Prediction of type 2 diabetes mellitus onset using logistic regression-based scoreboards

Type 2 diabetes mellitus (T2DM) accounts for ∼90% of all cases of diabetes which are estimated with an annual world death rate of 1.6 million in 2016. Early detection of T2D high-risk patients can reduce the incidence of the disease through a change in lifestyle, diet, or medication. Since populations of lower socio-demographic status are more susceptible to T2D and might have limited resources for laboratory testing, there is a need for accurate yet accessible prediction models based on non-laboratory parameters. This paper introduces one non-laboratory model which is highly accessible to the general population and one highly precise yet simple laboratory model. Both models are provided as an accessible scoreboard form and also as a logistic regression model. We based the models on data from 44,879 non-diabetic, UK Biobank participants, aged 40-65, predicting the risk of T2D onset within the next 7.3 years (SD 2.3). The non-laboratory prediction model for T2DM onset probability incorporated the following covariates: sex, age, weight, height, waist, hips-circumferences, waist-to-hip Ratio (WHR) and Body-Mass Index (BMI). This logistic regression model achieved an Area Under the Receiver Operating Curve (auROC) of 0.82 (0.79-0.85 95% CI) and an odds ratio (OR) between the upper and lower prevalence deciles of x77 (28-98). We further analysed the contribution of laboratory-based parameters and devised a blood-test model based on just five blood tests. In this model, we included age, sex, Glycated Hemoglobin (HbA1c%), reticulocyte count, Gamma Glutamyl-Transferase, Triglycerides, and HDL cholesterol to predict T2D onset. This logistic-regression model achieved an auROC of 0.89 (0.86-0.91) and a deciles’ OR of x87 (27-152). Using the scoreboard results, the Anthropometrics model classified three risk groups, a group with 1%(1-2%); a group with 9% (7-11%) probability, and a group with a 15% (7-23%) risk of developing T2D. The Five blood tests scoreboard model, further classified into four risk groups: 0.9% (0.7%-1%); 8%(6-11%); 18%(14-22%) and a high risk group of 38%(23-54%) of developing T2D. We analysed several more comprehensive models which included genotyping data and other environmental factors and found that it did not provide cost efficient benefits over the five blood tests model. The Five blood tests and anthropometric models, both in their logistic regression form and scoreboard form, outperform the commonly used non-laboratory models, the Finnish Diabetes Risk Score (FINDRISC) and the German Diabetes Risk Score (GDRS). When trained using our data, the FINDRISC achieved an auROC of 0.75 (0.71-0.78), and the GDRS auROC resulted in 0.58 (0.54-0.62), respectively.