At least two principal reasons exist for the difficulties these trials have had in showing efficacy, if indeed such efficacy exists. First, the sensitivity of FOBTs in asymptomatic patients is probably no better than 50% for cancer [10, 11] and much lower for adenomas [12, 13]. The efficacy of a screening test cannot be higher than its sensitivity. No benefit is gained by those patients whose cancers are missed, and it is unlikely that earlier detection will result in cure for all of those whose cancers are detected by screening. In view of this limitation, others [14-16] have speculated that the true efficacy of the FOBT as a screening tool for lowering mortality due to colorectal cancer in those screened is probably no greater than 10% to 30%.

A second problem with these trials is the substantial noncompliance with screening recommendations (particularly with follow-up screening) noted in the intervention groups. If screening does save lives, the noncompliance will diminish observed differences in mortality between intervention and control groups. "Crossover" screening in the control group, if it occurs, further reduces these differences. The combination of noncompliance and relatively low efficacy causes the mortality differences achievable in the trials to be small and therefore difficult to detect.

Nonrandomized, observational studies can provide direct estimates of the efficacy of screening and early treatment that are uninfluenced by noncompliance because persons are classified on the basis of actual exposure to screening rather than on the basis of "intention to screen." Case-control methods have been proposed for evaluating screening and other preventive practices [17, 18] and have been used to measure the benefit of several such practices, including sigmoidoscopy screening [19-22].

The price paid when randomization is abandoned, however, is the probable introduction of factors that may differ between screened and unscreened persons. If persons who comply with the FOBT differ from noncompliant persons with respect to risk for cancer mortality, then these differences will bias estimates of efficacy from observational studies. This report presents results of a case–control (nonrandomized) study of the efficacy of FOBT screening among members of the Kaiser Permanente Medical Care Program (KPMCP) of Northern California in which we sought to assess and control for several confounding factors.

Methods

Top Methods Results Discussion Author & Article Info References

Setting

Periodic FOBT screening with Hemoccult II slides (SmithKline Diagnostics, Inc., San Jose, California) for persons 50 years of age and older has been encouraged at most KPMCP facilities since 1979-1980. The KPMCP provides comprehensive inpatient and outpatient medical care to all members with no additional charges for periodic health checkups or laboratory tests such as the FOBT. For these reasons, it is unlikely that many members would obtain screening with the FOBT outside the health plan.

Identification of Cases

Potential cases included all KPMCP members diagnosed with adenocarcinoma of the colon or rectum between 1981 and 1987 who subsequently died before December 1988 as a result of this cancer. Persons 50 years of age or younger at diagnosis were excluded, because few, if any, would have been considered eligible for screening in the years before diagnosis. Patients who were members of the program for less than 1 year prior to diagnosis were also excluded.

Cases were identified from files provided to us annually by the Bay Area Resource for Cancer Control (the SEER registry for the San Francisco Bay Area). This registry reviews all cancer diagnoses from eight KPMCP facilities in the five counties of the Bay Area. Death was ascertained either directly from registry information or by automated linkage of remaining cases to California state death certificates (CAMLIS) [23].

Two chart reviewers reviewed a total of 979 potential cases. The first reviewer verified the histologic diagnosis of adenocarcinoma, its anatomic location, and the cause of the patient's death (using a written protocol and physician adjudication when needed) and recorded symptom status and mode of diagnosis. A total of 489 patients met the case definition. Nearly all the remaining cases were excluded, because they died of causes unrelated to their colorectal cancer. Three additional cases with a history of ulcerative colitis were excluded because FOBT is not recommended in the presence of this condition. A total of 486 cases remained eligible for our study.

Identification of Controls

One control participant was randomly selected for most cases from KPMCP membership lists at the same eight facilities covered by the tumor registry. Controls were matched with cases for age (± 1 year), sex, and date of health plan entry (± 1 year). Controls were required to have been living at the time the case patient died as shown by continued KPMCP membership. A previous history of adenomatous polyps or nonfatal colorectal cancer was not a grounds for exclusion of either case patients or controls. Among the 486 case patients, 96 were also eligible for a separate case–control study of the efficacy of screening sigmoidoscopy [22] by virtue of the distal location of their fatal lesion and the facility at which they were diagnosed. Four controls per case patient had been sought for that study. Data from all controls reviewed are included for these 96 case patients. Thus, among the final 486 case–control sets, 390 had a single control, 22 had two controls, 19 had three, 39 had four, and 16 had five controls, creating a total of 727 controls.

Documentation of the FOBT, Other Screening Tests, and Potential Confounding Variables

After verifying case or control eligibility, the first reviewer masked all records at the "index date," a date immediately before onset of the patient's symptoms or before the screening test that led to diagnosis. The second reviewer, who was blinded to the participant's case–control status, then reviewed each participant's outpatient medical records for the 10 years before the index date recording all instances of the FOBT, sigmoidoscopy, digital rectal examination, barium enema examination, and colonoscopy. Tests that led to the diagnosis of case patients were recorded by the first reviewer and included in all case–control comparisons of screening frequency. The FOBTs were characterized as "laboratory" FOBTs if laboratory reports of results were found, or as "office" FOBTs if the clinician reported doing the test after a rectal examination in the office. More than 50% of the laboratory FOBTs consisted of three specimens, which were counted as a single FOBT in our analyses. Dates, indication, and findings for each test were also recorded.

The indication for each FOBT was judged by examination of clinician notes. Tests were categorized according to whether they were done for screening, for possibly symptomatic patients, for patients with clear-cut symptoms, or for follow-up of another positive test result. Less than 1% of all tests fell into the ambiguous "possibly symptomatic" category. Double review of 144 records for quality control showed agreement on indication for 92% of 139 FOBT procedures. In all analyses reported here, only those tests judged to be done for screening are considered.

Findings were recorded as either a negative result; a positive result with subsequent discovery of cancer, an adenomatous polyp, or a hyperplastic polyp; a positive result with an adequate negative workup (a negative colonoscopic test result, a negative barium enema result, or a sigmoidoscopy that showed a non-neoplastic source of bleeding); or a positive result with an inadequate workup (none of the three adequate negative workups). Colonoscopy was the most frequent workup ordered for follow-up during these years.

Several possible confounders were also recorded. These confounders included a history of adenomatous polyps or colorectal cancer prior to the 10-year interval or a family history of colorectal cancer noted before the case patient's diagnosis. These factors could lead to both increased screening and an increased risk for colorectal cancer and thereby decrease or obscure a true protective effect. The number of periodic health checkups taken in the 10-year interval was recorded to be used as a marker for "lifestyle" factors that may be associated with a tendency to be screened and, independently, with a lower risk for mortality from colorectal cancer. The multiphasic health checkup (which accounted for a sizable number of periodic checkups in this study) has previously been shown to be inversely related to all-cause mortality among KPMCP members [24].

Analyses

Preliminary analyses confirmed that FOBT screening (in either the office or laboratory setting) was rarely done before 1979. We therefore considered only the 5-year interval before the index date for both case patients and their matched controls, because earlier years would have been before 1979 for nearly all study participants. For fatal cases of colorectal cancer detected by the screening FOBT, the FOBT that led to diagnosis was included with other screening FOBTs in all analyses. If a case patient or any control had less than the full 5 years of historical data, only those years of historical data available for all members of the case–control set were considered.

Conditional logistic regression models for matched sets were used to estimate odds ratios and to adjust analyses for potential confounders. Odds ratios were calculated for previous screening within 1, 2, and 5 years of case diagnosis compared with no screening examination in that interval. Analyses were adjusted for FOBT screening examinations done before the interval being studied to control for the possibility of "healthy screenee bias" [25]. For any screening test that leads to early detection, it is less likely that a case patient would have subsequent tests, because the earlier tests would have had some chance of detecting the cancer, thus making further screening impossible. Even if screening with the test is not efficacious, this bias leads to an inverse relation between the number of screening examinations and the likelihood of developing cancer. An adjustment for previous screening examinations controls for this bias. In addition, analyses were adjusted for the number of screening sigmoidoscopies and rectal examinations, for the number of periodic health checkups, and for a personal or family history of colorectal cancer. Interaction terms pertaining to gender, age at diagnosis (< 65 years compared with 65 years of age), and anatomic location (left-sided, transverse, and right-sided) were tested for statistical significance.

To address the question of an optimal screening interval, odds ratios were calculated for having one's most recent screening FOBT in each of the 5 years before case diagnosis (compared with no screening examination in that year or later). Separate models were used to assess each year, and persons having a more recent screening test were removed from the analysis. Nearly all these tests had negative results. Thus, these odds ratios provide estimates of the risk for developing fatal colorectal cancer 1 to 5 years after a negative screening test.

Results

Top Methods Results Discussion Author & Article Info References

At least 5 years of historical data were available for 390 of the 486 case–control sets (79%) and more than 2 years of data were available for all but 26 sets. Case patients and controls were matched on gender and age. The two groups were similar in terms of race and proportion with a family history of colorectal cancer (Table 1). Case patients were somewhat more likely than controls to have had a personal history of either colorectal cancer or polyps before the 5-year retrospective examination period.

The anatomic distribution of the fatal cancers included 22.7% originating in the rectum, 39.2% in the left colon, 12.3% in the transverse colon, and 25.2% in the ascending colon. Of these cancers, 98% had penetrated the bowel wall or had spread to regional lymph nodes or beyond at time of diagnosis (stage B2 or worse).

Case patients had been screened less frequently than controls by each colorectal cancer screening test and had received, on average, fewer periodic health checkups in the 10 years before diagnosis (Table 2). The number of screening FOBTs, both laboratory and office, were correlated with the numbers of screening sigmoidoscopies (r = 0.37 and 0.13 for laboratory and office FOBTs, respectively), digital rectal examination (r = 0.40 for both laboratory and office tests), and number of periodic health checkups (r = 0.49 and 0.33 for laboratory and office tests, respectively). Thus, these variables may confound case–control comparisons of the FOBT.

Relative to unscreened persons, the risk for fatal colorectal cancer was reduced by 10% to 30% among those who had an FOBT (Table 3). The magnitude of the reduction was slightly greater for laboratory than for office FOBTs and was not substantially altered by adjustment for potential confounders, including number of health checkups, screening sigmoidoscopies and rectal examinations, and a personal history of colorectal cancer or polyps. Most of the 5-year deficit in screening among case patients accrued during the 2 years immediately before case diagnosis.

The negative overall (5-year) association did not differ significantly by gender (odds ratios, 0.74 and 0.75 for men and women, respectively), by age at diagnosis (odds ratios, 0.66 and 0.75 for persons < 65 years of age and those 65 years of age, respectively), or by anatomic location (odds ratios, 0.74 for cancers of the distal sigmoid and rectum, 0.89 for cancers of the transverse and descending colon, and 0.56 for lesions of the ascending colon).

Risk increased gradually toward 1.0 after a patient's most recent screening FOBT (Table 4) as the interval between the test and the diagnosis lengthened. No suggestion of a reduced risk persisted at 3 years. A history of screening in the fifth year before diagnosis appeared to be associated with a sharply reduced risk. Screening frequency was low, however, in both controls and case patients in this interval because, for many persons, this period occurred before the introduction of FOBT screening. This apparently protective odds ratio probably results from the instability of estimates based on small numbers, as does the apparent increase in risk associated with screening seen in the sixth year.

Among the 486 case patients, 45 had a screening laboratory FOBT recorded in the 12 months before their fatal cancer was diagnosed. In 16 (36%), the result was positive, and in 10 of these 16, the FOBT led directly to the diagnosis. In four, the subsequent workup was negative, and in two the workup was judged to be inadequate. A total of 27 persons had screening office FOBTs in the same 12-month period; of these, 10 (37%) yielded a positive result. Three of these tests led to case diagnosis. An additional seven had positive results with either a negative workup (five patients) or an inadequate workup (two patients).

Among the controls, a total of 358 screening laboratory FOBTs were done in 216 persons during the 5 years before case diagnosis. Of these, 11 tests (3.1%) yielded positive results. Workups were negative for 10 of these positive results, and the eleventh yielded a 12-mm adenomatous polyp.

Discussion

Top Methods Results Discussion Author & Article Info References

These case–control data suggest an approximate 25% reduction in mortality due to colorectal cancer for persons who have a screening FOBT every 1 to 2 years. The confidence intervals around this estimate, however, were wide. The observed association could conceivably be due to chance, particularly because we tested a number of hypotheses related to differing intervals and types of FOBTs. Aside from chance, associations of this modest magnitude might also be explained by residual confounding factors that led persons with a lower risk for colorectal cancer mortality to have screening FOBTs more frequently. When we adjusted for measured confounders, odds ratio estimates generally increased slightly toward 1.0. Additional unmeasured factors may explain at least a portion of the remaining negative association.

Given these cautions, however, other aspects of the data appear to be more consistent with true efficacy. First, as stated previously, the mortality reduction expected from FOBT screening is small and of approximately the magnitude we observed. This expectation is based on the known performance characteristics of FOBT, most notably its sensitivity, which may be as low as 30% to 50% [10, 11]. In our data, only 36% of the FOBTs done within 1 year before diagnosis of fatal colorectal cancer yielded a positive result. Efficacy can be no better than this figure.

Second, the analysis of most recent screening examination (see Table 4) shows that case deficits in the FOBT occurred primarily during the 2 years immediately before diagnosis. For tests done more than 2 years earlier, little evidence of a similar deficit (that is, little evidence of reduced risk), was noted. If confounding factors explained the deficit of screening among case patients, no reason exists to expect the deficit to be confined to the years closest to case diagnosis. The observation that little or no reduction in risk exists more than 2 years after a negative FOBT result is consistent with current recommendations that screening be repeated at least biennially [1-3].

In addition to poor sensitivity, the low yield from workups of positive test results has also been cited as a drawback of FOBT screening [16, 26, 27]. False-positive results were seen frequently among controls in our study. Positive results (nearly all of which led to negative workups) were seen for 3.1% of all examinations, in agreement with positivity rates of 1.5% to 6% found in the randomized trials.

Obviously, screening would not be efficacious without appropriate workup and treatment (if indicated) of those persons with positive test results. Inadequate workup of positive FOBT results would diminish the apparent benefit of FOBT screening in our study. A total of eight case patients had an incomplete workup of a positive FOBT result (five in the office and three in the laboratory) during the 2 years before their diagnoses. Reasons for the incomplete workup varied. In two cases, other sources for the bleeding were thought to have been identified (hemorrhoids, diverticular disease). In two cases, positive office FOBT results were followed by negative laboratory FOBT results, and the workup was terminated. When these eight patients were removed and the data reanalyzed, results did not change substantially. For example, the odds ratio for exposure to the FOBT within 2 years declined slightly from 0.76 to 0.73 (CI, 0.53 to 1.01).

A 25% to 30% reduction in mortality due to colorectal cancer, as suggested by our data, would amount to a substantial saving of lives if this difference represents true efficacy. Given the relatively small size of the observed effect, however, the possibility exists that residual confounding could explain most, if not all, of the association. Randomized, controlled trials are best suited to removing this concern. Although efficacy may continue to prove difficult to detect in the ongoing trials of the FOBT, we believe it is important to await these results before recommending widespread screening using currently available FOBT methods.

Abbreviation

FOBT: Fecal occult blood test