OBJECTIVES: Multiple sclerosis (MS) subtypes are clinically relevant to inform treatment and health economic decisions. Various definitions are used to identify MS subtypes in administrative claims data, however coding practices have not been assessed.

METHODS: We used the MSDS-AOK PLUS database, which links claims data from the German AOK PLUS sickness fund and patient charts from the Multiple Sclerosis Management System 3D (MSDS-3D) to assess congruency of claims-based MS subtype coding (ICD-10-GM G35.-) and MSDS-3D patient records. We selected a cohort of 179 patients with at least one recorded subtype and date of recording (index) in MSDS-3D between 01/10/2015-30/06/2019. Claims data six months before and after each index was used to define the closest subtype diagnosis using a definition requiring one inpatient and/or two outpatient diagnoses in different quarters from any specialty or GP. The predictive performance of claims-based diagnoses was assessed against MSDS-3D through a multi-class confusion matrix. Subsequently, the positive and negative prediction values (PPV, NPV) were calculated for each subtype.

RESULTS: From 179 patients, there were 1,163 MS subtype recordings in MSDS-3D (mean 6.50), with only 18 patients (8.94%) presenting >1 different subtype within the inclusion period. For 148 observations (12.73%), an MS diagnosis could not be assigned in claims using our definition in the one-year period surrounding the index. For 465 cases (39.98%), only a G35.9 unspecified diagnosis was assigned vs. 0.69% unspecified cases in MSDS-3D. Congruency across all observations in the two datasets was 32.50% (378/1,163). PPV was highest for RRMS (G35.1, 0.953) and lowest for CIS (G35.0, 0.034), whereas NPV was highest for PPMS (G35.2, 0.983) and lowest for RRMS (0.108).

CONCLUSIONS: The overall low predictive performance of specific MS subtypes in claims data highlights the importance of linking to patient medical records to supplement detailed clinical information for real-world evidence studies in MS.