Skip to Main Content
 

Global Search Box

 
 
 
 

Files

File List


case1056547652.pdf (3.57 MB)

ETD Abstract Container

Abstract Header

A covariate model in finite mixture survival distributions

Soegiarso, Restuti Widayati
http://rave.ohiolink.edu/etdc/view?acc_num=case1056547652

Abstract Details

1992, Doctor of Philosophy, Case Western Reserve University, Epidemiology and Biostatistics.
Within a clinical trial setting, observations often come from different classes or groups that are related to the disease being studied. These classes may be different diagnoses or prognoses that may affect survival. Furthermore, some observations may have unknown group memberships, or new observations come along, and it is of interest to classify these observations into one of the groups based on their known survival times and covariates. This research addresses the problem of using finite mixture distributions with covariates to model such heterogeneous data. The survival distribution of these observations is considered as a mixture of two component exponential distributions, each representing a group. The hazard function of each component distribution is expressed as a log-linear function of the covariates. Maximum likelihood estimates of the covariate coefficients and the parameters of the component distributions are obtained. Five procedures that use the survival times as well as the covariates are constructed for classifying new observations into one of the component distributions. The classification procedures are compared with the neighborhood non-parametric procedure that uses the covariates alone. The methodology is illustrated using data from a prospective clinical trial in breast can cer based at Case Western Reserve University. These patients are considered to be a mixture of those with one to three positive nodes, and those with four or more positive nodes. Two covariates, estrogen receptor status and tumor diameter, are used in the model. The survival distributions from the proposed model seem to fit the actuarial distribution quite well. The proposed classification procedures also work quite well, and in fact they produce smaller error rates than the neighborhood procedure. To study the performances of the classification procedures when different sizes of correlations between the grouping variable and the covariates exist, the methodology is applied to simulated clinical trial data. The results from the simulation study suggest that when small or moderate correlation exists, the classification procedures that use the survival times and the covariates are more accurate than the neighborhood method. Furthermore, when the correlation is large, the proposed classification procedures yield better or comparable results to the neighborhood method.
Nahida Gordon (Advisor)
147 p.
Biology, Biostatistics
covariate model; mixture survival; distributions

Recommended Citations

Citations

  • Soegiarso, R. W. (1992). A covariate model in finite mixture survival distributions [Doctoral dissertation, Case Western Reserve University]. OhioLINK Electronic Theses and Dissertations Center. http://rave.ohiolink.edu/etdc/view?acc_num=case1056547652

    APA Style (7th edition)

  • Soegiarso, Restuti. A covariate model in finite mixture survival distributions. 1992. Case Western Reserve University, Doctoral dissertation. OhioLINK Electronic Theses and Dissertations Center, http://rave.ohiolink.edu/etdc/view?acc_num=case1056547652.

    MLA Style (8th edition)

  • Soegiarso, Restuti. "A covariate model in finite mixture survival distributions." Doctoral dissertation, Case Western Reserve University, 1992. http://rave.ohiolink.edu/etdc/view?acc_num=case1056547652

    Chicago Manual of Style (17th edition)

case1056547652
664
© 1992, all rights reserved.
This open access ETD is published by Case Western Reserve University School of Graduate Studies and OhioLINK.