DIAS-NIDDM--a model-based decision support system for insulin dose adjustment in insulin-treated subjects with NIDDM

A decision support system has been developed, Diabetes Insulin Advisory System for patients with non-insulin dependent diabetes mellitus (DIAS-NIDDM), assisting in the adjustment of insulin doses in insulin-treated subjects. DIAS-NIDDM uses a causal probabilistic network (CPN) model of carbohydrate metabolism to make stochastic predictions of blood glucose (BG) excursions. The CPN model is an extension of an existing model with an added component representing endogenous insulin secretion. A linear relationship between BG and insulin concentration due to BG stimulated insulin secretion is assumed. Model parameters (pancreatic sensitivity, insulin sensitivity, and time-to-peak of NPH insulin) are estimated by Bayesian probability updating from patient's specific data (food intake, insulin doses, BG measurements) recorded over a period of 4 days. The estimated parameters allow the system to be potentially used as a diagnostic tool to identify abnormalities of carbohydrate metabolism: impaired insulin secretion, insulin resistance and the severity of the impairments. DIAS-NIDDM was used to predict patient-specific BG profiles and advise on insulin doses during a pilot study in eight patients with NIDDM of whom five were treated with insulin. Compared to the administered insulin amount, daily insulin amount advised by DIAS-NIDDM was similar (within 4 U) in three patients, higher by 20% (19 U) in one patient and lower by 40% (18 U) and 50% (11 U) in two patients, respectively. The inter-day coefficient of variation of the daily insulin advice suggests that, at least according to DIAS-NIDDM criteria, day-to-day adjustment of insulin doses is necessary to maintain optimum control.