M. Tod and J.M. Rocchisani
Hôpital Avicenne, 93000 Bobigny, France
The most usual approach for optimization of sampling schedule in pharmacokinetics is the D-optimality criterion which consists in maximizing the determinant of the Fisher information matrix (max det F). However, D-optimal times depends on the values of the parameters to be estimated. Several optimization criteria that formally incorporates prior parameter uncertainty have been proposed. These criteria consist in finding the sampling schedule that maximizes the expectation (over a given parameter distribution) of det F (ED-optimality) or logdet F (API-optimality), or minimizes the expectation of 1/det F (EID-optimality). We introduce a new design criterion to optimize the sampling times when the parameters are determined by using a Bayesian Maximum A Posteriori estimator (MAP): the Bayesian-API criterion (BAPI), which consists in maximizing the expectation of the log of the determinant of an approximate Bayesian information matrix. The precision and the accuracy of parameter estimation after fitting a biocompartmental model to a small number of optimal data points (determined according to D, ED, EID, API and BAPI criteria) were compared in a Monte Carlo simulation study. Two bicompartmental models were considered, one with zero order infusion rate (4 parameters), the other with zero order absorption rate (5 parameters). Data were simulated 10 times for 100 subjects with both models. Interindividual variabilities in PK parameters were 20 to 50%. Optimal sampling times were calculated by using our software OSPOP 1.0. Five or six samples were allowed for parameter estimation by weighted least squares or MAP. Performances of each design criterion were evaluated in terms of mean prediction error (PE%), root mean squared error (RMSE%), and number of acceptable estimates (i.e. with a SE less than 30%), N%. With the infusion model, performances of the different sampling schedules were comparable although API provided a significantly higher N%. With the absorption model, the average optimal designs performed much better than D-optimality. BAPI provided the highest N%, followed by API, but precision and accuracy were in the order ED>API>EID>BAPI.
Reference: PAGE 5 () Abstr 583 [www.page-meeting.org/?abstract=583]
Poster: oral presentation