Search

Background: For conditions with rare clinical outcomes, real-world treatment comparisons are challenging to design and prone to confounding. Objectives: To present a robust methodologic approach for rigorous and transparent assessment of rare outcomes using real-world data. Methods: We emulated a target trial using an active comparator, new-user design to compare dronedarone to sotalol for rhythm control in atrial fibrillation (AF) as both are recommended for similar patient phenotypes. Using one protocol, a pre-specified stepwise approach was implemented across 4 datasets (Optum CDM; IBM MarketScan; Veterans Affairs Electronic Health Records; Swedish National Patient Register). Meta-analysis was used to ensure sufficient capture of specific, rare primary outcomes (cardiovascular (CV) hospitalization and ventricular proarrhythmia) and to evaluate consistency of findings across patient populations. Steps 1–3 focused on cohort selection, propensity score matching (PSM), baseline equipoise and residual confounding assessment via negative control outcome analyses. In steps 4–6, outcomes in the individual cohorts were analyzed using an as-treated approach and Cox proportional hazards models. Step 7 included a heterogeneity assessment, meta-analysis using fixed effects models, and hypothesis testing using a hierarchical approach. In step 8, sensitivity analyses, including E-values and Inverse Probability of Censoring Weighting, were conducted to verify the robustness of findings. Results: In step 1, 35,467 sotalol and 27,955 dronedarone patients with AF who were antiarrhythmic drug-naive were identified across databases. In steps 2–3, 23,275 dronedarone patients were PS-matched to 23,275 sotalol patients. Baseline covariates were well-balanced and little-to-no residual confounding was observed via the negative control analyses. Individual HRs were estimated in steps 4–6, and, when no significant heterogeneity between databases was observed, hazard ratios (HRs) were pooled

Survival analysis, or time-to-event analysis, is an important and widespread problem in healthcare research. Medical research has traditionally relied on Cox models for survival analysis, due to their simplicity and interpretability. Cox models assume a log-linear hazard function as well as proportional hazards over time, and can perform poorly when these assumptions fail. Newer survival models based on machine learning avoid these assumptions and offer improved accuracy, yet sometimes at the expense of model interpretability, which is vital for clinical use. We propose a novel survival analysis pipeline that is both interpretable and competitive with state-of-the-art survival models. Specifically, we use an improved version of survival stacking to transform a survival analysis problem to a classification problem, ControlBurn to perform feature selection, and Explainable Boosting Machines to generate interpretable predictions. To evaluate our pipeline, we predict risk of heart failure using a large-scale EHR database. Our pipeline achieves state-of-the-art performance and provides interesting and novel insights about risk factors for heart failure.