OBJECTIVES: For health care decision-makers, the use of computer simulation models requires transparency, precision and accuracy. Systematic comparisons of diabetes models, per Mount Hood Challenges, have shown significant variability in results between models. We developed and validated a new cost-effectiveness model (the DEtermination of Diabetes Utilities, Costs, and Effects [DEDUCE] model) in both type 1 and 2 diabetes mellitus (T1DM, T2DM) to evaluate sensor-based glucose monitoring.

METHODS: This Excel-based patient-level microsimulation model used a cost-utility approach to compare sensor-based glucose monitoring systems to self-monitoring of blood glucose (SMBG) testing over a specified time horizon (1 to 100 years) with yearly cycles. The model used the Sheffield risk engine for T1DM and the Risk Equations for Complications Of type 2 Diabetes (RECODe) risk engine for T2DM to predict macro- and microvascular events. Inputs, model architecture, and subsequent validation analyses were reviewed and informed by an advisory board of health economists, endocrinologists and diabetologists.

RESULTS: Internal validation (comparing model predictions to observed outcomes from studies from which the risk equations were derived) and external validation (predictions compared to external datasets) demonstrated high precision (R2 ≥ 0.98) and reasonable accuracy (mean absolute percentage error [MAPE] ranging from 7.64-68%) with regards to macrovascular outcomes for T1DM, and high precision (R2 = 0.94) and high accuracy (MAPE = 19.8%) with regards to all-cause mortality in T2DM. Cross validation (comparing model outcomes between DEDUCE and published results from models participating in previous Mount Hood Challenges) indicated that DEDUCE had the best accuracy (MAPE = 36%) and non-inferior precision (R2 = 0.16) relative to other T1DM models, and second-to-best accuracy (MAPE = 25.03%) and high precision (R2 = 0.95) relative to other T2DM models.

CONCLUSIONS: In both T1DM & T2DM, DEDUCE suitably predicted key outcomes and performed favorably compared with existing models that participated in the Mount Hood Challenges, including the Core Diabetes Model.