Surgical versus medical treatment and mortality in patients with infective endocarditis: The effect of selection bias, immortal time bias, and time dependent confounding
Background: Infective endocarditis (IE) is a serious infection of the endocardium. Available evidence on the comparative effect of medical versus surgical treatment of IE is limited by systematic error. It is unclear if surgery, when compared with medical treatment, is protective against long-term mortality. Objectives: 1) To complete a systematic review of the published studies that compare the effect of surgical versus medical IE treatment (modelled as time-dependent) on mortality. 2) To examine the effect of bias on the observed apparent long-term protective effect of surgery. 3) To determine whether surgical treatment of IE improves survival when compared to medical treatment. Methods: Using a retrospective population-based cohort study design, the effect of surgical versus medical treatment of IE on mortality was explored using advanced methods to examine and correct for bias including marginal structural models with inverse probability weighting, time-dependent exposure and confounders and the creation of adjusted survival curves. Results: The study cohort included 16,380 adult Ontarians with IE. In models with time-dependent treatment and confounders, the average risk of mortality at 30 days was increased in patients who underwent surgical versus medical treatment, (HR=1.22, 95% CI: 1.01-1.47), while those who underwent surgery had a decreased risk of mortality at one year (HR=0.75, 95% CI: 0.67-0.84) when compared to those treated medically. Conclusions: This study demonstrates that bias in published studies likely contributed to the discrepant evidence of surgical versus medical treatment on survival. We are also the first to show the effect of time-dependent confounding in IE research, and that the effect of time-dependent confounding may vary between males and females. Future research into the effect of surgical versus medical treatment of IE on mortality should aim to limit three important sources of bias present in previously published IE studies: immortal time bias, selection bias, and time-dependent confounding.