Simultaneously Dealing With Immortal Time Bias and Residual Confounding: A Case Study of a High-Dimensional Propensity Score Approach With a Nested Case–Control Framework in Multiple Sclerosis Research

Background
Observational studies of time-dependent treatments often face immortal time bias and residual confounding, complicating treatment effect estimation. We implemented a high-dimensional propensity score (hdPS) analysis within a nested case–control (NCC) framework to address both biases simultaneously.

Methods
We used a retrospective cohort of 19 360 individuals with multiple sclerosis (MS) in British Columbia, Canada, to examine the relationship between disease-modifying drugs (DMDs) and all-cause mortality. A 1:4 NCC analysis addressed immortal time bias, and hdPS was applied to handle residual confounding. Sensitivity analyses tested the robustness of findings across various hdPS parameters and matching strategies.

Results
We matched a total of 3209 cases to 12 293 controls in the NCC analysis, and demonstrated a 28% reduction in mortality risk associated with exposure to DMDs (hazard ratio [HR]: 0.72, 95% confidence interval [CI]: 0.62–0.84) in the NCC-hdPS analysis. Sensitivity analyses using different propensity score estimation techniques and control-matching strategies yielded consistent results, with HRs ranging between 0.70 and 0.77.

Conclusions
This study offers a practical framework for addressing immortal time bias and residual confounding simultaneously, improving the validity of effect estimates in real-world studies. We shared reproducible R codes for researchers to facilitate the adoption of this methodology in their research.