DESIGN AND ANALYSIS OF IN-VITRO ANTITUMOR PHARMACODYNAMIC STUDIES

The relationship between drug concentration (C), exposure time (t), an d the resulting effect (h) for a chemotherapeutic agent is expressed a s C-n x t = h, The value of n, derived from curve fitting of the C ver sus t plot, indicates the relative importance of concentration and exp osure time. The selection of concentrations and exposure times in a ph armacodynamic experiment may affect the precision and accuracy of para meter estimation. The use of optimal designs is even more critical whe n the numbers of experimental conditions are limited by tumor availabi lity (e.g., small size of surgical specimens from patients), The prese nt study used computer-simulated data to define the most efficient in vitro pharmacodynamic experimental designs and the optimal method of p harmacodynamic data analysis, All studies used Monte Carlo simulations to compare designs with varying numbers of drug concentrations, expos ure times, and replications, For each selected design, 50-100 error-co ntaining data sets were created by addition of experience-based random errors to expected concentration-response profiles. To compare method s of data analysis, the same 1250 simulated data sets were analyzed by two methods (i.e., surface response method and traditional method), T he results showed that simultaneous fitting of drug effect at all conc entrations and all exposure times by the surface response method yield ed n estimates that had greater precision and accuracy than a traditio nal method that required sequential determination of the effective inh ibitory concentration (e.g., IC50) and then the n value using the IC50 at different exposure times. Subsequent studies were analyzed using t he surface response method. To evaluate the effect of selection of con centrations and exposure times on the precision of n estimation, betwe en 1100 and 2200 simulated data sets, with 400 observations per data s et, were generated using different exposure times and drug concentrati ons. Because the number of observations was limited to 400, the number of replications at each condition varied depending on the total numbe r of selected conditions, The simulation results showed that the preci sion of n estimation decreased with an increased number of exposure ti mes at the expense of decreasing replication allocated to each experim ental condition The precision of n estimation significantly increased as the number of drug concentrations used at each exposure time increa sed from 2 to 3 but was not improved when the number of concentrations further increased from 3 to 20, A design constructed using D-optimal selection of 3 concentrations produced more precise n estimates than a standard design with 5 concentrations distributed in a log linear man ner, However, the D-optimal design requires a correctly selected pharm acodynamic model and prior knowledge of the error structure and model parameters, and is, therefore, not applicable to all situations. For d etermination of n, we recommend the use of the surface response method for data analysis, more than two exposure times evenly distributed on a log linear scale, and at least three concentrations spanning four o rders of magnitude, The data further show that a sufficient amount of specimen to provide 200 data points is needed to determine n with a co efficient of variation of less than 10%.