Approximate Confidence Intervals for Probabilities of Survival and Quantiles in Life-Table Analysis

摘要: For survival probabilities with censored data, Rothman (1978, Journal of Chronic Diseases 31, 557-560) has recommended the use quadratic confidence limits based on assumption that product 'effective' sample size at time t and life-table estimate probability past follows a binomial distribution. This paper shows proposed are asymptotically correct for continuous data. These intervals, as well those arcsine transformation, logit transformation log(--log) compared by simulation to Greenwood's formula--the usual method interval estimation in analysis. With large amounts alternatives Greenwood all produce acceptable intervals. On basis overall performance, intervals suggested preferred smaller samples. Any these methods may be used generate sets median or any other quantile.