Predicting Parkinson's With Biomarkers & Machine Learning AI

The scale of Parkinson's Disease diagnoses presents a significant challenge for healthcare systems worldwide.

According to the US non-profit Parkinson's Foundation, nearly 90,000 people are newly diagnosed with Parkinson's Disease each year in the United States.

Parkinson's is a progressive brain disorder that damages dopamine-producing neurons, leading to symptoms including tremors, stiffness, slow movement and non-movement issues, such as depression or sleep issues.

The condition's highly individual nature means healthcare teams face considerable uncertainty when developing long-term care pathways.

According to Parkinson's UK, "In 2025, around 28,000 people in the UK were predicted be diagnosed with Parkinson’s. That’s someone being diagnosed every 20 minutes."

The NHS writes: "In the early stages, your GP may find it difficult to say whether you definitely have the condition because symptoms are usually mild."

This diagnostic ambiguity underscores why healthcare providers have traditionally relied on reactive treatment approaches.

Though no cure exists, researchers at Adelaide University are using machine learning to move beyond reacting to symptoms and start predicting them, fundamentally changing how healthcare systems allocate resources.

Addressing the hidden burden

Parkinson's is largely characterised by its clinical presentation like the visible tremors or gait changes a doctor sees during an appointment. However, this focus on motor symptoms may be causing healthcare providers to overlook a substantial portion of the disease burden.

Up to 85% of people with the disease will be affected by neuropsychiatric symptoms, such as cognitive impairment and mood dysfunction.

These non-motor symptoms are often the silent "major predictors of quality of life and mortality", according to Adelaide University. For healthcare systems, this represents a significant gap in care delivery that may be contributing to poorer patient outcomes.

The challenge has always been the unknown of what the disease will look like for an individual five years down the line, making clinical management incredibly difficult and placing additional strain on specialist services.

Researchers from Adelaide University's School of Pharmacy and Biomedical Sciences and the Australian Institute for Machine Learning turned to the Parkinson's Progressive Markers Initiative database.

By analysing biomarkers found in cerebrospinal fluid and neuroimaging, the team discovered they could predict a patient's five-year trajectory with far greater accuracy than clinical exams alone.

"Across both studies, biomarkers had utility for improving prediction of how an individual's symptoms would present at the five-year follow-up, beyond clinical symptom presentation alone," says Associate Professor Lyndsey Collins-Praino, the senior author of the research published in the Journal of Geriatric Psychiatry and Neurology.

The findings revealed specific biological patterns for different types of progression.

For healthcare providers, this stratification could enable more targeted interventions and appropriate resource allocation based on predicted care needs.

Machine learning enhances care planning

The breakthrough lies in the application of machine learning to sort this data into clusters.

By using statistical and machine learning techniques, the team showed that a multi-modal panel significantly improves the ability to forecast an individual's future health.

"Such enhanced understanding has the potential to directly impact clinical management of Parkinson's, leading to enhanced monitoring, such as earlier specialist referral and more personalised management strategies," says Lyndsey.

The integration of wearable sensors and AI-driven blood tests is further refining this.

New tools like NeuroDiscovery AI and lightweight 1D convolutional neural network (CNN) models are now allowing for real-time monitoring of tremors and gait, catching subtle pathological signs months before they appear in a clinic.

By combining digital biomarkers with biological markers, researchers are working towards predictive models that address the unique progression of every individual.

For healthcare systems managing growing Parkinson's patient populations, these tools could represent a pathway towards more sustainable, proactive care models.