Neuroscience PhD Candidate Ricardo Diaz-Rincon recently had his paper, “CASCADE Conformal Prediction: Uncertainty-Adaptive Prediction Intervals for Two-Stage Clinical Decision Support,” accepted at the ICML 2026 Workshop on Statistical Frameworks for Uncertainty in Agentic Systems (AgenticUQ), held at the International Conference on Machine Learning, the leading internation machine learning conference.
This work introduces CASCADE, a novel framework that addresses a critical gap in AI-assisted Parkinson’s disease medication management. When AI systems guide dosing decisions, standard methods treat the question of whether a medication change is needed and the question of how much as independent decisions, discarding uncertainty from the first before it informs the second. CASCADE resolves this by mathematically linking both stages, automatically widening prediction ranges when the system is uncertain and sharpening them when it is confident.
Validated on a decade of deidentified electronic health records from inpatient admissions at UF Health, CASCADE achieved a four-fold improvement in its ability to differentiate between straightforward and complex cases while maintaining statistically guaranteed reliability.
By equipping neurologists with explicit uncertainty signals rather than false confidence, CASCADE has the potential to transform medication management in Parkinson’s care, moving from trial-and-error dosing toward evidence-based decisions that could meaningfully improve patient outcomes.