AI Predicts Chronic Diseases Years in Advance With 99% Accuracy Using New RiskPath Toolkit

Can we stop disease before it starts? A groundbreaking AI tool from the University of Utah reveals how explainable algorithms can forecast conditions like ADHD and hypertension long before symptoms strike, transforming early intervention and reshaping preventive medicine.

Study: RiskPath: Explainable deep learning for multistep biomedical prediction in longitudinal data. Image Credit: CHIEW / Shutterstock

Researchers at the University of Utah's Department of Psychiatry and Huntsman Mental Health Institute today published a paper introducing RiskPath, an open-source software toolkit that uses Explainable Artificial Intelligence (XAI) to predict whether individuals will develop progressive and chronic diseases years before symptoms appear, potentially transforming how preventive healthcare is delivered. XAI is an artificial intelligence system that can explain complex decisions in ways humans can understand.

The new technology represents a significant advancement in disease prediction and prevention by analyzing patterns in health data collected over multiple years to identify at-risk individuals with unprecedented accuracy of 85-99%. Current medical prediction systems for longitudinal data often miss the mark, correctly identifying at-risk patients only about half to three-quarters of the time. Unlike existing prediction systems for longitudinal data, RiskPath uses advanced timeseries AI algorithms and makes them explainable in order to deliver comprehensive models that provide crucial insights into how risk factors interact and change in importance throughout the disease development process.

"Chronic, progressive diseases account for over 90% of healthcare costs and mortality," says Nina de Lacy, MBA, MD, assistant professor in psychiatry at University of Utah Health and first author on the study. She adds, "By identifying high-risk individuals before symptoms appear or early in the disease course and pinpointing which risk factors matter most at different life stages, we can develop more targeted and effective preventive strategies. Preventative healthcare is perhaps the most important aspect of healthcare right now, rather than only treating issues after they materialize."

The research team validated RiskPath across three major long-term patient cohorts involving thousands of participants to successfully predict eight different conditions, including depression, anxiety, ADHD, hypertension, and metabolic syndrome. They say that the technology offers several key advantages:

• Enhanced Understanding of Disease Progression: RiskPath can map how different risk factors change in importance over time, revealing critical windows for intervention. For example, the study showed how screen time and executive function become increasingly important risk contributors for ADHD as children approach adolescence.
• Streamlined Risk Assessment: Though RiskPath can analyze hundreds of health variables, researchers found that most conditions can be predicted with similar accuracy using just 10 key factors, making implementation more feasible in clinical settings.
• Practical Risk Visualization: The system provides intuitive visualizations showing which time periods in a person's life contribute most to disease risk, helping researchers identify optimal times for preventive interventions.

The research team is now exploring how RiskPath could be integrated into clinical decision support systems, preventive care programs, and the neural underpinnings of mental illness. They plan to expand their research to include additional diseases and diverse populations.

The full study on RiskPath was published in the April issue of Patterns as "RiskPath: Explainable deep learning for multistep biomedical prediction in longitudinal data." The research was led by Nina de Lacy, Michael Ramshaw, and Wai Yin Lam from the Department of Psychiatry at the University of Utah. De Lacy serves on the One-U Responsible AI Initiative Executive Committee. The work was supported by the National Institute of Mental Health (grant number R00MH118359). Content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.