In a recent review of survival analyses published in cancer journals, Altman et al [l] found that about half of the papers did not include any summary of followup time. Only 31% of those papers that did report median follow-up specified the method used to compute it. Independently of Altman et al [I], we have surveyed articles in three medical journals-Journal of Clinical Oncology, Annals of Internal Medicine, and Nao England Journal of Medicine-within a 6-month period (January-June 1994) and identified 70 that used survival analysis. Among these, 47 (67%) included a statement regarding duration of follow-up, but 24 did not specify the method used to quantify followup (usually median follow-up). Among the 23 other articles, methods employed were (1) follow-up based only on censored times (14 articles); (2) specification of a minimum follow-up time (5 articles); (3) times from entry to death or last contact (1 article); (4) times from entry to end-of-study date (1 article); (5) other methods (2 articles). Median sample size in the surveyed articles was 236 (range 30-8331), and the median proportion of censoring was 60% (range 3-96%). This note shows that values of median follow-up may differ substantially depending on the method used. Results of survival analysis apply to the time frame in whichmost of the individuals were observed. In particular, standard analytical methods for survival data, such as the log-rank test 121, the generalized Wilcoxon test [3], or the proportional hazards model 141, estimate average effects [51 for the observed response times and test those effects for significance. Thus the current reporting of follow-up is unsatisfactory. The following methods have been used or suggested. We assume a medical study with staggered entry of all individuals between times T,, and T,, and analysis of the available data at a final end-of-study time, TX. For each individual i (1 G i s n), we observe the time of entry into the study, t,,, and the final recorded date, t2,. If tli is the date of death, the status indicator, Si, assumes a value of 1. For
[1]
D. G. Altman,et al.
Review of survival analyses published in cancer journals.
,
1995,
British Journal of Cancer.
[2]
N. Mantel.
Evaluation of survival data and two new rank order statistics arising in its consideration.
,
1966,
Cancer chemotherapy reports.
[3]
M. Schemper.
Cox Analysis of Survival Data with Non‐Proportional Hazard Functions
,
1992
.
[4]
E. Kaplan,et al.
Nonparametric Estimation from Incomplete Observations
,
1958
.
[5]
A. Cawkell.
Science Citation Index
,
1970,
Nature.
[6]
David R. Cox,et al.
Regression models and life tables (with discussion
,
1972
.
[7]
R. Prentice.
Linear rank tests with right censored data
,
1978
.
[8]
E. Korn.
Censoring distributions as a measure of follow-up in survival analysis.
,
1986,
Statistics in medicine.
[9]
Edmund A. Gehan,et al.
The performance of some two-sample tests in small samples with and without censoring
,
1969
.