Predicting future treatment of diabetes mellitus from characteristics available at presentation.

This study was designed to identify the clinical features of a newly diagnosed diabetic patient that are most useful in deciding treatment. A secondary aim was to formulate a statistical model for predicting subsequent treatment. The following features were considered in 289 patients: age, sex, severity and duration of symptoms, degree and duration of weight loss, glycosuria, ketonuria, blood glucose concentration, body mass index (BMI), and family history of diabetes. Three treatment groups, 6 months after diagnosis, were defined: diet alone, diet with oral hypoglycaemic agent(s), and insulin-treated. Univariate analysis showed that symptom severity, glycosuria, ketonuria, glucose concentration, weight loss, and BMI were significantly different between the three groups. Age and rate of weight loss were significantly different between the insulin-treated and non-insulin-treated groups. Multivariate analysis gave a model to calculate the probability of requiring each of the three treatments given certain characteristics. A second cohort of 174 patients was used to assess the accuracy of the model. The model predicted the actual treatment at 6 months correctly in 72%.