The following is a summary of “Machine Learning Characterization of a Rare Neurologic Disease Via Electronic Health Records: a proof-of-principle study on stiff person syndrome” published in the August 2024 issue of Neurology by Park et al.
Despite frequent diagnostic delays and statistical underpowering due to rarity, studying rare neurologic diseases (RND) and comorbidities, such as stiff person syndrome (SPS), an RND affecting 1-2/million annually with painful muscle spasms and rigidity, remains challenging.
Researchers conducted a retrospective study to manipulate underutilized EHR data and develop a machine-learning (ML)-based framework to identify clinical features characterizing SPS diagnosis.
They used an ML approach to analyze 319 features from the medical records of 48 individuals, 23 patients with SPS, and 25 controls who had high anti glutamic-acid-decarboxylase-65 (anti-GAD65). The algorithm identified critical features for diagnosing SPS using Support Vector Machines (SVM) and SHapley Additive exPlanation (SHAP) values, with results validated through cross-validation.
The results showed depression, hypothyroidism, gastroesophageal reflux disease (GERD), and joint pain were the most characteristic features of SPS. The SVM model achieved a precision of 0.817 (95% CI 0.795–0.840), sensitivity of 0.766 (95% CI 0.743–0.790), F-score of 0.761 (95% CI 0.744–0.778), AUC of 0.808, and accuracy of 0.775 (95% CI 0.759–0.790), providing a significant advancement in the understanding and diagnosis of SPS.
They concluded that the framework identified potential pathologic mechanisms of SPS linked to depression, hypothyroidism, and GERD, suggesting a novel approach to address diagnostic challenges in neurology through uncovering latent associations and generating hypotheses for RNDs.
Source: bmcneurol.biomedcentral.com/articles/10.1186/s12883-024-03760-7