Multiple sclerosis (MS) is a chronic illness in which the body’s immune system attacks myelin, the substance that surrounds and protects the nerve fibers of the central nervous system. The result is a disabling disease that causes damage to the brain and spinal cord over time. Symptoms vary from mild to severe and can include paralysis and loss of vision. MS impacts over 2.3 million people across the world, the majority being women, who are most often diagnosed between 20-50 years of age. Although disease-modifying treatments (DMTs) became available in the 1990s, to date there remains no cure.
Once diagnosed, clinicians use the Expanded Disability Status Scale (EDSS) to evaluate and measure functional levels in MS patients. Scores can help determine levels of care and treatment needed. The EDSS considers impairment in seven functional systems on an ordinal scale, with scores ranging from 0 to 10 in 0.5 increments. A score of 0 indicates normal neurological status while scores of 6 and higher indicate the need for increasing ambulatory aid. Although widely used in clinical trials and despite its importance in monitoring disease progression, use of the EDSS in routine clinical practice is limited due to the time required to complete the scale and the complexity of scoring.
Active prospective research contributes to our understanding of the disease, but clinical trials and collecting prospective longitudinal data via registries is costly and time intensive. Real-World Data (RWD), the data routinely collected in patient care, and Real-World Evidence (RWE), the clinical evidence derived from RWD, can help provide critically important retrospective information about the course of illness, subtypes of MS, and treatment response.
To address gaps in functional assessments of disability for patients with MS, OM1 used RWD and machine learning methods to generate estimated EDSS (eEDSS) scores. We started with data from the OM1 PremiOM™ MS Dataset that includes deep clinical information from patients followed by a neurologist, with a minimum of two diagnostic codes for MS or one diagnostic code and a prescription for a DMT for MS.
Nearly 14,000 MS patients with clinical notes from a neurologist were screened for a clinician EDSS score that was extracted from the notes using medical language processing (MLP). This method resulted in a test set of 684 patients with 3,489 scores that was further divided and randomly assigned to a model training cohort (75%) or a model validation cohort (25%). Demographics, patient characteristics, and common comorbidities were similar between the training and validation cohorts.
We then built a model based on the EDSS scores from the unstructured clinical notes from neurologists in the training cohort. Only notes that had clinical evaluation sections (including the history of the present illness, review of systems, physical exam, and clinical assessment with plan) with sufficient clinical detail to generate an eEDSS score were included. These clinical evaluation sections were chosen because they contain information related to the evolution and severity of MS and, as such, were deemed critical for generating the estimation model. The model features included explanatory terms and phrases indicating mobility impairments; the presence or negation of symptoms such as loss of balance, speech impairment, vision impairment, muscle spasms, and others. Explicit indications of disease progression such as improvement or deterioration of symptoms and mentions of medications were also among the key model features.
This data science methodology resulted in a model with extraordinary accuracy (see detailed results in Multiple Sclerosis Journal). We then used the model to amplify the number of available assessments from the original test set to nearly 190,000 scores on over 13,000 patients – a nearly 20x increase in the total available scores.
The combination of the available EDSS and eEDSS scores increase the ability to study disease progression, including insights into the transition from relapsing-remitting to secondary progressive MS and enables assessment of earlier treatment interventions. These insights ar important, as getting treatments to MS patients sooner has been demonstrated to improve clinical outcomes.
High quality, deep clinical data with specialist notes, medications and key outcomes are critical for building and using these types of models and studies. Innovative research methods, like OM1’s data science approach to amplify a critical outcome measure, demonstrates the opportunities that come from the use of RWD for addressing gaps in functional assessments, improving RWE insights, and understanding disease progression and treatment effectiveness for patients with MS and beyond.
With specialization in chronic conditions, OM1 is re-imagining real-world data and evidence by developing large electronically connected networks of clinicians and health data in specialty areas like neuroscience. Leveraging its extensive clinical networks and artificial intelligence (AI) platform, OM1 offers industry-leading enriched healthcare datasets, research analytics, data modeling, decision support, and retrospective and prospective clinical studies. With a focus on high-quality data and clinical outcomes, the offerings are used for accelerating research, demonstrating treatment effectiveness, supporting regulatory submissions, monitoring safety, and informing commercialization.
Learn more about the PremiOM MS Dataset here or email firstname.lastname@example.org to schedule a meeting with Dr. Marci.
Article Sources: National MS Society.org, FDA.gov
 Alves P, Green E, Leavy M, Friedler H, Curhan G, Marci C, Boussios C, “Validation of a Machine Learning Approach to Estimate Expanded Disability Status Scale Scores for Multiple Sclerosis,” Multiple Sclerosis Journal: Experimental, Translational and Clinical (June 2022); https://doi.org/10.1177%2F20552173221108635