Objective: To estimate unbiasedly the mean and median survival time of cancer patients with incomplete follow-up. To assess the effects of informative censoring on estimates by using the crude and a bias-reducing method, based on weighted averages of the age-specific results.
Study design and setting: Colon cancer patients diagnosed in 1970-1979 and thyroid cancer patients diagnosed in 1978-1987 were followed up until the end of each diagnosis period.
Results: Because of informative censoring, the crude estimates of the mean lifetime grossly overestimate the survival of the colon cancer patients and underestimate the survival of the thyroid cancer patients. Together with the most recent population life tables, the bias-reducing method succeeds in estimating the mean and the median lifetime accurately.
Conclusion: Stratifying by age is essential when the mean or median lifetime of the patients with a wide age range is to be estimated. The bias-reducing method should be used if a single summary estimate for the whole patient group is needed. The median is preferable if more than half of the patients die soon after diagnosis. Predicted population life tables should be used in extrapolation.