Disease progression in ALS is heterogeneous due to the varying presentation of clinical symptoms. This heterogeneity makes it difficult to accurately quantify longitudinal disease severity in people with ALS (pALS), making it difficult to determine the efficacy of therapeutic interventions. In this work, we explore a Bayesian Logistic Mixed-Effects model that can help predict individual trajectories in pALS. We used metrics extracted from 143 pALS who interacted with a cloud-based multimodal assessment platform comprising standard speaking exercises. We found that multimodal biomarkers can be predicted more accurately than the ALSFRS-R, the clinical gold standard to measure disease state, with dense and sparse training data. Such non-linear models have the potential to help with stratification of pALS into fast and slow progressors and thus inform treatment approaches. Patient stratification is also a key factor in designing clinical trials to test drug efficacy in slowing progression.