CLIN

ICA

LA

T

GASTROENTEROLOGY 2013;144:918–925

A Population-based Comparison of Immunochemical Fecal Occult BloodTests for Colorectal Cancer ScreeningTHIBAUT RAGINEL,1,2,3,4 JOSETTE PUVINEL,5 OLIVIER FERRAND,6 VERONIQUE BOUVIER,1,2,3,4 ROMUALD LEVILLAIN,7

ANGELA RUIZ,7 OLIVIER LANTIERI,7 GUY LAUNOY,1,2,3,4 and LYDIA GUITTET1,2,3,4

1Centre Hospitalier Universitaire (CHU) de Caen, Caen; 2Normandie University, Caen; 3Université de Caen Basse-Normandie (UCBN), Cancers and Preventions,4INSERM U1086, Caen; 5Association Bourbonnaise Interdépartementale de Dépistage des Cancers (ABIDEC), Moulins; 6Action de Dépistage Organisé des

Caen;

Cancers 18 (ADOC18), Saint-Doulchard; and 7Institut InterRégional pour la Santé (IRSA) de Tours, La Riche, France

cth

filSfe

See related article, Anderson AE et al, on page 526in CGH.

BACKGROUND & AIMS: Quantitative fecal immuno-chemical tests (FITs) identify individuals with colorectalcancer with greater levels of accuracy than guaiac tests. Wecompared the performances of 2 FITs in a populationundergoing screening for colorectal cancer. METHODS:We collected fecal samples from 19,797 individuals inFrance (age, 50�74 y) who participated in a colorectalcancer screening program, from June 2009 through May2011. Samples were analyzed using the Magstream (Fu-jirebio Inc, Tokyo, Japan) and OC Sensor (Eiken ChemicalCo, Tokyo, Japan) (2 samples each) FITs, as well as theHemoccult II guaiac test (SKD, Villepinte, France) (3 sam-ples each). Colonoscopies were performed for patientswith positive results from all 3 tests. The cut-off values forlevels of hemoglobin in buffer and stools were 55 ng/mLand 180 �g/g for the Magstream and 150 ng/mL and 30�g/g for the OC Sensor, respectively. Results from theFITs were compared with those from the guaiac test forcut-off values for stool samples, positivity rates, and thereceiver operating characteristic curve values. The num-bers needed to screen and the numbers needed to scope todetect an advanced neoplasia (cancer, adenoma �10 mm,or high-grade dysplasia) were calculated. RESULTS: Apositive test result was found in 1224 participants (6.2%);1075 (87.8%) underwent a colonoscopy examination. Ofthese, 334 were found to have advanced neoplasia. Con-sidering the cut-off values associated with the positivityrate of Hemoccult II (1.6%), the numbers needed to screenwere 239 for Hemoccult II, 166 for a 1-sample MagstreamFIT, and 129 for a 1-sample OC Sensor FIT; the numbersneeded to scope were 3.3, 2.3, and 1.8, respectively. For thesame false-positive rate as Hemoccult II (0.98%), the true-positive rates for Magstream and OC Sensor FITs were0.65% and 0.90% respectively, compared with 0.42% forHemoccult II. The OC Sensor FIT had a greater area underthe receiver operating characteristic curve value than theMagstream FIT. CONCLUSIONS: Based on results froma large, population-based study, the OC Sensor FIT iden-

tifies patients with colorectal cancer with greater accu-

racy than the Magstream FIT. ClinicalTrials.gov num-ber: NCT01251666.

Keywords: Colorectal Neoplasm; Early Detection; Sensitiv-ity; Specificity.

With roughly 1 million cases and 500,000 deathseach year worldwide, colorectal cancer is a major

public health concern, especially in industrialized coun-tries.1 The natural history of this cancer allows for earlydetection. Until now, 2 strategies have been shown toreduce the mortality associated with this cancer signifi-cantly: the regular guaiac fecal occult blood test2 andonce-only flexible sigmoidoscopy.3–5

Since the beginning of the 1990s, it has been knownthat immunochemical methods allow for improvement ofsensitivity of fecal occult blood tests.6 Automated quan-titative fecal immunochemical tests (FITs) make adapta-tion of the cut-off value possible to limit the loss inspecificity. Two randomized controlled trials, conductedin The Netherlands on roughly 10,000 average-risk indi-viduals each, found that the OC Sensor FIT (Eiken Chem-ical Co, Tokyo, Japan) had both better screening perfor-mances and higher acceptability than the guaiacHemoccult II test (SKD, Villepinte, France).7,8 One studyonducted in France on roughly 32,000 subjects foundhat the Magstream FIT (Fujirebio Inc, Tokyo, Japan) alsoad a better diagnostic accuracy than Hemoccult II.9,10

For both Magstream and OC Sensor FITs, it has beenshown that a gain in both sensitivity and specificity ispossible compared with the guaiac test Hemoccult II,providing an adequate cut-off value and using a smallnumber of samples.9,11–13 These results recently were con-

rmed in France.14 An indirect comparison based on aiterature review was in favor of slight superiority of OCensor over Magstream.15 However, this analysis sufferedrom limitations inherent to comparison between differ-nt populations.

Abbreviations used in this paper: FIT, fecal immunochemical test;FOBT, fecal occult blood test; FPR, false-positive rate; NNScope, num-ber needed to scope; NNScreen, number needed to screen; ROC, re-ceiver operating characteristic; TPR, true-positive rate.

© 2013 by the AGA Institute0016-5085/$36.00

http://dx.doi.org/10.1053/j.gastro.2013.01.042

ta

MtseFati

tFI

abssfspahnnwstbn

Ma

CLI

NIC

AL

AT

May 2013 COMPARISON OF PERFORMANCES OF 2 FITS 919

The aim of this study was therefore to comparedirectly the screening performances of Magstream andOC Sensor FITs in the targeted average-risk population,taking into account the cut-off value and the numberof samples.

Materials and MethodsStudy Population and DesignBetween June 2009 and May 2011, there were 19,797

residents of 2 French counties (Allier and Cher) who fulfilledinclusion criteria for the ongoing biennial organized screeningprogram (aged 50 –74 y, no digestive symptoms or personal orfirst-degree relatives with a history of colorectal cancer or ad-vanced adenoma, no colonoscopy in the past 5 years) who wereenrolled in a study comparing Magstream and OC Sensor FITs.Participation in the study was suggested at the time of central-ized invitation. The general practitioners consulted by the pa-tients checked the inclusion criteria and provided the guaiacHemoccult II test (SKD, France) (routine test) together as Mag-stream and OC Sensor FITs to all subjects agreeing to join thestudy. All participants signed a written informed consent.

For the guaiac test, 2 samples per stool on 3 consecutivebowel movements were required. For each of the FITs, 1 sampleper test on 2 consecutive bowel movements was required. Onlythe guaiac test was performed on the first stool. Then, for eachof the last 2 stools sampled, subjects were asked to perform all3 tests on the same stool, but no order between tests wasspecified. In Allier county, in accordance with the ongoing pro-gram, a specific diet (meatless for 3 days) was recommended.Subjects performed the 3 tests themselves at home (on the kitcard for Hemoccult II, in OC-Auto sampling bottle for OCSensor and in New Hemtube for Magstream), recorded the dateof each bowel movement involved, and sent the 3 tests by mailto the centralized Institut InterRégional pour la Santé labora-tory. The Hemoccult II reading was performed by trained staffwithout rehydration as planned in the national organizedscreening program, and blinded to the results of the FITs.Magstream samples were analyzed using the Magstream HTinstrument, and OC Sensor samples were analyzed using the OCSensor Diana instrument. As recommended, crude results ofMagstream and OC Sensor FITs were standardized by conver-sion into concentration of hemoglobin in the stools (see theSupplementary Appendix for conversion details derived fromprevious analytic studies).16 –18

Screening was considered positive if the Hemoccult II test waspositive (at least 1 of the 6 slots was positive), or at least 1 of the2 samples of OC Sensor reported a concentration of hemoglobinin the stool greater than 30 �g/g (eg, 150 ng/mL hemoglobin inhe buffer), or at least 1 of the 2 samples of Magstream reported

concentration of hemoglobin in the stool greater than 180�g/g (eg, 55 ng/mL in the buffer). The study cut-off value for

agstream was the cut-off value that led, in a former study, tohe same number of false-positive results as Hemoccult II, acenario compatible with financial resources and availability ofndoscopists characterizing health care systems such as therench one.9 A similar scenario was targeted for OC Sensor,dopting a cut-off value near the manufacturer’s one, limitinghe number of positive results for a 2-sample strategy but allow-ng comparison with other studies.7,12,19,20

Because Hemoccult II was the routine test in France, if it wasnegative and both FITs were either negative or unable to be

analyzed, the screening was considered to be negative. In addi-

ion, if Hemoccult II was not analyzed or not processed and noIT was positive, the subject was asked to repeat the HemoccultI only, without performing any FIT again.

The global screening test result was sent to both the patientnd general practitioner, without details on each fecal occultlood test (FOBT) result. If the screening was positive, theubject was encouraged to undergo a colonoscopy. The colono-copy results were recorded following the usual ongoing processor an organized screening program in the 2 areas. The colono-copy findings were classified according to the most advancedathologic lesion found. Advanced neoplasias included high-riskdenomas (adenomas measuring �10 mm or adenomas withigh-grade dysplasia) and invasive cancers (invasion of malig-ant cells beyond the muscularis mucosae). Intramucosal carci-omas were included in high-risk adenomas. Hyperplasic polypsere not included as neoplasia. Patients with incomplete colono-

copy, no invasive cancer diagnosed, and no further explorationo confirm the absence of any pathologic lesion (double-contrastarium enema or virtual colonoscopy) were considered as havingo colonoscopy result.

Statistical AnalysisThree strategies were used to compare the performances

of the FITs: at identical cut-off values of hemoglobin concen-tration in the stools (biochemical point of view), at an identicalpositivity rate implying a similar number of colonoscopies (en-doscopic workload point of view), and for either a similar num-ber of true-positive or false-positive results (patient’s harm orbenefit point of view).

To this purpose, the performances of each FIT were analyzed,varying the number of samples analyzed and the positivity cut-off value. For the 2-sample FIT strategy, the test was positive ifat least one of the samples contained a concentration of hemo-globin greater than the cut-off value. For the 1-sample FITstrategy, only the last sample was considered and compared withthe cut-off value. For each FIT, all concentrations of hemoglobingreater than the cutoff having determined colonoscopy in thestudy were assessed as alternative cutoffs. However, the cut-offvalue selected by the manufacturer to define Magstream as aqualitative test is smaller than the cut-off value used in ourstudy (80 �g/g instead of 180 �g/g of hemoglobin in the stools).

any of the subjects with one sample of Magstream between 80nd 180 �g/g of hemoglobin in the stools (positive at the

manufacturer’s cut-off value, but negative at the study cut-offvalue) were positive for either Hemoccult II, one of the OCSensor samples, or the other sample of Magstream at the studycut-off value and were hence offered a colonoscopy. Therefore,the performances of one sample Magstream at the manufactur-er’s cut-off value were derived from the study data, assumingthat the small proportion of missing colonoscopies would havedetected targeted lesions at a similar frequency as the performedones (see the Supplementary Appendix). The same extrapolationwas performed for 1-sample OC Sensor FIT at the manufactur-er’s cut-off value (20 �g/g of hemoglobin in the stools).

For each FOBT, the results of colonoscopies of subjects witha positive test result were classified as true-positive results (sub-jects with a targeted lesion) or false-positive results (subjectswith no targeted lesion). Then, the true-positive rate (TPR) wasdefined as the proportion of true-positive results among thesubjects who performed the FOBT. The false-positive rate (FPR)was defined as the proportion of false-positive results among thesubjects who performed the FOBT. In practice, the TPR is

directly proportional to the sensitivity of a screening test,

T

mtStNr

lWqH2aafb

TuFtontFvoNt1s

n

CLIN

ICA

LA

T

920 RAGINEL ET AL GASTROENTEROLOGY Vol. 144, No. 5

whereas the FPR is inversely proportional to its specificity. Thepaired design with blinding of details on test results allows oneto draw a receiver operating characteristics (ROC) curve, plottingthe TPR against the FPR.

The number needed to screen (NNScreen) to detect 1 subjectwith a targeted lesion was calculated as the inverse of the TPR.The number needed to scope (NNScope) to detect 1 subject witha targeted lesion was calculated as the inverse of the proportionof true-positive results among subjects with a positive test (in-verse of the positive predictive value). The analysis was con-ducted considering, first, advanced neoplasias, and, second, in-vasive cancers as targeted lesions.

Proportions were compared with chi-square tests. A sensitivityanalysis was conducted, first, drawing the ROC curves for detec-tion of advanced neoplasias excluding all subjects with at leastone nonanalyzable test, and, second, stratified on study area.

Statistical analysis was performed with R 2.12.121 (R Coreeam, Vienna, Austria) for Mac OS 10.6.

The study was approved by the local ethics committee (Co-ité de Protection des Personnes) and the French sanitary au-

hority (Agence Française de Sécurité Sanitaire des Produits deanté). All participants provided written informed consent. Therial was recorded on clinicaltrials.gov under the identifierCT01251666. All authors had access to the study data and

eviewed and approved the final manuscript.

ResultsFigure 1 presents the study flow chart. Among the

1224 subjects who were positive for at least one of thetests, an analyzable colonoscopy was performed for 1075subjects (87.8%), with a mean time interval between apositive screening and colonoscopy of 71 days (median, 56days). No colonoscopy complications were reported. Atotal of 334 participants were diagnosed with advancedneoplasia at colonoscopy.

In Table 1, 16.5% of participants underwent colorectalcancer screening for the first time. The median and meantimes between the last bowel movement for FOBT andlaboratory analysis were approximately 3 and 4 days, re-spectively. The rate of nonanalyzable tests decreased fromMagstream (4.4%) to Hemoccult II (2.0%) and OC Sensor(1.6%) (P � 10�3). A total of 1301 (6.6%) subjects had ateast 1 of the 3 tests nonanalyzable or not performed.

omen and older subjects were significantly more fre-uent among this population. The positivity rate ofemoccult II was 1.6% in both areas. When consideringsamples at the study cut-off values and only analyz-

ble tests, the positivity rates were identical for bothreas for OC Sensor, but higher in Cher than Allier areaor Magstream. The colonoscopy rate did not differetween tests (P � .73).The performances of the 3 FOBTs are compared in

able 2, for the study conditions and following the man-facturer’s guidelines. In the study conditions, for bothITs and with only 1 sample, the NNscreen was smallerhan with the guaiac test for detection of invasive cancers,r advanced neoplasias. Despite an increase in the totalumber of colonoscopies, the NNscope for detection ofhese lesions was also smaller than with the guaiac test.or each FIT, adding a second sample at the same cut-offalue decreased the NNscreen, by increasing the numberf colonoscopies at the expense of an increase in theNscope. The positivity rate of 1-sample Magstream at

he manufacturer cut-off value was similar to that with-sample OC Sensor at the study cut-off value, with aimilar positive predictive value.

Comparison of TPR and FPR for detection of advancedeoplasias according to the concentration of hemoglobin

Figure 1. Study flow chart.

fHazf(bMcnm

ffer

CLI

NIC

AL

AT

May 2013 COMPARISON OF PERFORMANCES OF 2 FITS 921

in the stools is provided in Supplementary Figure 1. Foreach FIT, the TPR and FPR decreased as the cut-off value,expressed as the concentration of hemoglobin in thestools, increased. Furthermore, adding a second sample ofthe same FIT at the same cut-off value increased both theTPR and the FPR. In the range of cut-off values in whichperformances of both FITs were evaluable, the TPR ofMagstream was greater than that of OC Sensor at thesame cut-off value and using the same number of sam-ples. However, at the same time, the FPR of Magstreamalso was greater than that of OC Sensor.

In Table 3, for a similar number of colonoscopies, theNNscreen for detection of advanced neoplasias was de-creased by 30% using 1-sample Magstream FIT instead ofthe guaiac test (166.4 instead of 238.5). At the same time,the NNscope also was decreased from 3.3 to 2.3. For thesame total number of colonoscopies, both the NNscreenand the NNscope associated with 1-sample OC SensorFIT were smaller than with 1-sample Magstream. Findingswere similar if the targeted lesion was invasive cancer only.For both the Magstream and OC Sensor, the same num-ber of positive subjects could result from use of 1 sample,or from use of 2 samples at a higher cut-off value. Then,for a similar number of colonoscopies, the NNscope and

Table 1. Demographic Characteristics and Test Results of Inc

All participants

Allier n (%) Cher n (%)

SexMen 6649 (44.6) 2125 (43.4)Women 8252 (55.4) 2771 (56.6)

Age, y50–54 2984 (20.0) 920 (18.8)55–59 3226 (21.7) 1051 (21.5)60–64 3591 (24.1) 1340 (27.4)65–69 2408 (16.2) 843 (17.2)70–74 2692 (18.1) 742 (15.2)

Screening roundFirst 2526 (17.0) 741 (15.1)Subsequent 12,375 (83.1) 4155 (84.9)

Test resultsHemoccult II

Positive 238 (1.6) 78 (1.6)Negative 14,428 (96.8) 4663 (95.2)Nonanalyzable 209 (1.4) 148 (3.0)Not performed 26 (0.2) 7 (0.1)

Magstreama

Positive 451 (3.0) 181 (3.7)Negative 13,747 (92.3) 4539 (92.7)Nonanalyzable 592 (4.0) 148 (3.0)Not performed 111 (0.7) 28 (0.6)

OC Sensorb

Positive 594 (4.0) 207 (4.2)Negative 14,038 (94.2) 4633 (94.6)Nonanalyzable 217 (1.5) 43 (0.9)Not performed 52 (0.3) 13 (0.3)

NOTE. Characteristics of the population and positivity rates for each taTwo-sample Magstream at a cut-off value of 55 ng/mL Hb in the bufbTwo-sample OC Sensor at a cut-off value of 150 ng/mL Hb in the bu

NNscreen for detection of advanced neoplasias were close w

with either 1 or 2 samples (providing a higher cut-offvalue), for both the Magstream and OC Sensor FITs.When fixing the positivity rate at a higher level (3.0%),for both the Magstream and OC Sensor the NNScreendecreased, and the NNscope increased compared with a1.6% positivity rate, while staying largely better thanwith the guaiac test. Both the NNscope and NNscreenobserved with the OC Sensor were better than with theMagstream.

Figure 2 compares the ROC curves associated with theguaiac test, Magstream, or OC Sensor FITs for the detec-tion of advanced neoplasias. Reading Figure 2 vertically,or a similar FPR of 9.8‰, the TPR increased fromemoccult II (TPR, 4.2‰), to 1-sample Magstream (7.3‰)

nd 1-sample OC Sensor (9.7‰). Reading Figure 2 hori-ontally, for a similar TPR of 4.2‰, the FPR decreasedrom Hemoccult II (FPR, 9.8‰) to 1-sample Magstream5.1‰) and 1-sample OC Sensor (2.8‰). Globally, the areaelow the curve of OC Sensor was greater than that ofagstream. Similar findings were observed for invasive

ancers. In the range of positivity rates evaluable for bothumber of samples given the study design, the perfor-ances for detection of advanced adenomas observed

ed Subjects

Excluding subjects with at least 1 nonanalyzable test

P value Allier n (%) Cher n (%) P value

.14 .196266 (45.0) 2002 (43.9)7665 (55.0) 2563 (56.1)

10�3 �10�3

2828 (20.3) 865 (19.0)3045 (21.9) 982 (21.5)3380 (24.3) 1265 (27.7)2240 (16.1) 777 (17.0)2438 (17.5) 676 (14.8)

10�2 �10�3

2365 (17.0) 675 (14.8)11,566 (83.0) 3890 (85.2)

10�3 .98225 (1.6) 74 (1.6)

13,706 (98.4) 4491 (98.4)— —— —

10�2 �.05444 (3.2) 173 (3.8)

13,487 (96.8) 4392 (96.2)— —

.01 .69558 (4.0) 189 (4.1)

13,373 (96.0) 4376 (95.9)— —

(study cut-off value and for 2-sample FITs).(�180 �g/g in the stools).(30 �g/g in the stools).

lud

�

�

�

�

estfer

ith either 1- or 2-sample FITs (at different cut-off values)

C

wi

NN

CLIN

ICA

LA

T

922 RAGINEL ET AL GASTROENTEROLOGY Vol. 144, No. 5

were very similar. An increase in TPR was possible byusing 2 samples, at the expense of increased FPR.

Results of the sensitivity analysis (available on request)did not change the findings.

DiscussionOur results, directly comparing the performances

of the 2 FITs in an average-risk population, strongly

Table 2. Comparison of Clinical Performances of Hemoccult IConditions, and According to Manufacturer Guideline

Guaiactest

Ma

1 sample

ut-off value (�g hemoglobin/g of stools) NA 180Cut-off value (ng hemoglobin/mL in the buffer) NA 55Positives 316 390

(positivity rate, %) (1.60) (1.97)Colonoscopies 277 344

(colonoscopy rate, %) (87.7) (88.2)Advanced neoplasias 83 146

(positive predictive value, %) (30.0) (42.4)(true-positive rate, %) (0.42) (0.74)

Number needed to scope 3.3 2.4Number needed to screen 238.5 135.6Invasive cancers 19 28

(positive predictive value, %) (6.9) (8.1)(true-positive rate, %) (0.10) (0.14)

Number needed to scope 14.6 12.3Number needed to screen 1041.9 707.0False positives 194 198

(false-positive rate, %) (0.98) (1.00)

aStudy cut-off value for Magstream is designed to equal the same numbStudy cut-off value for OC Sensor is designed to ensure comparability

Table 3. Comparison of Clinical Performances of Magstream

Guaiactest

1.6%

Ma

1 sample

Cut-off value (�g hemoglobin/g of stools) NA 206Cut-off value (ng hemoglobin/mL in the buffer) NA 63Positives 316 312

(positivity rate, %) (1.60) (1.58)Colonoscopies 277 273

(colonoscopy rate, %) (87.7) (87.5)Advanced neoplasias 83 119

(positive predictive value, %) (30.0) (43.6)(true positive rate, %) (0.42) (0.60)

Number needed to scope 3.3 2.3Number needed to screen 238.5 166.4Invasive cancers 19 24

(positive predictive value, %) (6.9) (8.8)(true-positive rate, %) (0.10) (0.12)

umber needed to scope 14.6 11.4umber needed to screen 1041.9 824.9

False positives 194 154

(false-positive rate, %) (0.98) (0.78)

suggest that the OC Sensor outperforms Magstream forboth the detection of invasive cancers alone or ad-vanced neoplasias as a whole; the 2 tests provide asubstantial improvement in cancer screening in com-parison with Hemoccult II. Strikingly, for the samenumber of positive results, 1-sample OC Sensor nearlydoubled the number of advanced neoplasias detectedcompared with the 3-sample guaiac test. For the same

agstream, and OC Sensor Fecal Occult Blood Tests in StudyExtrapolated)

Study cut-off valueaManufacturer cut-off

value

ream OC SensorFIT

MagstreamFIT

OC SensorFIT

2 samples 1 sample 2 samples 1 sample 1 sample

180 30 30 80 2055 150 150 20 100

632 551 801 548 694(3.19) (2.78) (4.05) (2.77) (3.51)554 488 712 484 615(87.7) (88.6) (88.9) (88.3) (88.6)210 225 290 209 275(37.9) (46.1) (40.7) (43.2) (44.7)(1.06) (1.14) (1.46) (1.06) (1.39)

2.6 2.2 2.5 2.3 2.294.3 88.0 68.3 94.7 72.036 37 44 37 44

(6.5) (7.6) (6.2) (7.6) (7.2)(0.18) (0.19) (0.22) (0.19) (0.22)15.1 13.2 16.2 13.1 14.0

549.9 535.1 449.9 535.1 449.9344 263 422 275 340(1.74) (1.33) (2.13) (1.39) (1.72)

of false-positive results as Hemoccult II, as derived from prior studies.th other studies, while limiting the colonoscopy workload.

OC Sensor FITs for 2 Positivity Rates

itivity rate (same as Hemoccult II) 3% Positivity rate

eam OC SensorFIT

MagstreamFIT

OC SensorFIT

2 samples 1 sample 2 samples 2 samples 2 samples

234 68 116 187 4872 340 580 57 240

330 316 317 599 595(1.67) (1.60) (1.60) (3.03) (3.01)295 282 285 526 523(89.4) (89.2) (89.9) (87.8) (87.9)123 154 160 203 241(41.7) (54.6) (56.1) (38.6) (46.1)(0.62) (0.78) (0.81) (1.03) (1.22)

2.4 1.8 1.8 2.6 2.2161.0 128.6 123.7 97.5 82.1

20 31 33 34 37(6.8) (11.0) (11.6) (6.5) (7.1)(0.10) (0.16) (0.17) (0.17) (0.19)

14.8 9.1 8.6 15.5 14.1989.9 638.6 599.9 582.3 535.1172 128 125 323 282

I, Ms (

gstFIT

er

and

Pos

gstrFIT

(0.87) (0.65) (0.63) (1.63) (1.42)

soe

avi

fis

otttOcpsmmt

aF

satcaMpbia

CLI

NIC

AL

AT

May 2013 COMPARISON OF PERFORMANCES OF 2 FITS 923

number of unnecessary colonoscopies (false-positive re-sults), the OC Sensor detected more advanced neopla-sias than Magstream.

Our study had several limitations. Only subjects posi-tive for at least one of the tests were explored by colono-scopy, preventing an estimation of sensitivity and speci-ficity of each test. Although a minority of studiesevaluating FOBT implied an endoscopy (colonoscopy orsigmoidoscopy) for all subjects,22,23 this limitation ishared by most of the studies comparing the performancef colorectal cancer screening programs, inasmuch as it isvaluated in the targeted population.7–9,14 Moreover, in

our study, plotting TPR against FPR provided a compre-hensive version of the ROC curve, with direct and practi-cal interpretation. Indeed, TPR in participants is the prod-uct of prevalence of the lesion, the sensitivity of the test,and the colonoscopy rate of positive subjects, and FPR isthe product of 1 minus the prevalence, 1 minus thespecificity, and colonoscopy rate. The paired design guar-antees that the prevalence is identical whatever the tests,whereas the delivery of positive screening results withoutspecification on which FOBT(s) was(ere) positive guaran-tees that the colonoscopy rate does not depend on thetest. Thus, comparing TPR or FPR between tests is equiv-alent to comparing sensitivity and specificity as in relativeROC curves.9,24

As in other studies, we evaluated indicators such aspositivity rate, detection rate among participants, andpositive predictive values, which reflect the performanceof the tests. To allow for standardization of our results, wetransformed the crude value given by the FIT automatsinto concentration of hemoglobin in the stools, but ourconversion equation for Magstream was not perfect.17

Finally, our evaluation of the performances of 1-sampleFIT at the manufacturer’s cut-off value implied the im-putation of only 13% of the total number of expectedcolonoscopies, resulting in reliable results. Too many un-

Figure 2. ROC curves for Hemoccult II, Magstream, and OC Sensortests, according to the number of samples of immunochemical tests.

available colonoscopies prevented us from performing a

similar imputation for 2 samples of FIT at the manufac-turer cut-off value.

In comparison with Hemoccult II, Magstream FITappeared less specific than expected based on our pre-vious study.9 This could be explained by changes inautomat (Magstream HT instead of Magstream 1000)and buffer (Magstream NewHemtube instead of Mag-stream Hemtube) for the Magstream test, or an appar-ent increase in the specificity of Hemoccult II owing todietary restrictions (although debated25) in Allier. In

ddition, the present study included subjects with pre-ious rounds of Hemoccult II, which may have anmpact on its specificity.

The superiority of OC Sensor over Magstream con-rms previous evidence based on an indirect compari-on relying on a literature review,15 and laboratory

analyses.17,18 However, this study suggests the superi-rity of the OC Sensor test, directly comparing the 2ests in a real-life scenario. The joint increase in sensi-ivity and the decrease in specificity of Magstream forhe detection of advanced neoplasias compared withC Sensor at the same cut-off value expressed in con-

entration of hemoglobin in the stools may be ex-lained by the poorer precision of the Magstream mea-urement because a much higher overlap in crude

easurements between different concentrations of he-oglobin in the feces had been observed for Magstream

han OC Sensor in our in-laboratory analysis.17 Fur-thermore, OC Sensor is based on latex agglutinationimmunoturbidimetry, whereas Magstream is based onmagnetic particle agglutination. Our results confirmthose of other studies that have shown that qualitativeFITs using the same cut-off value expressed in concen-tration of hemoglobin in the stools may be associatedwith different clinical characteristics (adapted fromBrenner et al26), supporting the need for clinical trials,

t least in high-risk subjects, to evaluate future newITs.The superiority of OC Sensor compared with Mag-

tream concerned both the detection of invasive cancersnd high-risk adenomas, hence offering the opportunityo increase both the reduction in colorectal cancer–spe-ific mortality and incidence, although not shown. Inddition, the rate of nonanalyzable tests was higher withagstream compared with OC Sensor test. This was ex-

lained both by failure to correctly paste the identifyingarcode due to the rounded form of the collecting tube, and

ntroduction of too much stool in the tube. These difficultiesre consistent with previous findings.18 In our study, how-

ever, and because Hemoccult II is the routine test in France,nonanalyzable FITs were not repeated, hence overestimatingthe rate of nonanalyzable tests.

As suggested by other studies,9,13,14 for moderate posi-tivity rates we found that both Magstream and OC SensorFITs led to identical performances with either 1 of 2samples (provided a different cut-off value is used),reflecting equivalence at the population level, rather

than at the level of the subject.10 Differences in bleed-

Se

psct

1

1

1

1

1

1

1

1

1

1

2

2

2

2

2

2

2

2

CLIN

ICA

LA

T

924 RAGINEL ET AL GASTROENTEROLOGY Vol. 144, No. 5

ing pattern between lesions (intensity and intermit-tence) underlies this finding. When limitation of thecolonoscopy workload is targeted, we would recom-mend the use of one sample of FIT. Alternatively, whenhigh positivity rates are affordable for the health caresystem, using 2 samples of FIT may increase the detec-tion rate of advanced neoplasias at an acceptable num-ber needed to scope.

Concerning the performance of FITs at moderate pos-itivity rates, our results are clearly in favor of 1-sample OCSensor. Selection of the cut-off value then could be ad-justed by politic decision makers to fit the possibility ofendoscopy. The data presented in this article provideresults for up to a 2.8% positivity rate for 1-sample OC

ensor (150 ng/mL in the buffer cut-off value which isquivalent to 30 �g/g hemoglobin in the stools), whereas

some studies provide results for lower cut-off values.11–13,20

However, although OC Sensor is more accurate than Mag-stream, it is also actually more expensive. Even if the costs ofcolorectal cancer screening programs are determined mainlyby the number of generated colonoscopies,27 for a fixed

ositivity rate, OC Sensor would be more effective (higherensitivity), less deleterious (higher specificity), but also moreostly. Cost-effectiveness analyses are required to guide fur-her decisions.

Supplementary Material

Note: To access the supplementary materialaccompanying this article, visit the online version ofGastroenterology at www.gastrojournal.org, and at http://dx.doi.org/10.1053/j.gastro.2013.01.042.

References

1. Parkin DM, Bray F, Ferlay L, et al. Global cancer statistics, 2002.CA Cancer J Clin 2005;55:74–108.

2. Hewitson P, Watson Glasziou P, et al. Cochrane systematic review ofcolorectal cancer screening using the fecal occult blood test (hemoc-cult): an update. Am J Gastroenterol 2008;103:1541–1549.

3. Atkin WS, Edwards R, Kralj-Hans I, et al. Once-only flexible sig-moidoscopy screening in prevention of colorectal cancer: a multi-centre randomised controlled trial. Lancet 2010;375:1624–1633.

4. Segnan N, Armaroli P, Bonelli L, et al. Once-only sigmoidoscopy incolorectal cancer screening: follow-up findings of the Italian Ran-domized Controlled Trial—SCORE. J Natl Cancer Inst 2011;103:1310–1322.

5. Schoen RE, Pinsky PF, Weissfeld JL, et al. PLCO Project Team.Colorectal-cancer incidence and mortality with screening flexiblesigmoidoscopy. N Engl J Med 2012;366:2345–2357.

6. Allison JE, Tekawa IS, Ransom LJ, et al. A comparison of fecaloccult-blood tests for colorectal-cancer screening. N Engl J Med1996;334:155–159.

7. van Rossum LG, van Rijn AF, Laheij RJ, et al. Random comparisonof guaiac and immunochemical fecal occult blood tests for colo-rectal cancer in a screening population. Gastroenterology 2008;135:82–90.

8. Hol L, van Leerdam ME, van Ballegooijen M, et al. Screening forcolorectal cancer: randomised trial comparing guaiac-based andimmunochemical faecal occult blood testing and flexible sigmoid-oscopy. Gut 2010;59:62–68.

9. Guittet L, Bouvier V, Mariotte N, et al. Performance of immuno-

chemical faecal occult blood test in colorectal cancer screening in

average-risk population according to positivity threshold andnumber of samples. Int J Cancer 2009;125:1127–1133.

0. Guittet L, Bouvier V, Guillaume E, et al. Colorectal cancer screen-ing: why immunochemical faecal occult blood test performs aswell with either one or two samples. Dig Liv Disease 2012;44:694–699.

1. van Rossum LG, van Rijn AF, Laheij RJ, et al. Cutoff determinesthe performance of a semi-quantitative immunochemical faecaloccult blood test in a colorectal cancer screening programme. Br JCancer 2009;101:1274–1281.

2. Hol L, Wilschut JA, van Ballegooijen M, et al. Screening for colo-rectal cancer: random comparison of guaiac and immunochemicalfaecal occult blood testing at different cut-off levels. Br J Cancer2009;100:1103–1110.

3. Park DI, Ryu S, Kim YH, et al. Comparison of guaiac-based andquantitative immunochemical fecal occult blood testing in a pop-ulation at average risk undergoing colorectal cancer screening.Am J Gastroenterol 2010;105:2017–2025.

4. Faivre J, Dancourt V, Manfredi S, et al. Positivity rates and perfor-mances of immunochemical faecal occult blood tests at differentcut-off levels within a colorectal cancer screening programme. DigLiver Dis 2012;44:700–704.

5. Guittet L, Bailly L, Bouvier V, et al. Indirect comparison of perfor-mances of two quantitative immunochemical faecal occult bloodtests in a population with average colorectal cancer risk. J MedScreen 2011;18:76–81.

6. Allison JE, Fraser CG, Halloran SP, et al. Comparing fecal immu-nochemical testing: improved standardization is needed. Gastro-enterology 2012;142:422–424.

7. Guittet L, Guillaume E, Levillain R, et al. Analytical comparison ofthree quantitative immunochemical fecal occult blood tests forcolorectal cancer screening. Cancer Epidemiol Biomarkers Prev2011;20:1492–1501.

8. Lamph SA, Bennitt WE, Brannon CR, et al. Evaluation report:immunochemical faecal occult blood tests. Centre for Evidence-based Purchasing 2009, Report no. 09042.

9. Castiglione G, Visioli CB, Ciatto S, et al. Sensitivity of latex agglu-tination faecal occult blood test in the Florence District population-based colorectal cancer screening programme. Br J Cancer 2007;96:1750–1754.

0. Grazzini G, Visioli CB, Zorzi M, et al. Immunochemical faecaloccult blood test: number of samples and positivity cutoff. What isthe best strategy for colorectal cancer screening? Br J Cancer2009;100:259–265.

1. R Development Core Team (2010). R: a language and environ-ment for statistical computing. Vienna, Austria: R Foundation forStatistical Computing. Available at: http://www.R-project.org/.

2. Allison JE, Sakoda LC, Levin TR, et al. Screening for colorectalneoplasms with new fecal occult blood tests: update on per-formance characteristics. J Natl Cancer Inst 2007;99:1462–1470.

3. Morikawa T, Kato J, Yamaji Y, et al. A comparison of theimmunochemical fecal occult blood test and total colonoscopyin the asymptomatic population. Gastroenterology 2005;129:422–428.

4. Cheng H, Macaluso M. Comparison of the accuracy of two testswith a confirmatory procedure limited to positive results. Epidemi-ology 1997;8:104–106.

5. Konrad G. Dietary interventions for fecal occult blood test screen-ing: systematic review of the literature. Can Fam Physician 2010;56:229–238.

6. Brenner H, Haug G, Hundt S. Inter-test agreement and quantitativecross-validation of immunochromatographical fecal occult bloodtests. Int J Cancer 2010;127:1643–1649.

7. Berchi C, Bouvier V, Réaud JM, et al. Cost-effectiveness analysisof two strategies for mass screening for colorectal cancer in

France. Health Econ 2004;13:227–238.

ICA

LA

T

May 2013 COMPARISON OF PERFORMANCES OF 2 FITS 925

Received August 30, 2012. Accepted January 22, 2013.

Reprint requestsAddress requests for reprints to: Lydia Guittet, MD, PhD, Inserm

U1086, Cancers and Preventions, CHU Caen, Centre FrançoisBaclesse, Avenue du Général Harris, 14000 Caen, France. e-mail:guittet-l@chu-caen.fr; fax: 33 (0)2 31 45 86 30.

AcknowledgmentsThe authors are grateful to all subjects who agreed to join the

study, to the general practitioners who included the patients, to theendoscopists who performed the colonoscopies, to the members ofthe 2 regional associations in charge of colorectal cancer screening,

ADOC18 and ABIDEC, to the Institut InterRégional pour la Santé

laboratory, and to the research team Cancers and Preventions whowere involved in the study.

Conflicts of interestThe authors disclose no conflicts.

FundingThe study was funded by the French National Institute for

Cancer (Institut National du Cancer), and the French NationalLeague against Cancer (Ligue Nationale Contre le Cancer); theautomat analyzers were provided on loan by Eiken (OC SensorDiana automat) and Fujirebio (Magstream HT automat)manufacturers. The sponsors and manufacturers had no role in

the study design or data analysis. C

LIN