Smart watches using AI to detect early signs of Parkinson’s

In the recent study, researchers examined data gathered by smart watches during a seven-day period, focusing on the participants' movement speed. By utilising artificial intelligence (AI), they were able to accurately forecast individuals who would subsequently develop Parkinson's disease.

According to the researchers, this breakthrough has the potential to serve as a novel screening tool for Parkinson's disease, allowing for early detection of the condition compared to existing diagnostic methods. The study was led by scientists from the UK DRI and the Neuroscience and Mental Health Innovation Institute (NMHII) at Cardiff University.

Parkinson's disease primarily impacts dopaminergic neurons, which are situated in the substantia nigra region of the brain. This neurodegenerative disorder manifests with motor symptoms including tremors, rigidity, and bradykinesia (slowness of movement). Unfortunately, by the time these distinctive symptoms become apparent, and a clinical diagnosis is possible, over 50% of the cells in the substantia nigra have already undergone cell death.

Hence, there is a pressing requirement for affordable, dependable, and readily available approaches to identify early alterations in order to intervene before the disease inflicts significant harm on the brain.

The study involved the analysis of data obtained from a large cohort of 103,712 participants in the UK Biobank. These individuals wore medical-grade smart watches for a period of seven days between 2013 and 2016. Throughout the entire week, the smart watches recorded and measured the average acceleration, which is indicative of the speed of movement.

The researchers conducted a comparison between two groups using the collected data. The first group consisted of participants who had already received a diagnosis of Parkinson's disease, while the second group included individuals who were diagnosed with Parkinson's up to seven years after the data from the smartwatches was collected. These two groups were then compared to a control group comprising healthy individuals who were matched in terms of age and sex.

The study demonstrated that by employing AI, it is feasible to detect individuals who would subsequently develop Parkinson's disease based on their smart watch data. Not only were these participants distinguishable from the healthy control group within the study, but the researchers also extended their findings to show that the AI model could identify individuals in the general population who would later develop Parkinson's disease. Remarkably, the AI-based approach outperformed any other risk factor or recognised early indicator of the disease in predicting its development. Furthermore, the model had the capability to predict the time until an individual would receive a diagnosis of Parkinson's disease.

A limitation of the study is the absence of replication using alternative data sources due to the unavailability of comparable datasets for conducting similar analyses. However, the researchers addressed this limitation by conducting thorough evaluations and implementing measures to minimise biases in their study. While replication using different datasets would strengthen the findings, the extensive evaluation and mitigation of biases contribute to the credibility of the study's results.

Study leader Dr Cynthia Sandor, Emerging Leader at the UK DRI at Cardiff, said: “Smart watch data is easily accessible and low-cost. As of 2020, around 30% of the UK population wear smart watches. By using this type of data, we would potentially be able to identify individuals in the very early stages of Parkinson's disease within the general population.

“We have shown here that a single week of data captured can predict events up to seven years in the future. With these results we could develop a valuable screening tool to aid in the early detection of Parkinson’s. This has implications both for research, in improving recruitment into clinical trials, and in clinical practice, in allowing patients to access treatments at an earlier stage, in future when such treatments become available.”

Dr Kathryn Peall, Clinical Senior Lecturer in the NMHII at Cardiff, said: “For most people with Parkinson’s disease, by the time they start to experience symptoms, many of the affected brain cells have already been lost. This means that diagnosing the condition early is challenging. Though our findings here are not intended to replace existing methods of diagnosis, smart watch data could provide a useful screening tool to aid in the early detection of the disease. This means that as new treatments hopefully begin to emerge, people will be able to access them before the disease causes extensive damage to the brain.”