If you have type 2 diabetes, a sudden drop in blood sugar can be scary and dangerous. Doctors sometimes use risk models to predict who might have these lows. But a new review of 25 studies finds these models are only moderately accurate.
The models had an average accuracy score of 0.815 out of 1.0. That sounds decent, but individual models ranged from 0.630 to 0.996, meaning some are much less reliable. Worse, 96% of the studies had a high risk of bias, meaning the results may not be trustworthy.
Many studies had small sample sizes, didn't handle missing data properly, and failed to test their models in new groups of patients. Only 12% of studies had low applicability concerns, so the models may not work well in real-world clinics.
This doesn't mean all risk models are useless. But it does mean doctors and patients should be cautious when using them to make decisions about diabetes care. More rigorous research is needed before these tools can be trusted widely.