Abstract 5836: The Effect of Survivor Bias on Surgical Outcomes in Infective Endocarditis
Introduction Recent studies have demonstrated conflicting results regarding the benefits of surgery in infective endocarditis (IE). In particular, studies have not accounted for potential survivor bias associated with surgery since patients who survive longer have more opportunities to undergo surgery.
Hypothesis We explored the effect of survivor bias on surgical outcomes in IE.
Methods We studied 223 consecutive patients admitted with left-sided IE to two centers between 1996–2006. All-cause mortality at one year was compared between medically-treated patients and those undergoing surgery using a Cox proportional-hazards model with adjustment for a patient’s propensity to surgery. To adjust for survivor bias, analyses were repeated with surgery as a time-dependent covariate.
Results Compared to medically-treated patients (n= 161), patients undergoing surgery (n= 62) had significantly lower mortality at one year (21.0% vs 38.5%, p= 0.017) with an unadjusted hazard ratio of 0.47 (95%CI 0.26–0.86, p= 0.014). However, patients undergoing surgery were significantly younger (46.9± 15.1 yrs vs 57.9± 19.2 yrs, p< 0.001) and had lower Charlson comorbidity scores (0.7± 1.5 vs 1.4± 1.9, p= 0.007), but were more likely to have severe heart failure (NYHA FC III-IV, 27% vs. 12%, p< 0.001). Baseline differences between groups were taken into account by developing a regression model for surgery and assigning a propensity score for surgery to each patient based on this model. Significant predictors of surgery were NYHA class (p< 0.001), aortic valve involvement (p< 0.001), severe regurgitation (p= 0.016), and lower hemoglobin at baseline (p= 0.012). After adjustment for the propensity score, there remained a significant benefit for surgery (hazard ratio 0.24, 95%CI 0.11–0.52, p< 0.001). However, the benefits of surgery were diminished and no longer significant when it was evaluated as a time-dependent covariate (hazard ratio 0.87, 95%CI 0.43–1.78, p= 0.87), suggesting that the apparent benefits of surgery may be related to survivor bias.
Conclusions Survivor bias may have a significant effect on the evaluation of outcomes associated with surgery in IE and such bias can be reduced with the use of time-dependent covariate analysis.