Researchers at Cardiff University have developed an AI tool that could help detect signs of Type 1 diabetes (T1D) in children before they experience dangerous complications. The new AI tool uses patterns in general practitioners’ (GP) records to identify symptoms associated with undiagnosed T1D, potentially accelerating diagnosis and allowing for timely treatment.
According to the research team, the early diagnosis tool could be instrumental in reducing cases of diabetic ketoacidosis (DKA), a life-threatening condition that can develop T1D goes untreated. Alarmingly, around 25% of young people with T1D are only diagnosed after they have already entered DKA. Early intervention, however, could prevent this scenario and save lives.
The team analysed electronic health records of over 1 million children in Wales to train the AI model. The tool examined various factors recorded in GP files, such as repeated urinary infections, bedwetting, family history of T1D, and antibiotic prescriptions. By analysing these factors, the tool identified combinations that might signal an underlying T1D diagnosis.
To validate its effectiveness, researchers tested the AI tool on an additional 1.5 million children’s records. They found it could successfully identify 72% of children who would develop T1D within the following 90 days. Importantly, the tool could alert doctors an average of nine days earlier than typical diagnostic timelines, potentially allowing children to begin insulin therapy before more severe symptoms develop.
The study authors noted the tool’s potential as a vital resource for early diagnosis but acknowledged that further refinements are needed to optimize its alert settings, balancing timely warnings with the avoidance of false positives. The research team plans to explore broader implementation strategies so that the AI tool can be adopted widely across primary care, helping GPs play a more active role in early detection.
This breakthrough promises a new avenue for protecting children’s health through early, proactive intervention in T1D.
Reference
Daniel R et al. Predicting type 1 diabetes in children using electronic health records in primary care in the UK: development and validation of a machine-learning algorithm. Lancet Digit Health. 2024;6(6):e386-95.
