Novel Artificial Intelligence Models Detect Type 1 Diabetes Risk Before Clinical Onset

Two studies at the American Diabetes Association’s 85th Scientific Sessions demonstrate how artificial intelligence can significantly improve early detection of type 1 diabetes, identifying risk up to a year before traditional screening methods and increasing diagnostic accuracy. These advancements aim to reduce late-stage diagnoses and enhance patient care.

In a promising advancement for type 1 diabetes detection, two studies utilizing artificial intelligence were presented at the 85th Scientific Sessions of the American Diabetes Association on June 20, 2025. These research efforts emphasize the capabilities of machine learning in identifying the risk of type 1 diabetes before clinical symptoms manifest. This early intervention is crucial, as around 64,000 Americans receive their diagnosis yearly, often only after experiencing severe symptoms requiring hospitalization.

The first study revealed machine learning models designed for two age groups, utilizing medical claims and lab test data from NorstellaLinQ. Researchers found that these AI models could identify individuals at risk as much as a year prior to traditional screening methods. Impressively, they achieved sensitivity rates of about 80% for younger people and 92% in adults, promising a significant drop in false positives compared to the standard approach, which only catches 0.3% of the population.

Laura Wilson, director of health economics outcomes research at Sanofi, expressed excitement over the findings. “We’re energized by the results of this study and what it could mean for early type 1 diabetes risk detection,” she said, shedding light on how AI can enhance early screening for a disease that can remain hidden until a serious health crisis occurs.

Meanwhile, the second study leveraged data from the Symphony Health Database, which includes information on 75 million patients. Researchers compared records of nearly 90,000 individuals diagnosed with type 1 diabetes against more than 2.5 million without. Their machine learning model distinguished potential diabetes risk significantly, boosting detection efficiency by over 18-fold. Alarmingly, they discovered that 29% of those with type 1 diabetes had been incorrectly classified previously, underscoring the critical need for accurate diagnostics.

The standout performer among the AI models was BERT, a tool initially meant for natural language processing. It correctly identified 80% of true type 1 diabetes cases, outperforming its counterparts with a much stronger predictive capability. Jared Joselyn, senior VP at Sanofi, noted that early identification could transform the entire care timeline, emphasizing that AI has the potential to enhance proactive health care significantly.

Further research is on the horizon, focusing on validating these approaches with various datasets from U.S. and international healthcare systems. Plans will also look into enhancing model performance through various AI techniques and incorporating longitudinal data to better support interventions. Key presentations will take place on Sunday, June 22, featuring insights from Wilson and Joselyn, offering a glimpse into the future of diabetes diagnostics.

The American Diabetes Association’s 85th Scientific Sessions is the largest global meeting dedicated to diabetes research, drawing thousands of professionals eager to see groundbreaking advancements in diabetes care. The ADA continues to commit itself to battling diabetes while celebrating 85 years of impactful research, as it supports the 136 million Americans living with diabetes or prediabetes.

For more information, individuals can visit diabetes.org or reach out via the ADA’s various social media platforms for updates and engagement about diabetes advocacy.

In essence, the emergence of artificial intelligence in type 1 diabetes risk detection signifies a forward step toward effective early diagnosis. With accurate machine learning models indicating risk as much as a year before standard methods and enhancing diagnostic accuracy, the potential for improving patient outcomes is substantial. These findings at the ADA Scientific Sessions demonstrate hope for more proactive healthcare strategies, a necessity in tackling diabetes effectively.

Original Source: www.prnewswire.com