Association of Colonoscopy and Death From Colorectal Cancer

Context The effect of colonoscopy on colorectal cancer (CRC) mortality is unknown. Contribution By using a Canadian province-wide data set, the authors identified 10292 case patients who died of CRC and, for each case patient, 5 matched controls who did not. A total of 7.0% of the case patients and 9.8% of the controls had colonoscopy. Colonoscopy was associated with fewer CRC deaths (odds ratio, 0.69 [95% CI, 0.63 to 0.74]). The odds ratios for death from CRC that developed in the left and right colon were 0.33 (CI, 0.28 to 0.39) and 0.99 (CI, 0.86 to 1.14), respectively. Caution This was an observational casecontrol study. The data set did not identify a reason for colonoscopy. Implication Colonoscopy may be much less effective in preventing death from CRC of the right colon compared with the left colon. The Editors Colorectal cancer (CRC) is the second most common cause of cancer death in North America (1). Although colonoscopy is untested in randomized trials, many lay organizations (2, 3) and specialty societies (46) advocate it as the preferred screening method (7, 8). In the United States, screening colonoscopy rates have rapidly increased (911). The evidence for effectiveness of screening colonoscopy is indirect, inferred from reductions in CRC deaths in randomized trials of screening with fecal occult blood testing (12), a test that is much less sensitive than colonoscopy (13). In casecontrol studies, large-bowel endoscopy was associated with a 50% reduction in CRC development and 60% reduction in CRC deaths (14, 15). However, most of these studies included few women and evaluated sigmoidoscopy. Casecontrol studies of the association of colonoscopy with the incidence of CRC (16, 17) have been conducted but are more prone to bias than those defining case status by cancer death (18). Colonoscopy has real-world limitations. In a population-based study, 6% of patients with newly discovered right-sided CRC had undergone colonoscopy 6 months to 3 years before diagnosis, indicating a substantial miss rate in the community setting (19). Detection rates vary by endoscopist and by endoscope withdrawal time (20). The accuracy of colonoscopy done in the real world is unknown but may be substantially less than that in published reports (2123). We studied colonoscopy throughout Ontario, Canada, hypothesizing that it would be associated with fewer CRC deaths, but to a lesser degree than estimated in the literature. Methods The research ethics board of St. Michael's Hospital, Toronto, Ontario, Canada, approved the study. Study Design We did a casecontrol study of the association of colonoscopy and CRC deaths. We measured the odds of exposure to colonoscopy in case patients (persons who died of CRC) and controls (persons who did not die of CRC) and calculated an odds ratio (OR) for exposure. Because this OR is mathematically equal to the ratio of the odds of being a case patient given colonoscopy exposure to the odds of being a control given colonoscopy exposure, it estimates the association between colonoscopy and CRC death. Data Sources We used 4 data sources. The Ontario Cancer Registry tracks all incident cancer cases diagnosed in Ontario since 1964; it is more than 95% complete (24). It is linked to the Mortality File, an electronic database maintained by the Registrar General of Ontario to tally deaths of Ontario residents. Data on vital status and cause of death were available through 31 December 2003. The Ontario Health Insurance Plan database contains information on claims billed by physicians for services and has been linkable since 1991, permitting identification of almost all medical procedures occurring in Ontario. The Registered Persons Database, a roster of all Ontario Health Insurance Plan beneficiaries, contains almost all Ontario residents. The Canadian Institute for Health Information hospital discharge abstract database, linkable since 1988, contains information on every patient discharged from a hospital or same-day surgery unit in Ontario and includes patient demographic information, major diagnoses, procedures, and discharge status. Identification of Case Patients We identified case patients from the Ontario Cancer Registry. They received a diagnosis of CRC (International Classification of Diseases, Ninth Revision, codes 153.0 to 153.4 or 153.6 to 154.1) from 1 January 1996 to 31 December 2001 and died of CRC or related causes between 1 January 1996 and 31 December 2003 (International Classification of Diseases, Ninth Revision, codes 159 [malignant neoplasm of other or ill-defined sites within the digestive organs and peritoneum], 197 [secondary malignant neoplasm of respiratory and digestive systems], or 199 [malignant neoplasm without specification of site]). Complete identification of CRC deaths by using the cause of death on Ontario death certificates requires including deaths from these closely related causes. We included only persons age 52 to 90 years so that all controls would be in the screening-eligible age range during the exposure period. We excluded persons who had received a previous diagnosis of CRC, lived where physicians do not bill directly (<5% of the sample), or were not continuously eligible for an Ontario Health Insurance Plan from 1 January 1992. For secondary analyses, we stratified case patients by age; sex; and site of primary CRC, as assessed by registrars of the Ontario Cancer Registry from pathology reports and hospital discharge summaries (right-sided cancer [proximal to the splenic flexure], left-sided cancer [splenic flexure to rectum], or unknown site of cancer). Identification of Controls From the Registered Persons Database, we selected persons continuously eligible for Ontario Health Insurance Plan coverage from 1 January 1992 to 31 December 2003 or death. Case patients were matched to controls for factors known to influence colonoscopy rates and risk for CRC death: sex (1, 25), socioeconomic status (25, 26), and age (25). To minimize bias from differing access to colonoscopy, we matched for geographic location. Consequently, each case patient had 5 controls matched for sex, income quintile (based on the mean household income of the enumeration area of residence), residence location by health care region during year of diagnosis, and calendar year of birth. For each potential match, we defined the referent date, the date on which case patients received a diagnosis of cancer, determined from the Ontario Cancer Registry, a source that uses a standardized hierarchy for date of diagnosis. The date of pathologic confirmation of cancer, if available, was considered the date of diagnosis. We then linked the cohort of potential matches to the Ontario Cancer Registry to detect and exclude potential matches with a diagnosis of CRC on or before the referent date or death from CRC before 31 December 2003. Controls were alive at the date of the case patients' death. We included controls who developed CRC after the referent date. By matching for calendar year of birth, case patients and controls had an equal period to be exposed to colonoscopy before the date of CRC diagnosis (referent date). By using random numbers, we randomly selected 5 controls from all matches and assigned them to a case patient to form the control group. Determining Exposure By using the Ontario Health Insurance Plan billing codes (Appendix Table), we detected exposure to colonoscopy from 1 January 1992 to an index date 6 months before case patients received a diagnosis of CRC. Every case patient and control had more than 42 months of potential exposure to colonoscopy and at least 18 months of potential exposure while of screening-eligible age. We defined 2 exposures: any colonoscopy regardless of completeness (Appendix Table) and complete colonoscopy (colonoscopy coded as reaching the cecum). We did not consider flexible sigmoidoscopy to be an exposure. We treated exposure to colonoscopy as a binary variable: Persons who had at least 1 colonoscopy were considered exposed, and if any colonoscopy was complete to the cecum, those persons were considered exposed to complete colonoscopy. For persons who had more than 1 colonoscopy, we included only the first complete colonoscopy. In persons with no complete colonoscopy, we included the first incomplete colonoscopy. From billing codes, we detected polypectomies done during colonoscopy. We ascertained the self-designated specialty of the physician who billed for the included colonoscopy from the Ontario Physicians Human Resources Data Centre. Appendix Table. Diagnostic and Procedure Codes Diagnostic colonoscopies (those done to evaluate CRC symptoms, confirm the diagnosis, or search for metachronous tumors) would be associated with greater odds of CRC death than screening colonoscopies, which could mask any effect of screening colonoscopy on preventing death from CRC. Because Ontario Health Insurance Plan billing codes do not distinguish between screening and diagnostic colonoscopies (27), we tried to increase the proportion of screening colonoscopies by excluding all colonoscopies done within 6 months before the date of CRC diagnosis (referent date). This strategy assumes that most diagnoses of CRC in the Ontario Cancer Registry would be made within 6 months of a colonoscopy to investigate symptoms, confirm diagnosis, or search for metachronous tumors. Previous investigators have used a 6-month window of exclusion (19), although some chose a shorter period (14). To validate our choice, we recalculated the OR for the association of colonoscopy and CRC death for windows varying from 0 to 12 months. Determining Comorbid Conditions We identified comorbid conditions at a hospital discharge occurring within 60 months before the index date. We calculated the Charlson Comorbidity Index score (28) as 0, 1, 2, or 3 or more. We assigned a score of 0 to those who were not hospitalized. Statistical Analysis We calculated descriptive statistics for study variables stra