These AI models can be deployed to increase the efficiency at which individuals with undiagnosed T2D and those at high risk for developing T2D are identified tenfold. The AI models constructed by Kokkorakis et al. meet clinical significance, outperform the currently available non-laboratory screening tools, and were validated across several ethnic groups. “The findings from our study have the potential to simplify much-needed population screening for Type 2 Diabetes and thereby accelerate diagnosis” says Prof. Bruce H.R. Wolffenbuttel of the University Medical Center Groningen.
Prof. Christos S. Mantzoros of Harvard Medical School adds: “Simplifying and customizing population screening using cutting edge technologies has the potential to alter how we practice medicine in the short term and promises to offer longer and healthier lives to individuals with Type 2 Diabetes” -.
T2D affects more than 462 million individuals worldwide and approximately 1 million in the Netherlands, with an estimated minimum of at least 250.000 undiagnosed individuals. Early detection is crucial to reduce complications and healthcare costs. The lifestyle and health data-based AI models developed in this study utilize only questionnaire data, making them valuable for population-wide screening. Deploying these AI models would allow identification of approximately 80.000 additional individuals with or about to developed T2D in the Netherlands alone, requiring a tenfold lower number of blood tests than without using these AI models.
A logical next step following this study would be to prospectively deploy these models using the latest data from the Lifelines cohort, a study tracking 170,000 individuals over 30 years, with minimal participant effort. Not only would this validate the effectiveness of this methodology in a real-world setting, but also identify individuals at high risk for or already having developing T2D for early diagnosis and intervention.
Deploying these models could lead to early preventive intervention, aligning with the healthcare’s growing emphasis on prevention. Specifically, it should be possible to remotely monitor individuals at high risk effectively, shedding light on the impact of personal actions, as well as allowing healthcare practitioners to take timely measures when needed. Effectively, this will reduce the number of individuals needing to rely intensely on the healthcare system at a later stage.
Ultimately, implementing this questionnaire-based risk stratification will help us avoid many complications resulting from undiagnosed T2D, such as kidney failure, which significantly impairs health and well-being. Simultaneously, this will translate into reduced healthcare burden and cost. Ultimately, deploying these models will contribute to increasing the number of years people live in good health.
This research resulted from a fruitful collaboration between the University Medical Center Groningen, Ancora Health, and several prominent institutions, including Beth Israel Deaconess Medical Center and Boston VA Healthcare System, both teaching affiliates of Harvard Medical School. The authors deployed AI to analyze data from two of the most extensive population cohorts in the world: the UK Biobank and Lifelines.
The full article was published today in the prestigious journal eClinicalMedicine from The Lancet Group:
