Elucidating symptoms clusters in multiple sclerosis using patient reported outcomes and unsupervised machine learning

项目总结

PROJECT SUMMARY Multiple sclerosis (MS) is a chronic disease affecting 900,000 persons in the U.S, and it is a leading cause of disability among young adults. Persons with MS (PwMS) experience wide-ranging symptoms across multiple domains, alone or in combination, with varied severity. Some of these symptoms include optic nerve dysfunction and vision problems, muscle weakness, bladder/bowel dysfunction, tremors, cognitive and emotional problems, and incoordination. The objectives of the current application are to identify and characterize symptom patterns and clusters in PwMS, which are aligned with PA-17-462, that states: “multiple sclerosis (is a) … model condition to advance (symptom) cluster research”. Thus, our analytical framework will inform research in other poly-symptomatic conditions. Stakeholders agree that the benefits of patient reported outcomes (PROs) and measures (PROMs) have not reached their full potential for PwMS. PROs, which provide invaluable insight into the patients’ perspective, are increasingly used in MS clinical trials and clinical practice as standard clinical measures fail to adequately measure impairment across domains or lack sensitivity to detect subtle but meaningful change. Aligning with ongoing global MS initiatives, the current proposal will focus on identifying and characterizing symptom patterns and clusters for PROs; which is also directly aligned with NOT-OD-20-079, a Notice of Special Interest to stimulate “research to improve the interpretation of PROs at the individual patient level for use in the clinical practice”. Furthermore, there is an additional incentive to maximize the use and interpretation of PROs considering the shift to telemedicine service in response to the COVID-19 pandemic. We have assembled a multi-disciplinary team of research scientists and clinical experts with access to two unparalleled data resources (discovery and validation data sets), the 1st being the North American Research Committee on Multiple Sclerosis (NARCOMS) Registry’s survey data for 21,558 PwMS spanning an average of 8.4 years (0.5 to 14 years) and 269,468 biannual surveys, and the 2nd being the structured electronic health records (EHRs) for 8,687 PwMS see at the Mellen Center for MS Treatment and Research (MCMS) at the Cleveland Clinic, spanning an average of 4 years (0.5-8.4 years) and 67,932 visits. In both resources, 11 MS-specific PROMs (MS-PROMs) were longitudinally captured including measures of mobility, dexterity, vision, fatigue, cognition, bladder/bowel, sensory, spasticity, depression, tremor/coordination, and pain. We propose the four complementary aims that will: 1. Characterize overall longitudinal impairment patterns for each 11 MS-PROMs; 2. Identify distinct clusters of Pw MS with similar symptom patterns within and across functional domains using machine learning approaches; 3. Develop new approaches to assess the strength of causal inference and identify sources of model prediction errors in unsupervised machine learning; and 4. Create a dynamic simulation dashboard for predicting MS phenotypes based on the findings of aims 1-3. With these aims, we seek to advance MS phenotyping to facilitate improvements in research, clinical care, and approaches to self-management. By focusing on PROMs, we will leverage the experience of PwMS which is independent of their location (i.e. applications to rural residents) and ideal for telemedicine. We hope that our findings will advance care and empower PwMS to engage in health decisions where personalized phenotypic characterization is necessary.