The Two-Tier Fecal Occult Blood Test: Cost-Effective Screening

The two-tier test represents a strategy combining HO Sensa and Hemeselect fecal occult blood tests (FOBTs) with the aim of greater specificity and consequent economic advantages. If patients register a positive result on any HO Sensa guaiac test, they are once again tested by a hemoglobin-specific Hemeselect test. This concept was applied to a multicentre study involving persons 40 years or older. One component of the study enrolled 573 high risk patients while the second arm recruited an additional 1301 patients (52% asymptomatic/48% symptomatic) stratified according to personal history and symptoms. The two-tier test produced fewer false positives than traditional tests in both groups evaluated in the study. In the high risk group, specificity (88.7% for two-tier versus 80.6% for Hemoccult and 69.5% for HO Sensa) was higher and false positive rates were lower (11.3% for two-tier versus 19.5% for Hemoccultand 30.5% for HO Sensa) for the two-tier test versus Hemoccult and HO Sensa FOBTs (95% CI for all colorectal cancers [CRCs] and polyps greater than 1 cm, α=0.05 ). No significant differences in sensitivity were observed between tests in the same group. Also, in the high risk group, benefits of the two-tier test outweighed the costs. Due to the small number of cancers and polyps in the second arm of the study, presentation of data is meant to be descriptive and representative of trends in a ‘normal’ population. Nevertheless, specificity of the two-tier test was higher (96.8% for two-tier versus 87.2% for Hemoccult and 69.5% for HO Sensa) and false positive rate lower (3.2% for two-tier versus 12.8% for Hemoccult and 22.3% for HO Sensa) than either the Hemoccult or HO Sensa FOBT (95% CI for all CRCs and polyps greater than 1 cm). This initial study, focusing on the cost-benefit relationship of increased specificity, represents a new way of economically evaluating existing FOBTs.


C ANCER OF THE COLON AND REC-
tum is one of the most common internal malignancies in western nations for men and women combined. In Canada, the lifetime probabi lity of developing colorectal cancer (CRC) and dying from it is approximately 6% and 3%, respectively (1). Figures are increased in persons with fami lial polyposis syndromes, a history of CRC in a first degree relative or a personal history of neoplasms (2).
Due to the high prevalence of CRC, economic interest in the benefits of screening has grown. Mathematical models outlining the cost-benefit relationship of various feca l occult blood tests (FOBTs) have attempted to quantify money spent per life-year or qua lity-adjusted life-year gained. Today, FOBTs must not only prove effective but cost-effective. This bas been raised by Mandel et al (3), who observed an "estimated benefit of a 33% reduction in mortality from colorectal cancer" after rehydrating Hemoccult test (Smith-Kline Diagnostics, California) (yearly screen over age 40), yet caution that the economics of a fourfo ld increase in positives must be addressed.
Three of the most common FOBTs are the Hemoccult guaiac test, HO Sensa guaiac test (SmithKline Diagnostics) and Hemeselect hemaglutination test (SmithKline Diagnostics). Problems with these tests include low specificity (guaiac-based tests) and high screening costs (Hemeselect). The two-tier test, first described by et d'en tirer des a vantages economiques. Si les patients obtiennent des resultats positifs au ga'iac, ils subissent alors l'epreuve Hemeselect, specifique a l'hemoglobine. Ce concept a ete applique a une etude multicentrique qui portait sur des personnes <le 40 ans ct plus. Le premier des bras de l'etude comportait 57  Cleator ( 4) in 1990, represents an attempt to increase specificity in screening (while maintaining sensitivity) and thus reduce costs associated with false positives. C learly, these two aims must be addressed by any FOBT if widespread acceptance is to be won in our costcon cious society.

SUBJECTS AND METHODS
Multiple sires (St Paul's Hospital, Vancouver, British Columbia, G eorge Washington University , W ashington DC, and several primary care site in Califomia, Florida and Penn ylvania) were part of this multicentre study involving gastroente rologists and primary care physicians. In the first arm of the study a high risk group (n=573 ) comprised patients (59% male, 41 % female) 40 yea rs of age or o lder, mean age 64.9 (range 40 to 90) . ln each of these cases, patients were referred to gastroenterologists fo r large bowel work-u p for one or more of the fo llowing reaons: prev ious po 1ttve tool tests; symptomatic case; personal history of CRC, polyps or inflammatory bowel disease; family history of colorectal neo-plasia or polyps; and iron deficiency anemia. O nly patients who completed all FOBTs were included in this report.
All patients, regardless of the FOBT results, underwent a complete large bowel work-up. All invas ive procedures were performed after the testing and results were correlated to FOBT results.
Of the 1301 patients (59% male, 41 % fe male) involved in the second arm of the study, all were 40 years of age or older, mean age 63. 1 (range 40 to 92). Fifty-two per cent were asy mptomatic without a history of gas tro intestinal disorders while 48% displayed either minor symptoms or expressed some personal or fa mily history of gas tro intestinal disorders.
Further description of these 130 1 patients is as fo llows: 678 asymptomatics with no personal history of gastro intestinal disease or family history of CRC or polyps; 23 4 patients with either a personal history of one or more of the fo llowing: hemorrhoids, anal fissures, upper gastro intestinal disease, diverticular disease, miscellaneous lower gastro intestinal disease, or nonspecific symptoms ( with re pect to CRC) such as CAN J GASTROENTEROL VOL 8 No 6 N OVEMBER 1994 diarrhea, nausea, bloating, fa tigue or incontinence; and 389 patients exhibiting personal history of nongastrointestinal cancer, or specific symptom · ( with respect to colon cancer) uch as visibl e blood in stool, melena, weight loss, abdominal pa in , anemia, rectal pa in, change in stool calibre or change in bowel habi ts, or asy mptomatic patients wi th fa mil y history of CRC or po lyps, or personal history of ulcerati ve colitis, C rohn 's disease or other inflammatory bowel disease, or personal hitory of CRC, adenomas or hyperplas tic polyps.
In this second arm of the tud y, patients were visiting a physician for routine physica l examination or for surve illance fo llow-up after personal history of CRC, polyps or inflammatory bowel disease. Patients may have had a personal or fa mily history of colorectal neoplas ia. O nly patients who perfo rmed all the FORTs fo r three days were included in this report. It is important to note that not all patients in this arm of the study can be considered average risk patients. In this pa rt of the study, patients who tested positive in any of the FOBTs were asked to unde rgo diagnostic fo llow-up. A complete fo llow-up consisted of full colonoscopy, although a combination of flex ible sigmoicloscopy and double air contrast barium enema was also considered adequate fo llow-up.
FOBTs studied were the Hemoccult, HO Sensa, Hemeselect and two-tier tests. The Hemoccult test i guaiacbased and detects ox idizing agents in feca l samples. It is simple and econ omica l but may register fa lse positives and fa lse negatives when meat and vitamin C , respecti vely, arc part of the subject's diet. HO Sensa is a more sensitive guaiac-based test than the Hcmoccul t test. It is simple to perform but less economical than the Hemoccult test. Hemese lect's advantages include fewe r fa lse negatives and positive due to the fact that it detects human hemoglobin in feca l matter. Unfo rtunately it i much more costl y -approximately seven times more -than Hemoccult or HO Sensa and requires laboratory analysis. The two-tier test, as previously observed, confirms po itives CRC bleeding, gastrointestinal disorders and asymptomatic cases. In the second arm of the study, findings were expressed in terms of an 'all risk' group (n= 1301) and asymptomatic subgroup (n=678) for patients recorded as asymptomatic or possessing no previou history of gastrointestinal disease or family history of CRC. Certain assumptions were made in the cost-benefit analysis of the two-tier test within a Canadian medical system. Cost of colono copy was set at $300, co t to repair perforations due to colonoscopy was estimated to be $6,000 ( including surgery and seven-to 10-day hospital stay), and the commonly accepted perforation rate of 1/1000 was assumed.

Statistics: Confidence intervals of95%
were set ( a=0.05) when comparing sensitivities, specificitie and pred ictive values, as well as false positive and false negative rates, of various FOBTs. Confidence limit for proportion were based on F distribution and binomial distribution. disease from HO Sen a tests with the Hemeselect test. This affords a high specificity without the significant cost of performing a Hemeselect test on every subject.
Ethical approval was obtained from respective study sites. Study participants were given sample kits containing three envelopes for FOBTs. Each envelope consisted of one Hemoccult test card, one HO Sensa test card, one Hemeselect te t card, a collection saddle, wooden applicator sticks, patient instructions and a return envelope. Fecal samples were then collected and appl ied to all sample cards by individuals over three consecutive bowel movements. Persons were given additional instructions by a nurse or physician's assistant. Individuals were asked to abide by a restricted diet during and three days before testing. This included no consumption of red meat, fruits and vegetable (including turnips, melons, radishe , horseradish, broccoli and cauliflower) and was done to avoid inaccurate testing as a result of red meat, high peroxidase vegetables, nonstcroidal anti-inflammatory drugs and vitamin C 364 (more than 250 mg/day). Females were also asked not to sample stool during menstruation and for the following three days. All cards were then analyzed by technicians who were blinded to physician's observations and, in the case of the two-tier test, results of HO Sensa when analyzing Hemeselect tests. Rehydration did not take place.
A two-tier approach was conducted when any one of the HO Sensa guaiac te ts registered a positive result. At that point, Hemeselect results from the same patient were examined. The relationship of these two tests was then assessed for sensitivity, specificity, predictive value and estimated costs for screening where possible.
As earlier mentioned, all patients in the high risk group underwent a complete bowel work-up following FOBT administration, while all patients registering positive on FOBTs in the second arm of the study were similarly examined.
Clinical findings were recorded for number (and stage) of CRCs, adenomatous and nonadcnomatous polyps, non-

RESULTS
As previously mentioned, the data stem from tudies that involved both gastroenterologists and primary care physicians at multiple sites. Stratification of the data based on risk code was performed. The first component (n=573) of the study comprised a high risk group, and the second component (n=l301) wa made up of asymptomatic and symptomatic subgroups reflecting the make-up of primary care practices involved in the study. An asymptomatic subgroup (n=678) from the econd component of the investigation wa studied in an identical fashion to the larger aims of the study to estimate costs associated with a normal screening population.
In the high risk group (n=573) 23 patients were determined to have CRC upon complete large bowel work-up (Table 1 ) A value of greater than 1.0 cm for polyps was selected for statistical compari on of tests (the rationale for this designation will be addres ed in the 'Discussion'). T he Hemoccu lt, HO Sensa, Hemese lect and two-tier FOBTs registered 79, 120, 59 and 46 fa lse positives, respectively, for this arm of the study. Many of these fa lse positives were as ociated with hemorrhoids or fissures, diverticular disease, C rohn's disease, inflammatory bowe l disease, ulcerative colitis, nonadenomatous polyps, and misce llaneous upper and lower gastrointestinal problem . The two-tier test produced fewer false positives than traditional tests in both groups evaluated in the study. In the high risk group, specificity (88. 7% for two-tier versus 80.6% for Hemoccu lt and 69.5% for HO Sensa) was significantly higher and fa lse pos itive rates were significantly lower (11.3% for two-tier versus 19.5% for Hemoccult and 30.5% for HO Sensa) fo r the twotier test ver us Hemoccu lt and HO Sensa FOBTs (95% Cl for all CRCs and polyps greater than 1 cm, a=0.05). No statistically significant differences in ensitivity were observed among tests in the ame group (Hemoccu lt 70.7%, HO Sensa 75.8%, two-tier 62%; a=0.5).
In the high risk group, the benefits of the two-tier test markedly outweighed the costs, even when one less detected cancer ( two-tier versus HO Sensa) was figured into calculations. In the high risk group the cost per detected cancer was $2,361, $3, 176 and $2,485 fo r Hemoccult, HO Sensa and Hemeselect tests, respectively ( Table  2). This is in contrast to the $1,842 for the two-tier FOBT. Even when determining cost per detected cancer as well as a<lenomas and polyps greater than 1 cm, the two-tier test remains the most cost-effective at $972.
Due to the small number of cancer and polyps in the second arm of the study, presentation of data is meant to be de criptive and representative of trends in a 'normal' popu lation. N evertheless, specificity of the two-tier test wa significantly higher (96.8% for two-tier versus 87.2% for Hemoccu lt and 69.5% for HO Sensa) and false Two-tier specificity and economic advantage  In the second arm (n=l301), benefits outweighed costs, and are perhaps more reflective of savings associated with screening a norma l popu lation ( though the limited number of cancer detected restricts what can be said of the data). Because HO Sensa detected three cancers versus two in all other FOBTs, total cost of detection is divided by three (for HO Sensa). As a result cost per detected cancer is $25,178 for Hemoccult, $27,165 for HO Sensa, $20,948 for Hemeselect and $10,825 for two-tier test ( incomplete findings were not calculated into cost per detected cancer) ( Table 3).
In the second arm (n=l301), fa lse positive rates were of great interest, particu larly because they were more reflective of a somewhat 'normal' population (Tables4,5). Diverticu lardisease, hemorrhoids and fissures, misce llaneous gastrointestinal problems, anemia and nonadenomatous po lyps resu lted in non-CRC bleeding in 108 Hemoccu lt, 173 HO Sensa, 44 Hemeselect and 28 two-tier tests. Additionally, confirmed or assumed confirmed asymptomatic fa lse positives added 32, 61, nine and four additional false positives for Hemoccu lt, HO Sensa, Hemese lect and the two-tier test, respectively. Additionally, in this arm of the study, three cancers were detected by HO Sensa and two by all other tests. Sensitivity (for all CRCs and polyps greater than 1 cm) figures did not differ significantly (at 95% Cl, a=0.05). In the asymptomatic ubgroup (n=678) from the second component of the study one CRC was detected by all FOBTs except Hemoccult. What the present authors were most interested   in, however, were the marked differences in false po itives. The number of false po itives chat were presumably related to diverticular disease, hemorrhoids and fissures, and nonadenomatous polyps varied substantially between tests. Hemoccult registered 17 false positives, HO Sensa 24, Hemeselect five and the two-tier test one. In addition to these results, false positives recorded in nonbleeding subjects made up an additional 26, 49, six and two case for Hemoccult, HO Sensa, Hemeselect and two-tier test, re pectively.

DISCUSSION
As previously noted, the costs and resu ltant benefits of CRC screening have been widely discussed in recent years. Eddy (5,6) ha initiated much of this study, proposing his own mathematical model. This research has dealt with cost-effectiveness of screening both high risk patients and an average risk population. Other contributors have included Walker and Whynes (7) who assessed 13 screening strategies and made their own economic evaluations of each filter type. They accurately note that though "cost-effective screening protocols following the administration of one or ocher of the FOB tests" have been researched, "little attention has thus far been paid to the selection of the most appropriate filter from among the range available" (7). Trehu and Cooper (8) have also contributed to our know ledge, prov iding a very recent estimate of $15,233 per case of colon cancer diagnosed through mass screening in the United States. Additionally, The Office of Technology A sessment (OTA) in the United Scates has a urned a lifetime price tag of $1.5 billion to screen those 2.1 mi Ilion individuals who are 65 years and older, resulting in 45,000 years being added to these person's lives (9). ln this assessment, costs per year of life added come very close to screening procedures such as mammography (9).
In summary, Eddy and the OTA have done the most comprehensive work in this field. Byers and Gorsky (10) recognize this in their article which reviews the contributions of both these groups. By the ir estimates the range of net cost per day of life gaine<l is between US$57 (Eddy) and US$1 18 (OTA).
What we fee l is missing in these assessments is a closer look at the economics of each FOBT, as Walker and Whynes uggest. It is at this stage that we present the reduction in fa lse positives associated with the two-tier test, which carries with it an economic advantage over individual Hemoccult, HO Sensa and Hemeselect tests. We argue that such economic benefits are found because of the innate specificity of HO Sen a and Hemeselect used in combination. Goosenberg ( 11) hints at this in a recent article, concluding that "an ideal test should combine the specificity of an immunologic test with the sensitivity of a guaiac test".
Because the two-tier approach, upon administration fo llowed by complete large bowel work-up, affords detection of a similar number of CRCs while reducing the number of fa lse positives (Table 1 ), we note that it thus po sesses a relationship in which benefits outweigh costs.
In our high risk group the cost per detected cancer was $2,361, $3,176 and $2,485 for Hemoccult, HO Sensa, and Hemeselect tests, respectively (Table  2), which i in contrast to the $1,842 for the two-tier FOBT. Even when determining cost per detected cancer a well as adenomas and polyps greater than 1 cm, the two-tier test remain the most cost-effective at $972.
In the second arm (n=l301), benefits outweigh costs and are perhaps more reflective of savings associated with screening a normal population (though the limited number of cancers detected restricts what can be said of the data). Due to the fact that HO ensa detected three cancer versu two in all other FOBTs, total cost of detection is divided by three. As a result cost per detected cancer is $25,178 for Hemoccult, $27, 165 for HO Sensa, $20,948 for Hemeselect and $10,825 for two-tier test (incomplete findings were not calculated into cost per detected cancer) (Table 3). Clearly, an apparent 'missed' cancer raises questions regard ing the two-tier test, which we answer later.
Examination of the a ymptomatic Cost of tests of larger polyps as other tests must be noted. C learly, this issue is most dependent on personal philosophies concerning the importance of detection and removal of polyps. We regard the detection of polyps less than 1.0 cm by the FOBT as minor in value and unnecessary to remove in the context of a popu lation receiving annual FOBTs.

CONCLUSIONS
We offer the two-tier test as a more cost-effective FOBT due to a reduction of false po itives. Potential 'missed' cancers, litigation issues and ability to detect polyps represent possible difficulties with this FOBT. Additionally, numbers in our study are relatively small and will require validation of these trends in future investigations. Kaiser Permanente (California) is conducting a much larger evaluation of the two-tier test and initial results from a screen of over 8000 patients are promising. Allison and colleagues ( 12) suggest that the two-tier actually detects more CRCs than Hemoccult II while limiting the number of follow-up investigations compared with HO Sensa.
Despite the limitations mentioned, we argue that fewer false positives and