# Covariate analysis of bivariate survival data

## Abstract

The methods developed are used to analyze the effects of covariates on bivariate survival data when censoring and ties are present. The proposed method provides models for bivariate survival data that include differential covariate effects and censored observations. The proposed models are based on an extension of the univariate Buckley-James estimators which replace censored data points by their expected values, conditional on the censoring time and the covariates. For the bivariate situation, it is necessary to determine the expectation of the failure times for one component conditional on the failure or censoring time of the other component. Two different methods have been developed to estimate these expectations. In the semiparametric approach these expectations are determined from a modification of Burke's estimate of the bivariate empirical survival function. In the parametric approach censored data points are also replaced by their conditional expected values where the expected values are determined from a specified parametric distribution. The model estimation will be based on the revised data set, comprised of uncensored components and expected values for the censored components. The variance-covariance matrix for the estimated covariate parameters has also been derived for both the semiparametric and parametric methods. Data from the Demographic and Healthmore »

- Authors:

- Publication Date:

- Research Org.:
- North Carolina Univ., Chapel Hill, NC (United States)

- OSTI Identifier:
- 6999822

- Resource Type:
- Miscellaneous

- Resource Relation:
- Other Information: Thesis (Ph.D.)

- Country of Publication:
- United States

- Language:
- English

- Subject:
- 99 GENERAL AND MISCELLANEOUS//MATHEMATICS, COMPUTING, AND INFORMATION SCIENCE; 63 RADIATION, THERMAL, AND OTHER ENVIRON. POLLUTANT EFFECTS ON LIVING ORGS. AND BIOL. MAT.; SURVIVAL CURVES; MULTIVARIATE ANALYSIS; DATA COVARIANCES; MORTALITY; PARAMETRIC ANALYSIS; MATHEMATICS; STATISTICS; 990200* - Mathematics & Computers; 560000 - Biomedical Sciences, Applied Studies

### Citation Formats

```
Bennett, L E.
```*Covariate analysis of bivariate survival data*. United States: N. p., 1992.
Web.

```
Bennett, L E.
```*Covariate analysis of bivariate survival data*. United States.

```
Bennett, L E. Wed .
"Covariate analysis of bivariate survival data". United States.
```

```
@article{osti_6999822,
```

title = {Covariate analysis of bivariate survival data},

author = {Bennett, L E},

abstractNote = {The methods developed are used to analyze the effects of covariates on bivariate survival data when censoring and ties are present. The proposed method provides models for bivariate survival data that include differential covariate effects and censored observations. The proposed models are based on an extension of the univariate Buckley-James estimators which replace censored data points by their expected values, conditional on the censoring time and the covariates. For the bivariate situation, it is necessary to determine the expectation of the failure times for one component conditional on the failure or censoring time of the other component. Two different methods have been developed to estimate these expectations. In the semiparametric approach these expectations are determined from a modification of Burke's estimate of the bivariate empirical survival function. In the parametric approach censored data points are also replaced by their conditional expected values where the expected values are determined from a specified parametric distribution. The model estimation will be based on the revised data set, comprised of uncensored components and expected values for the censored components. The variance-covariance matrix for the estimated covariate parameters has also been derived for both the semiparametric and parametric methods. Data from the Demographic and Health Survey was analyzed by these methods. The two outcome variables are post-partum amenorrhea and breastfeeding; education and parity were used as the covariates. Both the covariate parameter estimates and the variance-covariance estimates for the semiparametric and parametric models will be compared. In addition, a multivariate test statistic was used in the semiparametric model to examine contrasts. The significance of the statistic was determined from a bootstrap distribution of the test statistic.},

doi = {},

journal = {},

number = ,

volume = ,

place = {United States},

year = {1992},

month = {1}

}