Our client has a proprietary wearable device and has gathered a vast collection of clinical trial data from Parkinson's disease patients. Our client wanted to find ways to extract valuable insights to distill clinician decision-making insights, and assess the predictive power in their dataset.
The client’s data came from diverse trials, each with unique protocols and procedures, resulting in a wide variety of data types, including time series, surveys, written notes, and medication classifications. To realize the potential of this data, we needed to leverage AI/ML techniques to uncover digital biomarkers to characterize patients, learn from prior outcomes, and recommend the next best action within a digital therapeutic.
Our team started by identifying the most critical therapeutic objectives. We tested our hypotheses against the client's data to assess the predictive signal. We engineered a data ingest and merge pipeline within a standardized data science framework to connect and unify datasets from heterogeneous sources.
We engineered novel features to enhance predictive signal within the data. We achieved remarkable success, demonstrating strong signal with exceptional AUC scores in the 0.9 range.
These results underscore the potential of digital biomarker discovery and its application in improving disease management and treatment.