For a study, researchers sought to forecast the conditional average treatment effect (CATE) for people on anti-CD20 monoclonal antibodies and laquinimod for a study. For this, they used a multitask multilayer perceptron (MLP), a form of artificial neural network. ORATORIO, OLYMPUS, and ARPEGGIO, 3 clinical trials studying Disease Modifying Therapy (DMT)s in Primary Progressive Multiple Sclerosis (PPMS), collected baseline clinical and imaging data. These randomized, double-blind, placebo-controlled trials looked at ocrelizumab, rituximab, and laquinimod. A shuffled combination of anti-CD20-Abs from ORATORIO and OLYMPUS was divided into a training (70%) and testing (30%) dataset. ARPEGGIO data was used as an additional source of external validation. To calculate the CATE, a multitask MLP was trained to predict the rate of disability development on both active and placebo treatment. Across a range of anticipated impact sizes, the trained model distinguished between responders and non-responders. When selecting the model’s prediction for the top 25% of most responsive people (HR 0.442, P=0.0497), the average treatment effect for the anti-CD20 testing dataset was considerably greater (HR 0.442, P=0.0497), compared to HR 0.787 (P=0.292) for the full group. The same model was used to identify laquinimod responders, with a significant treatment impact found in the top 20% of responders (HR 0.275, P=0.028). In the responder subgroup, investigators found enrichment across a wide range of baseline characteristics: younger, more men, shorter illness duration, higher disability scores, and more lesional activity. The only feature that did not vary between responders and non-responders was normalized brain volume. Even though no substantial treatment effect can be identified at the whole-group level, it was possible to identify subgroups of individuals who respond favorably to anti-CD20-Abs or laquinimod depending on their baseline characteristics. Predicting treatment response in PPMS could aid with therapy selection in the clinic and increase the efficiency of future phase 2 clinical trials using predictive enrichment.
Source:www.abstractsonline.com/pp8/#!/10495/presentation/51
