Effects of Dose Selection and Sample Size on the Precision of Lethal Dose Estimates in Dose–Mortality Regression

Abstract

As a guide to experimental design for bioassays, the effects of total sample size and dose selection on the precision of LD50 and LD90 estimates were determined by Monte Carlo simulations with the logit model. A sample size of 120 appears to be the minimum necessary for reliable estimation. For increased precision, sample sizes of 240 or more are required. A computer procedure was used to search for the percentage of mortality resulting in minimum predicted 95% confidence interval lengths for the LD50 and LD90 in dose-response experiments with three to eight doses, equal numbers of test subjects at each dose, and total sample sizes of 120, 240, 480, and 720. Regardless of the number of doses, the most precise LD50 estimates were obtained when responses when evenly distributed between 25 and 75%. Precise LD90 estimates, however, required one or two responses ≤10% and the majority between 75 and 95%.