To improve the efficiency of ALS drug development clinical trials. 

We previously developed regression models for total ALSFRS-R score, ALSFRS-R subscores, vital capacity (VC), and percent expected VC, and time-to-event models for loss of speech, use of wheelchair, gastrostomy, use of NIV (using time to 50% expected VC as a surrogate) and survival.

ALS models were developed using the PRO-ACT database. We initially clean the data, run a preliminary random forest model for variable reduction, then refine the models using gradient boosting machines. The models are validated using internal and external datasets. Regression models were characterized using RMSD and bias analysis and time-to-event models were characterized using ROC curves, calibration (log-rank) and discrimination (C-index). The models were used to stratify patients and create drug development tools that were assessed using simulated trials with randomly drawn PRO-ACT data.

The models were utilized to create drug development applications, including a novel method of subgroup analysis, “detectable effect cluster analysis” (DEC) to identify subgroups with significant effect sizes. In addition, we used predicted outcomes to develop improved tools for enrichment, randomization and covariate adjustment. Finally, we developed virtual controls for situations without a placebo arm and where a placebo is not possible, ethically challenging or impossible to blind.

The detectable effect cluster (DEC) analysis, enrichment, randomization/covariate and virtual control tools find “hot spots” of patients with demonstrable benefit, decrease trial heterogeneity, lower sample size/increase power, and provide an objective measure of efficacy for drug development trials in ALS. DEC analysis shows great promise in identifying subgroups within a failed trial that could have formed a successful trial. These applications represent a significant paradigm shift with broad implications for the conduct of trials in ALS in particular and can be extended to a range of neurodegenerative diseases.