Descriptive statistics were used to characterize sample demographics. Comparisons between surgeons with recent SI and surgeons without recent SI were tested using Wilcoxon rank sum, Mann-Whitney, and Fisher exact tests. Such comparisons with approximately 7300 and 500 surgeons reporting in the 2 groups have 80% power to detect an average difference of 11% times the SD, a small effect size.26- 27 Accordingly, the P values in this report are not as important as the observed effect sizes. Consistent with recent advances in the science of QOL assessment,26 we a priori defined a 0.5 SD in QOL scores as a clinically meaningful effect size.26- 27 Linear regression was used to evaluate the incremental effect of each measure of distress on recent SI. In addition, the odds ratio (OR) for recent SI associated with screening positive for depression or each 1-point change in burnout or QOL score was calculated. The multivariable associations among demographic characteristics, professional characteristics, and distress with recent SI were assessed using logistic regression. Both forward and backward elimination methods were used to select significant variables for the models in which the directionality of the modeling did not affect the results. The independent variables used in these models included age, sex, relationship status, spouse/partner current profession, having children, age of children, subspecialty, years in practice, hours worked per week, hours per week spent in the operating room, number of nights on call per week, practice setting (private practice, academic medical center, Veteran's Affairs hospital, active military practice, not in practice or retired, or other), current academic rank, primary method of compensation (eg, salaried, incentive-based pay, or mixed), percentage of time dedicated to non–patient-care activities (eg, administration, education, or research), self-perceived medical error in the previous 3 months, depression, and burnout. All analyses were done using SAS version 9 (SAS Institute Inc, Cary, North Carolina) or R (R Foundation for Statistical Computing, Vienna, Austria; http://www.r-project.org). A likelihood ratio test was used to test the overall fit of the model. The likelihood ratio test compares the likelihood function of the final model with the likelihood of the reduced model. A significant P value for this test indicates that the expanded model fits the data better than the reduced model. Since the hazard ratio measures magnitude of risk rather than a model's ability to accurately classify individuals, the C statistic was also used to further evaluate the discriminatory value of the model for predicting SI.28 The C statistic estimates the proportion of correct predictions of the model (C = 1 indicates perfect discrimination between those with and without SI; C = 0.5 is equivalent to chance).