A significant federal grant of $15 million has been awarded to a pioneering team at Duke University to advance an artificial intelligence model designed to predict mental illness in adolescents. This groundbreaking tool, known as the Duke Predictive Model of Adolescent Mental Health (Duke-PMA), was co-developed by Professor of Psychiatry Jonathan Posner, Assistant Professor of Biostatistics & Bioinformatics Matthew Engelhard, and AI Health Fellow Elliot Hill. The Duke-PMA functions as an AI-based assessment tool that analyzes various factors relevant to adolescent mental well-being, predicting the likelihood of an individual developing a mental illness within a year.

Beyond merely predicting risk, the model also pinpoints the crucial factors driving these predictions, thereby offering invaluable insights for guiding targeted preventive interventions. Dr. Posner highlighted the transformative potential of this approach, stating that it represents a “paradigm change in psychiatry from a reactive to a proactive approach,” moving away from waiting for illness to manifest before initiating treatment.

The Duke-PMA has demonstrated remarkable efficacy, achieving an 84% accuracy rate in identifying adolescents aged 10 to 15 who are at risk for serious future mental health issues. Critically, this high level of performance remained consistent across diverse demographics, including socioeconomic status, race, and sex. Its accessibility is another key advantage; the model relies solely on questionnaires, eschewing the need for costly imaging or blood tests, thus making it a highly scalable and widely available assessment option.

Furthermore, the model maintains its accuracy even when its focus is narrowed to factors that clinicians can directly influence, such as sleep disturbances and family conflict. This capability ensures that the results provide clinicians with actionable insights, enabling them to formulate effective prevention and intervention strategies well before the onset of illness. Dr. Posner illustrated its practical application: a primary care doctor could use a quick assessment to receive a report indicating a child's 90% chance of developing an illness within a year, along with the specific factors driving that prediction.

The $15 million federal grant is a pivotal moment for the project's development. As Matthew Engelhard explained, “This is exactly the pathway to get it in (the clinicians’) hands and actually identify people early and connect them with services and support that can hopefully bend that trajectory.” The next phase of the project involves an observational study, enrolling 2,000 adolescents from rural clinics across North Carolina, Minnesota, and North Dakota. This focus on rural areas is deliberate, as these regions often have limited mental health resources, making an automated tool like Duke-PMA particularly beneficial.

During this study, participants will be assessed using the Duke-PMA to generate predictions. A year later, their families will be recontacted for detailed psychiatric evaluations to ascertain the accuracy of the model’s predictions. The application of artificial intelligence in healthcare, especially in sensitive areas like adolescent mental health, naturally elicits both enthusiasm and apprehension. To mitigate concerns about false positives, Elliot Hill emphasized that Duke-PMA is designed to be a supportive tool, complementing clinical judgment rather than replacing it.

The team is also deeply committed to patient privacy. Engelhard affirmed, “We’re very serious about protecting patients’ privacy, both in the context of the study that we’re doing, as well as more broadly, going forward.” This careful approach seeks to balance innovative technology with prudent caution, enhancing care while preserving the indispensable human element in clinical decision-making. Dr. Posner concluded by noting that such interdisciplinary research, requiring collaboration from many different fields, is an area where Duke University is exceptionally well-positioned.