(714f) Using a Semi-Mechanistic Model of Colorectal Cancer Development to Identify the Sources of and Quantify the Patient and Clinical Costs of Follow-up Colonoscopy Under- and Over-Use

Sherer, E. - Presenter, Roudebush VA Medical Center
Imperiale, T. - Presenter, Roudebush VA Medical Center
Ambedkar, S. - Presenter, Purdue University
Perng, S. - Presenter, Purdue University
Yih, Y. - Presenter, Purdue University

The development of colorectal cancer is somewhat unique among cancers in that precancerous neoplasia, or adenomas, can be both observed and removed during a colonoscopy. While this short-circuits the development of potential colorectal cancer, the cancer prevention via colonoscopy is not uniform across the population. Clinical studies of screening and follow-up colonoscopy have found that the prevalence of multiple or advanced adenomas at screening colonoscopy is related to an increased incidence of colon cancer and advanced adenoma at follow-up [1], and adenoma prevalence has been shown to have a stronger connection to subsequent advanced neoplasia than external risk characteristics such as race, gender, and family and smoking history [2]. As such, current follow-up colonoscopy guidelines, developed by the US Multi-Society Task Force and American Cancer Society, are stratified based on the adenoma findings at screening [3]. But these guidelines make quantitative recommendations based on what are essentially qualitative trends. For this abstract, we first introduce a mathematical model and framework ? developed to capture patient risks at multiple colonoscopies [4] ? and then apply this model to analyze the effects of recommended follow-up colonoscopy intervals. Specifically, we have identified regions of patient characteristics as well as colonoscopy findings and conditions with potential follow-up colonoscopy under- and over-use and have quantified the effect on patients (cost) and the endoscopy clinic (resources).

To incorporate the clinical risk factors, we developed an adaptive-predictive semi-mechanistic model ? in which the model predictions for an individual patient vary based on the findings and conditions of the colonoscopy measurement in addition to patient demographic characteristics - that captures the frequency of clinical relevant risk groups (no neoplasia, 1-2 non-advanced adenomas only, 3+ non-advanced adenomas only, any non-CRC advanced adenoma, and colorectal cancer) at first-, second-, and third-time colonoscopies. The model combines possible CRC development mechanisms into a basis set of ?patient types?. The a priori likelihood that an individual patient is of a particular patient type is a function of a patient's demographic characteristics and colonoscopy indications. These likelihoods for a particular patient are varied in a Bayesian manner in response to measurements of the colon and the conditions of those measurements which adjusts the model predictions and risk predictions for that patient. This model is tuned to longitudinal colonoscopy data which was manually abstracted from the procedure reports of >3,000 patients at the Roudebush VAMC.


[1] D. Lieberman, D. Weiss, et al. & D. Robertson, Gastroenterology 133 (2007) 1077-1085.

[2] M. Martinez, J. Baron, et al. & E. Greenberg, Gastroenterology 136 (2009) 832-841.

[3] S. Winawer, A. Zauber, et al. & D. Rex, Gastroenterology 130 (2006) 1872-1885.

[4] E. Sherer, S. Ambedkar, et al. & T. Imperiale, Proceedings of the 4th International Conference on Population Balance Modelling (2010).