Methods

Top Methods Results Discussion Author & Article Info References

Data Source

The Patient Treatment File consists of many individual data files. The data files are managed by the Department of Veterans Affairs' Central Automation Center in Austin, Texas. The main files contain records of all inpatient treatments from all Veterans Affairs hospitals throughout the United States. Files are available for each fiscal year since 1970. Although the contents of the files and their structure have changed over the past two decades, information on essential demographic and health characteristics are available in similar form from all main files. Since 1984, other files have been added to cover treatment outside Veterans Affairs facilities and in nursing homes. Each patient record contains one primary and up to nine secondary discharge diagnoses. From 1970 to 1980, the diagnoses were coded according to the 8th revision of the Clinical Modification of the International Classification of Diseases (ICD-8-CM); since 1981, the 9th revision of the ICD has been used [7]. Individual patients can be identified by their social security numbers. Diagnostic procedures and surgical procedures were initially included in the main files. Separate surgical and medical procedure files were initiated in 1984 and 1988, respectively. Procedures are also coded according to ICD-9-CM, the coding scheme that makes it possible to record in both the medical and surgical procedure files up to five procedures per day of hospitalization. Since 1990, outpatient procedures have been included in the Patient Treatment File. All outpatient procedures are listed by codes taken from the Current Procedural Terminology (CPT-code) [8].

Identification of Case-Patients and Controls

Information on all veterans who were discharged with a diagnosis of colon cancer (ICD codes 153.0 to 153.9, except 153.5) or rectal cancer (ICD codes 154.0 to 154.1) between 1988 and 1993 was extracted from the Patient Treatment File. Their social security numbers were then used to search all of their previous discharge records since 1970, and patients with a diagnosis of colon or rectal cancer before 1988 were excluded from the case population. The remaining patients were assumed to represent incident cases with a first diagnosis of colorectal cancer between 1988 and 1993. Patients who had had a previous diagnosis of Crohn disease (ICD code 555), ulcerative colitis (ICD code 556), or familial polyposis (ICD code 759.6) between 1970 and 1993 were also excluded from the case population. Ulcerative colitis and Crohn disease are both associated with an increased risk for colon cancer [9, 10]. Patients with inflammatory bowel disease have frequent endoscopies to assess colonic involvement, diagnose complications, and screen for dysplasia. Procedures in these patient populations are determined by considerations other than those in the general population.

For each case-patient, one control was selected from among the patients who were discharged during the same fiscal year that the case-patient was first diagnosed with colorectal cancer. Each control was matched to his or her case-patient based on age, sex, and race. Controls who had had any previous discharge diagnosis of colorectal cancer between 1970 and 1993 were excluded from the complete annual files of 1988 through 1993. Similarly, controls who had had any previous diagnosis of Crohn disease, ulcerative colitis, or familial polyposis were excluded. After the exclusions, the remaining patient records of each year from 1988 to 1993 were randomized. For each case-patient, the randomized annual file was scanned sequentially until a control of the same age, race, and sex as the case-patient was identified. Ten-year age categories were used for matching—for example, less than 25 years, 25 to 34 years, 35 to 44 years, and so on.

Extraction of Procedures and Previous Diagnoses

Detailed procedural codes based on the ICD-9-CM have been available in the Patient Treatment File since 1981. Therefore, for case-patients and controls recruited during the last 6 years, we were able to search the procedure files dating from 7 to 13 years before recruitment for a history of previous procedures of the large intestine. The procedure files from 1981 until the date on which cancer was first diagnosed (or corresponding time of hospitalization in controls) were searched for all of the procedural codes shown in Table 1. The procedure that was used to establish the first diagnosis of cancer was not considered in the analysis.

In a second search, we followed case-patients and controls backward through the annual files (all main files since 1970) to assess the overall duration of health coverage by the Veterans Affairs system and to determine the number of all hospital discharges before the first diagnosis of colorectal cancer. These two variables were created to adjust for the extent of medical care that an individual veteran received through the Veterans Affairs system.

In a third search, we followed case-patients and controls backward through the annual Patient Treatment Files of the 5 years preceding the first diagnosis of colorectal cancer to record all previous discharge diagnoses. Indicator variables were created for the following primary or secondary discharge diagnoses: colorectal polyps (ICD codes 211.3, 211.4, and 569.0), carcinoma in situ (codes 230.3 and 230.4), any form of anemia (codes 280 to 285), hemorrhoids and any form of gastrointestinal hemorrhage (codes 455, 569.3, 578.1, and 578.9), signs and symptoms that suggested intestinal obstruction (codes 560.0, 560.3, 560.9, 564.0, and 569.2), diverticulosis and diverticulitis of the colon (codes 562.10 to 562.13), various types of nonspecific gastrointestinal complaints (codes 558.9, 569.4, 569.8, 569.9, 789.0, and 789.9), abnormal weight loss (code 783.2), and abdominal or pelvic swelling or mass (code 789.3). These diagnoses may often be associated with a diagnosis of colorectal cancer, or they may precede its definitive diagnosis; it is highly unlikely that cancer would be misdiagnosed, for example, as hemorrhoids or bleeding peptic ulcer, for more than 5 years. Arthritis and related disorders (codes 714 to 716 and 720 to 721) were common conditions in the Patient Treatment File that were likely to be treated with nonsteroidal anti-inflammatory drugs. Such drugs have been shown to reduce the risk for death from colorectal cancer [11, 12]. Currently, no separate ICD code identifies patients with hereditary nonpolyposis colon cancer (the Lynch syndrome).

Statistical Analysis

Except for the duration of coverage by the Veterans Affairs system and the number of hospital discharges, all variables were coded as dichotomous variables: Code 1 was used to indicate the attribute of interest, and code 0 was used to indicate its absence. The development of colon or rectal cancer served as the outcome variable. Any type of flexible endoscopy done before the first diagnosis of colorectal cancer served as the main predictor variable. Different types of discharge diagnoses that preceded the diagnosis of colorectal cancer (for example, polyps or anemia) represented the other predictor variables. The duration of coverage (in years) by the Veterans Affairs system and the number of hospital discharges were transformed into categorical predictor variables according to their quartile ranges. For each pair of predictor variables, the Kendall correlation coefficient or the Spearman correlation coefficient was calculated.

The proportional hazards regression procedure of the Statistical Analysis System was used for all types of univariate and multivariate conditional logistic regression analyses [13]. We did univariate conditional logistic regression analyses in which we considered each predictor variable separately. We also did multivariate conditional logistic regression analyses in which we considered all predictor variables simultaneously [14]. Besides main-effect models, the possibility of interaction terms among the variables were assessed.

We did separate analyses to identify the efficacy of specific procedures. During their follow-up in the Veterans Affairs system, many of the case-patients and controls had the same procedure more than once and different colorectal procedures in varying temporal sequences. Also, the different codes of the ICD-9-CM do not represent completely distinct procedures, whereas some identical procedures, such as colonoscopy or flexible sigmoidoscopy, may be assigned multiple procedural codes. For these reasons, it was difficult to compare the protective influence of individual types of procedures. To sort out the exclusive contribution of individual procedures, three separate multivariate analyses considered ICD code 45.23 and CPT code 453.78 to represent colonoscopy, the ICD code 45.24 and CPT code 453.30 to represent flexible sigmoidoscopy, and the ICD codes 45.41, 45.42, and 48.35 and CPT codes 453.33 and 453.85 to represent polypectomy. Of the procedures done before the first diagnosis of colorectal cancer, only the most recent in each category was considered in the regression analyses. To assess the contribution of other endoscopic procedures on the estimated effect of each procedure type, an indicator variable for all other procedures was included in each of the three separate regression analyses that examined flexible sigmoidoscopy, colonoscopy, and polypectomy, respectively. For instance, to control for the confounding influence of other endoscopic procedures in the regression analysis of colonoscopy, a new variable was created that indicated whether the individual patient had had any colorectal procedure other than colonoscopy, such as flexible sigmoidoscopy, rigid proctosigmoidoscopy, polypectomy, or barium enema. Similarly, in the regression analyses of flexible sigmoidoscopy and polypectomy, variables were created that indicated whether a patient had had any additional colorectal procedure.

Separate multivariate analyses were done to evaluate the influence of procedures in the outpatient and inpatient settings. To assess the duration of the protective effect that the endoscopic procedures provided, we did a series of 10 conditional multiple logistic regression analyses. Each regression analysis excluded case-patients and controls who had had a procedure within a given number of years before the first diagnosis of colon or rectal cancer. Exclusion points were set at 1-year intervals over the 10-year period.

Results

Top Methods Results Discussion Author & Article Info References

General Characteristics

During the period 1988 to 1993, colon cancer was diagnosed in 8722 patients and rectal cancer in 7629 patients. During the same time period, 1.9 million persons were treated in Veterans Affairs hospitals nationwide. Compared with the general veterans population, the patients with colon or rectal cancer were older and were more likely to be men. The population of patients with colon or rectal cancer also consisted of more whites than the general veterans population (Table 2).

Of the patients with a first diagnosis of colorectal cancer between 1988 and 1993, we excluded 21 patients who had had a previous diagnosis of inflammatory bowel disease and 4 patients who had had familial polyposis. The final case population consisted of 1093 patients who had previously had colorectal polyps, 70 patients who had had carcinoma in situ, and 1200 patients who had had arthritis or related diagnoses. Ninety-one patients with inflammatory bowel disease and 6 patients with familial polyposis were excluded from the control population. The control population consisted of 369 patients who had previously had colorectal polyps, 7 patients who had had carcinoma in situ, and 1399 patients who had had arthritis or related diagnoses.

Before the first diagnosis of cancer, patients with colon cancer had been followed in the Veterans Affairs system for an average of 6.8 ± 7.2 years (median, 4.1 years), and patients with rectal cancer had been followed for 6.1 ± 7.1 years (median, 2.4 years). The average duration of follow-up in the control population was 7.1 ± 7.4 years (median, 5.4 years). Before the first diagnosis of cancer, case-patients with colon cancer had been hospitalized 3.7 ± 5.0 times (median, 2.0 times), and case-patients with rectal cancer had been hospitalized 3.3 ± 4.8 times (median, 1.4 times). The corresponding frequency with which hospitalizations occurred in the control population was 3.7 ± 5.1 (median, 2.0). The differences in the number of hospitalizations between case-patients and controls were not statistically significant.

Comparison of Colorectal Procedures in Case-Patients and Controls

Of the case population of 16 351 patients, 1051 patients had had endoscopic procedures involving the large intestine during the period preceding the diagnosis of colorectal cancer. These procedures included 535 flexible sigmoidoscopies, 359 colonoscopies, and 291 polypectomies. In the control population of 16 351 persons, 1166 had had endoscopic procedures, including 638 flexible sigmoidoscopies, 467 colonoscopies, and 250 polypectomies.

Each pair of predictor variables yielded a nonsignificant correlation coefficient of less than 0.4, except between the duration of Veterans Affairs coverage and the number of discharges (Spearman correlation coefficient, = 0.73). Because duration of coverage proved to be a better predictor variable that was associated with a larger change in the log of the likelihood, the number of hospital discharges was excluded from multivariate analyses.

Table 3 shows the outcomes of the multivariate analyses with respect to colon cancer and rectal cancer, respectively. In the overall analysis, endoscopic procedures were associated with a reduced risk for subsequent development of cancer of the colon or rectum. A previous history of polyps or carcinoma in situ was associated with a markedly elevated risk for cancer. On the other hand, any previous diagnosis of arthritis or related diagnoses associated with chronic use of nonsteroidal anti-inflammatory drugs had a protective influence against the subsequent development of cancer. These associations applied similarly to colon cancer and rectal cancer. In patients with colon cancer, nonspecific diagnoses of anemia, gastrointestinal hemorrhage, intestinal obstruction, or abdominal mass frequently preceded the true diagnosis of cancer. Such patterns were less obvious in patients with rectal cancer, because we found only gastrointestinal hemorrhage to be associated with a significantly increased risk for cancer.

The multivariate analyses of colon cancer showed that both inpatient and outpatient procedures were associated with a protective influence (Table 4). Similarly, in patients with rectal cancer, we found both outpatient and inpatient procedures to be protective if analyzed separately. Table 4 also shows the protective contribution of specific procedures. We found that each procedure exerted a significant influence on its own. However, the odds ratios of individual procedures were about the same, and their confidence intervals overlapped. As already indicated in the Methods section, the nature of the data may have rendered any rigorous comparison of various endoscopic procedures very difficult.

As shown by the odds ratios in Figure 1, endoscopic procedures reduced the risk for both cancer types consistently during a period of 5 years before their first diagnosis. The protective influence of endoscopic procedures appeared to wear off after 8 years in patients with colon cancer and after 6 years in those with rectal cancer. When these sequential analyses were repeated for specific procedures, the general pattern of results was similar. All procedures had a protective influence during a period of 5 to 6 years before cancer diagnosis.

View larger version (19K): [in this window] [in a new window]   Figure 1. Odds ratios associated with endoscopic procedures. Odds ratios are for case-patients with colon cancer (top panel) or rectal cancer (bottom panel) compared with controls. Odds ratios less than 1.0 indicate a reduced risk for cancer. Bars indicate 95% CIs.

Discussion

Top Methods Results Discussion Author & Article Info References

The data from our study showed that U.S. military veterans who did not have colorectal cancer were about 50% more likely to have had endoscopic procedures than matched controls with colorectal cancer. These data provide strong support for the efficacy of endoscopy as a tool for reducing the risk for colorectal cancer. We found no significant difference among the protective influence of various types of endoscopic procedures. This lack of significant difference may reflect the fact that most patients have multiple procedures and that the protective contribution of an individual procedure is therefore difficult to assess [15]. A previous diagnosis of colorectal polyps or carcinoma in situ was a strong risk factor for the subsequent diagnosis of colon or rectal cancer. Diseases associated with the chronic use of aspirin or of other nonsteroidal anti-inflammatory drugs, on the other hand, were associated with a decreased risk for colon and rectal cancer. Such drugs have been shown to reduce the risk for adenomatous polyps and death from colorectal cancer [11, 12]; their action is mediated by inhibiting the mucosal arachidonic acid cascade and reducing the proliferative growth of the colorectal mucosa [16]. Our data corroborate previously observed epidemiologic patterns of colorectal cancer, which strengthens their overall reliability. In addition, the protective influence of endoscopic procedures was consistent with regard to both colon and rectal cancer and with regard to both inpatients and outpatients, providing further validation for our findings.

The concept of the polyp-cancer sequence proposes that cancers of the colon and rectum develop from benign adenomatous polyps of the large intestine [17]. This concept is supported by morphologic data [17, 18], molecular biology [19], animal studies [18], epidemiologic evidence [20, 21], and various clinical observations [1-5]. The most convincing evidence to support the concept of the polyp-cancer sequence has come recently from clinical studies that show that the removal of adenomatous polyps reduces the risk for death from cancer. In a retrospective study from the Mayo Clinic, Stryker and colleagues [1] followed 226 patients with large polyps that were detected by barium enema before colonoscopy was available, over a period of 5 to 20 years. Follow-up showed that 21 carcinomas developed at the polyp site and that 11 carcinomas developed at another site, with a cumulative risk for cancer after 20 years of 24%. In a second retrospective study from St. Mark's Hospital in London, Atkin and colleagues [2] followed for an average of 14 years 1618 patients who had had rectosigmoid adenomas removed during proctoscopy. Because these patients were not followed after the initial procedure, rectal cancer developed at a rate similar to that in the general population. However, 80% of the patients in whom cancer developed had had polyps that were incompletely resected initially. The risk for subsequent cancer in these patients was particularly high: the risk was 3.6 times greater in patients with large or villous adenomas. In another study (the National Polyp Study), patients had a clearing colonoscopy and were subsequently followed for at least 3 years [3]. Compared with historical controls, a significant reduction in the rate of colon cancer was noted. In a similar study from Norway, Gilbertsen and Nelms [4] found that the removal of rectal adenomas reduced the risk for rectal carcinoma when compared with historical controls from the same geographic area.

In a case–control study by Selby and colleagues [5], 261 patients with fatal colorectal cancer were compared with 868 age-matched controls. The investigators found that screening with a rigid instrument resulted in a significant reduction of colon cancer in the region of the large intestine that had been reached with the instrument. Mortality from cancer of other colonic sites was not affected. Although mortality from rectosigmoidal cancer was reduced by 60%, overall mortality from colorectal cancer decreased by only 30%. Most of the significant outcomes were based on a comparison with 210 controls who had rigid proctosigmoidoscopies, with only 23 of the patients having had this procedure before their cancer was diagnosed. The negative association between rigid proctosigmoidoscopy and cancer applied even for the period of 8 to 10 years before the diagnosis of cancer. The reduction in mortality could be traced back to only 12 polypectomies done in the control population. The aim of our study was to confirm these data in a different population and test the benefit of other colorectal procedures that affect the region of the colon that extends beyond the proctosigmoid.

It is important to address several shortcomings of our study and its use of the Patient Treatment File as its database. The primary limitation of the case–control design, as with all outcomes research, is the fact that procedures were not randomly assigned and may be confounded by other factors that could not be fully accounted for in the analysis. The duration of coverage by the Veterans Affairs system served to adjust for the potentially confounding influences of health-seeking behavior and the extent of medical care provided to an individual veteran. Similarly, recording each patient's previous history with respect to specific diagnoses adjusted for differences in health status before the cancer diagnosis. These adjustments were designed to overcome some of the limitations inherent to a case–control study. Another shortcoming of our study is that the Patient Treatment File is limited in its ability to record all pertinent health care activity. No records exist to document the variety of medical care that the individual case-patients and controls may have received outside the Veterans Affairs system. Undoubtedly, veterans receive part of their medical care outside of the Veterans Affairs facilities, or they may drift in and out of the Veterans Affairs system at different points in their medical history. Therefore, the activity documented in the Patient Treatment File represents only a fraction of all medical care that any an individual veteran receives. Many screening tests, such as anoscopy or rigid proctoscopy, are done as outpatient procedures. Our inability to document outpatient activity that occurred before 1990 introduced a systematic error to both groups of case-patients and controls.

The endoscopic procedures used to diagnose colorectal cancer were excluded from the analysis. If such a procedure was the only endoscopy done in a particular case-patient, this patient entered the analysis as if he or she had not had endoscopy at all. Because controls are not excluded on this basis, this may have led to some bias in favor of a protective influence of endoscopy. Finally, the data of the Patient Treatment File deal exclusively with veterans, a population that consists primarily of men with low family income and increased exposure to environmental risk factors such as alcohol consumption and smoking. Their health behavior and medical care therefore differ from those of other Americans, and, in general, their health awareness and concern about prevention may be less developed than in the general U.S. population. Therefore, the uniqueness of the Veterans Affairs system and the peculiarities of the population it serves may have led to results that do not readily apply to other populations in the United States.

Outcomes research tests the effectiveness of the medical service provided by average physicians to all patients under ordinary circumstances [22]. The data generated through outcomes research reflect the achievements of common medical practice rather than those of centers of excellence, university hospitals, or dedicated specialists. Therefore, such data represent realistic predictors of outcome in a majority of patients. The randomized controlled trial, by contrast, provides information about the efficacy of medical procedures or interventions when they are applied under ideal conditions by skilled practitioners to a highly selected group of patients. The difference between efficacy and effectiveness represents the gap between the theoretical optimum of health care that could be achieved under ideal conditions and the general quality provided in real practice [23]. Our study describes the outcome of colorectal procedures in the veterans population. It provides a picture of how colorectal procedures were used and shows that an endoscopy with a flexible instrument provided protection that was similar in magnitude to that previously observed and that lasted for about 6 years. Our data extend the previous findings by showing that the beneficial effect of endoscopy with respect to cancer is not limited to reducing mortality due to cancer but also includes the prevention of cancer. Moreover, the beneficial effect of endoscopy applies to both the distal and proximal areas of the large bowel.