The following is a summary of “Identifying Multiple Sclerosis Relapses from Clinical Notes Using Combined Rule-based and Deep Learning Methodologies,” published by Chin, et al.
This study aims to create an algorithm to extract MS relapse episodes from the American Academy of Neurology Axon Registry clinical notes, an electronic health record (EHR) -based neurology-specific patient registry. The frequency of relapses is an important indicator of disease activity in multiple sclerosis patients. Mechanisms to automatically extract this information will better facilitate real-world evidence (RWE) studies. MS relapses are generally captured in unstructured clinical notes rather than organized areas in the EHR.
Using structured ICD codes and the mention of MS in clinical notes, the Axon Registry identified 46,600 MS patients as of May 2022, out of 3.2 million patients with 24 million patient visits. The relapse status of these individuals (present relapse, no relapse, discussion of past relapse only, or unknown) at a given contact was classified using a rule-based, deep learning (DL) technique. Notes from one thousand MS patients were randomly selected and searched for terms indicative of relapse. A clinical expert assigned relapse labels to these notes, which were then used to create training, validation, and testing sets (70-15-15 split). Using the training and validation sets, a DL model was trained to categorize notes into one of the relapse statuses. The test set served as the basis for evaluation.
The total accuracy of the model was 0.88. There was an 83% sensitivity and 97% specificity for determining whether a relapse occurred recently. Identified MS patients averaged 0.58+/-0.55 relapses/year, and the proportion of patients with relapses decreased over time (2014: 14.74% vs. 2021: 9.86%), consistent with clinical expectations. Researchers employed a hybrid rule-based and DL approach to identify relapses in patient records. Performance metrics and clinically consistent patterns support this scalable approach for RWE research.
Source: index.mirasmart.com/aan2023/PDFfiles/AAN2023-002829.html
