GHREG is a computer program for modeling short-term and long-term effects of potentially time-dependent covariates on right-censored failure time data in cohort studies. Standard statistical analysis of failure time data using the Cox model can yield very misleading results when the proportional hazards assumption of constant relative risk is violated. GHREG implements valid and efficient statistical methods that allow non-constant hazard ratios over time, as described in Diao, Zeng and Yang (Biometrics, 2013). The current release performs semiparametric regression analysis of right-censored failure time data with time-independent covariates. We are working intensely to improve the capabilities of GHREG, so please check back frequently for updates.
The program is written in C language. Executable file for Linux/Mac platform is available in the download section below.
On Linux/Mac systems:
The program requires two separate groups of input: the control parameter input and the data input. The data input file is a plain text file with n rows and m columns. The rows represent subjects. The columns are separated by one or more white spaces (the space character or the TAB character). The columns have fixed meanings, as described below:
Column 1: Failure or censoring time. This column contains the minimum of failure time and censoring time.
Column 2: Censoring indicator. This column indicates by the values 1 versus 0 whether the failure time is observed or right censored.
Column 3 column m: Covariates. All covariate values are numerical.
The control parameters input file contains the file name of data output, file name of program output, total number of subjects, and total number of covariates. The rows have fixed meaning, as described below:
Row 1: File name of data input. Character.
Row 2: File name of program output. Character.
Row 3: Total number of subjects. Integer.
Row 4: Total number of covariates. Integer.
Computational results are written to the output file specified by the user. For each covariate, the output shows the maximum likelihood estimate of the short-term and long-term effects (denoted by beta and gamma, respectively), their standard errors, the Wald test statistics and the two-sided p-values for testing (1) H_0: beta=0; (2) H_0: gamma=0; (3) H_0: beta=gamma=0; and (4) H_0:beta=gamma.
contains a simulated data set with 200 subjects and 2
covariate. The control parameters are given in
ghreg.dat and output file