Development and validation of a method to compare outcomes between switchers and continuers in routine clinical practice

Aim: In clinical practice, patients may initiate and continue with a treatment (continuers) or switch to another treatment (switchers). Comparison of clinical outcomes between these cohorts can be challenging owing to several factors including differences in risk profiles, particularly when switching is related to the occurrence of a clinical event of interest and the time-varying nature of the risk of an event. Analyses may be biased if these factors are not considered, including determining the appropriate start point to evaluate outcomes. We developed and validated a method to address these issues. Materials & methods: The proposed method (SMARTS) assigned random pseudo-switching times to continuers, matching the distribution of actual switching times among switchers. Baseline characteristics at (pseudo-) switching time were balanced using propensity score matching, inverse probability of treatment weighting (IPTW) or standardized morbidity ratio weighting. For validation, we conducted a factorial simulation with two clinical scenarios: sicker switchers where switching helps (true post-switching HR = 0.7) and healthier switchers where it harms (true post-switching HR = 1.5), crossed with three hazard trends (constant, increasing, decreasing). A time-varying confounder followed different trajectories for switchers and continuers, with five analysis methods evaluated each condition. The utility of the developed method was assessed in a real-world study. Results: Under increasing hazard, the conventional approach (evaluating continuers from treatment initiation) showed substantial timing bias. For the sicker-switchers scenario (post-switching HR = 0.7), conventional IPTW HR was 1.06 (bias = +0.36); SMARTS IPTW reduced this to 0.78 (bias = +0.08). For the healthier-switchers scenario (post-switching HR = 1.5), conventional IPTW HR was 1.80 (bias = + 0.30); SMARTS IPTW reduced this to 1.41 (bias = -0.09). Under decreasing hazard, timing bias reversed direction: notably, for the healthier-switchers scenario (post-switching HR = 1.5) appeared protective using the conventional approach (IPTW HR = 0.61); SMARTS correctly identified the harmful effect (IPTW HR = 1.46). Under constant hazard, both approaches performed well, confirming that SMARTS specifically addresses timing bias. Application of the method improved results in a real-world study comparing outcomes between continuers and switchers. Conclusion: When hazard varies over time, evaluating continuers from treatment initiation introduces timing bias that can reverse the apparent direction of treatment effect. SMARTS, combined with confounder adjustment via propensity score matching, IPTW or standardized morbidity ratio weighting, substantially reduced this bias across diverse clinical scenarios and hazard trends.

We developed a method to address confounding effects in comparing clinical outcomes between patients who continue or switch treatments. Our approach improves accuracy using pseudo-switching times. #Pseudo-switching #TreatmentSwitching #BiasReduction