Survivor treatment selection bias and outcomes research: a case study of surgery in infective endocarditis

Recent studies in infective endocarditis have suggested an association between surgery and reduced mortality. However, these studies did not account for survivor treatment selection bias, which is an underrecognized source of error in observational studies. Therefore, we sought to evaluate the effects of survivor bias on surgical outcomes in infective endocarditis.

We studied 223 patients admitted with left-sided infective endocarditis between 1996 and 2006 and compared all-cause mortality between surgically treated and medically treated patients using Cox regression analysis. Propensity scores were used to account for selection bias, and time-dependent analyses were performed to account for survivor bias. Compared with medical patients (n=161), surgical patients (n=62) had lower mortality during a median follow-up of 5.2 years (32% versus 51%; P=0.02) with an unadjusted hazard ratio of 0.54 (95% CI, 0.33 to 0.88, P=0.01). After adjustment for baseline differences in propensity for surgery and risk of mortality, there remained a significant benefit for surgery (hazard ratio, 0.50; 95% CI, 0.28 to 0.88; P=0.02). However, this was diminished after time-dependent analysis (hazard ratio, 0.77; 95% CI, 0.42 to 1.40; P=0.39). Conditional Kaplan-Meier analyses confirmed the effect of survivor bias because the apparent benefit of surgery was primarily attributable to excess mortality in the medical group during early hospitalization when surgery was not frequently performed.

Survivor bias significantly affects the evaluation of surgical outcomes in infective endocarditis, and it should be considered in other areas of outcomes research where randomized controlled trials are not feasible. Survivor bias is not corrected by propensity analysis alone but may be reduced by time-dependent survival analysis.