APPLICATION OF CAUSAL FOREST TO IDENTIFY IMPORTANT FEATURES TO EVALUATE HETEROGENEOUS TREATMENT EFFECTS OF DISEASE-MODIFYING THERAPIES IN MULTIPLE SCLEROSIS
Author(s)
Yinan Wang, PhD, MPP, Jieni Li, PhD, MPH, Ying Lin, PhD, Rajender R. Aparasu, PhD, FAPhA.
University of Houston, Houston, TX, USA.
University of Houston, Houston, TX, USA.
OBJECTIVES: Selecting the initial treatment strategy is critical in managing multiple sclerosis (MS). Initiation of moderate-efficacy disease-modifying therapies (meDMTs) is the most common practice for MS patients, while high-efficacy disease-modifying therapies (heDMTs) are increasingly considered due to greater availability and better effectiveness. This study applied a causal forest model to identify important features for evaluating heterogeneous treatment effects (HTEs) of DMTs for MS patients.
METHODS: We used 2014-2024 Merative MarketScan Commercial healthcare claims data to identify MS patients. The study selected patients who initiated heDMTs or meDMTs with a 12-month washout period. One-to-one propensity score matching was used to balance baseline characteristics between patients in heDMTs and meDMTs. The outcome of interest was any MS relapse within a 12-month follow-up after the index date. We developed two causal forest models to detect important features for evaluating HTE. The first model used the “grf” package in R and modeled the absolute risk of relapse. The other model used the “rrcf” package and modeled the relative risk of relapse. Hyperparameters of the first model were tuned using cross-validation through functions within the “grf” package. The second model used the default settings of hyperparameters in the “rrcf” package.
RESULTS: After matching, we included 2,814 matched MS patients. The proportion of MS relapse for the heDMT and meDMT cohorts was 11.23% and 10.87%, respectively. The top five important features identified by the absolute risk model were obesity status, brainstem symptoms, anticonvulsant medication use, hypothyroidism, and bladder/bowel symptoms. The top five features identified by the relative risk model were other neurological disorders, full-time employment, brainstem symptoms, obesity status, and a high number of MS-related outpatient visits.
CONCLUSIONS: Comorbidities and MS-related clinical features may influence the comparative effectiveness of treatment between heDMTs and meDMTs. Further research is needed to evaluate the role of these factors in HTE of DMTs.
METHODS: We used 2014-2024 Merative MarketScan Commercial healthcare claims data to identify MS patients. The study selected patients who initiated heDMTs or meDMTs with a 12-month washout period. One-to-one propensity score matching was used to balance baseline characteristics between patients in heDMTs and meDMTs. The outcome of interest was any MS relapse within a 12-month follow-up after the index date. We developed two causal forest models to detect important features for evaluating HTE. The first model used the “grf” package in R and modeled the absolute risk of relapse. The other model used the “rrcf” package and modeled the relative risk of relapse. Hyperparameters of the first model were tuned using cross-validation through functions within the “grf” package. The second model used the default settings of hyperparameters in the “rrcf” package.
RESULTS: After matching, we included 2,814 matched MS patients. The proportion of MS relapse for the heDMT and meDMT cohorts was 11.23% and 10.87%, respectively. The top five important features identified by the absolute risk model were obesity status, brainstem symptoms, anticonvulsant medication use, hypothyroidism, and bladder/bowel symptoms. The top five features identified by the relative risk model were other neurological disorders, full-time employment, brainstem symptoms, obesity status, and a high number of MS-related outpatient visits.
CONCLUSIONS: Comorbidities and MS-related clinical features may influence the comparative effectiveness of treatment between heDMTs and meDMTs. Further research is needed to evaluate the role of these factors in HTE of DMTs.
Conference/Value in Health Info
2026-05, ISPOR 2026, Philadelphia, PA, USA
Value in Health, Volume 29, Issue S6
Code
EPH202
Topic
Epidemiology & Public Health
Topic Subcategory
Safety & Pharmacoepidemiology
Disease
SDC: Neurological Disorders