Construction and Evaluation of a Statistical Model for a Probit Method Simulator in Pharmacological Education

Purpose: As animal welfare becomes increasingly important, there is a corresponding desire to reduce the number of animals used in experiments. Recently, we reported on statistical models for a local anaesthetic simulator and developed a simulator for use in pharmacology education. In this study, we aimed to create a simulator for bioassay. Methods: Mice were intraperitoneally injected with a set concentration of lidocaine, and the time to the onset of convulsions or death was measured. Judgment times were set at 10 s intervals from 3 to 10 min. Parameter values were estimated by probit analysis based on the presence or absence of a reaction at each judgment time. The distributions and 95% confidence intervals (CI) of the estimated parameter values were confirmed using a nonparametric bootstrap method. Additionally, the generalization performance of the statistical model was confirmed using a five-fold cross-validation method. Monte Carlo simulations were performed using the estimated parameters from this model, and the average and distribution of the toxic dose 50% (TD50) and lethal dose 50% (LD50) were compared to those obtained from the animal experiments. Results: The parameters were properly estimated at each judgment time, and their 95% CIs were relatively narrow. The TD50 and LD50 values were similar across the five folds. Monte Carlo simulations demonstrated that the average and distribution of TD50 and LD50 were comparable to those obtained from animal experiments. Conclusions: These results suggest that a simulator based on this model is useful as an alternative to animal experiments. Therefore, our strategy will further reduce the number of experimental animals. Moreover, the method used in this study can be applied to other experiments that measure reaction time from treatment.