Chronological Diagnostic Algorithm Predicting Neuropathology in Parkinsonism.

Abstract

Objective: Pre-mortem diagnosis of parkinsonism is often challenging due to atypical presentations, overlapping syndromes, and co-pathologies. This study aimed to develop a machine learning-based algorithm predicting neuropathology in parkinsonism using chronological clinical presentations, which has previously been underexplored.

Methods: Clinical information was automatically abstracted from medical records of the Mayo Clinic Brain Bank using fine-tuned Generative Pre-trained Transformer 4 models. Patients who developed parkinsonism within 3 years of disease onset were included. Six machine learning models were trained with age, sex, family history, and 197 clinical presentations paired with onset information to predict neuropathologic diagnoses, including co-pathologies.

Results: Among 7,825 donors, 949 met inclusion criteria, representing 9 neuropathologic categories: Lewy body disease (LBD; n = 128), LBD with Alzheimer's disease (AD; n = 136), progressive supranuclear palsy (PSP; n = 303), PSP with AD (n = 56), PSP with LBD (n = 27), multiple system atrophy (MSA; n = 120), corticobasal degeneration (CBD; n = 99), AD (n = 43), and frontotemporal lobar degeneration (FTLD; n = 37). The CatBoost algorithm achieved an area under the receiver operating characteristic curve of 0.83 across the 9 diagnostic categories at 3 years after onset. Important predictors included age at onset, restricted eye movement, and tremor. The model remained robust to incomplete data, requiring only 23 of 200 parameters for reliable predictions with an area under the curve of 0.80. The algorithm was implemented into a user-friendly program providing diagnostic probabilities with visualizations of parameter contributions.

Interpretation: This neuropathology-confirmed diagnostic algorithm provides a cost-effective and interpretable screening tool for parkinsonism, bridging biomarker testing and molecular-targeted therapies. ANN NEUROL 2026.