A team from the University of Miami has developed machine-learning models that can identify Parkinson’s disease patients at higher risk for rapid cognitive or motor decline three to five years before symptoms worsen — using data that neurologists already collect during routine visits. Published in npj Parkinson’s Disease, the study trained models on 1,602 participants from the Parkinson’s Progression Markers Initiative and validated them on an independent cohort of 541 patients. The models achieved AUROC values above 0.80, demonstrating strong predictive accuracy. Surprisingly, structural MRI data — which researchers expected to be a key predictor — added relatively little value beyond what routine clinical assessments already provided. For motor decline, the most informative predictors were a synuclein seed amplification assay (which detects abnormal alpha-synuclein biology) and the rate of change in motor scores during the first year after diagnosis. For cognitive decline, the strongest predictors included early cognitive worsening and motor decline trajectory. “One of the most difficult things about Parkinson’s disease is not knowing how it will go,” said Dr. Ihtsham ul Haq, the study’s lead. “We wanted to use AI’s ability to analyze many kinds of information at once to see if it could predict who would have a more rapid decline”. The team now plans to investigate whether similar approaches can predict rapid decline in Alzheimer’s and other neurodegenerative diseases.
About the Author
