Design of sample surveys to estimate the prevalence of rare diseases: three unbiased estimates.