Evaluation of Glucose Sensor Simulated Use as a Predictor of Device Performance

The goal of this project is to characterize the relationship between devicesimulated use and device performance. Toward that end, we analyzed trends in performance of device simulated use testing and device performance separately, as well as the associativity between these two. Statistical methodologies involve multivariate correlation estimation, McNemar’s test, Chi Square test, linear and logistic regression modeling and selection, and survival analysis. In selecting models that characterize the relationship between the simulated use testing and the device performance data, various model comparison criteria (AIC, BIC, MSE, norm ratio, etc.) were explored. Model searching strategies include stepwise search and exhaustive search. Cross-validation was employed to evaluate and choose models with the highest predictive accuracy.