Surveillance endoscopy does not improve survival for patients with local and regional stage colorectal cancer†
This article is a US Government work and, as such, is in the public domain in the United States of America.
Endoscopic surveillance is recommended and widely practiced after definitive treatment for colorectal cancer, yet to the authors' knowledge there is little evidence supporting its benefit. The purpose of the current study was to estimate the impact of endoscopic surveillance on colorectal cancer–specific survival for persons with localized or regional colorectal cancer. The population included Medicare patients (age ≥65 years) who were diagnosed with local or regional stage colorectal cancer between 1986 and 1996.
The current study was a retrospective case–control study. Cases were defined as those individuals who died of colorectal cancer and controls were defined as those with colorectal cancer who did not die of colorectal cancer; controls were frequency matched to cases. Surveillance was defined as the use of colonoscopy, flexible sigmoidoscopy, or barium enema ≥6 months after diagnosis. Logistic regression was used to control for endoscopic procedure, race, comorbidity index at the time of diagnosis, and types of initial treatments after surgery.
The analysis group contained 8130 cases (29%) and 20,079 controls (71%). The average time to first bowel surveillance for those with at least 1 surveillance examination was 15.9 months after the diagnosis (median, 13 months). In the regression analysis, surveillance endoscopy was not found to be associated with improved colorectal cancer–specific survival (odds ratio of 1.01; 95% confidence interval, 0.95‒1.06 [P = 0.85]). Setting the surveillance interval to 12 months and 15 months rather than 6 months after diagnosis did not appear to influence the results.
Surveillance endoscopy does not appear to influence colorectal cancer–specific mortality in patients age >65 years who are diagnosed with localized or regional stage colorectal cancer. Cancer 2007. Published 2007 by the American Cancer Society.
Colorectal cancer is the third most common cancer in the U.S., with 148,600 new cases expected in 2006.1 Approximately 70% of patients diagnosed with colorectal cancer have no evidence of distant metastases, and for these individuals surgery and adjuvant chemotherapy when indicated with curative intent is the treatment of choice.2–4 After definitive treatment, periodic surveillance with endoscopy is often recommended on the grounds that it will lead to the earlier identification and management of recurrent disease or second primary tumors, with a corresponding survival benefit.5–9 Although evidence from individual studies has been conflicting,10–13 a recent review and updated clinical guideline for colonoscopy surveillance after cancer resection suggested that the evidence is considered sufficient to warrant a colonoscopy 1 year after resection to detect metachronous colorectal neoplasms.14 Although this recommendation is based on reports of a high incidence of metachronous second tumors after resection, to our knowledge to date there have been no large-scale randomized controlled trials documenting the efficacy of postoperative endoscopic monitoring in reducing mortality.
Direct imaging of the colon and rectum is costly and carries some risk for adverse events. Consequently, the purpose of the current study was to use a large, population-based database of older enrollees with new colorectal cancer diagnoses to evaluate the correlation between ≥1 lower intestinal tract endoscopies or barium enemas (hereafter referred to as “endoscopy”) and survival for persons who have been diagnosed with local or regional disease. We tested the null hypothesis that the use of endoscopy has no impact on colorectal cancer–specific survival for individuals diagnosed with localized or regional colorectal cancer.
MATERIALS AND METHODS
Database and Patient Selection Criteria
The data used for this study were drawn from the National Cancer Institute's database of the Surveillance, Epidemiology, and End Results (SEER) cancer registry15 and linked to Medicare claims. The SEER Program is considered as the standard for quality among cancer registries around the world. Each year, studies are conducted in the SEER areas to evaluate the quality and completeness of the data being reported (SEER's standard for case ascertainment is 98%).16 Vital status is determined through death certificate records. The linkage, described in detail elsewhere,17 includes 94% of all SEER cases age ≥ 65 years.
Patients were included in the study if they were diagnosed with primary local or regional (local includes TNM Stages I and II disease and regional includes TNM Stage III disease) colon or rectal cancer between January 1, 1986 and October 31, 1996 and received surgical treatment for their disease. For reference and future studies, we include in Table 1 the histology and site codes used to identify colorectal cancer cases. Exclusion criteria were as follows: colorectal cancer diagnosed prior to age 65 years, diagnosis made by death certificate or autopsy, or diagnosis occurring in an unknown month; metastatic disease (Stage IV) at the time of diagnosis; enrollment in hospice prior to or at the time of diagnosis; prior or simultaneous diagnosis of any noncutaneous malignancy; diagnosis of carcinoma in situ or cancer of the appendix; and the diagnosis of Crohn disease or ulcerative colitis. The latter 2 conditions were identified by searching hospital and physician claims data for the appropriate International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) diagnosis codes. Patients receiving total colectomy as their initial treatment were also excluded. Finally, Medicare beneficiaries enrolled in a health maintenance organization (HMO) during the month of cancer diagnosis were excluded because the Centers for Medicare and Medicaid Services (CMS) does not require HMOs to submit individual claims for services.
Table 1. Diagnostic and Procedural Codes Used to Determine Surgery or Endoscopic Removal of Colorectal Cancer, Presence of Ulcerative Colitis or Crohn Disease, Recurrent Disease, and Bowel Surveillance in Patients with Local and/or Regional Colorectal Cancer Who Were Age ≥65 Years
|Surgical removal of colorectal cancer||45.7X, 48.41, 48.49, 48.5, 48.61–48.69||44140, 44141, 44143–44147|| |
|Nonsurgical removal of colorectal cancer||45.41–45.43, 48.31–48.36||45308, 45309, 45315, 45320, 45333, 45336, 45338, 45339, 45383–45385|| |
|Ulcerative colitis|| || ||556.9|
|Crohn disease|| || ||555.9|
|Chemotherapy||964XX, 965XX (X = any number)||V58.1, 99.25 (or NEC99.25), J9000-9999, Q0083-Q0085|| |
|Radiation therapy||V6671, V672, V58.0, 92.21–92.29||77XXX (X = any number), 0330, 0333|| |
|Sigmoidoscopy||45.24, 48.22, 48.23||45300, 45305, 45308, 45309, 45310, 45315, 45320, 45330, 45331, 45333, 45336, 45338, 45339|| |
|Colonoscopy||45.23||44388, 44389, 44392, 44393, 44394, 45335, 45378, 45380, 45383, 45384, 45385|| |
|Endoscopy with destruction||45.42, 45.43, 48.31–48.36|| || |
|Barium enema||87.64||74270, 74280|| |
To determine the index date of initial colorectal surgery, Medicare hospital, outpatient, and physician claims files were searched over the 2 months after the SEER month and year of diagnosis (Table 2). If no date could be determined from the claims data, we used the midpoint of the month of diagnosis from the SEER data. Once the month of initial surgery was identified, the index date was set as the 15th day of the month to allow for variation in the exact date of surgery on the Medicare claim.
Table 2. Baseline Characteristics of Population with Colorectal Cancer, 1986–1996
| Male||4019 (49%)||9467 (47%)||13,486 (48%)|
| Female||4111 (51%)||10,612 (53%)||14,723 (52%)|
|Age at diagnosis, y|
| 65–69||1811 (22%)||4141 (21%)||5952 (21%)|
| 70–74||2014 (25%)||4802 (24%)||6816 (24%)|
| 75–79||1876 (23%)||4688 (23%)||6564 (23%)|
| 80–85||1429 (18%)||3740 (19%)||5169 (18%)|
| ≥85||1000 (12%)||2708 (13%)||3708 (13%)|
|Stage at diagnosis|
| Localized||2377 (29%)||6656 (33%)||9033 (32%)|
| Regional||5753 (71%)||13,423 (67%)||19,176 (68%)|
|Year of diagnosis|
| 1986–1990||3987 (49%)||8764 (44%)||8130 (29%)|
| 1991–1996||4143 (51%)||11,315 (56%)||20,079 (71%)|
| White||6901 (85%)||17,281 (86%)||24,182 (86%)|
| Black||586 (7%)||1103 (6%)||1689 (6%)|
| White Hispanic||267 (3%)||600 (3%)||867 (3%)|
| Other||376 (5%)||1095 (5%)||1471 (5%)|
| Yes||1803 (22%)||3513 (17%)||5316 (19%)|
| No||6327 (78%)||16,566 (83%)||22,893 (81%)|
| Yes||855 (11%)||1071 (5%)||1926 (7%)|
| No||7275 (89%)||19,008 (94%)||26,283 (93%)|
|Time in cohort|
| 1–2 y||3297 (41%)||4027 (20%)||7324 (26%)|
| 2–3 y||1805 (22%)||4124 (21%)||5929 (21%)|
| 3–4 y||1149 (14%)||3296 (16%)||4443 (16%)|
| 4–5 y||753 (9%)||2309 (11%)||3062 (11%)|
| 5–6 y||457 (6%)||1763 (9%)||2220 (8%)|
| 6–7 y||250 (3%)||1435 (7%)||1685 (6%)|
| 7–8 y||150 (2%)||798 (4%)||948 (3%)|
| 8–9 y||87 (1%)||755 (4%)||842 (3%)|
| 9–10 y||69 (0.9%)||581 (3%)||650 (2%)|
| ≥10y||113 (1%)||991 (4%)||1104 (4%)|
| 0||6984 (86%)||16,647 (83%)||23,631 (84%)|
| 1||802 (10%)||2283 (11%)||3085 (11%)|
| 2||245 (3%)||761 (4%)||1006 (4%)|
| ≥3||99 (0.9%)||388 (1%)||487 (1%)|
|Reason for censoring|
| End of file†|| ||3608 (18%)||3670 (13%)|
| Death|| ||7729 (38%)||10,385 (37%)|
| Recurrent disease|| ||2196 (11%)||5329 (19%)|
| Entry into HMO|| ||3867 (19%)||4441 (16%)|
| Second malignancy|| ||1832 (9%)||2096 (7%)|
| Went to hospice|| ||509 (3%)||1687 (6%)|
| Second primary tumor|| ||306 (2%)||547 (2%)|
| >1 reason for censoring|| ||32 (0.2%)||54 (0.2%)|
Each individual in the cohort was followed from the date of his/her initial colorectal surgery until death or censoring, or until the last date for which Medicare claims data were available (December 31, 2003). Patients were censored when they lost Medicare A/B entitlement, enrolled in an HMO, or enrolled in hospice at any time after their date of diagnosis. Patients were also censored if they developed recurrent colorectal cancer or a second primary colorectal tumor, or had a noncutaneous malignancy other than colorectal cancer. Recurrent disease is not identified in the SEER registry, and therefore was identified from Medicare files. Recurrence was defined as receiving chemotherapy ≥16 months after the index date and/or radiation therapy ≥12 months after the index date (Table 2). The first date of chemotherapy or radiation therapy identified in the Medicare files was used as the recurrence date.
Determination of Bowel Surveillance Patterns
Medicare claims were searched for bowel surveillance procedures from the date of diagnosis until the date of censoring or the end of the observation period. Bowel surveillance procedures included colonoscopy, sigmoidoscopy, endoscopy, and barium enema (Table 1). Endoscopic procedures performed within 6 months of diagnosis were excluded because they were likely to be related to treatment-related monitoring or the management of complications due to therapy. If a second surveillance procedure of the same type was detected within 1 month of an identified surveillance, these events were consolidated into a single surveillance event because this may have represented a delay in submitted charges. The combination of either flexible sigmoidoscopy followed within 2 months by colonoscopy, or barium enema followed within 2 months by colonoscopy, was considered to represent 1 bowel surveillance event. This was done based on the assumption that the second surveillance was likely representative of a complete evaluation of an inconclusive first surveillance test. The date of the first event was chosen as the surveillance date for analysis.
The study was designed as a case–control study with the goal of evaluating the impact of surveillance endoscopy on colorectal cancer-related survival. Cases were defined as those who died of colorectal cancer. Controls were defined as those with colorectal cancer who did not die of colorectal cancer. Controls were selected by 1:3 frequency matching for year of diagnosis (before or after 1991), stage of disease at diagnosis, gender, age at the year of the case's diagnosis (within 5-year increments, starting at age 65 years), and survival time. One strong design feature of our study is the use of matching survival time. This aspect of design addresses the potential length time bias that was not considered in the previous studies. Because this was a retrospective case–control study, measures were taken to decrease bias. We matched on case–control survival time to allow controls as much opportunity as the cases to undergo surveillance, by searching for surveillance procedures up to the matched cases' time of death or censoring. To be eligible for the analysis, controls had to have survived at least as long as their matched case so as to allow cases and controls a similar opportunity for endoscopic surveillance after diagnosis.
The exposure of interest was the presence of bowel surveillance. In the main analysis, cases who had undergone total colectomy were excluded together with their matched controls because these patients are unlikely to receive further endoscopic procedures. We defined surveillance as the presence of ≥1 endoscopic procedures during the interval beginning 6 months from the time of diagnosis and up to 6 months prior to either death due to colorectal cancer or censoring. The interval over which surveillance was ascertained was the same for controls as for their matched case. To minimize the likelihood of including endoscopic procedures that had not been performed for routine surveillance purposes, we excluded procedures occurring within 6 months prior to death or censoring, with the exception of circumstances in which patients switched from Medicare fee–for–service into a Medicare HMO plan. For these individuals, endoscopic procedures were counted until the date of enrollment in an HMO.
We used a logistic regression model to determine the association between surveillance and colon cancer-related survival. Independent variables included receiving an endoscopic procedure, race, comorbidity index at the time of diagnosis, and types of initial treatments after surgery (chemotherapy, radiation therapy, and a combination of the 2). Comorbidity was calculated using Medicare data on claims for diagnoses or procedures that were made within 1 year prior to the diagnosis of colorectal cancer. We used the Charlson comorbidity index, created by Charlson et al.18 and later modified by Romano et al.,19 which employs the ICD-9-CM diagnosis and procedure codes.20 We calculated the odds ratios and 95% confidence intervals (95% CI) for exposure (surveillance) relative to nonexposure (no–surveillance). Because endoscopies in the early months after diagnosis may also have been performed to evaluate symptoms or complications of therapy versus surveillance, we ran secondary analyses with the endoscopic surveillance interval beginning at 12 months and 15 months after diagnosis.
We identified 34,328 patients with colorectal cancer. Of these, 3047 (9%) were excluded due to having undergone a total colectomy as primary treatment (1,248 cases [41%] and 1799 controls [59%]). The remaining 3071 unmatched cases and controls were removed prior to analysis. The final analysis group contained 8130 cases and 20,079 controls.
The average period of follow-up for cases and controls, starting at 3 months after diagnosis to the end of the surveillance interval, was 27.8 months (standard deviation [SD] 23.8 months; median, 20 months). The average survival time from the date of diagnosis to censoring or noncancer-related death was 33.3 months for cases (SD 20.4 months) and 52.0 months for controls (SD 31.9 months). Reasons for censoring were death (27%), recurrent disease (19%), end of file (22%), entering into an HMO (16%), second malignancy (7 %), enrolling in hospice (6%), and the development of a second primary tumor (2%).
For all patients over the observation period, 47% had no surveillance, 25% had 1 procedure, 14% had 2 procedures, 7% had 3 procedures, and roughly 7% had >3 procedures. Among those with at least 1 surveillance procedure, the average time to first bowel surveillance was 16 months after diagnosis (median, 13 months).
Among cases, 3964 (48.8%) had received bowel surveillance over the observation period. Among controls, 10,873 (54.2 %) received bowel surveillance, yielding an unadjusted odds ratio of 1.04 (95% CI, 0.99–1.09). After controlling for race, comorbidity index at the time of diagnosis, and types of initial treatments after surgery in the regression analysis, surveillance endoscopy was not found to be associated with improved colorectal cancer–specific survival over the observation period (odds ratio of 1.01; 95% CI, 0.95–1.06 [P = .85]) (Table 3).
Table 3. Logistic Regression Models Evaluating the Influence of Surveillance Endoscopy on the Odds of Dying from Colorectal Cancer, Controlling for Race, Comorbidity, and Initial Treatments (Chemotherapy, Radiation Therapy, and Combination of Chemotherapy and Radiation Therapy)*
|Black vs. white (non–Hispanic) Race||1.4||(1.2–1.5)||1.4||(1.2–1.5)||1.4||(1.2–1.5)|
|Hispanic vs. white (non–Hispanic) Race||1.1||(0.95–1.3)||1.1||(0.95–1.3)||1.1||(0.95–1.3)|
|Other vs. white (non–Hispanic) Race||0.85||(0.75–0.96)||0.85||(0.75–0.96)||0.85||(0.75–0.96)|
|Adjuvant treatment (chemotherapy plus radiation)||0.53||(0.44–0.65)||0.53||(0.44–0.65)||0.44||(0.44–0.66)|
In secondary analyses, changing the beginning of the endoscopic surveillance interval to 12 and 15 months after diagnosis did not appear to influence the results substantially. Changing the sample to include patients who had initially undergone total colectomy did not affect the influence of surveillance on cancer–specific survival (odds ratio of 0.98; 95% CI, 0.95–1.06 [P = .93]).
Finally, to address the concern that cancer–specific survival might not capture important differences (or lack of differences) in overall survival, we reran the primary analysis using all-cause mortality in place of cancer–specific mortality as the primary endpoint, with corresponding reselection of cases and controls. In this analysis, endoscopic surveillance was associated with improved survival (odds ratio of 0.62; 95% CI, 0.60–0.65 [P < .001]).
Lifelong periodic endoscopic surveillance is widely recommended for patients with local and regional stage colorectal cancer who have undergone definitive treatment. In this retrospective case–control study of Medicare enrollees diagnosed with localized colorectal cancer, we found no evidence that ≥1 surveillance endoscopy procedures after definitive therapy provides a colorectal cancer–specific survival benefit.
The size and composition of the cohort, how well the sample population was represented in relation to the population of adults age ≥65 years, and the quality of the SEER database in accurately identifying and staging persons with colorectal cancer strengthen the validity of the findings of the current study. Nevertheless, we need to highlight important limitations in the analysis. The primary limitation of the current analysis was our inability to distinguish symptom-based evaluation from true postcancer surveillance (evaluation in the absence of symptoms). In addition, the average duration of follow-up (31.4 months) was relatively short and may not have afforded adequate time to reveal a longer–term survival benefit.
A third limitation concerns the reliability with which we ascertained disease recurrence. If, for some individuals, it was not possible to identify disease recurrence from the administrative data (eg, patients who had disease recurrence but did not receive chemotherapy), then follow-up would continue to be counted beyond the true point of disease recurrence for these individuals. Inadequate ascertainment of recurrence among cases would bias the results against the intervention. Conversely, inadequate ascertainment of recurrence among controls would bias results in favor of the intervention.
We took several steps to exclude symptom-driven endoscopic procedures from the analysis. Endoscopic procedures that occurred immediately after diagnosis and 6 months prior to censoring for a variety of reasons that could be related to disease recurrence (eg, transfer to a hospice) were not included. To further minimize the likelihood of disease recurrences that were not conducted for routine surveillance purposes, individuals were censored if their SEER records indicated a second primary colorectal cancer, or if their records showed evidence of colon cancer-related chemotherapy beginning 16 months after the initial treatment or radiation therapy beginning 12 months after the initial treatment. Individuals were also censored if they had evidence of these treatments 6 months prior to death. When stratified by stage of disease at diagnosis, recurrence rates for both cases and controls were higher for those individuals with regional disease (12.7%) compared with those with localized disease (5.5%).
To our knowledge, the current study is the first to explicitly consider colorectal cancer mortality as an endpoint in the evaluation of bowel surveillance after the diagnosis of colorectal cancer. This contrasts with a recent case–control study of 3500 military veterans that used all-cause mortality and found a 43% reduction in mortality for persons undergoing surveillance endoscopy.21 This study ascertained cases using administrative data (ICD-9-CM codes and Veteran Administration death records), and therefore could not distinguish the stage of the cancer at the time of diagnosis, or the cause of death. Smaller prospective studies that have used all-cause mortality as an endpoint have found no significant benefit from endoscopy.22, 23 Although it has been hypothesized that small sample size was the primary reason for the apparent lack of benefit,21 we believe that the discrepancy can be better explained by the fact that physicians are not performing surveillance procedures on patients with significant concomitant illnesses who are at increased risk of death. Large, prospective controlled studies are needed to determine the survival benefit from surveillance endoscopy for persons with definitively treated colorectal cancer.
Guidelines for endoscopic surveillance after definitive therapy for colorectal cancer acknowledge that the primary benefit may exist from the detection of second primary tumors. Furthermore, guidelines do not address older patients, possibly because the potential survival benefit from identifying metachronous lesions is likely to be substantially less than for younger colon cancer survivors. Nevertheless, surveys of endoscopy practice after definitive treatment of localized colorectal cancer in persons age ≥65 years in the U.S. showed that 46% received endoscopy within 14 months of diagnosis, and 14% had received at least annual endoscopy up to 50 months after diagnosis.24
There are several reasons that might explain why endoscopic surveillance did not confer a colorectal cancer–specific survival benefit in this population. First and foremost, the population was elderly (age ≥65 years) at the time they were diagnosed with cancer. Although persons with colorectal cancer are at increased risk for developing a second primary tumor, the sojourn time for developing new malignant polyps may be so long that competing causes of death overwhelm any benefit from identifying and removing them. In support of this hypothesis is our finding that only 2% of patients experienced a second primary malignancy over the follow-up period. A recent study of SEER registries also found a comparatively low incidence of second primary colorectal cancers in persons age ≥65 years.25 Second, recurrence of the primary cancer may have a particularly poor prognosis in an elderly population. If, in this population, the identification of recurrent cancer by endoscopy versus symptoms or other measures (eg, serum carcinoembryonic antigen) affords little or no survival benefit, then this could explain the observed lack of association. Finally, the sporadic surveillance that has been observed in the Medicare population may not achieve benefits that could be attributed to screening at recommended intervals.24
The current study findings suggest that, in the setting of standard clinical practice, older patients receive no disease-specific mortality benefit from endoscopic surveillance after surgery for localized and regional colorectal cancer. These findings suggest that prospective controlled trials are needed to establish the value of surveillance for this population.
Supported by Grant R01 HG002941 from the National Cancer Institute.