A note on quantifying follow-up in studies of failure time

摘要: In a recent review of survival analyses published in cancer journals, Altman et al [l] found that about half the papers did not include any summary followup time. Only 31% those report median follow-up specified method used to compute it. Independently [I], we have surveyed articles three medical journals-Journal Clinical Oncology, Annals Internal Medicine, and Nao England Journal Medicine-within 6-month period (January-June 1994) identified 70 analysis. Among these, 47 (67%) included statement regarding duration follow-up, but 24 specify quantify (usually follow-up). 23 other articles, methods employed were (1) based only on censored times (14 articles); (2) specification minimum time (5 (3) from entry death or last contact (1 article); (4) end-of-study date (5) (2 articles). Median sample size was 236 (range 30-8331), proportion censoring 60% 3-96%). This note shows values may differ substantially depending used. Results analysis apply frame whichmost individuals observed. particular, standard analytical for data, such as log-rank test 121, generalized Wilcoxon [3], proportional hazards model 141, estimate average effects [51 observed response significance. Thus current reporting is unsatisfactory. The following been suggested. We assume study with staggered all between T,, available data at final time, TX. For each individual i G s n), observe into study, t,,, recorded date, t2,. If tli death, status indicator, Si, assumes value 1.