Purpose: Recent diabetes trials evaluating very intensive glucose control have produced results that appear inconsistent with prior epidemiological studies. Forecasting models can help evaluate the relationships between epidemiological studies and trials. We utilize a forecasting model of diabetes complications to assess the predictability of recent trial results based on prior epidemiology.
Methods: Models derived from United Kingdom Prospective Diabetes Study (UKPDS) was[V1] used to make outcome predictions based on the trial designs and patient populations for the Action to Control Cardiovascular Risk in Diabetes (ACCORD) trial, Action in Diabetes and Vascular Disease (ADVANCE) trial, and Veterans Affairs Diabetes Trial (VADT). These predictions were then compared to the actual results. Predicted event rates (for both intensive and control arms) and relative risks were compared and judged on whether they fell within a calculated 95% confidence interval from the actual trials.
Result: For ADVANCE, the predicted event rates fell within the 95% confidence interval for 7 out of 18 compared outcomes. For VADT, this number was 5 out of 14. For ACCORD, model predicted event rates did not match any of the actual 12 event rates. The predicted relative risks fell within the 95% confidence interval for 8 out of 9 ADVANCE outcomes, 6 out of 7 VADT outcomes, and 2 out of 6 ACCORD outcomes.
Conclusion: The ACCORD trial represents a “black swan” trial, producing results that are unpredictable in relation to past diabetes epidemiology. These unpredictable results demand a renewed focus on the overall health effects of glucose lowering medication especially among the elderly and those with long-standing diabetes.