The median lethal dose - Design and estimation

Abstract

A biological assay from a pharmaceutical company is simulated with various designs and a range of parameter values. Three estimation techniques - Fieller's method, profile likelihood and the bootstrap - of finding confidence intervals for the median lethal dose LD50 are compared. Correcting for bias does not substantially improve the maximum likelihood point estimator of LD50. The choice of the number of dose levels and sample size at each dose is seen to be critical in terms of the accuracy and precision of the estimation techniques.