SUMMARY Comparison of survival for two groups of patients when adjusting for covariates is a standard problem in clinical trials. In this paper the general life table models developed by Cox (1972) and the stratified log rank test (Mantel & Haenszel, 1959; Peto & Peto, 1972) are compared. An analytical expression and numerical results for the asymptotic relative efficiency of the two tests are given when there is no censoring. Stratification on the covariates is asymptotically as efficient as the test arising from Cox's model provided (i) there is no treatment effect, (ii) the covariates are balanced across treatment groups and (iii) the hypotheses underlying Cox's model are satisfied. However, if the proportional hazard model does not hold for the covariates then Cox's model leads to a biased estimate of the difference between the two treatments.
[1]
J. Peto,et al.
Asymptotically Efficient Rank Invariant Test Procedures
,
1972
.
[2]
N. Mantel.
Evaluation of survival data and two new rank order statistics arising in its consideration.
,
1966,
Cancer chemotherapy reports.
[3]
J. Kalbfleisch,et al.
The Statistical Analysis of Failure Time Data
,
1980
.
[4]
David R. Cox,et al.
Regression models and life tables (with discussion
,
1972
.
[5]
W. Haenszel,et al.
Statistical aspects of the analysis of data from retrospective studies of disease.
,
1959,
Journal of the National Cancer Institute.
[6]
B. Efron.
The Efficiency of Cox's Likelihood Function for Censored Data
,
1977
.