To estimate the conditional average treatment effect (CATE) of disease modifying therapies (DMTs) in primary progressive multiple sclerosis (PPMS) using machine learning.

Only one DMT, ocrelizumab, was found to slow disability progression in PPMS. Modeling CATE could help identify a subgroup of individuals who experience a greater benefit from DMTs, allowing for predictive enrichment to increase the power of future phase 2 clinical trials.

Baseline clinical and imaging data was obtained as part of three clinical trials investigating DMTs in PPMS: ORATORIO, OLYMPUS and ARPEGGIO. These placebo-controlled trials investigated ocrelizumab, rituximab and laquinimod, respectively. A shuffled mix from ORATORIO and OLYMPUS (anti-CD20-Abs) was separated into a training (70%) and testing (30%) dataset. Data from ARPEGGIO served as additional external validation. An ensemble of multitask multilayer perceptrons was trained to predict the rate of disability progression on both active treatment and placebo to estimate the CATE.

The model could separate responders and non-responders across a range of predicted effect sizes. The average treatment effect for the anti-CD20 testing dataset was significantly greater when selecting the model’s prediction for the top 25% most responsive individuals (HR 0.442, p=0.0497), compared to HR 0.787 (p=0.292) for the entire group. The same model could also identify responders to laquinimod, finding a significant treatment effect in the top 20% most responsive individuals (HR 0.275, p=0.028). Simulating a 1 year study where the 50% most responsive individuals were randomized led to a 6-fold reduction in the number of participants needed to achieve 80% power to detect a significant difference.

Subgroups of individuals with PPMS who respond favourably to DMTs can be identified based on their baseline characteristics, even when no significant treatment effect can be found at the whole-group level. Doing so allows for predictive enrichment of future clinical trials and personalized treatment selection in the clinic.