JMIR Publications

ABSTRACT

Background:

Micro- and macrovascular complications are a major burden for individuals with diabetes and can already arise in a prediabetic state. To allocate effective treatments and to possibly prevent these complications, identification of those individuals at risk is essential.

Objective:

The aim of this study was to build machine learning (ML) models, which predict the risk of developing a micro- or macrovascular complication in individuals with (pre)diabetes.

Methods:

We made use of electronic health records (EHRs) to identify individuals with diabetes or prediabetes. Subsequently, we analyzed which of these individuals developed a microvascular (retinopathy, nephropathy, neuropathy) or one of the following macrovascular complications: peripheral vascular disease (PVD), cerebrovascular disease (CeVD), and cardiovascular disease (CVD). We chose a five year forecast horizon. The EHRs contained information about demographics, biomarkers, medications, and disease codes. In this work, we compared two ML models, namely logistic regression and gradient boosted decision trees (GBDTs).

Results:

Overall, 13,904 and 4,259 individuals with prediabetes and diabetes, respectively, were identified in our underlying dataset. For individuals with prediabetes, the areas under the receiver operating characteristic (AUROCs) for logistic regression and GBDTs were 0.669 and 0.702 (retinopathy), 0.806 and 0.814 (nephropathy), 0.718 and 0.664 (neuropathy), 0.737 and 0.728 (PVD), 0.690 and 0.692 (CeVD), 0.706 and 0.699 (CVD). For individuals with diabetes, the AUROCs were 0.676 and 0.731 (retinopathy), 0.771 and 0.780 (nephropathy), 0.736 and 0.761 (neuropathy), 0.698 and 0.710 (PVD), 0.652 and 0.640 (CeVD), 0.688 and 0.668 (CVD).

Conclusions:

ML models allow for an identification of individuals with (pre)diabetes who are at increased risk of developing micro- or macrovascular complications. This could help to offer adequate treatment and, hence, may prevent the onset of these complications.