(A-210) Validation and improvement of an algorithm identifying relapses in multiple sclerosis (MS) within the French nationwide claims database

Background: MS is a neurological disease with varying degrees of disability for patients. Detecting accurately MS relapses is a key step for conducting (pharmaco)epidemiological studies on this disease in claims databases.

Objectives: The objective of this study was to validate and improve a recently developed algorithm to identify MS relapses in patients initiating disease-modifying therapy (DMT) and naïve to any other MS treatments within the SNDS nationwide claims database.

Methods: Clinical data from the French MS registry (OFSEP) linked to the SNDS were used. A SNDS cohort included all MS patients who received their first DMT (dimethyl fumarate, teriflunomide, fingolimod, injectable immunomodulators) between 07/2015 and 12/2017, naive to any other MS treatment during the 4.5 years prior to inclusion, and followed until 12/31/2018 (n=1,640). The initial algorithm was based on high-dose corticosteroid prescriptions and hospitalizations with an ICD-10 MS relapse compatible diagnosis (MS, encephalitis, myelitis, optic neuritis, etc.). The first relapse for each patient was identified by the algorithm during the follow-up in SNDS and compared to the OFSEP clinical data considered as the gold standard. Then, the algorithm’s criteria were adjusted based on expert opinion through a detailed review of healthcare consumption related to MS for false positive and false negative patients. The final algorithm was also tested adding non-naive MS treatment patients at inclusion (n=9,966). Algorithm’s performances were estimated using sensitivity (Se), specificity (Sp), positive predictive value (PPV), and negative predictive value (NPV).

Results: The performances of the initial algorithm were 85.2% for Sp, 74.0% for Se, 89.8% NPV and 65.3% for PPV. After modifications of the algorithm’s criteria regarding the minimum and maximum thresholds for corticosteroid dosages and hospitalization durations, performances of the final algorithm slightly improved: 85.0%, 75.4%, 90.2%, and 65.3%, respectively. When considering all patients on MS treatment (naïve or non-naïve), the Sp and NPV remain high (81.8% and 91.6%), while the Se and PPV decrease (to 75.1% and 55.3%, respectively).

Conclusions: The optimized algorithm can be used in the SNDS with a high level of accuracy in MS patients initiating a DMT and naive to any other MS treatment. Although the modification of the algorithm did not improve performances, they are clinically sound. The reduced frequency of visits to OFSEP centers by patients with older diseases following a relapse provides a partial explanation for the increase in the number of false positives in non-naïve patients. Performances in this subgroup are more likely to be underestimated.