The trained model was able to estimate weekly domain completion within ±7.86% over the course of therapy. Prediction error increased with duration during the first month of therapy, but stabilized to ±11% from 6-12 weeks, demonstrating the model’s ability to predict at longer durations. A patient’s first week score was the most important model feature, with model accuracy decreasing by an average of 59.5% (±4.2%) with its exclusion. Age and the average number of days practiced per week were both associated with slower therapeutic progress, and were important model features (5% decrease in predictive accuracy with their exclusion).