Leveraging electronic health records to optimize treatment selection and response in multiple sclerosis

点击翻译按钮获取中文摘要

PROJECT SUMMARY AND ABSTRACT The rapid expansion of approved multiple sclerosis (MS) disease-modifying therapies (DMTs) and the diverse individual variation in treatment response contribute to the critical unmet need for individually tailored treatment strategy for the nearly 3 million persons with multiple sclerosis (pwMS) worldwide. The shift towards a precision medicine approach to guide treatment selection based on individual profiles will improve patient outcome by ensuring prompt initiation of effective DMTs while avoiding ineffective DMTs. To fill the knowledge gaps due to the absence of randomized clinical trial evidence and to advance precision medicine for pwMS, it is crucial to harness available clinical data and develop approaches deployable at the point of care. Electronic health records (EHR) data contain a wealth of longitudinal real-world clinical information and provide a complementary platform for clinical discovery. Building on our prior research efforts, the proposed study has the overall goal to optimize DMT selection and patient outcomes in pwMS using EHR data. We will use EHR data from two academic healthcare systems, both ideally positioned as they contain longitudinal clinical information of thousands of pwMS and hold crucial linkage to MS research registries that provide the ground truth. For additional validation, we will use integrated claims and EHR data from a large population of commercially insured pwMS. Aim 1: Compare relapse outcomes across DMTs. We will test the hypothesis that confounder correction using full EHR features yields more robust and consistent results in DMT effectiveness comparison analysis than expert-selected covariates. We will test the generalizability by using a transfer learning approach. Aim 2: Identify patient clusters based on DMT prescription sequences over time. We will test the hypothesis that DMT prescription sequences inform differential patient clusters and MS outcomes. We will apply a covariate-adjusted mixture Markov Model. Aim 3: Identify optimal DMT sequences that predict favorable treatment response. We will test the hypothesis that optimized DMT prescription sequence(s) through reinforcement learning could improve MS outcomes (i.e., relapse rate, patient-reported outcomes). This research will close knowledge gaps due to absent randomized clinical trial evidence and limited real-world evidence to guide optimal MS treatment selection. It will bring precision medicine closer to pwMS by developing clinically deployable strategies to optimize treatment selection. This project is consistent with the mission of the NINDS to reduce the burden of neurological diseases such as MS.