Off-Hour Presentation and Outcomes in Acute MI
Off-Hour Presentation and Outcomes in Acute MI
This study was conducted according to guidance from the Cochrane Handbook of Systematic Reviews and is reported according to PRISMA (preferred reporting items for systematic reviews and meta-analyses) recommendations.
We did a comprehensive search of several databases from database inception to April 2013, any language. The databases included Ovid Medline in-process and other non-indexed citations, Ovid Medline, Ovid Embase, Ovid Cochrane Database of Systematic Reviews, and Scopus. An experienced librarian designed and conducted the search strategy with input from the principal investigator. He used controlled vocabulary supplemented with keywords to search for comparative studies of off-hour effects for patients with acute myocardial infarction. The strategy is available in the Web Appendix (Table A). We also manually searched PubMed, Ovid Medline, and references in pertinent articles that were identified in the screening processes.
We considered all studies published in any language, with any study design, that evaluated the association between time of presentation and mortality or door to balloon times among adult patients who presented with acute myocardial infarction. Studies were eligible if they compared the outcomes between patients with off-hour versus regular hour presentation. We categorized the comparison of off-hours versus regular hours as weekend and night versus weekday regular hours, weekend versus weekday, or night versus daytime (Appendix Figure A). Time of presentation could be measured by arrival at the hospital, admission to hospital, or start of percutaneous coronary intervention. Eligible mortality outcomes included in-hospital mortality and 30 day mortality. Initial screening of abstracts excluded non-relevant or non-original studies. We then used full text screening to assess eligibility. Whenever reports pertained to the same set of patients, we retained the one with the most recent year of publication to obtain the most updated data. Two investigators (AS and AA) independently screened reports. Studies with discrepant decisions in screening of the abstract proceeded to full text screening. We resolved discrepancies in full text screening through consensus. We calculated a Κ statistic to quantify the agreement between the two reviewers on study selection.
We recorded information on studies’ characteristics and demographics such as authors, publication year, country, years of enrollment of the cohort, data source, the definition of off-hours and regular hours, time of presentation measured for allocation, and the inclusion and exclusion criteria, as well as per group sample size, characteristics of the population, variables adjusted for, and outcomes. We recorded mortality by either number or proportion of deaths in each group and odds ratio or hazard ratio with confidence intervals. We recorded door to balloon time by either mean with standard deviation or median with interquartile range or by the proportion of patients whose door to balloon time was 90 minutes or less. For studies that implemented an intervention that could affect the outcome, we abstracted data from the observational (pre-intervention) period. One investigator abstracted data, which a second investigator independently verified. The discrepancies found in the verification process were solved by consensus or further review by a third investigator. SRS, DAR, and KMT abstracted data, and AS and AA verified data.
We assessed the methodological quality of the included studies by using the Newcastle-Ottawa scale. This scale consists of three domains (cohort selection, comparability, and outcome) and evaluates the study’s overall risk of bias. The maximum score for an observational study is 9 points. Two investigators independently assessed quality, and another resolved discrepancies.
Mortality Outcomes. We used in-hospital or 30 day mortality as the main outcome. For studies without in-hospital mortality results, we used 30 day mortality when available. We did the main analysis for all studies combined. We also separately analyzed each mortality outcome (in-hospital versus 30 day). For the main outcome, we did subgroup analyses by diagnosis of patient cohort (STEMI versus non-STEMI), type of off-hours (weekend and night versus weekend versus night), measured time of presentation (arrival versus admission versus start of percutaneous coronary intervention), data source (clinical registry versus administrative data), region (North America versus Europe versus others), and outcome adjustment (adjusted versus unadjusted). To evaluate the possibility of a time trend effect of mortality across studies, we did meta-regression using the mid-year of enrollment of the cohort as the independent variable and the natural log of the effect size as the dependent variable. Owing to concern about potential overlapping patient sets, we did sensitivity analyses by excluding each single cohort and by including only one cohort from each study. We also did sensitivity analyses by excluding studies that expressed results as a hazard ratio.
Door to Balloon Time. We analyzed the proportion of patients with STEMI whose door to balloon time was less than 90 minutes and the mean or median door to balloon times. For mean or median door to balloon times, we did subgroup analyses by type of off-hours determination, measured time of presentation, and region, as well as meta-regression using the mid-year of enrollment of the cohort to evaluate time trends in door to balloon times. We also did sensitivity analyses limiting to studies that included only patients who were directly admitted to the hospital and excluding interventional studies.
For the mortality outcome, we retrieved or calculated the adjusted odds ratio and corresponding 95% confidence interval from each study. When adjusted estimates were unavailable, we used unadjusted ones. When a hazard ratio was reported, we considered it to approximate the relative effect measure reported in other studies that used odds ratios and tested this assumption in sensitivity analysis. For door to balloon times, we retrieved or calculated odds ratios with 95% confidence intervals for the proportion of patients who received percutaneous coronary intervention within 90 minutes. We estimated the mean difference and 95% confidence interval from the mean and standard deviation. When the standard deviation was not reported, we imputed it by using the methods reported in the Cochrane Handbook for Systematic Reviews.
We used the I statistic to estimate the percentage of total variation across studies due to heterogeneity rather than chance (ranging from 0% to 100%). I values of 25% or less, 50%, and 75% or greater represent low, moderate, and high inconsistency. We used the random effect model to pool results across studies, accounting for between study variance. We chose this model because of anticipated significant heterogeneity between studies in terms of population and methods used to ascertain outcomes. We used the Q statistic to assess the presence of statistically significant heterogeneity. To assess the potential effect of publication bias, we inspected funnel plots for asymmetry and used the Duval and Tweedie trim and fill method and the Begg and Mazumdar rank correlation test.
We used Comprehensive Meta-Analysis, version 2 (Englewood, New Jersey) for statistical analysis. All P values are two tailed, and we set P<0.05 as the threshold for significance.
Method
This study was conducted according to guidance from the Cochrane Handbook of Systematic Reviews and is reported according to PRISMA (preferred reporting items for systematic reviews and meta-analyses) recommendations.
Data Sources and Search Strategies
We did a comprehensive search of several databases from database inception to April 2013, any language. The databases included Ovid Medline in-process and other non-indexed citations, Ovid Medline, Ovid Embase, Ovid Cochrane Database of Systematic Reviews, and Scopus. An experienced librarian designed and conducted the search strategy with input from the principal investigator. He used controlled vocabulary supplemented with keywords to search for comparative studies of off-hour effects for patients with acute myocardial infarction. The strategy is available in the Web Appendix (Table A). We also manually searched PubMed, Ovid Medline, and references in pertinent articles that were identified in the screening processes.
Study Selection
We considered all studies published in any language, with any study design, that evaluated the association between time of presentation and mortality or door to balloon times among adult patients who presented with acute myocardial infarction. Studies were eligible if they compared the outcomes between patients with off-hour versus regular hour presentation. We categorized the comparison of off-hours versus regular hours as weekend and night versus weekday regular hours, weekend versus weekday, or night versus daytime (Appendix Figure A). Time of presentation could be measured by arrival at the hospital, admission to hospital, or start of percutaneous coronary intervention. Eligible mortality outcomes included in-hospital mortality and 30 day mortality. Initial screening of abstracts excluded non-relevant or non-original studies. We then used full text screening to assess eligibility. Whenever reports pertained to the same set of patients, we retained the one with the most recent year of publication to obtain the most updated data. Two investigators (AS and AA) independently screened reports. Studies with discrepant decisions in screening of the abstract proceeded to full text screening. We resolved discrepancies in full text screening through consensus. We calculated a Κ statistic to quantify the agreement between the two reviewers on study selection.
Data Extraction
We recorded information on studies’ characteristics and demographics such as authors, publication year, country, years of enrollment of the cohort, data source, the definition of off-hours and regular hours, time of presentation measured for allocation, and the inclusion and exclusion criteria, as well as per group sample size, characteristics of the population, variables adjusted for, and outcomes. We recorded mortality by either number or proportion of deaths in each group and odds ratio or hazard ratio with confidence intervals. We recorded door to balloon time by either mean with standard deviation or median with interquartile range or by the proportion of patients whose door to balloon time was 90 minutes or less. For studies that implemented an intervention that could affect the outcome, we abstracted data from the observational (pre-intervention) period. One investigator abstracted data, which a second investigator independently verified. The discrepancies found in the verification process were solved by consensus or further review by a third investigator. SRS, DAR, and KMT abstracted data, and AS and AA verified data.
Assessment of Methodological Quality
We assessed the methodological quality of the included studies by using the Newcastle-Ottawa scale. This scale consists of three domains (cohort selection, comparability, and outcome) and evaluates the study’s overall risk of bias. The maximum score for an observational study is 9 points. Two investigators independently assessed quality, and another resolved discrepancies.
Outcome Definition and Subgroup Analyses
Mortality Outcomes. We used in-hospital or 30 day mortality as the main outcome. For studies without in-hospital mortality results, we used 30 day mortality when available. We did the main analysis for all studies combined. We also separately analyzed each mortality outcome (in-hospital versus 30 day). For the main outcome, we did subgroup analyses by diagnosis of patient cohort (STEMI versus non-STEMI), type of off-hours (weekend and night versus weekend versus night), measured time of presentation (arrival versus admission versus start of percutaneous coronary intervention), data source (clinical registry versus administrative data), region (North America versus Europe versus others), and outcome adjustment (adjusted versus unadjusted). To evaluate the possibility of a time trend effect of mortality across studies, we did meta-regression using the mid-year of enrollment of the cohort as the independent variable and the natural log of the effect size as the dependent variable. Owing to concern about potential overlapping patient sets, we did sensitivity analyses by excluding each single cohort and by including only one cohort from each study. We also did sensitivity analyses by excluding studies that expressed results as a hazard ratio.
Door to Balloon Time. We analyzed the proportion of patients with STEMI whose door to balloon time was less than 90 minutes and the mean or median door to balloon times. For mean or median door to balloon times, we did subgroup analyses by type of off-hours determination, measured time of presentation, and region, as well as meta-regression using the mid-year of enrollment of the cohort to evaluate time trends in door to balloon times. We also did sensitivity analyses limiting to studies that included only patients who were directly admitted to the hospital and excluding interventional studies.
Statistical Analysis
For the mortality outcome, we retrieved or calculated the adjusted odds ratio and corresponding 95% confidence interval from each study. When adjusted estimates were unavailable, we used unadjusted ones. When a hazard ratio was reported, we considered it to approximate the relative effect measure reported in other studies that used odds ratios and tested this assumption in sensitivity analysis. For door to balloon times, we retrieved or calculated odds ratios with 95% confidence intervals for the proportion of patients who received percutaneous coronary intervention within 90 minutes. We estimated the mean difference and 95% confidence interval from the mean and standard deviation. When the standard deviation was not reported, we imputed it by using the methods reported in the Cochrane Handbook for Systematic Reviews.
We used the I statistic to estimate the percentage of total variation across studies due to heterogeneity rather than chance (ranging from 0% to 100%). I values of 25% or less, 50%, and 75% or greater represent low, moderate, and high inconsistency. We used the random effect model to pool results across studies, accounting for between study variance. We chose this model because of anticipated significant heterogeneity between studies in terms of population and methods used to ascertain outcomes. We used the Q statistic to assess the presence of statistically significant heterogeneity. To assess the potential effect of publication bias, we inspected funnel plots for asymmetry and used the Duval and Tweedie trim and fill method and the Begg and Mazumdar rank correlation test.
We used Comprehensive Meta-Analysis, version 2 (Englewood, New Jersey) for statistical analysis. All P values are two tailed, and we set P<0.05 as the threshold for significance.
