Swedish researchers have developed an artificial intelligence system that can identify people at high risk of melanoma by mining existing medical records, a breakthrough that could reshape how the disease is detected before it becomes deadly.
The study analyzed health data from nearly 6.1 million Swedish adults over five years, during which 38,582 people, or 0.64% of the population, developed melanoma. Instead of starting from scratch, the team tapped into information already stored in healthcare systems: patient age, sex, diagnoses, medications, and socioeconomic status.
The most sophisticated AI model achieved 73% accuracy at distinguishing future melanoma patients from those who would not develop the disease. That compares to just 64% accuracy using only age and sex as predictors. The gap matters enormously in clinical practice, where better predictions mean fewer false alarms and more efficient resource use.
The real power emerged when researchers narrowed their focus to smaller, high-risk groups identified by the AI. Within those targeted populations, the five-year melanoma incidence jumped to about 33% from the baseline 0.64%. That concentration of risk creates an ideal target for selective screening.
"Registry data that's already available within healthcare systems can be used to identify individuals at higher risk of melanoma," said Martin Gillstedt, a doctoral student at the University of Gothenburg's Sahlgrenska Academy and statistician at the hospital's dermatology department. "This form of decision support isn't yet available in routine healthcare, but our results signal that registry data can be deployed more strategically in the future."
The findings suggest dermatologists and primary care physicians could soon shift from a one-size-fits-all approach to something more precise. Instead of screening everyone equally, clinicians could prioritize patients whose medical histories and demographic profiles place them in the high-risk tier identified by the AI.
"Selective screening of small, high-risk groups could lead to both more accurate monitoring and more efficient use of healthcare resources," said Sam Polesie, the study's lead researcher and an associate professor of dermatology at the University of Gothenburg. "This brings population data into precision medicine and supplements clinical assessments."
Melanoma kills thousands of people annually, but survival rates improve dramatically with early detection. Current screening relies largely on visual examination by dermatologists, a bottleneck in many healthcare systems. A predictive tool that flags high-risk patients before symptoms appear could accelerate diagnosis and treatment.
The researchers caution that additional validation studies and policy decisions must come before this approach enters routine practice. But the trajectory is clear: large-scale registry data combined with machine learning offers a practical path toward personalized melanoma screening that existing healthcare infrastructure can support.
The work was a collaboration between the University of Gothenburg and Chalmers University of Technology, published in Acta Dermato-Venereologica.
Author Jessica Williams: "The real test will be whether hospitals actually implement this when they have the chance, or whether it ends up gathering dust as another promising study that never reaches patients."
Comments