Development of a parametric simulation model for forecasting goal-oriented treatment outcomes

Abstract

Background: Treatment-to-goal (TTG) analyses are frequently used to predict guideline-directed population control rates for drug therapies based on mean efficacy data. Nevertheless, estimates are commonly inaccurate because variability in efficacy is not considered. A new methodology was developed to improve TTG forecasting. Methods: Patient-level blood pressure (BP) lowering data sets, designed to simulate clinical trial results, were generated for testing from three underlying distributions: normal, lognormal, and beta. To emulate real-world conditions where patient-level data are unavailable, two approaches were considered: parametric-simulated BP lowering data were generated using the mean and standard deviation of the test data sets; and point-estimate-BP lowering was uniformly assigned as the mean lowering. BP control (systolic BP < 140 and diastolic BP < 90 mmHg) was forecasted by subtracting values generated by these two methods from baseline BP values in untreated hypertensive patients (n = 2483) from the Third National Health and Nutrition Examination Survey. Estimated control rates were compared to analyses where the patient-level data sets were bootstrapped. Results: We assumed mean (± SD) BP lowering of 20 (12) mmHg systolic and 14 (7) mmHg diastolic. Parametric method predicted a BP control rate of 66.9% [95% confidence interval (CI) 65.7-67.9], similar to the bootstrapping approach (67.3%, 95% CI 65.9-68.8). The control rate projected based on the point-estimate method was 75.5%. The point-estimate method frequently led to substantially different results under a wide range of model assumptions. Conclusions: A new parametric-based forecasting method, which addresses underlying variability, improves on estimates based on mean efficacy only. In the absence of patient-level data, this method is generalizable to different therapeutic areas.