Validation of limited sampling models (LSM) for estimating AUC in therapeutic drug monitoring--is a separate validation group required?

Limited sampling models (LSM) for estimating AUC in therapeutic drug monitoring are usually validated in a separate group of patients, according to published guidelines. The aim of this study is to evaluate the validation of LSM by comparing independent validation with cross-validation using the patient data from the development group.

The design of the Monte Carlo simulation study was similar to a study described in the literature, i.e. a development group of 20 patients receiving cyclosporine orally every 12 h. Blood samples were taken at 10 fixed time points. In total 20,000 patient data sets were generated by Monte Carlo simulation, taking into account interindividual variability and measurement errors. Accuracy (mean error, ME) and precision (root mean squared error, RMSE) were calculated for evaluation of the validation procedures, varying the time points of the samples used for the estimation of AUC to identify the optimal sampling time points. In addition, the influence of the number of samples and the number of subjects was investigated.

Cross-validation resulted in values for ME and RMSE almost identical to values using a separate validation group with the same number of subjects as the development group.

A separate validation group is not needed. The most efficient method is to use all patient data for the development of the LSM, and to assess the accuracy and precision by cross-validation.