Modeling disease severity in multiple sclerosis using electronic health records.

Area of Science:

• Neurology
• Biomedical Informatics
• Health Data Science

Background:

• Electronic Health Records (EHR) offer scalable data for clinical research.
• Accurate patient identification and extraction of meaningful clinical measures are crucial for leveraging EHR data.
• Multiple Sclerosis (MS) serves as a model for complex neurological disorders.

Purpose of the Study:

• To develop and validate an algorithm using EHR data to identify patients with MS.
• To extract clinically relevant surrogate measures of MS disease severity from EHR data.
• To demonstrate the utility of EHR data for research-grade disease assessment.

Main Methods:

• A cross-sectional observational study involving 5,495 MS patients identified via EHR.
• An algorithm combining codified and narrative EHR data with natural language processing (NLP) was employed.
• Routinely collected EHR data was used to derive the Multiple Sclerosis Severity Score (MSSS) and Brain Parenchymal Fraction (BPF).

Main Results:

• The EHR-based MS identification algorithm achieved an Area Under the Curve (AUC) of 0.958, with 83% sensitivity and 92% positive predictive value.
• EHR-derived MSSS and BPF showed moderate correlations with true values (R²=0.38 and R²=0.22, respectively).
• Derived MSSS effectively differentiated disease severity between relapsing-remitting and progressive MS phenotypes (p < 1.56×10⁻¹²).

Conclusions:

• Sophisticated analysis of codified and narrative EHR data enables accurate identification of MS patients.
• EHR data can provide estimations of established MS severity indicators like MSSS, previously limited to research settings.
• This methodology is adaptable for identifying patients and assessing disease severity in other complex neurological conditions.