Identification of empirical dynamic models from type 1 diabetes subject data

Empirical linear dynamic models have been identified from ambulatory data from two type 1 diabetes subjects in order to determine approximately how far into the future the models could be expected to make reasonably accurate predictions. For a prediction horizon of 30 minutes, FIT values (related to R² values) of the model predictions for validation data were 46% for one subject and 60% for the other subject. These FIT values correspond to root mean square errors of 14 and 24 mg/dL, respectively. Longer prediction horizons resulted in substantially worse predictions for these ambulatory subject data.