Cox Regression for Survey Data
(Written by Doug Midthune)
This IBM PC program fits Cox's proportional hazards model to sample survey data, allowing for time-dependent covariates and left-truncated data (subjects entering at time >0). Since the program allows left-truncated data, age can be modeled as the time variable as described in Korn EL, Graubard BI, Midthune D (1997), Am J Epidemiology, 145, pp. 72-80.
For complex sampling designs, the program maximizes a weighted partial log-likelihood and calculates robust design-based standard errors (assuming with-replacement sampling at the first stage). For further details see Binder, DA (1992), Biometrika, 79, pp. 139-47.
For simple random sampling, the method reduces to the standard (unweighted) partial log-likelihood analysis.
The program uses Breslow's method for handling tied event times.
- coxreg.zip (m, 367kb) - Zip file containing PC program, documentation and example.