New machine learning model predicts Parkinson’s disease risk up to 15 years in advance

A recent study published in Neurology suggests that individuals at high risk for Parkinson’s disease can be identified up to 15 years before symptoms appear. Researchers used machine learning to analyze blood proteins and clinical data, developing a predictive model that could detect Parkinson’s risk early, offering the potential to prevent or delay disease progression. Parkinson’s, the second most common neurodegenerative disease, typically presents motor symptoms like tremors and stiffness after significant brain damage has already occurred. The disease often has a long preclinical phase marked by non-motor symptoms such as sleep disturbances and depression, which are not always recognized as early signs. The challenge in treating Parkinson’s lies in its late diagnosis, when brain damage is irreversible. Current treatments mainly manage symptoms rather than slowing disease progression. However, identifying high-risk individuals early could enable more effective interventions. Led by Jian-Feng Feng and Wei Cheng at Fudan University, the study analyzed data from over 50,000 participants in the UK Biobank, focusing on the levels of 1,463 proteins in blood plasma. The researchers trained a machine learning model on this data, predicting Parkinson’s risk based on protein levels and clinical information, such as age and history of brain injury. The model achieved high accuracy and identified 22 key proteins associated with Parkinson’s, including neurofilament light (NfL), a marker of brain damage. The study also revealed that changes in certain proteins, such as rising NfL levels, began over a decade before a Parkinson’s diagnosis. While promising, the study has limitations, such as a lack of diversity in participants and potential misdiagnoses based on medical records. Further research is needed to refine the model and confirm its effectiveness in diverse populations. Despite these challenges, the study represents a significant step toward non-invasive, early detection of Parkinson’s. Early identification of at-risk individuals could lead to interventions that slow disease progression, improving long-term outcomes.