ACCOUNTING FOR TIME-VARYING CONCOMITANT MEDICATION USE IN TREATMENT EFFECT ESTIMATION USING MARGINAL STRUCTURAL MODELS (MSMS): CASE STUDY IN PROSTATE CANCER
Author(s)
Neal D Shore, MD1, Elaine Gallagher, BA, MSc2, Daisy Gaunt, PhD3, Philip Orishaba, MSc3, Howard Thom, MSc, PhD4, Noman Paracha, MSc5, David Aceituno, MD, PhD6, Alicia Morgans, MD, MPH7.
1Carolina Urologic Research Center, Myrtle Beach, SC, USA, 2Bayer, Zürich, Switzerland, 3Clifton Insight, Bristol, United Kingdom, 4University of Bristol, Bristol, United Kingdom, 5Bayer, Basel, Switzerland, 6MD, PhD, Clifton Insight, London, United Kingdom, 7Dana-Farber Cancer Institute, Boston, MA, USA.
1Carolina Urologic Research Center, Myrtle Beach, SC, USA, 2Bayer, Zürich, Switzerland, 3Clifton Insight, Bristol, United Kingdom, 4University of Bristol, Bristol, United Kingdom, 5Bayer, Basel, Switzerland, 6MD, PhD, Clifton Insight, London, United Kingdom, 7Dana-Farber Cancer Institute, Boston, MA, USA.
OBJECTIVES: Concomitant medications can affect therapeutic responses and thus clinical outcomes, and medication use can vary throughout the patient journey. Using proton pump inhibitor (PPI) use in prostate cancer as a case study, we applied marginal structural models (MSMs) to account for the dynamic use of co-medications when estimating treatment effect of darolutamide.
METHODS: We analyzed anonymized individual participant data (IPD) from three prostate cancer trials: ARAMIS, ARASENS and ARANOTE. Baseline PPI use was first assessed using inverse probability of treatment weighting (IPTW), adjusted survival curves, and interaction models. Because PPI use changed over time, we then applied trial-specific marginal structural models (MSMs) to account for time-varying exposure. These models used stabilized IPTW for PPI exposure and censoring, followed by weighted Cox models. The primary outcome was overall survival (OS).
RESULTS: The pooled IPD cohort included 2052 patients. Baseline weighted analyses showed no OS difference between PPI users and non-users (weighted log-rank p=0.40). Accounting for time-varying PPI use did not change the original treatment effect estimates for OS in ARASENS, ARANOTE, or ARAMIS. Treatment-by-PPI interaction p-values were 0.19, 0.94 and 0.55, respectively. Stabilized weights were generally acceptable, supporting feasibility despite low PPI prevalence in two trials.
CONCLUSIONS: MSMs provide a useful causal framework for evaluating concomitant medications when exposure might change during the patient journey. Across three phase 3 trials, PPI use did not impact darolutamide’s survival benefit, with no treatment-by-PPI interaction observed.
METHODS: We analyzed anonymized individual participant data (IPD) from three prostate cancer trials: ARAMIS, ARASENS and ARANOTE. Baseline PPI use was first assessed using inverse probability of treatment weighting (IPTW), adjusted survival curves, and interaction models. Because PPI use changed over time, we then applied trial-specific marginal structural models (MSMs) to account for time-varying exposure. These models used stabilized IPTW for PPI exposure and censoring, followed by weighted Cox models. The primary outcome was overall survival (OS).
RESULTS: The pooled IPD cohort included 2052 patients. Baseline weighted analyses showed no OS difference between PPI users and non-users (weighted log-rank p=0.40). Accounting for time-varying PPI use did not change the original treatment effect estimates for OS in ARASENS, ARANOTE, or ARAMIS. Treatment-by-PPI interaction p-values were 0.19, 0.94 and 0.55, respectively. Stabilized weights were generally acceptable, supporting feasibility despite low PPI prevalence in two trials.
CONCLUSIONS: MSMs provide a useful causal framework for evaluating concomitant medications when exposure might change during the patient journey. Across three phase 3 trials, PPI use did not impact darolutamide’s survival benefit, with no treatment-by-PPI interaction observed.
Conference/Value in Health Info
2026-11, ISPOR Europe 2026, Vienna, Austria
Value in Health, Volume 29, Issue 12S
Code
MSR22
Topic
Clinical Outcomes, Methodological & Statistical Research, Study Approaches
Topic Subcategory
Confounding, Selection Bias Correction, Causal Inference
Disease
Oncology