Use of regression modeling to simulate patient-specific decision analysis for patients with nonvalvular atrial fibrillation.

Abstract

To create a Web-based decision support tool that uses a simple regression equation to simulate performance of patient-specific decision analysis (PSDA) for patients with nonvalvular atrial fibrillation. Patient-level data were used, along with decision model estimates of the gain in quality-adjusted life expectancy associated with anticoagulant therapy to train regression models. Models involving successively higher order polynomial functions were evaluated. Quadratic (R2 = 0.89) and cubic (R2 = 0.97) regression models provided incremental benefit over a simple linear model (R2 = 0.56). For the cubic model, 95% of estimates were within 0.26 QALYs of decision model estimates. The cubic model accurately predicted actual decision model recommendations (AUROC of 0.957). Regression modeling can be used to simulate the performance of PSDA for patients with atrial fibrillation. This approach can be used to create fast, reliable, and portable decision support tools to improve patient care.