It is well-known that bias such as lead-time and length distort studies of screening efficacy whether survival or incidence is of interest. A third bias, usually called overdiagnosis bias, occurs when an individual is only diagnosed with disease before death from a different cause because he/she is screened. These forms of bias can also arise in observational studies where the proportion screened and screening rates vary by risk-factor strata. This difference in screening behaviors influences corresponding case ascertainment or case enrollment probabilities which can lead to erroneous conclusions about the size of the risk-factor effect on the disease. It has been suggested that classic confounding occurs in such risk-factor studies when screening is efficacious; therefore, it can be addressed by conventional analyses such as stratification or confounder adjustment in regression models. However, even if the test is not efficacious, screening creates changes in case ascertainment probabilities which must be addressed using alternative methods. Recurrence-time models, long used for screening programs, can be adapted to model the affect screening use has on risk-factor studies. These models can be used to study the magnitude of potential bias, but may also be adapted to provide an analytic approach to correct estimates for such bias. The risk-factor studies nested in the PLCO trial are potentially affected by such bias, and this randomized study also provides a structure within which models of screening bias may be tested and validated. To validate our model, a variety of nested case-control studies will be developed that measure the effect smoking has on lung cancer and the degree to which the bias affecting those estimates change based on the study design will be determined. This process will include a) expanding a previously developed lead-time bias model to incorporate length and overdiagnosis; b) incorporating a more flexible and realistic model of screening that can incorporate the patterns documented in the PLCO trial; c) exploring if the mathematical model is valid using varied nested study designs within PLCO and comparing resulting logistic regression estimates to simulated results; and d) using the validated models to produce correction factors for use in other nested risk-factor studies. Results indicate that the mathematical model is highly sensitive to overdiagnosis as increasing rates increase expected bias, but relatively insensitive to using different screening test sensitivities. Increasing screening behavior differential during the study, preclinical duration length, and selecting from the intervention group are associated with increasing expected screening bias. Increasing screening behavior before the study and selecting from the usual-care group are associated with a decreasing expected screening bias. Although the mathematical model couldn't be validated as a correction factor here, the results suggest using a shorter preclinical duration distribution for the model may produce more accurate screening bias values. The focus of this work was to identify if chest x-ray screening could modify the estimated risk of smoking on lung cancer diagnosis. An additional goal was to develop a usable method for adjusting observational studies of lung cancer for the bias arising from differential chest x-ray screening between ever and never smoking groups. In a boarder sense, this work has provided an explanation of the effect screening use may have on an observational risk-factor study and an example of how to implement the mathematical technique. Additionally, this project has provided a more general method for doing sensitivity analyses on the screening related assumptions involved with these studies, whether nested in a randomized trial or sampled from the population at large.
University of Minnesota Ph.D. dissertation. September 2009. Major: Environmental Health. Advisor: Dr. Timothy R. Church. 1 computer file (PDF); xiii, 139 pages, appendix pages 123-139.
Jansen, Ricky Jeffrey.
Evaluating bias caused by screening in observational risk-factor studies of lung cancer nested in the PLCO randomized screening trial..
Retrieved from the University of Minnesota Digital Conservancy,
Content distributed via the University of Minnesota's Digital Conservancy may be subject to additional license and use restrictions applied by the depositor.