AI models identify high-risk melanoma patients before disease develops
Swedish researchers have successfully utilized artificial intelligence to spot high-risk melanoma patients before they develop the disease, relying entirely on health data from existing medical records.
A new study from the University of Gothenburg demonstrates that artificial intelligence can identify individuals at significantly higher risk of developing this serious, rapidly spreading skin cancer within five years.
The research team analyzed clinical registry data from approximately six million Swedish adults who lived in the country between 2005 and 2014.
By feeding AI models detailed clinical information beyond the usual age and gender metrics, including medication history and other prior diagnoses, the system achieved remarkable accuracy.
The most advanced AI model accurately identified individuals who later developed melanoma with a success rate of approximately 73 percent, a significant improvement over the 64 percent accuracy achieved using only age and gender.
By combining various sociodemographic and medical data points, researchers pinpointed small groups facing a 33 percent risk of developing melanoma within five years. Doctoral student Martin Gillstedt explained that while this decision support is not yet available in routine healthcare, the results provide a clear signal that registry data can be used strategically.
Melanoma, primarily caused by ultraviolet light from the sun or sunbeds, accounted for four percent of all new cancer diagnoses in the European Union in 2020.
Lead author Sam Polesie suggested that selective screening of these high-risk groups could ensure more accurate monitoring and efficient use of healthcare resources.
Clinicians could focus on follow-ups and invite individuals predicted to be at high risk for screening through digital outreach.
Ultimately, while AI models trained on large registry data demonstrate significant potential for personalized risk assessments, the researchers stress that further policy decisions are necessary before implementation.
