Aspects of quantitative cancer dose-response modeling and the role of the lognormal distribution /
| Main Author: | |
|---|---|
| Other Authors: | , , |
| Format: | Thesis Book |
| Language: | English |
| Published: |
1988.
|
| Subjects: | |
| Online Access: | ProQuest, Abstract Link to OAKTrust copy |
| Abstract: | Quantitative cancer dose-response models have been the subject of much research. The multistage model is currently the most widely used cancer dose-response model. Several characteristics of the multistage model iterative estimation procedure are derived; namely, the concavity of the likelihood function, the use of Akaike's information criterion to select the degree of the model, the derivation of initial parameter estimates, and the use of Kuhn-Tucker conditions as stopping criteria. Current dose-response models including the multistage model have limited biological rationale and tend to greatly oversimplify the carcinogenic process. To incorporate more science into the cancer dose-response models, Sielken (1987) proposed a new family of models. The lognormal is an important part of these new models. To make the new models more tractable, approximations and alternatives to the lognormal density are proposed. Often there are problems in choosing between the lognormal distribution and the normal distribution. Hypothesis testing procedures based on the likelihood ratio test for these two non-nested families are suggested and examined. In a large scale simulation study, the critical values and corresponding powers of the suggested procedures were obtained. Some of the consequences of assuming normality when the underlying distribution is lognormal are determined and illustrated. |
|---|---|
| Item Description: | Typescript (photocopy). Vita. "Major subject: Statistics." |
| Physical Description: | xvii, 139 leaves : illustrations ; 29 cm |
| Bibliography: | Includes bibliographical references. |