recurrent metastatic lung tumor; (3) long-term steroid exposure; or (4) impossible follow-up due to incomplete information. Anesthesia, Surgery, and Perioperative Management General anesthesia with double-lumen endobronchial tube intubation was performed for all patients. Anesthesia was induced with intravenous anesthetics (propofol and/or etomidate) and opioids (fentanyl or sufentanil), and maintained with inhalational anesthetics (sevoflurane or isoflurane, with or without nitrous oxide) and opioids (fentanyl and/or sufentanil). Epidural anesthesia with local anesthetics (lidocaine and/or ropivacaine) was also performed in some patients (combined with general anesthesia) according to the preference of anesthesiologists. For some patients, dexamethasone (5–10 mg) was administered for the prevention of postoperative nausea and vomiting (PONV), and flurbiprofen axetil (50–100 mg) was administered as a supplemental analgesia according to the decision of anesthesiologists during anesthesia.

according to the preference of anesthesiologists. For some patients, dexamethasone (5–10 mg) was administered for the prevention of postoperative nausea and vomiting (PONV), and flurbiprofen axetil (50–100 mg) was administered as a supplemental analgesia according to the decision of anesthesiologists during anesthesia. Lung resections were performed through a standard posterolateral thoracotomy. Lobectomy with mediastinal lymph node dissection was the standard procedure, but other surgical procedures (such as pneumonectomy, wedge resection, bronchial sleeve resection, or even tumor resection/sampling, with or without mediastinal lymph node dissection) were also performed according to the situations of tumor and decisions of surgeons. Postoperative analgesia was provided with patient-controlled analgesia pumps, which were established with ropivacaine (with or without fentanyl) for epidural analgesia or opioids (morphine or sufentanil) for intravenous analgesia. For some patients, flurbiprofen axetil (100–200 mg) was added into the intravenous analgesia pump according to the decision of anesthesiologists. PONV were treated with dexamethasone (5–10 mg), 5-hydroxytryptamine-3 receptor antagonists, or metoclopramide. Other perioperative managements were performed according to routine practice.

nts, flurbiprofen axetil (100–200 mg) was added into the intravenous analgesia pump according to the decision of anesthesiologists. PONV were treated with dexamethasone (5–10 mg), 5-hydroxytryptamine-3 receptor antagonists, or metoclopramide. Other perioperative managements were performed according to routine practice. Data Collection Data were collected using the medical record system and included demographic characteristics (age, gender, height, and weight), preoperative information (surgical diagnosis, comorbidity, American Society of Anesthesiologists classification, tumor location, and history of chemotherapy for cancer), anesthesia-related information (type of anesthesia, uses of anesthetic drugs, intraoperative blood infusion, uses and doses of glucocorticoids, and NSAIDs), surgery-related information (type of surgical procedure, duration of surgery, performance of mediastinal lymph node dissection), and postoperative information (occurrence of complications, maximal tumor size, pathological diagnosis, grade of tumor cell differentiation, chemotherapy, and/or radiotherapy). The maximal tumor size was the results reported by the pathologists by measuring the resected tumor specimens. The total consumption of opioids during and after surgery was converted into fentanyl equivalents (1 μg fentanyl equals to sufentanil 0.1 μg or morphine 110 μg).8

entiation, chemotherapy, and/or radiotherapy). The maximal tumor size was the results reported by the pathologists by measuring the resected tumor specimens. The total consumption of opioids during and after surgery was converted into fentanyl equivalents (1 μg fentanyl equals to sufentanil 0.1 μg or morphine 110 μg).8 Postoperative Follow-up All patients were followed up by specially assigned personnel from the Department of Medical Records and Statistics of Peking University Cancer Hospital. Follow-up was performed in the way of outpatient review, telephone inquiry, or letter communication. Patients were followed up every 6 months within the first year after surgery, and once a year thereafter. The living status and the recurrence of tumor were confirmed during each follow-up. In case of tumor recurrence, the time of diagnosis was recorded; in case of patient death, the time of death was also documented. Tumor recurrence including local recurrence or distant recurrence/metastasis was confirmed by imagological examination.9 The time of recurrence was the earliest date of clinical diagnosis made by surgeons or radiologists according to imagological evidence. The time of death was the date appeared in the medical certificate of death. Follow-up was terminated when patients died or were lost to follow-up. The recurrence-free survival (RFS) and overall survival (OS) were calculated according to the follow-up results. Data assignment was performed according to a previous published study similar to the current one.10

in the medical certificate of death. Follow-up was terminated when patients died or were lost to follow-up. The recurrence-free survival (RFS) and overall survival (OS) were calculated according to the follow-up results. Data assignment was performed according to a previous published study similar to the current one.10 The primary outcome was OS, which is defined as the duration from the date of surgery to the date of death from any cause. The secondary outcome was RFS, which is defined as the duration from the date of surgery to the date of recurrence or death for any cause, whichever happened first.

in the medical certificate of death. Follow-up was terminated when patients died or were lost to follow-up. The recurrence-free survival (RFS) and overall survival (OS) were calculated according to the follow-up results. Data assignment was performed according to a previous published study similar to the current one.10 The primary outcome was OS, which is defined as the duration from the date of surgery to the date of death from any cause. The secondary outcome was RFS, which is defined as the duration from the date of surgery to the date of recurrence or death for any cause, whichever happened first. Statistical Analysis Numeric data with abnormal distribution were presented as median (interquartile range [IQR]). Categorical data were presented as numbers (%). Patients with missing data were presented as numbers (%). Univariate analyses were performed using the Kaplan–Meier survival analysis against outcomes (RFS or OS), with comparisons between layers of baseline and perioperative variables performed with log-rank tests. For continuous variables, the choices of cutoff points between layers were made according to clinical significance, literatures, or median values.11–13 Factors that were possibly associated with the outcomes (set as P < .20 in log-rank tests or were regarded as clinically important) were included in the Cox proportional hazard model for multivariable analysis to identify independent factors that were associated with outcomes (P < .05). The survival curves in patients with 3 combinations of perioperative flurbiprofen axetil and/or dexamethasone were compared to a reference group (no use of both) by log-rank test, and the criterion of significance was adjusted with Bonferroni correction (P < .05/3 = .0167). All tests were 2-sided. Statistical analysis was performed using the SPSS version 14.0 (SPSS, Inc, Chicago, IL).

e flurbiprofen axetil and/or dexamethasone were compared to a reference group (no use of both) by log-rank test, and the criterion of significance was adjusted with Bonferroni correction (P < .05/3 = .0167). All tests were 2-sided. Statistical analysis was performed using the SPSS version 14.0 (SPSS, Inc, Chicago, IL). Although no formal sample size calculation was performed beforehand, the high number of events (nearly 300 deaths) compared with the number of Cox model variables (15 or 16) indicated that the “ten events per variable” rule was exceeded, implying the reliability of the regression estimates.14 RESULTS Patient Recruitment A total of 676 patients underwent lung cancer surgery between January 1, 2006, and December 31, 2009; 632 patients met the inclusion/exclusion criteria; 588 patients completed long-term follow-up and were included in the final statistical analysis (Figure 1). The postoperative follow-up was ended on December 31, 2015. Figure 1. Flowchart of the study.

RESULTS Patient Recruitment A total of 676 patients underwent lung cancer surgery between January 1, 2006, and December 31, 2009; 632 patients met the inclusion/exclusion criteria; 588 patients completed long-term follow-up and were included in the final statistical analysis (Figure 1). The postoperative follow-up was ended on December 31, 2015. Figure 1. Flowchart of the study. Results of Follow-up Baseline, perioperative, and follow-up data were presented in Supplemental Digital Content, Tables A–E, http://links.lww.com/AA/C285. The median interval from the date of surgery to the last follow-up was 5.2 years (IQR, 2.0–6.8). At the time of the last follow-up, 297 patients (50.5%) died. Among survivors, 32 patients had tumor recurrence; resulting an overall recurrence/death rate of 56.0%. The median duration of RFS was 53.5 months (IQR, 14.9–78.1), and the median duration of OS was 64.3 months (IQR, 28.5–81.6). The RFS and OS rates at the first, third, and fifth year after surgery were 81.6% (standard error, ±1.6%), 59.2% (±2.0%), 48.1% (±2.1%), and 90.8% (±1.2%), 70.0% (±1.9%), 57.1% (±2.0%), respectively.

median duration of RFS was 53.5 months (IQR, 14.9–78.1), and the median duration of OS was 64.3 months (IQR, 28.5–81.6). The RFS and OS rates at the first, third, and fifth year after surgery were 81.6% (standard error, ±1.6%), 59.2% (±2.0%), 48.1% (±2.1%), and 90.8% (±1.2%), 70.0% (±1.9%), 57.1% (±2.0%), respectively. Risk Factor Analysis Recurrence-Free Survival. Fifteen factors that were identified by univariate analysis (P < .20) (Table 1) or considered clinically important were included in the multivariable Cox proportional hazard model. Multivariable analysis identified 7 independent factors; among them increasing tumor size (hazard ratio [HR], 1.23; 95% confidence interval [CI], 1.12–1.34; P < .001), postoperative radiotherapy (HR, 1.72; 95% CI, 1.06–2.78; P = .027), and postoperative chemotherapy (HR, 1.60; 95% CI, 1.27–2.02; P < .001) were associated with early recurrence, whereas high body mass index (BMI) grade (HR, 0.85; 95% CI, 0.72–1.00; P = .046), highly differentiated tumor (HR, 0.61; 95% CI, 0.40–0.92; P = .019), mediastinal lymph node dissection during surgery (HR, 0.52; 95% CI, 0.35–0.77; P = .001), and perioperative use of dexamethasone (HR, 0.70; 95% CI, 0.55–0.89; P = .004) were associated with delayed recurrence (Table 2). Table 1. Predictors of Survival (Kaplan–Meier Univariate Analyses) Table 2. Predictors of Survival (Multivariate Cox Proportional Hazard Model)

Risk Factor Analysis Recurrence-Free Survival. Fifteen factors that were identified by univariate analysis (P < .20) (Table 1) or considered clinically important were included in the multivariable Cox proportional hazard model. Multivariable analysis identified 7 independent factors; among them increasing tumor size (hazard ratio [HR], 1.23; 95% confidence interval [CI], 1.12–1.34; P < .001), postoperative radiotherapy (HR, 1.72; 95% CI, 1.06–2.78; P = .027), and postoperative chemotherapy (HR, 1.60; 95% CI, 1.27–2.02; P < .001) were associated with early recurrence, whereas high body mass index (BMI) grade (HR, 0.85; 95% CI, 0.72–1.00; P = .046), highly differentiated tumor (HR, 0.61; 95% CI, 0.40–0.92; P = .019), mediastinal lymph node dissection during surgery (HR, 0.52; 95% CI, 0.35–0.77; P = .001), and perioperative use of dexamethasone (HR, 0.70; 95% CI, 0.55–0.89; P = .004) were associated with delayed recurrence (Table 2). Table 1. Predictors of Survival (Kaplan–Meier Univariate Analyses) Table 2. Predictors of Survival (Multivariate Cox Proportional Hazard Model) Overall Survival. Sixteen factors that were identified by univariate analysis (P < .20) (Table 1) or considered clinically important were included in the multivariable Cox proportional hazard model. Multivariable analysis identified 6 independent factors; among them limited resection (HR, 1.46; 95% CI, 1.08–1.98; P = .013) and increasing tumor size (HR, 1.29; 95% CI, 1.17–1.42; P < .001) were associated with shortened survival, whereas high BMI grade (HR, 0.82; 95% CI, 0.69–0.97; P = .021), highly differentiated tumor (HR, 0.59; 95% CI, 0.37–0.93; P = .024), mediastinal lymph node dissection during surgery (HR, 0.45; 95% CI, 0.30–0.67; P < .001), and perioperative use of dexamethasone (HR, 0.70; 95% CI, 0.54–0.90; P = .006) were associated with prolonged survival. Perioperative use of flurbiprofen axetil (HR, 0.80; 95% CI, 0.62–1.03; P = .086) was not associated with prolonged survival (Table 2).

issection during surgery (HR, 0.45; 95% CI, 0.30–0.67; P < .001), and perioperative use of dexamethasone (HR, 0.70; 95% CI, 0.54–0.90; P = .006) were associated with prolonged survival. Perioperative use of flurbiprofen axetil (HR, 0.80; 95% CI, 0.62–1.03; P = .086) was not associated with prolonged survival (Table 2). Joint Effects of Perioperative Flurbiprofen Axetil and Dexamethasone The survival curves in patients with 4 combinations of perioperative flurbiprofen axetil (yes, no) and dexamethasone (yes, no) showed a possible additive effect in improving survival (Figure 2). Patients who received both flurbiprofen axetil and dexamethasone had a better survival than those who did not received both (χ2 = 11.494; P = .001), with the estimated 5-year survival rates of 68.0% and 46.1%, respectively (Table 3). After adjustment with confounding factors, administrating both flurbiprofen axetil and dexamethasone was associated with prolonged OS when compared to no use of both (adjusted HR, 0.57; 95% CI, 0.38–0.84; P = .005) (Table 4). Table 3. The Overall Survival Rates of Patients With 4 Combinations of Perioperative FA and DXM Table 4. Joint Effects of Perioperative FA and/or DXM on Overall Survival (Cox Proportional Hazard Model) Figure 2. Survival curves of patients with or without the perioperative administration of FA and DXM alone or in combination. Patients who received both FA and DXM showed survival better than those who received none of them (P = .001). The criterion of significance after Bonferroni correction was P < .0167. DXM indicates dexamethasone; FA, flurbiprofen axetil.

or without the perioperative administration of FA and DXM alone or in combination. Patients who received both FA and DXM showed survival better than those who received none of them (P = .001). The criterion of significance after Bonferroni correction was P < .0167. DXM indicates dexamethasone; FA, flurbiprofen axetil. DISCUSSION In this retrospective cohort study, 588 patients with primary NSCLC after surgical resection were followed up for a median of 5.2 years. The OS rates were 90.8%, 70.0%, and 57.1% at the first, third, and fifth year after surgery. Multivariable Cox proportional hazard analysis showed that limited resection and increasing tumor size were associated with shortened survival, whereas high BMI grade, highly differentiated tumor, mediastinal lymph node dissection, and perioperative use of dexamethasone were related to longer survival after surgery. Perioperative use of flurbiprofen axetil was not associated with longer survival; however, combined administration of dexamethasone and flurbiprofen axetil showed additive effect in prolonging survival.

iastinal lymph node dissection, and perioperative use of dexamethasone were related to longer survival after surgery. Perioperative use of flurbiprofen axetil was not associated with longer survival; however, combined administration of dexamethasone and flurbiprofen axetil showed additive effect in prolonging survival. Long-term survival remains low in patients after lung cancer surgery. In a systematic review by Whitson et al,15 the mean 5-year OS rate was 65.6% (95% CI, 56.7–74.4) in patients after thoracotomy lobectomy for early-stage NSCLC. A later systematic review showed that the 5-year survival ranged from 58% to 97% after thoracotomy for lung cancer.16 In our patients, the 5-year survival rate (57.1%) was slightly lower than the previously reported mean results, but it was in accord with results in Chinese patients, that is, a 5-year survival rate of 54.0% after surgery for stage I lung cancer.17

ar survival ranged from 58% to 97% after thoracotomy for lung cancer.16 In our patients, the 5-year survival rate (57.1%) was slightly lower than the previously reported mean results, but it was in accord with results in Chinese patients, that is, a 5-year survival rate of 54.0% after surgery for stage I lung cancer.17 It is reported that BMI is inversely proportional to the risk of lung cancer.12 Indeed, a previous study18 reported that higher BMI was associated with improved OS after surgical resection of NSCLC patients. In young patients with advanced NSCLC, BMI <25 kg·m2 was a negative prognostic factor.19 Our results also showed that high BMI grade was associated with delayed recurrence and longer OS. Histological grade has been known to be a significant prognostic factor for survival of NSCLC patients.20 In line with previous studies, we found that highly differentiated tumor was associated with a low risk of recurrence and a long OS. Tumor size is also a well-known prognostic factor for many cancers including NSCLC, with larger size predicting a worse prognosis in most cases.11 In the present study, tumor size was stratified into 5 grades (ie, ≤1.0, 1.1–2.0, 2.1–3.0, 3.1–4.0, and ≥4.1 cm), and we found that each grade increase was associated with significant higher risks of both recurrence and death.

many cancers including NSCLC, with larger size predicting a worse prognosis in most cases.11 In the present study, tumor size was stratified into 5 grades (ie, ≤1.0, 1.1–2.0, 2.1–3.0, 3.1–4.0, and ≥4.1 cm), and we found that each grade increase was associated with significant higher risks of both recurrence and death. Intraoperative lymph node scavenge is closely related to postoperative outcome; complete mediastinal lymph node dissection reduces the risk of recurrence and death after surgery.21 However, for patient with lymph node stages N0 or N1 (less than hilar) NSCLC, complete lymphadenectomy during pulmonary resection did not improve survival.22 The present study enrolled patients with various stages of NSCLC. Our results showed that mediastinal lymph node dissection during surgery was associated with delayed recurrence and prolonged survival. Lobectomy or greater resection remains the treatment of choice for patients with early-stage NSCLC and is associated with better outcome.23 Our results also showed that, compared to lobectomy or pneumonectomy, limited resection was associated with short survival. Other reasons that might have lead to our results were that limited resection was usually performed in patients with more severe comorbidity (such as decreased pulmonary function) or advanced-stage NSCLC.

lts also showed that, compared to lobectomy or pneumonectomy, limited resection was associated with short survival. Other reasons that might have lead to our results were that limited resection was usually performed in patients with more severe comorbidity (such as decreased pulmonary function) or advanced-stage NSCLC. Dexamethasone is frequently used to prevent PONV.24 High-dose dexamethasone (30 mg·kg−1) can cause significant immunosuppression,25 but the dose of dexamethasone for the purpose of PONV prevention is usually 8–10 mg or less. Glucocorticoids have long been used to treat hematological malignancies via apoptotic pathway.26 However, data from cancer cell lines derived from various “solid” tumors suggested that glucocorticoids, for example, dexamethasone, can inhibit chemo-induced cancer apoptosis and promote cancer cell growth.27 The underlying molecular mechanisms of these phenomena remain unknown but may be somehow attributed to the functional loss or less expression of glucocorticoid receptor depending on certain cancer type.28 For example, in patients after colorectal cancer surgery, perioperative dexamethasone was associated with an increased risk of recurrence29,30; in patients with endometrial cancer, the administration of dexamethasone was not associated with an increased risk of recurrence.31 In contrast, in patients with pancreatic cancer, perioperative dexamethasone was associated with improved long-term survival.32 Our results showed that the use of dexamethasone was associated with a prolonged RFS and OS after lung cancer surgery.

dexamethasone was not associated with an increased risk of recurrence.31 In contrast, in patients with pancreatic cancer, perioperative dexamethasone was associated with improved long-term survival.32 Our results showed that the use of dexamethasone was associated with a prolonged RFS and OS after lung cancer surgery. NSAIDs are commonly used during the perioperative period to improve analgesia and reduce opiate consumption by inhibiting the activity of cyclooxygenase (COX) and the synthesis of prostaglandins. Prostaglandin E2 can selectively inhibit the activity of macrophage, neutrophils, T-helper cells, and natural killer cells,33 whereas blocking prostaglandins synthesis with COX inhibitors decreases tumor angiogenesis and induces tumor cell apoptosis.4 Retrospective studies showed that intraoperative use of ketorolac or diclofenac was associated with better outcome in patients undergoing breast cancer surgery.34,35 In lung cancer patients, the observational study by Forget et al35 showed that NSAIDs use at the beginning of the surgery was independently associated with a lower metastases risk; the retrospective study by Choi et al36 found that ketorolac administration was slightly associated with better OS (P = .05) in univariate analysis. In this study, flurbiprofen axetil (a nonselective COX inhibitor with high binding affinity to the site of lesion) was the only NSAID administrated during the perioperative period. Although the use of flurbiprofen axetil was not associated with long-term survival, we found an additive effect of dexamethasone and flurbiprofen axetil in prolonging postoperative survival, possibly due to relieved immunosuppression action in patients after cancer surgery.37–39 Considering the widespread use of these 2 drugs during the perioperative period, their effects in particular long-term effects on lung cancer patients indeed need to be explored further.

axetil in prolonging postoperative survival, possibly due to relieved immunosuppression action in patients after cancer surgery.37–39 Considering the widespread use of these 2 drugs during the perioperative period, their effects in particular long-term effects on lung cancer patients indeed need to be explored further. Our results also showed that postoperative radiotherapy and chemotherapy were associated with short RFS. We attributed the phenomena to the use of these therapies in patients with advanced-stage NSCLC,40 a common practice during the study period. Except the retrospective nature of our study, several other limitations also exist. First, cancer stage was not included in the analyses because these data were lacking in many patients. However, we included tumor size and differentiation grade in the regression model, which are closely related with cancer staging and patients’ outcomes. Second, some perioperative data such as intraoperative body temperature and blood glucose level were not included because these parameters were not routinely monitored during that period of time when surgeries were performed. Third, as a monocenter study, our results may not be extrapolated to patients in other centers. Nevertheless, our results showed for the first time that perioperative dexamethasone and flurbiprofen axetil in combination might produce a synergic effect in improving survival although these observational findings require further clarification with randomized clinical trials.

xtrapolated to patients in other centers. Nevertheless, our results showed for the first time that perioperative dexamethasone and flurbiprofen axetil in combination might produce a synergic effect in improving survival although these observational findings require further clarification with randomized clinical trials. CONCLUSIONS Our results showed that, for NSCLC patients, the OS rates at the first, third, fifth year after surgery were 90.8%, 70.0%, and 57.1%, respectively. Limited resection and increasing tumor size were associated with a shortened OS, whereas high BMI grade, highly differentiated tumor, mediastinal lymph node dissection during surgery, and perioperative use of dexamethasone were associated with a prolonged OS after surgery. Although no association was found between perioperative use of flurbiprofen axetil and long survival, combined administration of dexamethasone and flurbiprofen axetil showed an additive effect in prolonging survival. Considering the small sample size and retrospective nature of the study, the aforementioned findings should be interpreted with caution. Further studies in particular to determine whether combined use of perioperative dexamethasone and NSAIDs therapy improves patients’ long-term survival after lung cancer surgery are urgently needed. ACKNOWLEDGMENTS The authors gratefully acknowledge Dr Yue Yang, MD (Professor, Department of Thoracic Surgery II, Peking University Cancer Hospital, Beijing 100142, China) for his help in data collection. DISCLOSURES Name: Wen-Wen Huang, MD.

CONCLUSIONS Our results showed that, for NSCLC patients, the OS rates at the first, third, fifth year after surgery were 90.8%, 70.0%, and 57.1%, respectively. Limited resection and increasing tumor size were associated with a shortened OS, whereas high BMI grade, highly differentiated tumor, mediastinal lymph node dissection during surgery, and perioperative use of dexamethasone were associated with a prolonged OS after surgery. Although no association was found between perioperative use of flurbiprofen axetil and long survival, combined administration of dexamethasone and flurbiprofen axetil showed an additive effect in prolonging survival. Considering the small sample size and retrospective nature of the study, the aforementioned findings should be interpreted with caution. Further studies in particular to determine whether combined use of perioperative dexamethasone and NSAIDs therapy improves patients’ long-term survival after lung cancer surgery are urgently needed. ACKNOWLEDGMENTS The authors gratefully acknowledge Dr Yue Yang, MD (Professor, Department of Thoracic Surgery II, Peking University Cancer Hospital, Beijing 100142, China) for his help in data collection. DISCLOSURES Name: Wen-Wen Huang, MD. Contribution: This author helped design the study; perform data collection; and analyze, draft, and revise the manuscript. Conflicts of Interest: None. Name: Wen-Zhi Zhu, MD. Contribution: This author helped design the study, perform data collection, and analyze and draft the manuscript. Conflicts of Interest: None. Name: Dong-Liang Mu, MD.

Contribution: This author helped design the study; perform data collection; and analyze, draft, and revise the manuscript. Conflicts of Interest: None. Name: Wen-Zhi Zhu, MD. Contribution: This author helped design the study, perform data collection, and analyze and draft the manuscript. Conflicts of Interest: None. Name: Dong-Liang Mu, MD. Contribution: This author helped conceive and design the study. Conflicts of Interest: None. Name: Xin-Qiang Ji, MD. Contribution: This author helped collect and analyze the data. Conflicts of Interest: None. Name: Xiao-Lu Nie, MSc. Contribution: This author helped in statistical analysis. Conflicts of Interest: None. Name: Xue-Ying Li, MSc. Contribution: This author helped in statistical analysis. Conflicts of Interest: None. Name: Dong-Xin Wang, MD, PhD. Contribution: This author helped conceive and design the study, review the original data and the results of analyses, and critically revise the manuscript. Conflicts of Interest: D.-X. Wang has received travel funding for overseas lectures from Beijing Tide Pharmaceutical Co., Ltd. Name: Daqing Ma, MD, PhD, FRCA. Contribution: This author helped revise the manuscript. Conflicts of Interest: D. Ma is a board member of British Journal of Anaesthesia. This manuscript was handled by: Scott M. Fishman, MD. Supplementary Material Published ahead of print March 6, 2018.

Conflicts of Interest: D.-X. Wang has received travel funding for overseas lectures from Beijing Tide Pharmaceutical Co., Ltd. Name: Daqing Ma, MD, PhD, FRCA. Contribution: This author helped revise the manuscript. Conflicts of Interest: D. Ma is a board member of British Journal of Anaesthesia. This manuscript was handled by: Scott M. Fishman, MD. Supplementary Material Published ahead of print March 6, 2018. Funding: This study was supported by departmental funding from the Department of Anesthesiology and Critical Care Medicine, Peking University First Hospital, Beijing, China. D.M. is supported by British Oxygen Company Chair grant, Royal College of Anaesthetists, and British Journal of Anaesthesia Fellowship grant, London, United Kingdom. Conflicts of Interest: See Disclosures at the end of the article. Supplemental digital content is available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal’s website. W.-W. Huang and W.-Z. Zhu contributed equally and share first authorship. Reprints will not be available from the authors.

pain and anxiety with an effect size of 0.35.48 Because VR exposure as a preparation tool for medical procedures is a fairly unexplored area of research, it is not (yet) possible to compare effect sizes for VR preparation to other forms of preparative interventions to reduce pain and anxiety during medical procedures. The studies in the current systematic review and meta-analysis varied in quality. Most studies applied randomization and clearly described their inclusion and exclusion criteria. However, concealed treatment allocation was often not guaranteed and intention-to-treat analyses were often not performed. Also, very few studies focused on possible moderating factors of VR effectivity, such as anxiety sensitivity and temperament.

KEY POINTS Question: Is virtual reality (VR) effective in reducing pain and anxiety in pediatric patients undergoing medical procedures? Findings: VR was most often used as a distraction method during medical procedures and was found to be significantly more effective in reducing pain (14 studies) and anxiety (7 studies), with large effect sizes, than care as usual (CAU). Meaning: VR can be used effectively as a distraction method in clinical practice, but more research is needed to establish evidence on VR exposure as a preparation tool for medical procedures. Medical procedures often evoke pain, distress, and anxiety.1 Especially in children, these feelings not only severely affect comfort levels during medical procedures but are also associated with adverse consequences, such as attempts to escape,2 poor recovery,3 eating and sleeping disturbances,3 and posttraumatic stress symptoms.4 Furthermore, as pain and anxiety can lead to avoidance of health care,5,6 interventions are needed to address pain and anxiety in pediatric patients.

res but are also associated with adverse consequences, such as attempts to escape,2 poor recovery,3 eating and sleeping disturbances,3 and posttraumatic stress symptoms.4 Furthermore, as pain and anxiety can lead to avoidance of health care,5,6 interventions are needed to address pain and anxiety in pediatric patients. Distraction is a commonly applied intervention during medical procedures. For example, the use of music7,8 and movies9,10 has been proven efficacious in reducing pain and anxiety. Virtual reality (VR) is a relatively new technique to provide distraction and might be more effective than traditional methods. VR consists of a computer-generated environment, in which orientation and 3-dimensional interaction are possible. This environment is projected right in front of the user’s eyes via advanced head-mounted displays (HMDs), including a wide field of view and motion tracking systems.11 VR can create full immersion, which is a feeling of presence in the virtual environment.11,12 Importantly, more immersion is related to more pain reduction, because less attention is available for pain perception.13,14 VR is especially engaging for children, as they often become truly captivated by imaginative play.15 Beyond providing distraction, VR can also alleviate pain and anxiety by providing exposure. Recently, VR exposure has been applied in a more preventive manner, to make patients feel at ease and increase their familiarity with the medical procedures and environments.16,17 This preprocedural application of VR has not been thoroughly evaluated yet.

n, VR can also alleviate pain and anxiety by providing exposure. Recently, VR exposure has been applied in a more preventive manner, to make patients feel at ease and increase their familiarity with the medical procedures and environments.16,17 This preprocedural application of VR has not been thoroughly evaluated yet. While the amount of research investigating the effect of VR on alleviating pain and anxiety has increased over the past years, studies are often small and encompass a wide variety of medical procedures. This emphasizes the need for a systematic evaluation of VR in pediatric populations. Although some reviews are available on the effectiveness of VR on pain,18,19 the effectiveness on anxiety has received little attention. This is remarkable, because anxiety can intensify pain.20 Only 1 meta-analysis is available on VR interventions,21 but no meta-analysis has specifically focused on children. This distinction is important, because children are potentially even more affected by discomfort of medical procedures and might experience VR differently than adults. In this meta-analysis, we will collate evidence on the effectiveness of VR as either a distraction or an exposure tool, compared to standard care, on pain and anxiety in pediatric patients undergoing medical procedures. METHODS We followed the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) guidelines for the reporting of meta-analyses of randomized controlled trials (RCTs).22

In this meta-analysis, we will collate evidence on the effectiveness of VR as either a distraction or an exposure tool, compared to standard care, on pain and anxiety in pediatric patients undergoing medical procedures. METHODS We followed the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) guidelines for the reporting of meta-analyses of randomized controlled trials (RCTs).22 Selection Criteria Studies reporting on the effect of VR on reducing pain and/or anxiety in pediatric patients ≤21 years of age undergoing medical procedures were considered eligible for the systematic review. VR was defined as a fully immersive 3-dimensional computer-generated environment displayed in surround stereoscopic vision on an HMD. Studies that used 360° videos, which are not computer generated, displayed on a VR HMD were considered eligible as well. Studies were included in the meta-analysis if they had at least the following data available: a mean or median score for pain or anxiety during the procedure, as well as a measure of dispersion, for both the intervention and standard care groups. If not available, we requested these data by contacting the authors. Exclusion criteria were the application of VR in nonsomatic patients samples, audiovisual glasses that offer visual and audio stimulation but do not allow interaction between the user and the computer-generated world, or no distinction made between pediatric and adult patients. Reviews, meta-analyses, single-case studies, dissertations, conference papers, and abstracts were excluded as well.

es, audiovisual glasses that offer visual and audio stimulation but do not allow interaction between the user and the computer-generated world, or no distinction made between pediatric and adult patients. Reviews, meta-analyses, single-case studies, dissertations, conference papers, and abstracts were excluded as well. Search Strategy Table 1. Literature Search Terms Used for Keywordsa An exhaustive search in the following electronic databases was established and conducted by a biomedical information specialist on April 25, 2018 for articles published in English: EMBASE, MEDLINE, CENTRAL, PubMed, Web of Science, and PsycINFO. No date limit was applied to the search. The search terms “VR” and “children” or “adolescents” were used. For each database, different search strategies were developed. Table 1 gives an overview of the search terms that were used. Data Extraction Figure 1. Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) flowchart of study selection.

An exhaustive search in the following electronic databases was established and conducted by a biomedical information specialist on April 25, 2018 for articles published in English: EMBASE, MEDLINE, CENTRAL, PubMed, Web of Science, and PsycINFO. No date limit was applied to the search. The search terms “VR” and “children” or “adolescents” were used. For each database, different search strategies were developed. Table 1 gives an overview of the search terms that were used. Data Extraction Figure 1. Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) flowchart of study selection. A detailed overview of the study selection process is shown in Figure 1. The search yielded 2889 articles. Two of the authors (R.E. and P.F.A.d.N.) first assessed the identified studies for compliance with the inclusion and exclusion criteria, independently. Discrepancies (2%) were discussed until consensus was reached. Based on title and abstract, 44 of the 2889 studies were included. Next, both authors screened the full texts of these articles, independently. Discrepancies (16%) were discussed until consensus was reached. We excluded 27 of the 44 studies. Most of these studies (n = 11) were excluded because they did not use VR. Other reasons included, but were not limited to, overlap with a different age group or no inclusion of pediatric patients (see Figure 1). The final 17 studies were included.

discussed until consensus was reached. We excluded 27 of the 44 studies. Most of these studies (n = 11) were excluded because they did not use VR. Other reasons included, but were not limited to, overlap with a different age group or no inclusion of pediatric patients (see Figure 1). The final 17 studies were included. Assessment of Study Quality Two authors (R.E. and P.F.A.d.N.) independently evaluated the included studies with the Delphi list23 (Table 2) to evaluate their methodologic quality. The Delphi list is often used in systematic reviews and is able to measure internal validity, external validity, and statistical aspects.23 The Delphi list contains of 9 items, with equal weights, which can be evaluated as satisfactory (yes: scored 1) or nonsatisfactory (no: scored 0). Discrepancies in scores (17%) were discussed until consensus was reached. Table 2. Delphi List for Quality Assessment of Randomized Clinical Trials For our assessment, criterion 7 (“Was the patient blinded?”) was omitted, as it is impossible to be blinded to wearing a VR HMD or not. Consequently, the maximum possible score for studies in this review was 8 points. Criteria 5 (“Was the outcome assessor blinded?”) and 6 (“Was the care provider blinded?”) also concern blinding but were not omitted, as these criteria could be either applicable (when VR was applied before the medical procedure and outcome assessment) or nonapplicable (when VR was applied during the medical procedure and outcome assessment).

come assessor blinded?”) and 6 (“Was the care provider blinded?”) also concern blinding but were not omitted, as these criteria could be either applicable (when VR was applied before the medical procedure and outcome assessment) or nonapplicable (when VR was applied during the medical procedure and outcome assessment). Synthesis of Results For the purpose of this systematic review and meta-analysis, we did not include data on distress, maladaptive behavior, nor physiological measures of arousal, such as heart rate. We only included data on pain and anxiety outcomes based on behavioral observations, self-reports, or questionnaires.

come assessor blinded?”) and 6 (“Was the care provider blinded?”) also concern blinding but were not omitted, as these criteria could be either applicable (when VR was applied before the medical procedure and outcome assessment) or nonapplicable (when VR was applied during the medical procedure and outcome assessment). Synthesis of Results For the purpose of this systematic review and meta-analysis, we did not include data on distress, maladaptive behavior, nor physiological measures of arousal, such as heart rate. We only included data on pain and anxiety outcomes based on behavioral observations, self-reports, or questionnaires. Mean scores and SDs for pain and anxiety during the procedure in VR intervention and standard care conditions were either extracted from articles, calculated using median scores and interquartile ranges, or received from authors. Other non–VR intervention conditions were not taken into account in our analyses. Data were entered into a worksheet in Comprehensive Meta-analysis software version 2 (Biostat Inc, Englewood, NJ) by 2 authors (R.E. and B.D.). The following data were also collated and entered into Comprehensive Meta-analysis: first author, publication year, title of study, sample size per condition, mean age per condition, medical procedure, assessment instruments, quality score, informant, and study design. We used patients as primary source of data within each study, because pain and anxiety are subjective experiences. Observations of pain and anxiety made by caregivers and professionals (eg, nurse or researcher) were also entered into the worksheet. Assessment instruments for pain and anxiety were classed as either visual scales (ie, visual analog, graphic rating, and different faces scales) or questionnaires. Study design was divided into parallel or crossover designs. For crossover designs, data from the first period only, that is, before crossover, were included when available. When authors merely provided combined data from both periods, as if groups were parallel, these data were used. When data were available on different components of pain (eg, cognitive, affective, and sensory pain) the sensory component of pain was used in the meta-analysis.

fore crossover, were included when available. When authors merely provided combined data from both periods, as if groups were parallel, these data were used. When data were available on different components of pain (eg, cognitive, affective, and sensory pain) the sensory component of pain was used in the meta-analysis. Pain and anxiety were analyzed separately. Effect sizes were generated as standardized mean difference (SMD) by calculating the mean difference on pain or anxiety outcomes between VR and standard care conditions during the procedure and dividing the result by the pooled SD. Meta-analyses for either pain or anxiety were conducted for overall effect sizes of VR compared to control conditions. Because of the heterogeneity of studies, a random-effects model was used. Sensitivity analyses were performed by removing the study with the largest effect size and studies with low methodological quality (ie, a quality score of 0–2) from both meta-analyses. Separate sensitivity analyses were run for type of medical procedure. Furthermore, we investigated whether informant affected VR effectivity. To achieve a more reliable estimate of effect sizes, we also excluded outlying and low-quality studies from these analyses. To explore if young children respond differently to VR interventions than older children, a meta-regression analysis was performed with mean age of the study samples as predictor and a random-effects model (with methods of moments).

of effect sizes, we also excluded outlying and low-quality studies from these analyses. To explore if young children respond differently to VR interventions than older children, a meta-regression analysis was performed with mean age of the study samples as predictor and a random-effects model (with methods of moments). Heterogeneity was assessed using the I2 statistic, with values ≥75% indicating substantial heterogeneity.24 In case of substantial heterogeneity, subanalyses were performed to explore sources of heterogeneity. Publication bias was assessed with funnel plot asymmetry and Egger tests.25 All analyses were performed using Comprehensive Meta-analysis software version 2. RESULTS Study Characteristics Table 3 summarizes the main characteristics and results of the studies. We organized the final 17 studies based on the type of medical procedure. In 16 studies, VR was applied as a distraction technique during dental care (n = 2),26,27 burn care (n = 6),28–33 oncological care (n = 4),34–37 or venous access (n = 4).38–41 Oncological care includes quite heterogeneous procedures (ie, lumbar puncture),35 port access (piercing of the skin to access a previously implanted catheter in the chest for chemotherapy),36,37 or chemotherapy.34 Only 1 study applied VR preprocedurally, before elective surgery under general anesthesia (n = 1).42 The studies were conducted between 1999 and 2018. The number of included patients of the studies varied between 7 and 143, with a median of 38.

implanted catheter in the chest for chemotherapy),36,37 or chemotherapy.34 Only 1 study applied VR preprocedurally, before elective surgery under general anesthesia (n = 1).42 The studies were conducted between 1999 and 2018. The number of included patients of the studies varied between 7 and 143, with a median of 38. Table 3. Characteristics and Results of Included Studies That Report on the Effectiveness of Virtual Reality on Pain and Anxiety in Pediatric Patients Undergoing Medical Procedures (n = 17) Fourteen studies were RCTs, of which 10 used a parallel design and 4 studies a crossover design. All RCTs compared the VR intervention group to care as usual (CAU). CAU was often not well defined. However, CAU varied widely and could involve either no distraction or rather intensive distraction, such as watching television or listening to music. Moreover, not all studies made clear whether or not parents remained present during the procedure, nor which pharmacological analgesia were used. Three RCTs added a third condition to their designs: movie distraction,32 playing a non–VR computer game,36 or applying external cold and vibration.40 The 3 non–RCTs trials were quasi-experimental, of which 2 did not use randomization,26,39 while the other study used an interrupted time series design with removed treatment.34 The age range of participants for 16 of the 17 studies varied between 4 and 21 years. One study reported a mean age of 6.5 years but did not indicate the age range.29 Studies were heterogeneous regarding VR environments (software) and VR hardware.

Fourteen studies were RCTs, of which 10 used a parallel design and 4 studies a crossover design. All RCTs compared the VR intervention group to care as usual (CAU). CAU was often not well defined. However, CAU varied widely and could involve either no distraction or rather intensive distraction, such as watching television or listening to music. Moreover, not all studies made clear whether or not parents remained present during the procedure, nor which pharmacological analgesia were used. Three RCTs added a third condition to their designs: movie distraction,32 playing a non–VR computer game,36 or applying external cold and vibration.40 The 3 non–RCTs trials were quasi-experimental, of which 2 did not use randomization,26,39 while the other study used an interrupted time series design with removed treatment.34 The age range of participants for 16 of the 17 studies varied between 4 and 21 years. One study reported a mean age of 6.5 years but did not indicate the age range.29 Studies were heterogeneous regarding VR environments (software) and VR hardware. Study Quality Assessment We assessed all included studies with the Delphi list23 to evaluate their methodologic quality. Blinding of the outcome assessor and caregiver (criteria 5 and 6 of the Delphi list) was only applicable to the study of Ryu et al42 because they applied VR before, instead of during, the medical procedure. Therefore, the maximum possible score for this study was 8, while for the other studies, the maximum possible score was 6 (as the 2 criteria regarding blinding were not applicable).

the Delphi list) was only applicable to the study of Ryu et al42 because they applied VR before, instead of during, the medical procedure. Therefore, the maximum possible score for this study was 8, while for the other studies, the maximum possible score was 6 (as the 2 criteria regarding blinding were not applicable). The included studies varied in quality, as the quality scores ranged between 0 and 6 (see Table 3 for quality scores). The average quality score was 3.5 (SD = 1.7). Most studies had moderate quality, whereas 5 studies had high quality (ie, a maximum score, or 1 point below maximum). Four studies had poor quality (ie, a score of 0–2). Even though in 76% (n = 13) a method of randomization was performed, only 18% (n = 3) of the studies guaranteed a concealed treatment allocation. The majority of studies stated that a randomization scheme or table was used, but not enough information was provided to ensure that the allocation procedure was not transparent before assignment. In more than half of the studies, groups were similar at baseline regarding characteristics such as age, sex, and degree of injury (n = 10, 59%). Inclusion and exclusion criteria were not described precisely enough for 6 studies (35%). Seven studies (41%) included intention-to-treat analysis.

sparent before assignment. In more than half of the studies, groups were similar at baseline regarding characteristics such as age, sex, and degree of injury (n = 10, 59%). Inclusion and exclusion criteria were not described precisely enough for 6 studies (35%). Seven studies (41%) included intention-to-treat analysis. Other specific findings that could have influenced study quality were as follows: initially, Das et al28 (burn care) only included patients who experienced burns for the first time, but they let some patients participate more than once (ie, 11 trials were undertaken from 7 patients). Piskorz and Czub39 (venous access) let children play a VR game. If they enjoyed it, these children were included in the VR condition. Afterward, the authors collected data for the control group (who had not tried out the VR game). Gerceker et al40 excluded all unsuccessful phlebotomy attempts from their analyses (ie, when there was no blood flow into the tube within 5 seconds during the first attempt). Ryu et al42 observed less anxiety during the preoperative period but did not assess anxiety during induction of anesthesia, when anxiety peaks.

. Gerceker et al40 excluded all unsuccessful phlebotomy attempts from their analyses (ie, when there was no blood flow into the tube within 5 seconds during the first attempt). Ryu et al42 observed less anxiety during the preoperative period but did not assess anxiety during induction of anesthesia, when anxiety peaks. Virtual Reality and Pain Management As shown in Figure 2, effect sizes for patient-reported pain could be generated for 14 of the 17 studies. For 2 studies, means and SDs were calculated using median values and interquartile ranges.32,35 Calculated effect sizes were positive when VR reduced pain more than CAU. Across all studies, using a random-effects model, the weighted effect size of VR on pediatric pain during a medical procedure was large (SMD = 1.30; 95% CI, 0.68–1.91; P < .001). This indicated a substantial clinical benefit, but heterogeneity of study effects was high (I2 = 93.3%). A sensitivity analysis was performed by excluding the outlying study, that is, the study with the largest effect size (Gerceker et al40) and studies with low methodological quality.29,39 This analysis still suggested effects of VR with an attenuated but still medium to large effect size, which indicated a robust effect (SMD = 0.73; 95% CI, 0.35–1.11; P < .001). Though, still substantial, this analysis had lower heterogeneity (I2 = 78.3%). Figure 2. Random-effects meta-analysis for the effect of VR on patient-reported pain during a medical procedure compared to CAU. Note: study effect for Gerceker et al40 is out of range. CAU indicates care as usual; VR, virtual reality.

Virtual Reality and Pain Management As shown in Figure 2, effect sizes for patient-reported pain could be generated for 14 of the 17 studies. For 2 studies, means and SDs were calculated using median values and interquartile ranges.32,35 Calculated effect sizes were positive when VR reduced pain more than CAU. Across all studies, using a random-effects model, the weighted effect size of VR on pediatric pain during a medical procedure was large (SMD = 1.30; 95% CI, 0.68–1.91; P < .001). This indicated a substantial clinical benefit, but heterogeneity of study effects was high (I2 = 93.3%). A sensitivity analysis was performed by excluding the outlying study, that is, the study with the largest effect size (Gerceker et al40) and studies with low methodological quality.29,39 This analysis still suggested effects of VR with an attenuated but still medium to large effect size, which indicated a robust effect (SMD = 0.73; 95% CI, 0.35–1.11; P < .001). Though, still substantial, this analysis had lower heterogeneity (I2 = 78.3%). Figure 2. Random-effects meta-analysis for the effect of VR on patient-reported pain during a medical procedure compared to CAU. Note: study effect for Gerceker et al40 is out of range. CAU indicates care as usual; VR, virtual reality. The following sensitivity analyses were performed after removal of the outlying study40 and low-quality studies29,39 to achieve a more reliable estimate of effect sizes. Sensitivity analyses were run for caregivers and professionals as observers of pediatric pain. We found significant results based on both types of informants (caregivers31,33,36,41: SMD = 0.47; 95% CI, 0.22–0.72; P < .001; I2 = 0.0%, professionals31,33,36: SMD = 0.82; 95% CI, 0.48–1.15; P < .001; I2 = 0.0%). Finally, we ran sensitivity analyses on self-reported pain for each type of medical procedure, when data from >1 study were available. We found significant effects for burn care28,30–33 (SMD = 0.66; 95% CI, 0.40–0.91; P < .001; I2 = 0.0%) and venous access38,41 (SMD = 0.32; 95% CI, 0.01–0.62; P = .046; I2 = 0.0%) but not for oncological care35–37 (SMD = 0.65; 95% CI, −0.26 to 1.57; P = .159; I2 = 76.3%). The suggested effect of VR for observed pain and for self-reported pain during burn care and venous access was associated with decreased effect sizes, but also with zero heterogeneity.

2; 95% CI, 0.01–0.62; P = .046; I2 = 0.0%) but not for oncological care35–37 (SMD = 0.65; 95% CI, −0.26 to 1.57; P = .159; I2 = 76.3%). The suggested effect of VR for observed pain and for self-reported pain during burn care and venous access was associated with decreased effect sizes, but also with zero heterogeneity. A random-effects model (with methods of moments) was used for the meta-regression analysis with age as a predictor. The results suggested that VR interventions for pain reduction were more efficacious for younger than for older children (P = .015). More specifically, the effect size of VR on pain decreased with 0.26 when age increased with 1 year. After removing the study with the largest effect size,40 age was still a significant predictor of the effect of VR on pain (P < .001).

pain reduction were more efficacious for younger than for older children (P = .015). More specifically, the effect size of VR on pain decreased with 0.26 when age increased with 1 year. After removing the study with the largest effect size,40 age was still a significant predictor of the effect of VR on pain (P < .001). Virtual Reality and Anxiety Management Effect sizes for patient-reported anxiety could be generated for 7 of the 17 studies (Figure 3). For 1 study, mean and SD were calculated using median value and interquartile range.42 Using the random-effects model, a large effect size was found for VR on anxiety (SMD = 1.32; 95% CI, 0.21–2.44; P = .020). This indicated substantial clinical benefit, but heterogeneity of study effects was high (I2 = 96.6%). A sensitivity analysis was performed by excluding the outlying study (Asl Aminabadi et al27) and studies with low methodological quality.34,39 This analysis still suggested effects of VR (SMD = 0.50; 95% CI, 0.20–0.79; P = .001) with an attenuated but still medium effect size, which indicated a robust effect. Moreover, heterogeneity decreased significantly in this analysis (I2 = 22.4%). Figure 3. Random-effects meta-analysis for the effect of VR on patient-reported anxiety during a medical procedure compared to CAU. Note: study effect for Asl Aminabadi et al27 is out of range. CAU indicates care as usual; VR, virtual reality.

Virtual Reality and Anxiety Management Effect sizes for patient-reported anxiety could be generated for 7 of the 17 studies (Figure 3). For 1 study, mean and SD were calculated using median value and interquartile range.42 Using the random-effects model, a large effect size was found for VR on anxiety (SMD = 1.32; 95% CI, 0.21–2.44; P = .020). This indicated substantial clinical benefit, but heterogeneity of study effects was high (I2 = 96.6%). A sensitivity analysis was performed by excluding the outlying study (Asl Aminabadi et al27) and studies with low methodological quality.34,39 This analysis still suggested effects of VR (SMD = 0.50; 95% CI, 0.20–0.79; P = .001) with an attenuated but still medium effect size, which indicated a robust effect. Moreover, heterogeneity decreased significantly in this analysis (I2 = 22.4%). Figure 3. Random-effects meta-analysis for the effect of VR on patient-reported anxiety during a medical procedure compared to CAU. Note: study effect for Asl Aminabadi et al27 is out of range. CAU indicates care as usual; VR, virtual reality. The following sensitivity analyses were performed after removal of the outlying study27 and low-quality studies34,39 to achieve a more reliable estimate of effect sizes. Unfortunately, very limited data were available for caregivers and professionals as observers of pediatric anxiety. We were only able to run a separate analysis for caregiver as informant,36,41 which did not yield a significant result (SMD = 0.31; 95% CI, −0.02 to 0.63; P = .067; I2 = 0%). Regarding different types of medical procedures, only for oncological care, enough data were available to run a sensitivity analysis on self-reported anxiety,36,37 which yielded a significant result (SMD = 0.53; 95% CI, 0.10–0.96; P = .015; I2 = 0.0%). The effect of VR during oncological care was associated with a decreased effect size but also with zero heterogeneity.

ly for oncological care, enough data were available to run a sensitivity analysis on self-reported anxiety,36,37 which yielded a significant result (SMD = 0.53; 95% CI, 0.10–0.96; P = .015; I2 = 0.0%). The effect of VR during oncological care was associated with a decreased effect size but also with zero heterogeneity. A random-effects model (with methods of moments) was used for the meta-regression analysis with age as a predictor. The results suggested that VR interventions for anxiety reduction were more efficacious for younger than for older children (P = .023). More specifically, the effect size of VR on anxiety decreased to 0.35 when age increased with 1 year. After removing the study with the largest effect size,27 age was still a significant predictor of the effect of VR on anxiety (P = .037). Publication Bias and Heterogeneity Funnel plots for pain and anxiety showed asymmetry, but Egger regression asymmetry tests did not confirm the presence of a significant publication bias for pain (P = .105) nor anxiety (P = .282). Funnel plots indicated that there was one clear outlier for pain40 and one for anxiety.27 These outliers correspond to the studies with the largest effect sizes which we have removed from the sensitivity analyses.

did not confirm the presence of a significant publication bias for pain (P = .105) nor anxiety (P = .282). Funnel plots indicated that there was one clear outlier for pain40 and one for anxiety.27 These outliers correspond to the studies with the largest effect sizes which we have removed from the sensitivity analyses. As discussed above, substantial heterogeneity of study effects was found for the overall meta-analysis on pain (I2 = 93.3%) and anxiety (I2 = 96.0%). We found that the outlying and low-quality studies were important sources of heterogeneity, because removal of these studies was associated with decreased heterogeneity (I2 = 78.3% for pain and I2 = 22.4% for anxiety). Moreover, the available data suggested that the different medical procedures were an important source of heterogeneity as well because the study effects of these sensitivity analyses were associated with zero heterogeneity. DISCUSSION This is the first systematic review and meta-analysis that specifically focused on VR in pediatric patients. Our meta-analysis, based on 14 studies for pain and 7 studies for anxiety, showed VR to be an effective tool to diminish patient-reported pain (SMD = 1.30) and anxiety (SMD = 1.32) during a range of medical procedures. The effect of VR on pediatric pain was also significant when observed by caregivers or professionals. For anxiety, limited observer data were available on VR effectivity. Due to small groups, it was difficult to compare VR effectivity in different types of medical procedures. VR was most often applied during burn care.

The effect of VR on pediatric pain was also significant when observed by caregivers or professionals. For anxiety, limited observer data were available on VR effectivity. Due to small groups, it was difficult to compare VR effectivity in different types of medical procedures. VR was most often applied during burn care. Our results showed that VR interventions for pain and anxiety were potentially more efficacious for younger than for older children. A possible explanation is that younger children tend to have higher levels of anxiety before medical procedures.43,44 A different possible explanation is that VR is especially engaging for younger children, as they are often more engaged in magical thinking45 and become truly captivated by imaginative play.15 However, because the relationship of age with VR efficacy on pain or anxiety could be different within each study compared to across studies, the relationship shown between age and VR efficacy in the meta-regression may not represent the true relation. This phenomenon is called ecological fallacy.46

Our results showed that VR interventions for pain and anxiety were potentially more efficacious for younger than for older children. A possible explanation is that younger children tend to have higher levels of anxiety before medical procedures.43,44 A different possible explanation is that VR is especially engaging for younger children, as they are often more engaged in magical thinking45 and become truly captivated by imaginative play.15 However, because the relationship of age with VR efficacy on pain or anxiety could be different within each study compared to across studies, the relationship shown between age and VR efficacy in the meta-regression may not represent the true relation. This phenomenon is called ecological fallacy.46 VR was found to be significantly more effective in reducing pain and anxiety than CAU. However, it remains difficult to differentiate between the added value of VR over other forms of distraction, for example, watching television, and no distraction, because CAU was often not well defined. The high weighted effect sizes we found suggest that VR distraction is possibly more effective than other distraction interventions during medical procedures. For example, a Cochrane review47 found an effect size of 0.61 for the impact of distraction (eg, games, music, and toys) on self-reported pain during needle-related procedures. Similarly, a meta-analysis including trials on music therapy as distraction during different types of medical procedures (eg, dental care, magnetic resonance imaging scans, and venipuncture) showed a significant reduction in pain and anxiety with an effect size of 0.35.48 Because VR exposure as a preparation tool for medical procedures is a fairly unexplored area of research, it is not (yet) possible to compare effect sizes for VR preparation to other forms of preparative interventions to reduce pain and anxiety during medical procedures.

omization and clearly described their inclusion and exclusion criteria. However, concealed treatment allocation was often not guaranteed and intention-to-treat analyses were often not performed. Also, very few studies focused on possible moderating factors of VR effectivity, such as anxiety sensitivity and temperament. An important area of focus is immersion, which is influenced by interaction with the virtual environment by means of translation (changing position), rotation (changing orientation), point of view (perspective), and field of view.19,49 Non–VR content, that is, regular (cartoon) videos or 360° videos, creates less immersion, because the user is limited to the filmmaker’s movements and progress of the video. This difference in content is important, as it has been hypothesized that more immersion is related to more pain reduction, because less attention is available for pain perception.13,14 Even though some studies included questions about subjective feelings of immersion, it is difficult to objectively analyze this phenomenon. During certain medical procedures, for example, dental treatment, patients were required to keep their head still, which may have limited immersion as well. True VR creates a more compelling illusion of presence in the virtual world than more passive audiovisual glasses and non–VR (360°) videos. However, the supposed superiority of VR over audiovisual glasses and non–VR content regarding efficacy in medical care has yet to be proven.11 Therefore, the role of immersion should be a focus of future research.

illusion of presence in the virtual world than more passive audiovisual glasses and non–VR (360°) videos. However, the supposed superiority of VR over audiovisual glasses and non–VR content regarding efficacy in medical care has yet to be proven.11 Therefore, the role of immersion should be a focus of future research. Implications VR distraction has a large impact on pediatric pain and anxiety during medical procedures, especially for younger children. This easy-to-use tool can be used effectively in clinical practice. More research like the study of Ryu et al42 is needed to establish evidence on VR exposure as preparation to reduce pain and anxiety during medical procedures. This is crucial, because anticipatory anxiety can lead to more pain and distress during the medical procedure itself.50,51

sed effectively in clinical practice. More research like the study of Ryu et al42 is needed to establish evidence on VR exposure as preparation to reduce pain and anxiety during medical procedures. This is crucial, because anticipatory anxiety can lead to more pain and distress during the medical procedure itself.50,51 Limitations The following limitations should be taken into account when interpreting the results of the current review and meta-analysis. First, effect sizes for patient-reported anxiety could be generated for only 7 studies. Second, limited observer data were available, especially for anxiety outcomes. Third, means and SDs were estimated using median values and interquartile ranges for 3 studies.32,35,42 This was necessary to pool all data, but is unclear how reliable these estimations are. Fourth, substantial heterogeneity was present in the findings. We have identified outlying and low-quality studies as important sources of heterogeneity. Moreover, there was a difference in effect of VR for different medical procedures, so one should be careful when generalizing the suggested effect for VR to clinical practice. However, in our opinion, the mean pooled effect of all medical procedures still provides the most useful information, especially because certain procedures have not been studied extensively or have not been studied at all, regarding VR interventions. Finally, the included studies applied various kinds of VR software, which could have influenced the amount of immersion and VR effectivity. On the other hand, it is also possible that VR software only plays a small role, as Kenney and Milling21 found no differences in their meta-analysis between commercially available VR games and VR software that was specifically developed for distraction.

could have influenced the amount of immersion and VR effectivity. On the other hand, it is also possible that VR software only plays a small role, as Kenney and Milling21 found no differences in their meta-analysis between commercially available VR games and VR software that was specifically developed for distraction. CONCLUSIONS This systematic review and meta-analysis indicate that pediatric patients undergoing a range of medical procedures benefit from VR as a tool to reduce pain and anxiety. Due to limited available observer data, we could not provide insight into possible differences in perspective between patients, caregivers, and professionals. VR research in pediatrics has mainly focused on VR as a distraction tool. Using VR exposure as a preparation tool could be an innovative way to decrease anxiety and pain before and during medical procedures. However, further research into this field is needed. ACKNOWLEDGMENTS We would like to thank biomedical information specialists Gerdien B. de Jonge, MSc, and Wichor M. Bramer, MSc, of the Medical Library, Erasmus Medical Center, University Medical Center Rotterdam, Rotterdam, the Netherlands, for their assistance in conducting the systematic literature search. DISCLOSURES Name: Robin Eijlers, MSc. Contribution: This author helped perform the literature searches, study selection, and statistical analyses; extract the data; write the protocol draft, which was revised for important intellectual content by all other authors; and approve the final version of the manuscript. Name: Elisabeth M. W. J. Utens, PhD.

DISCLOSURES Name: Robin Eijlers, MSc. Contribution: This author helped perform the literature searches, study selection, and statistical analyses; extract the data; write the protocol draft, which was revised for important intellectual content by all other authors; and approve the final version of the manuscript. Name: Elisabeth M. W. J. Utens, PhD. Contribution: This author, as principal investigator of this project, helped provide important intellectual content and approve the final version of the manuscript. Name: Lonneke M. Staals, MD, PhD. Contribution: This author helped provide important intellectual content and approve the final version of the manuscript. Name: Pieter F. A. de Nijs, MD, PhD. Contribution: This author helped perform the literature searches and study selection, provide important intellectual content, and approve the final version of the manuscript. Name: Johan M. Berghmans, MD. Contribution: This author helped provide important intellectual content and approve the final version of the manuscript. Name: René M. H. Wijnen, MD, PhD. Contribution: This author helped provide important intellectual content and approve the final version of the manuscript. Name: Manon H. J. Hillegers, MD, PhD. Contribution: This author helped provide important intellectual content and approve the final version of the manuscript. Name: Bram Dierckx, MD, PhD. Contribution: This author helped extract the data, perform statistical analyses, provide important intellectual content, and approve the final version of the manuscript. Name: Jeroen S. Legerstee, PhD.

Contribution: This author helped provide important intellectual content and approve the final version of the manuscript. Name: Bram Dierckx, MD, PhD. Contribution: This author helped extract the data, perform statistical analyses, provide important intellectual content, and approve the final version of the manuscript. Name: Jeroen S. Legerstee, PhD. Contribution: This author helped interpret the results of statistical analyses, provide important intellectual content, and approve the final version of the manuscript. This manuscript was handled by: James A. DiNardo, MD, FAAP. Published ahead of print 23 May 2019. Funding: This research is funded by the Zilveren Kruis foundation (project No. 2015233) and the Coolsingel foundation (project No. 401). The authors declare no conflicts of interest. Reprints will not be available from the authors.

Recent research has identified that a significant proportion of adolescents undergoing major surgery go on to develop chronic postsurgical pain and prolonged opioid use at home.1,2 At the same time, management of adolescents undergoing major surgery continues to follow a biomedical model, with attention on short-term outcomes such as inpatient pain scores and analgesic consumption and variables that focus on purely physiological and medical risk factors. Currently, there is a paucity of data regarding biopsychosocial factors that may have a significant role in the development of persistent postoperative pain and opioid use. The aim of this The Open Mind article is therefore to present an innovative perioperative biopsychosocial model for delivering care to adolescents undergoing major surgery. This article will focus on a model that is aimed at improving long-term recovery after surgery, reducing opioid use at home, and preventing chronic postsurgical pain. We hope to challenge our field to evolve the current approach to perioperative care to address these important issues facing adolescent patients.

urgery. This article will focus on a model that is aimed at improving long-term recovery after surgery, reducing opioid use at home, and preventing chronic postsurgical pain. We hope to challenge our field to evolve the current approach to perioperative care to address these important issues facing adolescent patients. Background and Rationale Despite biomedical advances in perioperative care, rates of severe pain after pediatric surgery have remained similar over the past 2 decades.3,4 Further, the United States currently faces pain and opioid crises, with rising rates of disabling chronic pain and an epidemic of opioid addiction and overdose.5 Neurobiological, psychological, behavioral, and social changes occurring during adolescence make adolescents particularly vulnerable to chronic pain6 and drug addiction,7 and recent data indicate that opioids are now the most frequently overdosed drug in this age group.8 Chronic pain and the evolving opioid crisis have significant implications for the perioperative care we provide for adolescents. Emerging perioperative research has identified that adolescents experience high prevalence of chronic postsurgical pain and persistent opioid use after surgery.1,2 After extensive adolescent surgeries such as spinal fusion, up to 80% of the patients report severe pain at home during the initial weeks after surgery,9 and 20% go on to report chronic postsurgical pain, defined as postsurgical pain lasting >3 months.2 These youth experience significant distress,10 and impairment in physical and psychosocial health-related quality of life,11 which may persist months to years after surgery. A recent study in a large national cohort of adolescents and young adults1 found that 4.8% of opioid-naive youth undergoing surgery (2.7%–15.2% across procedures) developed new persistent opioid use after surgery as compared to 0.1% of nonsurgical counterparts. In susceptible adolescents, long-term opioid use after an opioid prescription may be associated with risk for opioid addiction and overdose.5 Based on recent research indicating that psychosocial and behavioral factors play an important role in the high rates of both persistent pain2 and opioid use12 after surgery, we propose a biopsychosocial framework of perioperative care for this vulnerable patient population.

ith risk for opioid addiction and overdose.5 Based on recent research indicating that psychosocial and behavioral factors play an important role in the high rates of both persistent pain2 and opioid use12 after surgery, we propose a biopsychosocial framework of perioperative care for this vulnerable patient population. Biopsychosocial Model of Perioperative Care for Adolescents Figure. Psychosocial perioperative care model for adolescents undergoing major surgery. In 2010, Chorney and Kain13 proposed a framework for family-centered pediatric perioperative care. The framework spans the preoperative period (preparation for surgery), the intraoperative period (management strategies), and the postoperative period (pain management and recovery at home), and incorporates family and system variables that influence family-centered care delivery. Building on this conceptual framework, we present a biopsychosocial model of perioperative care, adapted to the population of adolescents undergoing major surgery (Figure). In the model described in the Figure, we identify adolescent and parent risk factors for chronic postsurgical pain and persistent opioid use and highlight perioperative interventions that can be implemented to improve short- and long-term outcomes after surgery.

ulation of adolescents undergoing major surgery (Figure). In the model described in the Figure, we identify adolescent and parent risk factors for chronic postsurgical pain and persistent opioid use and highlight perioperative interventions that can be implemented to improve short- and long-term outcomes after surgery. Preoperative Period During the preoperative period, both adolescent factors such as high anxiety, poor sleep patterns, and poor pain coping efficacy, and parent/family factors such as high parental anxiety and distress have been identified as risk factors for acute pain and chronic postsurgical pain .9,11,14–17 Further, high depressive symptoms and adolescent anxiety about pain before surgery have also been identified as risk factors for increased opioid consumption after hospital discharge.12 Based on these findings, we submit that an opportunity exists during preoperative preparation to deliver family-based psychosocial interventions targeting these adolescent and parent risk factors and teaching adaptive coping strategies to adolescents and their parents to promote recovery and prevent pain persistence.

2 Based on these findings, we submit that an opportunity exists during preoperative preparation to deliver family-based psychosocial interventions targeting these adolescent and parent risk factors and teaching adaptive coping strategies to adolescents and their parents to promote recovery and prevent pain persistence. The perioperative home model of care can serve as an entry point to identify and intervene for at-risk adolescents in advance of surgery.18 Major musculoskeletal surgeries are commonly planned for the mid to late adolescent period; for example, surgery for pectus deformities and spinal fusion for idiopathic scoliosis are often anticipated for several years. Patients are referred to the perioperative home for preoperative planning, and families are typically followed at multiple appointments before surgery. This period offers ample opportunity to prepare adolescents and their families for surgery and subsequent recovery. At a minimum, we need to start providing adolescents and their families with realistic expectations of pain and recovery. In the setting of a pediatric center, pediatric psychologists may be available to provide behavioral preparation for surgery.19 Detection of clinical levels of depression or anxiety should trigger referral for psychological treatment of these conditions.

their families with realistic expectations of pain and recovery. In the setting of a pediatric center, pediatric psychologists may be available to provide behavioral preparation for surgery.19 Detection of clinical levels of depression or anxiety should trigger referral for psychological treatment of these conditions. Postoperative Period While advances have been made in maximizing nonopioid treatments such as multimodal analgesia and regional anesthesia, nonpharmacologic approaches need to be developed and integrated into postoperative care for adolescents undergoing major surgery. Psychological strategies such as deep breathing and self-regulation techniques to cope with pain, and cognitive strategies to reduce anxiety and modify thoughts and beliefs about pain, may reduce pain and enhance recovery from surgery. Evidence supports efficacy of interventions teaching distraction, imagery, and relaxation in reducing children’s acute postsurgical pain; however, data on the effects of psychological interventions targeting long-term pain outcomes are still needed.19 Although adolescents are more involved in their own health care and require their own skills to effectively manage pain, involving parents is important to increase their willingness to support behavioral skill use by their child.

fects of psychological interventions targeting long-term pain outcomes are still needed.19 Although adolescents are more involved in their own health care and require their own skills to effectively manage pain, involving parents is important to increase their willingness to support behavioral skill use by their child. That said, it is essential that a biopsychosocial perioperative care model extend well beyond early recovery from surgery. Pain and recovery need to be assessed during the days, weeks, and months after surgery, to identify youth who are struggling with recovery and who need additional intervention to prevent chronic pain. Multidimensional assessment of pain and recovery ideally includes characteristics and extent of pain as well as measures of functional impact.20 Brief functional measures that are appropriate for use in adolescents recovering from major surgery should be administered alongside pain intensity assessments. For example, the Youth Acute Pain-Functional Ability Questionnaire assesses function relevant in the acute setting, in a 24-hour time frame, to monitor daily change in function during recovery, with a short form developed for use in adolescents admitted to the hospital after major surgery.21 Patients experiencing delayed pain recovery may need surgical evaluation to rule out other etiology (eg, infection) before a diagnosis of chronic postsurgical pain.22 Important long-term outcomes include pain outcomes such as chronic pain status, opioid use including opioid misuse (ie, use of prescription opioids without prescription or in a manner other than prescribed) and opioid use disorders (ie, addiction, abuse, or dependence), as well as broader patient-reported health outcomes such as health-related quality of life.

outcomes such as chronic pain status, opioid use including opioid misuse (ie, use of prescription opioids without prescription or in a manner other than prescribed) and opioid use disorders (ie, addiction, abuse, or dependence), as well as broader patient-reported health outcomes such as health-related quality of life. Spurred by the opioid epidemic, efforts are being undertaken at state and national levels to reduce opioid overprescribing and curtail potential for opioid diversion. This includes guidelines limiting opioid prescription, and the prescription drug–monitoring program that tracks controlled substance prescriptions in a state. Restrictions on opioid prescriptions make it imperative that we maximize nonopioid pain treatment and incorporate nonpharmacologic pain management strategies into perioperative care. As experts in perioperative pain management, it is critical that we provide adolescent patients and their parents with education on the danger of opioid misuse, and detailed instructions on appropriate use, tapering, discontinuation, and safe disposal of opioid medications.23 In addition, adolescents who have a history of opioid misuse before presenting for surgery may need careful consideration of risk/benefit, closer monitoring, and intervention to reduce risks related to opioid use.

led instructions on appropriate use, tapering, discontinuation, and safe disposal of opioid medications.23 In addition, adolescents who have a history of opioid misuse before presenting for surgery may need careful consideration of risk/benefit, closer monitoring, and intervention to reduce risks related to opioid use. System Factors On a systems level, organizational commitment to a biopsychosocial approach as a standard of practice for perioperative care is critical to success of such endeavors. Infrastructure needs to be expanded to support access to psychologists for delivery of pre- and postoperative biopsychosocial interventions to improve patient outcomes. Electronic medical record systems and patient portals should be leveraged to facilitate evaluation of postoperative pain and health outcomes, both in the hospital and once patients go home. This will serve not only to monitor individual patients’ recovery, but also to evaluate changes in perioperative management pathways for quality improvement.

ystems and patient portals should be leveraged to facilitate evaluation of postoperative pain and health outcomes, both in the hospital and once patients go home. This will serve not only to monitor individual patients’ recovery, but also to evaluate changes in perioperative management pathways for quality improvement. FUTURE RESEARCH DIRECTIONS The federal pain research strategy, released in 2017, highlighted the current need for research to improve pain management and reduce reliance on opioids, to prevent transition from acute to chronic pain, and to reduce the burden of high-impact chronic pain.24 Top research priorities identified include studies examining susceptibility and resilience factors underlying acute to chronic pain transition, and research developing optimal approaches for self-management strategies to prevent, cope with, and reduce pain. The perioperative period provides a unique opportunity to study these important questions in youth. Longitudinal research is needed: (1) identifying high-risk adolescents before surgery to enable implementation of primary prevention; (2) developing behavioral interventions and nonpharmacologic strategies to manage pain and reduce opioid use; and (3) monitoring youth during early recovery to identify adolescents experiencing problems with recovery to enable secondary prevention. Research currently being conducted in this area will inform interventions to reduce opioid exposure and decrease overall incidence of chronic postsurgical pain in adolescents. A further gap in research that urgently needs to be addressed is regarding long-term opioid use after surgery. Specifically, research is lacking examining risk factors and preventive strategies for long-term opioid use. It is also currently unknown whether maximizing opioid-sparing treatments that are recommended for acute postoperative pain management (ie, multimodal analgesia, regional anesthesia)25 could reduce the risk for long-term opioid use after surgery.

xamining risk factors and preventive strategies for long-term opioid use. It is also currently unknown whether maximizing opioid-sparing treatments that are recommended for acute postoperative pain management (ie, multimodal analgesia, regional anesthesia)25 could reduce the risk for long-term opioid use after surgery. Finally, while the concepts of family-centered pediatric perioperative care are also applicable to youth undergoing nonmajor surgery, work is needed to understand which of these youth are at risk for persistent pain and opioid use. This will inform the adaptation of this biopsychosocial framework developed to address persistent pain and opioid use after major surgery to the context of nonmajor surgeries and the outpatient setting.

g nonmajor surgery, work is needed to understand which of these youth are at risk for persistent pain and opioid use. This will inform the adaptation of this biopsychosocial framework developed to address persistent pain and opioid use after major surgery to the context of nonmajor surgeries and the outpatient setting. CONCLUSIONS The current pain crisis and opioid epidemic threaten the health of the adolescents we provide care for. In our role as leaders of perioperative care, anesthesiologists not only have the opportunity but also the responsibility to address these important issues facing our community. A biopsychosocial approach to perioperative care including interventions targeting risk factors for chronic postsurgical pain, nonpharmacologic strategies to manage pain at home, and assessment of broader outcomes of perioperative care is critically needed for adolescents undergoing major surgery. Developing partnerships with psychologists in the perioperative home model of care will be essential to this approach. This is an opportunity for pediatric perioperative medicine to innovate our practice, evolving our model of care, to improve outcomes for our vulnerable patients. DISCLOSURES Name: Jennifer A. Rabbitts, MBChB. Contribution: This author helped conceive and draft the manuscript, and approve the final manuscript. Conflicts of Interest: None. Name: Zeev Kain, MD, MA (hon), MBA, FAAP. Contribution: This author helped conceive, edit, and approve the final manuscript.

CONCLUSIONS The current pain crisis and opioid epidemic threaten the health of the adolescents we provide care for. In our role as leaders of perioperative care, anesthesiologists not only have the opportunity but also the responsibility to address these important issues facing our community. A biopsychosocial approach to perioperative care including interventions targeting risk factors for chronic postsurgical pain, nonpharmacologic strategies to manage pain at home, and assessment of broader outcomes of perioperative care is critically needed for adolescents undergoing major surgery. Developing partnerships with psychologists in the perioperative home model of care will be essential to this approach. This is an opportunity for pediatric perioperative medicine to innovate our practice, evolving our model of care, to improve outcomes for our vulnerable patients. DISCLOSURES Name: Jennifer A. Rabbitts, MBChB. Contribution: This author helped conceive and draft the manuscript, and approve the final manuscript. Conflicts of Interest: None. Name: Zeev Kain, MD, MA (hon), MBA, FAAP. Contribution: This author helped conceive, edit, and approve the final manuscript. Conflicts of Interest: Z. Kain is a speaker for Edwards Lifesciences and Huron Consulting. He is also the founder and President of the American College of Perioperative Medicine. This manuscript was handled by: James A. DiNardo, MD, FAAP. Published ahead of print 1 February 2019.

Contribution: This author helped conceive, edit, and approve the final manuscript. Conflicts of Interest: Z. Kain is a speaker for Edwards Lifesciences and Huron Consulting. He is also the founder and President of the American College of Perioperative Medicine. This manuscript was handled by: James A. DiNardo, MD, FAAP. Published ahead of print 1 February 2019. Funding: This work was supported by the National Institute of Arthritis, Musculoskeletal, and Skin Diseases under Award No. R01AR073780 (Principal Investigator: J.A.R.) and Eunice Kennedy Shriver National Institute of Child Health & Human Development under Award No. R01HD091286 (Principal Investigator: Z.K.). Conflicts of Interest: See Disclosures at the end of the article. Reprints will not be available from the authors.