An Instrumental Variable Approach to Account for Informative Treatment Switching in Real-world Evidence

2026-07-01

AI Summary

This study addresses the bias in treatment effect estimation arising from informative treatment switching in real-world evidence studies. The authors propose a doubly robust instrumental variable estimator that treats the unobserved potential outcomes under switching as a confounder. Building upon a structural cumulative survival model, the method constructs estimating equations using baseline treatment assignment as an instrumental variable to identify causal effects. Notably, it does not require the existence of a never-switch subgroup and integrates baseline survival adjustment with a cross-fitting strategy, allowing compatibility with flexible machine learning nuisance models. Simulation studies demonstrate that the proposed approach substantially outperforms existing benchmarks across various scenarios, effectively reducing both bias and inconsistency. The method is successfully applied to a real-world comparative effectiveness analysis of second-line therapies for multiple sclerosis.

Abstract

Reproducible and generalizable assessment of treatment decisions requires principled handling of subsequent treatment switching that may inform expected outcomes and shift across cohorts and over time. To effectively account for informative treatment switching, we propose an instrumental variable approach that characterizes the poorly documented expected outcomes at switching as unmeasured confounding. After establishing the baseline treatment as a viable instrumental variable, we constructed an estimating equation based on the association between the centered instrumental variable and a martingale style residual process that identifies the treatment effect under structural cumulative survival model. Our proposed method is doubly robust, i.e., valid whenever either of baseline propensity model or no-switching outcome model is consistently estimated. A co-training of treatment effect parameter and survival outcome regression model eliminated the requirement of observing a no-switching subset under semi-parametric additive hazards models. We further developed an baseline-survival-corrected cross-fitting approach to incorporate general machine learning models for estimating nuisance models. Numerical results demonstrated the validity of our method in various settings when a basket of benchmark solutions produced biased or contradictory results. We applied our method to comparison of high-efficacy vs standard efficacy disease modifying treatments as the second line therapy of multiple sclerosis.

Problem

• informative treatment switching
• real-world evidence
• unmeasured confounding
• treatment effect estimation
• instrumental variable

Innovation

• instrumental variable
• informative treatment switching
• doubly robust
• structural cumulative survival model
• cross-fitting