Cancers take many years to develop, and death due to a cancer may occur many years after diagnosis.
Therefore, it is important to use innovative methods of analysis to make optimal use of all exposure data
both before and after diagnosis. We propose four types of innovations in the current application.
Lethal cancer models: Of highest public health interest is what risk factors predispose a disease-free
subject to die due to a specific cancer in the future. We have previously developed models for lethal
cancer by integrating models for cancer incidence with models for prognosis after a cancer diagnosis. In
this application, we propose to extend this approach in the case of colorectal cancer (CRC) by
considering three stages in the carcinogenic process: (a) development of advanced polyp, (b)
development of incident CRC after advanced polyp have been identified and removed, and (c) death due
to CRC among patients with incident CRC.
Latency models: Some prospective studies have risk factor data available at several points in time. One
issue is how these data should be optimally used to predict cancer incidence. One approach is to use the
most recent exposure; a 2nd approach is to use the total duration of exposure; a 3rd approach is to
introduce a lag between exposure and outcome assessment. In this application, we propose a latency
model to estimate the optimal weighting of previous exposures to predict cancer incidence; these
models enhance understanding of biological mechanisms for specific risk factors.
Cure Models: There have been many studies of risk factors predicting mortality among cancer patients.
However, for some cancers, if patients do not die from their cancer over a given period of time (e.g.,
within 5 years for CRC), then they are unlikely to ever die due to their cancer, and will probably die due
to another cause (i.e., they are cured). But what are the risk factors that predict cure? Although cure
models have been used before, they mostly are based on post-diagnostic risk factors. To our knowledge,
this is the 1st proposal to consider pre-diagnostic risk factors as predictors of cure.
Assessing Effects of Screening on CRC risk models: Colonoscopy is the current standard for CRC
screening. It is unique, in that if pre-cancerous lesions (i.e., adenomas) are found, then they are
removed, and the natural history of CRC progression is interrupted. However, even if these lesions are
removed, for some subjects, these lesions are more likely to develop again at a future time. Thus, it is
challenging how to control for effects of screening in CRC risk models. In this application, we propose an
innovative approach to control for screening by both assessing effects of a risk factor on adenoma
incidence, and effects of adenoma incidence on CRC risk.