To identify distinct clusters of persons with multiple sclerosis (PwMS) with shared impairment patterns across 11 functional domains.

PwMS experience wide-ranging symptoms across multiple domains, alone or in combination, with varied severity. We sought to characterize naturally occurring clusters of MS using seemingly heterogeneous longitudinal patient-reported outcome measure (PROMs) responses, which are increasingly used in clinical trials and practice.

From the electronic health records of 6,619 MS patients with extended care (≥2 encounters 6 months apart) at a tertiary referral center between 2008-2015, we abstracted responses for 11 validated MS-PROMs include 8 MS Performance Scales© (mobility, hand function, vision, fatigue, cognition, bladder/bowel, sensory, spasticity) and 3 MS Functional Scales (pain, depression, tremor/coordination). We applied unsupervised machine learning through mixture modeling (latent profile analysis) to the 11 MS-PROMs at baseline, to identify a set of discrete and non-overlapping clusters of MS patients with similar impairment patterns.

Nine patient clusters were detected, and were differentiated by low (4 clusters), medium (1 cluster), and high (4 clusters) levels of mobility impairment. The low mobility impairment clusters were differentiated by low impairment across domains (31.7%), to moderate fatigue (18.7%), to moderate fatigue + moderately-high sensory dysfunction (7.9%), and to moderately-high fatigue + pain (10%). The medium mobility cluster had moderately-high impairment across domains (10.2%). The clusters with high mobility impairments varied with mild impairment (8.3%) to high impairment (4.1%) across all domains, and two clusters with moderate impairment across domains but not sensory dysfunction (low [3.9%] vs high [5.1%]). 

We leveraged PROMs to innovatively deep phenotype MS patients into distinct subgroups with varying levels of disability and symptomatology.