OBJECTIVES: Diabetes, a metabolic disorder resulting in chronic hyperglycemia, is mainly classified as type 1 or 2. The study aimed to establish a statistical model that can discriminate between and predict diabetes types, based on the consumption of pharmaceutical products.

METHODS: A cohort of 73,498 diabetics patients was identified in 2021 in THIN®— a medicalized primary care database. We divided them into two groups: 51,449 patients constituting the working database and 22,049 patients to validate the resulting algorithm.

RESULTS: Most of patients were DT2 (92%). The average age was 67 and 42% were women. The most significant variables (p<.0000) were combined therapies (rapid-acting insulin, insulin combination, slow-active insulin, metformin, other oral anti-diabetics (OADs)) and age (≥57 and <57). The data showed a clear divide between OAD and insulin. Compared with rapid-acting insulin, metformin increased the likelihood of being DT2, administered alone or in combination with other OADs. Likewise, the other OADs increased the likelihood of being DT2. The more insulin treatment alone was rapid-acting insulin, the lower was the likelihood of being DT2. The results with combined OAD and insulin treatments were more varied, but with a higher likelihood of being DT2. Age risk was less discriminating but still significant. This model was applied to the validation sample. The accuracy rate was 95% for DT2 patients. By changing the different parameters, this model can serve as a modular basis depending on the type of diabetes and the level of precision required.

CONCLUSIONS: This study shows that by combining a logistic regression model with a sensitivity and specificity test, we can predict the type of diabetes based on the consumption of pharmaceutical products and patient age. The interest of this type of model lies in the fact that it makes it possible to enrich databases without clinical data.