The Effectiveness of Colonoscopy in Reducing Mortality From Colorectal Cancer

IN RESPONSE:

Our recent study investigating the association between colonoscopy and colorectal cancer mortality has generated substantial interest. The media attention in response to publication of our study has been due, at least in part, to a misinterpretation of our methodology and primary outcome. We did not evaluate the rate of missed cancer cases at colonoscopy, and the degree to which missed cancer cases (or missed precursor lesions) may have contributed to our findings—that colonoscopy is significantly associated with a reduction in mortality from left-sided colorectal cancer but not right-sided colorectal cancer—is unknown. There has been considerable speculation with respect to the influence of procedural quality in Ontario on our findings; however, our understanding of the quality of colonoscopy in the general population in any jurisdiction is limited and is almost certainly lower than that published in series from expert centers.

In addition, this line of argument reveals an unwavering belief in the potential of colonoscopy to prevent the overwhelming majority of colorectal cancer cases and colorectal cancer deaths. However, our understanding of the molecular basis of colorectal cancer carcinogenesis is evolving, and it seems likely that a considerable proportion of cancer cases do not originate from easily identified adenomas that have a slow rate of progression (1). Although improvements in the quality of colonoscopy and the use of other screening methods may increase our ability to detect colorectal cancer or precursor lesions (such as the serrated adenoma), a rigorous evaluation of the relative effectiveness of colorectal cancer screening methods in a randomized trial is needed. Specifically, we need to know the marginal benefit of colorectal cancer screening with colonoscopy compared with flexible sigmoidoscopy in terms of reducing deaths from this disease.

Case–control studies are a challenging methodology and prone to numerous biases. We agree with Dr. Romagnuolo and colleagues that selection of case patients and control participants is important and can be difficult, particularly in evaluation of maneuvers that may have a role in screening, investigation of symptoms, and surveillance of patients after treatment. We did not include control participants in whom colorectal cancer was diagnosed before the date of diagnosis of their matched case patient because the appropriate exposure interval in such individuals is unclear, given that most would be undergoing regular surveillance colonoscopy after diagnosis. Because exposure in our study was based on administrative data, our information on exposure to colonoscopy was complete for all case patients and control participants and accurate with respect to timing, necessary prerequisites to minimize bias when only those without disease are selected as control participants (2). The length of follow-up after diagnosis might be considered too short to ensure that enough time had elapsed after diagnosis to identify all cases (that is, all colorectal cancer deaths in those with colorectal cancer diagnosed in our study interval); however, this would tend to bias the study toward an overestimate of the strength of the association between colonoscopy and mortality from colorectal cancer.

Far more important in terms of threat to the validity of our study results, we could not determine the indication for colonoscopy necessitating the use of an exclusion window, in which colonoscopy performed close to the date of diagnosis (whether for screening or investigation of symptoms) was not considered an exposure. We agree with Drs. Weiss and Doria-Rose that, because of this, it would be ill-advised to consider our odds ratio estimates of the association between colonoscopy and colorectal cancer death as precise or generalizable to other populations. Although many assume that the limitations of our study have resulted in an underestimate of the strength of the association between colonoscopy and colorectal cancer mortality, Drs. Weiss and Doria-Rose correctly highlight that, because of the necessity of the 6-month exclusion window, our study may have overestimated the strength of the association. However, despite the limitations, there is no reason to conclude that potential biases due to the study design would have influenced the major findings of this study: that colonoscopy is strongly associated with a reduction in colorectal cancer mortality, but that the association is not uniform throughout the colon. Although our study does not provide an explanation for the lack of association between colonoscopy and prevention of right-sided colorectal cancer deaths, our findings seem to be consistent with an emerging literature (3–5).