A new study applying explainable machine learning to real-world, multicentric clinical data from 247 patients with Parkinson’s disease highlights the significant challenges inherent in forecasting the onset of motor complications. The research focused on predicting the development of levodopa-induced dyskinesias and motor fluctuations over a three-year period. By employing a transparent prognostic framework, researchers sought to establish a more reliable approach to risk stratification for patients currently free of these specific complications.
The performance evaluation of the predictive models yielded moderate results, with a Matthews correlation coefficient (MCC) of 0.28 for levodopa-induced dyskinesias and 0.32 for motor fluctuations. To assess the interpretability of these models, the team utilized SHAP-based techniques. This analysis confirmed that the model predictions were consistent with established clinical knowledge, identifying levodopa duration and the total Levodopa Equivalent Daily Dose as the primary drivers of risk. Specifically, the data indicated that the risk for new-onset complications increases significantly when the dosage surpasses a threshold of 300–400 mg.
A critical component of the study involved an ablation test, which underscored the necessity of including diverse clinical presentations in training datasets. When researchers excluded patients who already suffered from pre-existing complications from the training process, the model’s sensitivity collapsed. This finding demonstrates that access to the full spectrum of disease severity is essential for developing robust risk stratification tools, as focusing solely on patients who have yet to develop complications appears to limit the model’s predictive utility.
The methodological stress-test conducted by the research team ultimately reveals that reliance on baseline clinical features alone results in limited absolute sensitivity. These findings emphasize that while machine learning can provide insight into risk factors, the complexity of Parkinson’s disease trajectory requires more than static baseline data. The authors suggest that integrating dynamic, longitudinal patient information is a necessary evolution to achieve the level of clinical-grade individualized prediction required for optimizing long-term disease management.
Subject of Research: Prognostic machine learning models for motor complications in Parkinson's disease
Article Title: Explainable machine learning reveals the challenges of predicting new-onset motor complications in Parkinson’s disease
Article References: Endrizzi, W., Campese, N., Ragni, F., Moroni, M., Bovo, S., Longo, C., Gios, L., Uccelli, A., Giometto, B., Jurman, G., Osmani, V., Malaguti, M. C., the NeuroArtP3 Network, Falini, A., Castellano, A., Rossi, A., Tortora, D., Parodi, C., Verrico, A., … Bacchin, R. (2026). Explainable machine learning reveals the challenges of predicting new-onset motor complications in Parkinson’s disease. npj Parkinson's Disease. https://doi.org/10.1038/s41531-026-01550-1
Image Credits: AI Generated
DOI: 10.1038/s41531-026-01550-1
Keywords: Parkinson’s disease, machine learning, explainable AI, motor complications, levodopa, dyskinesia, prognostic modeling, clinical data, real-world evidence, risk stratification, neurodegenerative disorders, predictive analytics, SHAP, levodopa equivalent daily dose, clinical decision support
News Source: Diana Fleming. (November 16, 2026). Explainable Machine Learning Reveals Challenges in Predicting Parkinson’s Motor Complications. Scienmag.



