Predicting diabetes-related conditions in need of intervention: Lolland-Falster Health Study, Denmark.

In the Danish population, about one-in-ten adults have prediabetes, undiagnosed, poorly or potentially sub-regulated diabetes, for short DMRC. It is important to offer these citizens relevant healthcare intervention. We therefore built a model for prediction of prevalent DMRC. Data were derived from the Lolland-Falster Health Study undertaken in a rural-provincial area of Denmark with disadvantaged health. We included variables from public registers (age, sex, age, citizenship, marital status, socioeconomic status, residency status); from self-administered questionnaires (smoking status, alcohol use, education, self-rated health, dietary habits, physical activity); and from clinical examinations (body mass index (BMI), pulse rate, blood pressure, waist-to-hip ratio). Data were divided into training/testing datasets for development and testing of the prediction model. The study included 15,801 adults; of whom 1,575 with DMRC. Statistically significant variables in the final model included age, self-rated health, smoking status, BMI, waist-to-hip ratio, and pulse rate. In the testing dataset this model had an area under the curve (AUC) = 0.77 and a sensitivity of 50% corresponding to a specificity of 84%. In a health disadvantaged Danish population, presence of prediabetes, undiagnosed, or poorly or potentially sub-regulated diabetes could be predicted from age, self-rated health, smoking status, BMI, waist-to-hip ratio, and pulse rate. Age is known from the Danish personal identification number, self-rated health and smoking status can be obtained from simple questions, and BMI, waist-to-hip ratio, and pulse rate can be measured by any person in health care and potentially by the person him/her-self. Our model might therefore be useful as a screening tool.