Authors: Sara Riggare, Jamie Luckhaus, Therese Scott Duncan, Uppsala University, Sweden

Reviewed by: Ali Saad, PhD, Anigma Technologies. Mónica M. Kurtis. MD, Movement Disorders Unit Director,Hospital Ruber Internacional, Madrid, Spain

Every person with Parkinson’s experiences a unique journey. Some people's symptoms progress slowly over many years, while others develop significant disability and more quickly. “What is going to happen to me?” is often one of the first questions asked by someone diagnosed with Parkinson's. The idea of using artificial intelligence (AI) to predict the course of an individual’s PD is compelling. In theory, machine learning algorithms could analyze large – including symptoms, genetic markers, brain scans, and lifestyle factors – to identify patterns and predict how quickly someone's condition may worsen or which symptoms may develop. Such predictions could help patients and doctors plan treatments and life adjustments in advance. However, despite progress in AI and data collection, accurately predicting PD progression remains an enormous challenge, and it raises important practical and ethical questions we must confront before clinical use.

The appeal of prognostic algorithms

Researchers are using studies such as the Parkinson's Progression Markers Initiative (PPMI), which collects data from thousands of participants to train AI systems that identify groups of patients with different progression patterns. Tech companies, including IBM with support from the Michael J. Fox Foundation, used PPMI to build models that may help identify early signs of aggressive Parkinson’s by discovering data-driven state and support more personalized care  [1].

AI is also making it easier to remotely monitor symptoms between clinic visits. At UC San Francisco, researchers created a system that uses smartphone video to analyze movements like walking or finger tapping [2]. This tool produces objective scores that can reveal subtle changes over time and confirm whether treatments are working. Such data-driven monitoring could help doctors adjust medications earlier and providing more personalized care.

Other projects use “passive” monitoring, that do not require user input, such as monitoring breathing during sleep. In one Nature Medicine study, researchers developed an AI model that analyzed signals from a simple belt or contactless sensor and assessed Parkinson’s severity as accurately as a neurological exam [3]. This technology requires no effort from patients, can be used at home, and may make expert care accessible even in remote areas.

Beyond monitoring, some AI models are being trained to predict outcomes like memory decline. One PPMI-based  model combining brain scans with clinical measures predicted significant cognitive problems within five years with about 89% accuracy [4]. Other models forecast motor changes, such as when a patient might need walking aids.

No biomarker, no crystal ball

Despite these advances, major challenge remains: Parkinson's still lacks a clear and validated biological marker of progression. Unlike, say, cholesterol for heart disease or viral load for HIV, there is no single blood test, brain scan, or measurement that definitively tracks how Parkinson’s is advancing. Doctors mainly rely on clinical rating scales (like the UPDRS), observed milestone events (e.g. needing a wheelchair), or composite scores to gauge progression – all imperfect proxies. This makes training an AI tricky: the “ground truth” it tries to predict is hard to define and measure consistently. Different studies may label progression differently, and patients vary widely in symptom patterns. An AI might learn to predict a certain outcome (for example, onset of dementia or rate of motor decline), but that outcome may not capture every individual’s experience of “progression”.

Moreover, models developed so far are not yet ready for clinical use. Machine learning models from PPMI data show promise but are still in research phases. They may work well on the dataset they were trained on, but could perform less accurately in the real-world clinic with more diverse patients. In short, AI predictions of PD progression are still experimental. People understandably want to know how their condition is likely to progress, but it remains extremely difficult to predict how Parkinson's will develop in any individual, and neurologists often refrain from answering these questions.

Do people want to know the future?

Suppose for a moment that we did have a highly accurate PD progression predictor – the next question is, do patients want this information? The answer isn’t straightforward. Many patients do ask about their prognosis, as it could help with life planning and setting expectations. Knowing one might face significant disability in, say, five years could influence decisions about retirement, finances, or bucket-list goals. And if a slower course is predicted, it could provide peace of mind. There is value in knowledge.

On the other hand, Parkinson’s is a case where detailed prognostic knowledge can be a double-edged sword. Not everyone may want to peer into that crystal ball. Parkinson’s advocate Sara Riggare reflected that had she been told as a teen that she had an “old person’s disease” like PD, she likely would have avoided pursuing university, a career, or starting a family – decisions which, in hindsight, she’s glad she made despite the diagnosis. Knowing the worst-case scenario early on might have curtailed her ambitions unnecessarily. Another person with PD, columnist Sherri Woodbridge, observed that none of her neurologists ever spelled out how bad Parkinson’s could get – and she isn’t sure she would have wanted them to. “I was doing quite well getting [to a dark place] on my own” with fear, she writes, “I didn’t need my neurologist to assist me further into the deep dark” (https://parkinsonsnewstoday.com/columns/knowing-preparing-what-future-holds/). Crucially, she notes doctors aren’t clairvoyant; any prediction is essentially an educated guess. “They don’t have the ability to tell your future… We think we want them to tell us what our future holds. Or do we?”. Her perspective highlights the emotional burden that a prognosis – especially a probabilistic one – can bring. It could cause anxiety or despair “borrowing trouble from tomorrow”, especially given that not every patient will  experience all potential symptoms of PD.

Thus, even if an AI could estimate that someone is a “fast progressor,” delivering that news must be done with great care and with the patient’s desires in mind. Some people will prefer to know all information, while others might choose not to hear grim predictions that might never fully materialize. This is analogous to genetic testing situations (for example, the choice to know one’s risk of Huntington’s disease or Alzheimer’s); personal preference is paramount.

Looking ahead: cautious optimism

The long-term goal is for AI-driven prediction tools to support a more proactive approach to Parkinson's care. Instead of a reactive model (waiting for significant symptom worsening before changing therapy), doctors and patients could anticipate changes and act early. For example, an AI might forecast that based on current trends, a patient’s walking ability will significantly decline within the next year – prompting the care team to intensify physical therapy and implement fall-prevention strategies now, rather than after falls occur. For patients, having an idea of their likely future could guide personal decisions and coping strategies, although it’s always important to remember that these predictions are probabilistic, not guarantees.

AI and big data are likely to play an increasingly important role in understanding Parkinson's. They might help uncover hidden patterns or subsets of PD that respond differently to treatments. In the future, if disease-modifying therapies become available, predicting progression could help target those therapies to the patients who need them most urgently. For now, though, AI-based progression prediction is in its infancy, and both patients and clinicians should view any such forecasts as tentative. The lack of a definitive biomarker and the variability of PD mean that current models have wide margins of error.

Even without predictive algorithms, there are many ways healthcare professionals, family members, support networks, and people living with Parkinson's can work together to support quality of life. While we cannot yet reliably predict how Parkinson's will progress in an individual, access to appropriate treatment, rehabilitation, social support, and healthy lifestyle opportunities may help maintain daily functioning and well-being. The most appropriate approach will differ from person to person.

In conclusion, the idea of using AI to predict the course of Parkinson's is promising and is no longer science fiction. Research is advancing rapidly, but important scientific, practical, and ethical challenges remain. Until reliable progression markers are available and the benefits for patients are clearly demonstrated, these tools should complement—not replace—shared decision-making between patients and healthcare professionals. Any prediction about the future course of an illness must be communicated with compassion, consent, and appropriate context. Ultimately, technology is most valuable when it supports the needs and preferences of the person at its centre. In Parkinson's, resilience, support, and self-management remain just as important as any prediction an AI system can provide.

References

[1] Severson, K. A. et al. Discovery of Parkinson’s disease states and disease progression modelling: a longitudinal data study using machine learning. Lancet Digit Health 3, e555–e564 (2021). https://doi.org/10.1016/S2589-7500(21)00101-1

[2] Deng, D. et al. Interpretable video-based tracking and quantification of parkinsonism clinical motor states. npj Parkinsons Dis. 10, 122 (2024).

[3] Yang, Y. et al. Artificial intelligence-enabled detection and assessment of Parkinson’s disease using nocturnal breathing signals. Nat Med 28, 2207–2215 (2022).

[4] Gorji, A. & Fathi Jouzdani, A. Machine learning for predicting cognitive decline within five years in Parkinson’s disease: Comparing cognitive assessment scales with DAT SPECT and clinical biomarkers. PLoS ONE 19, e0304355 (2024).