EMR Data Can Be Used to Expand Research on Multiple Sclerosis

Electronic medical records (EMRs) collect clinical information for sharing, but the original goal of EMRs was not to be used in research. However, information of large real-world cohorts can potentially be used to expand clinical research opportunities in multiple sclerosis (MS), according to a new study published in Multiple Sclerosis Journal.

Although EMRs were not created with the primary goal of being used in research, clinicians and researchers increasingly use the data in them for that purpose, the authors explained.

“For [MS], several studies have used EMR extraction in various contexts and have proven that MS patients can be identified and various clinical data extracted from the EMR,” they wrote.

The authors identified and classified patients with MS in an EMR system at a large academic medical center, extracted the demographic and clinical data, assessed the reliability of these extraction processes, and validated the data by comparing the information with information collected in an MS research cohort database.

A total of 4142 patients with MS were identified in the EMR data and were compared with 337 patients in the research cohort. Patients were classified into 4 groups: Well-Defined MS Group; Probably MS Group A; Probably MS Group B; and Probably MS Group C.

Using the text from the visit notes of the 4142 patients in the EMR, the authors were able to extract 25,260 values for MS clinical variables: Expanded Disability Status Scale (EDSS) score, timed-25-foot walk (T25FW), MS subtype (relapsing remitting [RR], secondary progressive [SP], and primary progressive), and disease onset. The patients in the Probable Groups were less likely to have values for MS variables recorded compared with the patients in the Well-Defined MS Group, the authors noted. The authors calculated the intraclass correlation coefficient (ICC), which describes how strongly units in a group resemble each other, for EDSS, T25FW, and onset.

The authors found an ICC of 0.79 for T25FW and of 0.87 for EDSS, which indicates a high concordance of these values between the 2 databases. For onset year, there was high similarity between the 2 databases with an ICC of 0.87.

There was also substantial agreement between the 2 databases for subtype (Cohen’s kappa coefficient of 0.65). There was a slight imbalance between the 2 databases for patients in the SP subtype group. In the EMR, these patients were more likely to be categorized as RR, although they might be categorized as SP for research purposes, the authors explained.

There were limits with potential information biases of EMR-based data, according to the authors. As the average EDSS of a patient increased, the raw counts of encounters and extracted values increased.

“These associations may result from the correlation between total time of follow-up, severity of the disease, and number of encounters,” the authors noted.

Ultimately, the authors concluded that extraction of clinical data can happen with high reliability and that the EMR data is generally consistent with research data, which is important because EMR data will likely be increasingly used for research and for personalized medicine. Some potential applications for EMR data that the authors highlighted include observational analyses identification of candidates to participate in traditional clinical research trials.

“We are still at the beginning of systematic use of clinical ‘big-data’ in medical research,” the authors wrote. “Increasing access to large amounts of electronically available data generated by clinical activity will open up exciting possibilities to bridge observations made using traditional research cohorts and controlled trials with observations made using ‘real-world’ clinical data.”

Reference

Damotte V, Lizée A, Tremblay M, et al. Harnessing electronic medical records to advance research on multiple sclerosis. Mult Scler J. 2019;25(3):408-418. doi: 10.1177/1352458517747407.