Updating risk engine for diabetes progression and mortality in the United States
Description
Background: The prediction of diabetes-related comorbidities and mortality over lifetime has significant clinical and policy implications. A prediction model can be used for economic evaluation on diabetes medications, comparative effectiveness review (CER) over different therapeutic plans, and estimation of the expected long-tern outcomes for different treatment goals (e.g., HbA1c). Most of the current diabetes prediction models heavily relied on the UKPDS risk engine and Framingham equation, which used data from 1970s on European populations. These populations were significantly different from current US population in various ways including race, different health related concept, treatment algorithm, screening method of comorbidities and even definition of diabetes. In addition, UKPDS risk engine does not include impact of hypoglycemia, which emerged as an important issue in the management of diabetes due to its impact on quality of life, cardiovascular events and mortality. Furthermore, with the advancement of the medical technology and innovation in redefining treatment guideline during the last decades, the rates of cardiovascular events, all-cause mortality and event related mortality have fundamentally changed, especially the survival rates from CVD events has substantially increased. There is an urgent needs to develop a new risk engine that more adaptable to the current US population. Objective: The objective of this study was to update risk engine using a cohort of patients with type 2 diabetes in the United States. Methods: A total of 21 equations for forecasting diabetes-related microvascular and macrovascular events, hypoglycemia, mortality, and progression of diabetes risk factors were estimated using data on 10,251 patients from the Action to Control Cardiovascular Risk in Diabetes (ACCORD) trial. Left truncated proportional hazard model was applied to fit each event equation using diabetes duration as time index, and a variety of distributions including Weibull and Gompertz distribution were tested. 10-folds cross-validation or bootstrapped validation was applied to account for overfitting issue. Predicted cumulative incidence rates was plotted against the observed cumulative incidence to serve as internal validation to evaluate the prediction accuracy of the BRAVO risk engine on ACCORD data. External validation was performed through applying the BRAVO risk engine onto population from other clinical trials. Results: The BRAVO risk engine’s forecast felled within the 95% confidence interval for the occurrence of observed events at each time point through 40 years after diabetes onset. The model prediction provides accurate prediction according to the internal validation and external validation process, and good face validity on risk factors were established by endocrinologists. Severe hypoglycemia was found to be an important risk factor for congestive heart failure (CHF), myocardial infarction (MI), angina, blindness, and associated with increased mortality. Racial factor was included in more than half of the events equations (e.g. MI, revascularization surgery, blindness, SPSL, hypoglycemia). Therefore, the BRAVO risk engine can capture racial difference on diabetes outcomes among US population, as a significant improvement over UKPDS risk engine. Conclusion: The BRAVO risk engine for the US diabetes cohort has a good internal validity to simulate events that closely match observed outcomes in the ACCORD trial. And it is also capable of accurately predict diabetes comorbidities in other US and non-US based clinical trials. The risk engine can be extrapolated over lifetime and provide long-term effect evaluation. The BRAVO risk engine can potentially provide more accurate prediction over a range of long-term outcomes than other current models, thus assist making clinical and policy decisions.