A patient is diagnosed with multiple sclerosis once they meet the McDonald criteria of dissemination in space and time. Studies of cohorts of patients with multiple sclerosis need a reproducible way to determine an accurate date of diagnosis. We developed an automatic data-driven algorithm to determine the date when the MacDonald criteria are met, which we validated with the Registre Lorrain des Scleroses en Plaques (ReLSEP), a regional French registry of patients with multiple sclerosis.

We developed an algorithm to determine the date of diagnosis based on clinical and paraclinical data adapted from the four versions of the McDonald criteria. For validation, the dates of diagnosis generated by the algorithm were compared with those determined by an expert physician using the patients’ files as the gold standard. We calculated the sensitivity and specificity of the algorithm to provide a date, then we tested the equivalence between the dates of the gold standard and the algorithm (two-one-sided-t-test).

The algorithm used every possibility of determining dissemination in space and time according to the four sets. The sensitivity of the algorithm was 100% for the four sets, and specificity ranged between 95 and 100%. The difference between the dates of diagnosis found by the physician and the algorithm was usually less than 2 weeks (equivalence test P<0.0001).

The algorithm appears to be an efficient surrogate to accurately determine dates of diagnosis of multiple sclerosis in datasets of patients.