PDB6 LONG-TERM ASSOCIATION OF OBSERVED VERSUS PREDICTED MICRO- AND MACROVASCULAR COMPLICATIONS IN TYPE 1 DIABETES MELLITUS: EXTERNAL VALIDATION OF THE CORNERSTONE DIABETES SIMULATION MODEL (CDSM)

Abstract

To assess the suitability of the CDSM in replicating observed long-term micro- and macrovascular complications in Type 1 Diabetes Mellitus (T1DM) following best model validation practices. The CDSM is a patient-level microsimulation model for T1DM, incorporating published risk equations and transition probabilities derived from the Dose Adjustment For Normal Eating (DAFNE) program. Validation was conducted by populating the CDSM with the baseline characteristics and treatment effects of major T1DM datasets (FinnDiane, EuroDiab, WESDR, DCCT/EDIC). Estimated measures of association between modeled and observed complications were assessed via the coefficient of correlation (R2) or the difference of observed vs predicted outcomes. CDSM predicted outcomes yielded a high level of correlation for cardiovascular and nephropathic events (R2 > 98% in all cases). Prediction of cardiovascular disease (CVD) events within the FinnDiane, WESDR and EuroDiab datasets (at 7, 20 and 15 years respectively) varied by less than 2% between estimated and observed dataset outcomes. At 30-years, the model underestimated DCCT/EDIC trial incidence by 5% for the same endpoint. Neuropathy incidence exhibited a modest 1.8% overestimation in the CDSM (8.5% vs 6.7%), whereas nephropathic mean prevalence, displayed a slight underestimation of 0.8% at 15 years (4.3% to 5.1%) and 2.3% at 30-years (15% to 17.3%). This external validation exercise suggests the CDSM to be a suitable tool for long-term predictions of T1DM complications. The model could benefit from further research, notably additional validation steps and its use for cost-effectiveness evaluations.