Censoring distributions as a measure of follow-up in survival analysis.