A mixture model for differentiating longitudinal courses of multiple sclerosis

Abstract

Background: Multiple sclerosis has a broad spectrum of clinical courses. Early identification of patients at greater risk of accumulating disability is essential.

Objectives: Identify groups of patients with similar presentation through a mixture model and predict their trajectories over the years.

Methods: Retrospective study of patients from 1994 to 2019. We performed a latent profile analysis followed by a latent transition analysis based on eight parameters: age, disease duration, EDSS, number of relapses, multi-topographic symptoms, motor impairment, sphincter impairment, and infratentorial lesions.

Results: We included 629 patients, regardless of the phenotypical classification. We identified three distinct groups at the beginning and end of the follow-up. The three-classes model disclosed the "No disability regardless disease duration" (NDRDD) class with low EDSS and younger patients, the "Disability within a short disease duration" (DSDD) class with the worse disability besides short illness, and the "Disability within a long disease duration" (DLDD) class that achieved high EDSS over a long disease duration. EDSS, disease duration, and no sphincter impairment had the best entropy to distinguish classes at the initial presentation. Over time, the patients from NDRDD had a 52.1 % probability of changing to DLDD and 7.7 % of changing to DSDD.

Conclusions: We identified three groups of clinical presentations and their evolution over time based on considered prognostic factors. The most likely transition is from NDRDD to DLDD.

Keywords: Disability; Latent profile analysis; Latent transition analysis; Mixture model; Multiple sclerosis; Prognosis.