Real-world data (RWD) are an important complement to randomized, controlled and registry datasets in defining a disease course longitudinally. There is growing interest in understanding the insidious progression in multiple sclerosis (MS) that can occur despite aggressive relapse prevention, as well as how diversity and comorbidities impact multiple sclerosis (MS) patients, particularly in the era of the coronavirus (COVID19) pandemic.
We aim to derive RWD from a diverse cohort of approximately 4,000 MS patients in Northern California to pair with biomarkers from the Sutter-wide Precision Medicine Biobank – a longitudinal biorepository with a healthy aging comparator cohort. This pilot of 34 patients evaluates the integration of several data sources to extract key information about disease course. From the EHR, we use a combination of text processing, automated data element extraction, manual chart curation, and patient- and physician-targeted questionnaires to form a real-world dataset of interpretable outcome metrics.
This is an ambidirectional cohort study of subjects at least 18 years old, with a defining diagnosis of MS from at least one hospitalization or two outpatient encounters. Data elements including demographics, medication orders and comorbidities were directly extracted from the EHR. MRI reports in text format were stored in an Epic Clarity database, and neurology notes were mined for terms indicating stability versus worsening. Manual curation was used to transform prose clinician notes into tabular-format outcome scores.
We curated 9930 total encounters, 136 brain MRI reports and 137 spine MRI reports. We found 7.5 years (+/- 3.3) of data per patient in this pilot of 34 patients. 79% of patients were female, 21% male; 68% white, 26% black and 6% other/not disclosed. The most common disease-modifying therapies used were natalizumab, dimethyl fumarate and glatiramer acetate. 68% of patients had at least one comorbidity, 35% specifically had hypertension. Using automated and manual data methods, we were able to compile metrics of clinical and radiographic worsening versus stability from information in the EHR.
Our methods may be used to generate interpretable data on a system-wide scale from the comprehensive, longitudinal data of an EHR. These RWD can be paired with biospecimens, research assessments, and other datasets to add to the diversity of data on MS natural history and medication response.