Application of the Parametric g-Formula to Characterizing Counterfactual Time-to-Event Disability Progression Outcomes in Multiple Sclerosis

• Published in: Statistics in biosciences Vol. 17; no. 2; pp. 501 - 527
• Main Authors: Yiu, Sean, Wang, Qing, Mercier, Francois, Dahlke, Frank, Model, Fabian
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.07.2025; Springer Nature B.V
• Subjects: Biostatistics; Case Studies and Practice Articles; Computation; Health Sciences; Inference; Mathematics and Statistics; Medicine; Multiple sclerosis; Statistics; Statistics for Life Sciences; Theoretical Ecology/Statistics; Estimands; Path-specific effects; Repeated regression; Mediation analysis; Time-varying confounding
• ISSN: 1867-1764; 1867-1772
• DOI: 10.1007/s12561-024-09426-9

Disentangling the effects of disease activity associated with focal inflammation, e.g., as characterized by relapses, and other pathophysiological processes leading to disability progression in multiple sclerosis (MS) is of central importance. This is because it will provide an enhanced understanding of the pathology of MS, and may have implications for treating patients. Recently, this endeavor has lead researchers to characterize progression independent of relapse activity (PIRA), which has been defined as a counterfactual outcome representing disability progression without relapses. To date, several methods based on data pre-processing have been proposed to characterize PIRA, e.g., censoring onset of disability progression at onset of relapses. However, these methods can be subject to severe bias if relapse-progression confounders are present and lead to highly imprecise inference if most of the data is unused. In this article, we perform simulations to demonstrate that, unlike these methods, the parametric g-computation formula can provide unbiased inference on PIRA in the presence of relapse-progression confounders. Additionally, we utilize new results on the g-computation formula to estimate treatment effects that provide complementary information, and require less extrapolation to be estimated than the treatment effect on PIRA. Finally, we apply our proposed methodology to two phase 3 studies of MS to highlight the benefits and additional insights that it generates over standard methods.