Use this URL to cite or link to this record in EThOS: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.603774
Title: Estimation of disease progression parameters from retrospective or cross-sectional data
Author: Harrison, D. A.
Awarding Body: University of Cambridge
Current Institution: University of Cambridge
Date of Award: 2003
Availability of Full Text:
Full text unavailable from EThOS. Please contact the current institution’s library for further details.
Abstract:
Cross-sectional or case-control studies offer an attractive alternative to the follow-up of large numbers of asymptomatic subjects when the disease of study is rare. However, in the field of stochastic modelling very little work has taken place in developing techniques for parameter estimation from such studies. Some simple two-parameter models have been fitted to nested case-control data, but the scope for fitting more complex models is not clear. There is no clear knowledge a priori of which parameters are estimable for a given design. This thesis is concerned with developing techniques for such estimation and discovering the limitations of various designs for identification of particular aspects of disease progression. The following designs are studied: - A case-control study of mammographic patterns and breast cancer screening. Estimation of progression parameters and covariates by the conditional likelihood approach becomes difficult as the model complexity increase. Introducing epidemiological modelling techniques enables to draw qualitative conclusions on the interaction between mammographic pattern and malignancy grade. - A case-control study of cervical cancer screening. To account for the likelihood that many premalignant lesions do not process to invasive cancer, models are fitted involving mover-stayer mixtures of Markov chains and frailties. - A cross-sectional sample, within a cohort undergoing colonoscopy, of subjects with no disease, premalignant polyps and colorectal cancer. Markov chain models are fitted, with parameter estimates based on a maximum likelihood approach using the known sampling fractions. It is concluded that non-prospective studies do offer scope, albeit limited, for estimation of disease progression parameters. A generic approach is not possible, but a general rule is to start with simple models and work towards more complex and realistic models. Deterministic models may succeed where stochastic models fail. Results of estimation in the applications above illuminate various aspects of disease progression and may inform design of future studies of disease prevention and early detection.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.603774  DOI: Not available
Share: