American Journal of Gastroenterology ISSN 0002-9270C© 2006 by Am. Coll. of Gastroenterology doi: 10.1111/j.1572-0241.2006.00390.xPublished by Blackwell Publishing

Polyp Miss Rate Determined by Tandem Colonoscopy:A Systematic Review

CME

Jeroen C. van Rijn, M.D.,1 Johannes B. Reitsma, M.D., Ph.D.,1 Jaap Stoker, M.D., Ph.D.,2

Patrick M. Bossuyt, Ph.D.,1 Sander J. van Deventer, M.D., Ph.D.,3 and Evelien Dekker, M.D., Ph.D.31Departments of Clinical Epidemiology & Biostatistics; 2Radiology; and 3Gastroenterology, Academic MedicalCenter (University of Amsterdam), Amsterdam, The Netherlands

BACKGROUND Colonoscopy is the best available method to detect and remove colonic polyps and therefore servesAND AIMS: as the gold standard for less invasive tests such as virtual colonoscopy. Although gastroenterologists

agree that colonoscopy is not infallible, there is no clarity on the numbers and rates of missedpolyps. The purpose of this systematic review was to obtain summary estimates of the polyp missrate as determined by tandem colonoscopy.

METHODS: An extensive search was performed within PUBMED, EMBASE, and the Cochrane Library databasesto identify studies in which patients had undergone two same-day colonoscopies with polypectomy.Random effects models based on the binomial distribution were used to calculate pooled estimatesof miss rates.

RESULTS: Six studies with a total of 465 patients could be included. The pooled miss rate for polyps of any sizewas 22% (95% CI: 19–26%; 370/1,650 polyps). Adenoma miss rate by size was, respectively, 2.1%(95% CI: 0.3–7.3%; 2/96 adenomas ≥10 mm), 13% (95% CI: 8.0–18%; 16/124 adenomas 5–10mm), and 26% (95% CI: 27–35%; 151/587 adenomas 1–5 mm). Three studies reported data onnonadenomatous polyps: zero of eight nonadenomatous polyps ≥10 mm were missed (0%; 95% CI:0–36.9%) and 83 of 384 nonadenomatous polyps <10 mm were missed (22%; 95% CI: 18–26%).

CONCLUSIONS: Colonoscopy rarely misses polyps ≥10 mm, but the miss rate increases significantly in smaller sizedpolyps. The available evidence is based on a small number of studies with heterogeneous studydesigns and inclusion criteria.

(Am J Gastroenterol 2006;101:343–350)

INTRODUCTION

Colorectal cancer is the second leading cause of cancerdeath in the United States (1). It has been shown that asignificant reduction of the incidence of colorectal cancercan be achieved by removing adenomatous polyps beforetheir potential progress to adenocarcinomas (2). Currently,colonoscopy is the best available method to detect and re-move colonic polyps and can therefore be considered as thegold standard for this purpose (3, 4). However, colonoscopy isnot infallible and although most gastroenterologists are awarethat polyps can be missed during colonoscopy, numbers andrates remain uncertain.

Furthermore, as population-based studies on colorectalcancer screening are progressing, it is conceivable that inthe near future many western countries adopt colorectal can-cer screening of all individuals aged 50 or above, leading toa vast increase in the number of colonoscopic examinations.Within the debate on mass screening for colorectal cancer, a

Evelien Dekker is a fellow of the Dutch Cancer Foundation.

To access a continuing medical education exam for this article, please visit www.acg.

gi.org/journalcme.

quantification of polyp miss rate is essential for future cost-benefit analyses.

Besides its use in clinical practice or screening,colonoscopy also serves as the reference standard in manystudies on the evaluation of less invasive tests such as virtualcolonoscopy or fecal tests. Misclassification of the referencestandard, however, can influence the perceived accuracy ofthe test under evaluation (5). Without accurate estimates ofcolonoscopic miss rates, the impact on the test accuracy can-not be deduced.

The most reliable method to investigate polyp miss rateis by “tandem” or “back-to-back” colonoscopy, a method inwhich two same-day colonoscopies, both with polypectomy,are performed within each patient. In such studies, the missrate is usually expressed as the number of polyps detectedonly during the second colonoscopy relative to the numberof polyps found during both examinations.

The purpose of this systematic review was to appraise theevidence from all available studies on tandem colonoscopyand to obtain summary estimates of the colonoscopic polypmiss rate. We also investigated possible sources of hetero-geneity in miss rate within and between studies.

343

344 van Rijn et al.

METHODS

Our aim was to identify studies in which patients had un-dergone two same-day colonoscopies with polypectomy. Wealso aimed to determine the miss rate of optical colonoscopyindependent of other diagnostic tests. This was done becausein the optimal study design in diagnostic test evaluation stud-ies, readers of the reference standard should be blinded forthe results of the test under evaluation (6). In most studieson the evaluation of less invasive tests to detect polyps orcancer, colonoscopy is used as the reference standard. There-fore, studies in which virtual colonoscopy or other tests wereused to improve the colonoscopic detection of polyps wereexcluded.

Search StrategyAn electronic search of PUBMED, EMBASE and theCochrane Library (January 1984 through June 2005) wasperformed. The following keywords and medical subjectheading (MeSH) terms were combined: “polyp,” “large in-testine tumor,” “colonic neoplasms,” “colonic polyps,” and“colonoscopy.” We used no language restrictions. One re-viewer (J.C.vR.) screened all abstracts of the initial searchwhether tandem colonoscopy could have been performed.Case-reports, letters, abstracts only, reviews, and nonhumanstudies were excluded at this stage.

Inclusion CriteriaWe retrieved the full text from potentially relevant abstractsand two reviewers (J.C.vR., reviewer 1 and E.D., reviewer 2)independently assessed the full paper for inclusion. Studieswere included in the meta-analysis if the study performed asame-day tandem colonoscopy and reported on the raw dataof the polyp miss rate.

A manual cross-reference search of eligible papers wasperformed to identify additional relevant articles. No attemptswere made to identify unpublished data.

Data CollectionThe two reviewers independently evaluated the selected fullpapers and extracted data on design, study population, endo-scopists, and polyp miss rates, if possible by type and size.We also evaluated the number of patients with a complete ex-amination in each study (cecum reach), as well as the patient-based negative predictive value defined as the proportion ofpatients without polyps at both colonoscopic examinations.Disagreements about the inclusion of studies and data extrac-tion were resolved in a consensus discussion.

Statistical AnalysisOur primary outcome was the polyp miss rate, defined asthe proportion of polyps detected only during the secondcolonoscopy relative to the total number of polyps found dur-ing the first and second examinations. The 95% confidenceinterval around the miss rate for individual studies was cal-culated using the Wilson method for binomial proportions

(7). To test for heterogeneity in miss rates, the Q-test wasused. p Values lower than 0.05 were considered statisticallysignificant.

We used a random effects approach for the pooling of missrates across studies. Our basic model assumes that the truemiss rates (logit transformed) vary from one study to another.These deviations from the overall mean value are assumed tohave an approximately normal distribution with an unknownamount of between-study variability. Our model also takesinto account that the miss rate observed in each study hasbeen measured with imprecision due to sampling error. Thisimprecision depends largely on the size of the study, or moreprecisely on the total number of polyps in that study. Thisimprecision is used to weigh the results when calculating apooled estimate (inverse variance method). The NLMIXEDprocedure of SAS (version 9.1) was used to estimate the pa-rameters of this nonlinear mixed model.

Covariates can be added to the model to estimate and testwhether the pooled miss rate differs between studies or be-tween subgroups of patients. Two subgroup analyses werecarried out: by type (adenomatous or nonadenomatous) andby polyp size (three categories: 1–5 mm, 5–9 mm, and 10 mmor more).

Cecum reach was defined as the proportion of patientswith a colonoscopic examination completed to the cecum.A pooled estimate was calculated using the same model thatwas used for the miss rate.

RESULTS

The initial search yielded 6,439 hits. Figure 1 shows the flowchart of the selection procedure of the papers. There was ahigh level of agreement between the independent reviewers.All 33 abstracts selected for detailed evaluation were inde-pendently selected by both reviewers. No papers were addedto the search based on cross-checking references of the in-cluded studies. After the second consensus meeting, a totalof six papers using same-day tandem colonoscopy met ourinclusion criteria. One paper reported data on two differentcohorts (12); we analyzed these cohorts as separate studies.Hixson et al. published data on the same cohort of patients intwo journals (8, 9); we combined the data from these papersto obtain the maximal amount of information on this cohort.In total, 6 cohorts involving 465 patients were included in ourreview.

General and Clinical CharacteristicsGeneral and clinical characteristics of the included studiesare presented in Table 1. The age distribution in the differentstudies was comparable, but there was considerable variationin the male/female ratio (Table 1). The study by Hixson et al.included only one woman because patients were recruited atthe Tucson Veterans Affairs Medical Center. An explanationfor the high percentage of men (76%) in the study by Rexet al. was not reported.

Polyp Miss Rate Determined by Tandem Colonoscopy 345

Hits after initial searchin PUBMED, EMBASE

and The Cochrane Library (n=6,439)

Exclusion of case reports, no abstract, non-human studies,

letters and reviews (n=6,409), performed by reviewer 1

Potentially relevant abstracts (n=430)

Excluded after first consensus meeting (n=397) because no

tandem colonoscopy was performed or colonic polyps were not an

objective

Selected by reviewer 1

(n=38 / 430)

Selected by reviewer 2 (n=43 / 430)

Full papers retrieved for detailed evaluation

(n=33)

Excluded after second consensus meeting (n=27) because no same-

day tandem colonoscopy was performed or no data on miss rate

were available

Papers included in meta-analysis (n=6)

Figure 1. Flow chart of papers included in the review.

All studies applied the following exclusion criteria: historyof surgical resection, anti-coagulant use, no informed con-sent, and poor medical condition not allowing two colono-scopies. There were two studies (12, 13) in which patientswith inflammatory bowel disease (IBD) were excluded. Onlyone study (8) explicitly reported inadequate bowel prepara-tion (>5% of mucosal area obscured) as an exclusion crite-rion.

The distribution of indications for colonoscopy is givenin Table 1. The proportion of screenees among the differentstudies varied from 0% in the study of Rex et al. and Mat-sushita et al. to 41% in the study by Harrison et al.; it wasnot reported in two studies. The study of Matsushita et al. in-cluded only patients in whom polyps were detected by previ-ous examinations. Based on the indications for colonoscopy,we judged the population as high risk for polyps in two studies(8, 11) medium to high risk in one study (10), and one studyas medium to low risk (13). No qualification was possible intwo studies because of insufficient reporting on indicationsfor colonoscopy.

A randomized design comparing different colonoscopictechniques (using the miss rate determined by tandemcolonoscopy as the primary outcome) was used in five of the

six cohorts in our study. Colonoscopic techniques that wereevaluated included: standard forward view versus retroflexedview (13), wide angle versus standard view (12), with trans-parent cap versus without (11), and a change in body posi-tion (10) of patients during colonoscopy. None of these ran-domized studies reported a significant difference between thetechniques being compared. Given the absence of significanteffects on miss rate, we combined the data of the different in-tervention arms in each trial. The study by Hixson et al. wasthe only “standard” cohort study in which patients sched-uled to undergo routine outpatient colonoscopy received twocolonoscopies without randomization.

We observed substantial differences in the prevalence ofpolyps between the study populations. The mean number ofpolyps per patient ranged from 1.2 to 5.0 (Table 1). In thestudy by Harrison the colonoscopic examination was limitedto the proximal colon. The number of polyps detected inthe distal colon was not reported, which largely explains thelow number of polyps per patient. None of the other studieselucidated whether the high mean numbers of polyps wereattributable to a small number of patients with many polyps orwhether the polyps were evenly distributed among patients.

Miss Rate All PolypsThe miss rates of the individual studies are presented in Figure2. The six studies reported a total of 1,650 polyps of all typesand all sizes. Overall, 370 of the 1,650 polyps were seen atthe second colonoscopy only. The pooled miss rate was 21%(95% CI: 14–30%), but Figure 2 shows a substantial amountof heterogeneity (Q-test for heterogeneity p value < 0.0001).

One possible explanation for the variation in miss rateacross studies is a difference in the prevalence. To test whetherthere was an association between polyp prevalence and missrate, we included this parameter as a covariate in our mixedmodel. The study by Harrison was excluded from the analy-sis because we were not able to retrieve the total number ofpolyps in the whole colon. There was a moderate, but non-significant effect (p value = 0.12) between the mean numberof polyps per patient and the observed miss rate.

Miss Rate by TypeInformation on nonadenomatous polyps was missing in thereport by Matsushita et al. In Figure 3, the miss rate in theremaining five cohorts is presented stratified by type (adeno-matous and nonadenomatous). The pooled miss rate of non-adenomatous polyps was higher (27%) compared to adeno-mas (22%), but this difference was not statistically significant(p value = 0.49).

Miss Rate by SizeThe most commonly reported categories of polyp size were1–5 mm, 5–10 mm, and ≥10 mm, which were available inall cohorts except for the study by Matsushita et al. None ofthe studies reported the miss rate of nonadenomatous polypsusing these three size categories. Therefore, Figure 4 presentsthe pooled miss rate with 95% confidence intervals by size

346 van Rijn et al.

Table 1. General Characteristics and Clinical Features of the Six Included Studies Involving 465 Patients

Rex Matsushita Rex Rex HarrisonHixson et al. et al. et al. (I) et al. (II) et al.

StudyYear 1991 1997 1998 2003 2003 2004Center Multi Single Single Single Single MultiSetting Academic/veterans Academic Academic Academic Academic Academic/privateStudy Design Cohort Trial Trial Trial Trial TrialNumber of patients 90 183 24 20 50 98Mean age 65 62 58 Nr 61 58Range 37–92 Nr 42–77 Nr 39–87 42–83Male 99% 62% 63% Nr 76% 46%Number of endoscopists 2 1 1 3 2 3Anatomical Region Whole colon Whole colon Whole colon Whole colon Whole colon Proximal colonNumber of incomplete exam 0 4 3 0 2 2Coecum reach 100% 98% 88% 100% 96% 98%

Indications for colonoscopyScreening 0 7% 0 Nr Nr 41%Family history of colon carcinoma 0 0 0 Nr Nr 6%Follow-up of polyps 32% 38% 0 Nr Nr 29%Follow-up of colon carcinoma 4% 0 0 Nr Nr 0Hematochezia 6% 11% 0 Nr Nr 17%Change of bowel habits/diarrhea 0 3% 0 Nr Nr 4%Rectal/abdominal pain 1% 6% 0 Nr Nr 0Constipation 0 0 0 Nr Nr 0Fe deficiency 9% 4% 0 Nr Nr 0IBD 2% 0 0 Nr Nr 0Other signs & symptoms 1% 2% 0 Nr Nr 0Positive heme test in stool 17% 16% 0 Nr Nr 1%Polyps on sigmoidoscopy 28% 14% 100% Nr Nr 0

or barium enemaEstimated population risk∗ High Medium to high Maximal Nr Nr Low to MediumMean no. of polyps/patient 4.6 3.8 3.8 4.0 5.0 1.2

Nr = not reported, IBD = inflammatory bowel disease.∗Qualitative estimation of pretest risk of the polyp prevalence in each study, based on the distribution of indications for colonoscopy.

for adenomas only. The adenoma miss rate increased signifi-cantly with smaller size from 2% (large adenomas ≥10 mm)to 13% for adenomas 5–10 mm (p value = 0.012), and to 26%for small adenomas 1–5 mm (p value = 0.0001). From a totalof 96 large adenomas (≥10 mm), two were detected duringthe second examination only. Both were reported within thesame study by Rex et al. in which the total number of largeadenomas was 32 (miss rate 6%). The remaining studies re-ported a total of 64 adenomas ≥10 mm, all detected at thefirst examination and therefore, a miss rate of zero.

For nonadenomatous polyps, two studies involving 376polyps in 160 patients used a categorization of large (≥10mm) versus small (<10 mm). No large nonadenoma-tous polyps were missed during the first colonoscopy. Themiss rate for nonadenomatous polyps <10 mm was 22%(83/376).

Cecum ReachTable 1 provides the number of patients in whom the colono-scopic examination was incomplete either at the first or thesecond examination. The total number of patients with in-complete examinations was 11 of 473 (2.3%). In all studiesexcept for the study by Matsushita et al., these patients wereexcluded from the analysis. Therefore, of the 465 patients

in the analysis of polyp miss rate (all types and all sizes;Figure 2), 462 patients had two complete colonoscopies and3 patients (0.6%) had an incomplete first and/or second exam-ination. The summary estimate for the proportion of patientswith complete colonoscopy was 98% (95–99%).

Negative Predictive ValueThe only study to accurately report on patient-based missrate was the study by Rex et al. The proportion of patientswithout adenomas (of any size) at the first examination was43% (79/183). In 13 of these 79 patients, at least one ade-noma of any size was detected at the second examination,which corresponds with a patient-based negative predictivevalue of 84%. Rex et al. also reported an increase in missrate with more adenomas detected during the first examina-tion: the miss rate was 16% in patients with 0 adenomas atfirst examination, 30% in patients with 1 adenoma, increas-ing to 47% in patients with 2 or more adenomas at the firstexamination.

DISCUSSION

In this systematic review we analyzed the colonoscopicmiss rate for polyps by type and size. The available evi-dence, however, was based on a relatively small number of

Polyp Miss Rate Determined by Tandem Colonoscopy 347

58 / 411

178 / 697

7 / 91

63 / 250

19 / 80

45 / 121

21% (14 to 30)

Exam 2 / Total

Pooled

Harrison, 2004

Rex, 2003 (II)

Rex, 2003 (I)

Matsushita, 1998

Rex, 1997

Hixson, 1991

Miss rate (%)

0 20 40 60 80 100

Figure 2. Miss rate of all polyps. Note: Figure 1 shows significantheterogeneity between studies (Q-test p value <0.0001).

heterogeneous studies including a total of 465 cases. Our re-sults do not support the widely appreciated assumption thatsignificant proportions of polyps ≥10 mm are missed duringcolonoscopy. For smaller sized polyps, we can conclude thatthe miss rate significantly increases, inversely related to size.

The study by Rex et al. was the only study to report twomissed large adenomas. The first was a slightly raised “flat”15 mm lesion and the second was a polyp of exactly 10 mmwith a focus of well differentiated adenocarcinoma, locatedbehind a rectal valve. The question arises whether this evi-dence is sound enough to conclude that colonoscopy misseslarge polyps in general.

The studies in our systematic review used different classi-fications of polyps regarding size-categories and histologicaltypes. Most studies used three size-categories: 1–5 mm, 5–10mm, and ≥10 mm. Although currently most guidelines ad-vise removal of polyps ≥10 mm, it is conceivable that futureguidelines will recommend removal of polyps smaller than 10mm. As many gastroenterologist currently remove all polyps,we believe that reporting sizes of polyps smaller than 10 mmis meaningful. The significant miss rate of 13% for polyps5–10 mm could become a serious issue in clinical practice.We therefore support the recommendation by Pickhardt (14)to report medium-sized polyps (5–10 mm) by mm in futurestudies of polyp detection.

The method of tandem colonoscopy cannot be consideredas 100% accurate to estimate the polyp miss rate. In gen-eral, back to back colonoscopies will tend to underestimatethe miss rate as polyps in awkward positions are more likelyto be missed during both examinations. A second factor in-

24 / 221

89 / 378

23 / 102

10 / 26

23 / 80

22% (15 to 32)

34 / 190

89 / 319

40 / 148

9 / 54

22 / 41

27% (19 to 37)

Exam 2 / Total

Adenoma

Non-adenoma

Pooled

Harrison, 2004

Rex, 2003 (II)

Rex, 2003 (I)

Rex, 1997

Hixson, 1991

Pooled

Harrison, 2004

Rex, 2003 (II)

Rex, 2003 (I)

Rex, 1997

Hixson, 1991

Miss rate (%)

0 20 40 60 80 100

Figure 3. Polyp miss rate by type.

fluencing the estimated miss rate is observer variability. Inthe study by Harrison et al., the initial examination was per-formed by a fellow and the second examination by an expertgastroenterologist. This may have inflated the miss rate. Inthe study by Rex et al., both examinations were performedby a single observer. Hereby, the miss rate was not liable tointerobserver variation. However, the miss rate reported inthis study was likely to be underestimated because the endo-scopist was specifically selected as the endoscopist with thelowest miss rate from a group of 26 observers.

Another potential source of variability in studies concern-ing polyp detection is the accuracy of size measurement. In1997, Schoen reported that the endoscopists’ estimates ofpolyp size are often unreliable. In 5 of 11 instances, the in-accuracy altered the classification of polyps across the 1 cmthreshold (15). There is evidence available suggesting thatthis inaccuracy can be reduced by using a calibrated lin-ear probe instead of an open forceps (16). To reduce ob-server variability, the optimal method to prospectively in-vestigate polyp miss rate, independent of other tests, wouldbe by tandem-colonoscopy performed by a single observerwithin a nonrandomized consecutive series of patients. Acommonly accepted colonoscopic protocol should be usedwith accurate reporting of the bowel preparation method, thedegree of bowel preparation, sedative use, type of scope,position of the patient and the examination time. Valuable

348 van Rijn et al.

17 / 106

81 / 298

23 / 92

8 / 20

22 / 71

26% (21 to 30)

7 / 57

6 / 48

0 / 8

2 / 5

1 / 6

13% (8 to 20)

0 / 58

2 / 32

0 / 2

0 / 1

0 / 3

2% (1 to 8)

Exam 2 / Total

Polyps 1-5mm

Polyps 5-9mm

Polyps >=10mm

Pooled

Harrison, 2004

Rex, 2003 (II)

Rex, 2003 (I)

Rex, 1997

Hixson, 1991

Pooled

Harrison, 2004

Rex, 2003 (II)

Rex, 2003 (I)

Rex, 1997

Hixson, 1991

Pooled

Harrison, 2004

Rex, 2003 (II)

Rex, 2003 (I)

Rex, 1997

Hixson, 1991

Miss rate (%)

0 20 40 60 80 100

Figure 4. Adenoma miss rate by size.

information might be lost if patients with incomplete colono-scopic examination are excluded from the study becausepolyps can be missed in the visualized part of the colon. Theproportion of these patients and the causes of cecal nonreachshould be reported. All adenomatous and nonadenomatouspolyps should be removed. In order to compare results, ade-quate description of the frequencies of histological subtypesof adenomas and nonadenomatous polyps is also essential.For reporting diagnostic studies in general, we would like torefer to the STARD-check list (17).

The included studies showed no consensus on the catego-rization of histo-pathological subtypes of nonadenomatouspolyps. Only two studies reported a specified description.Harrison et al. reported five categories: “hyperplastic polyp,”“benign mucosa,” “pseudo polyp,” “inflammatory polyp,”and “abscess.” Hixson et al. reported four categories: “hy-perplastic polyp,” “inflammatory polyp,” “lymphoid,” “fi-broma,” and “no pathological diagnosis.” In the remainingstudies there was no specific description of nonadenomatouspolyps.

The growing evidence on nonadenomatous polyps (e.g.,hyperplastic polyps) being predecessors of cancer requiresthat study reports should not be limited to adenomas only(18). Another reason for providing more information on allpolypoid lesions detected by colonoscopy (e.g., subtype, size,location, frequency), is that virtual colonoscopy does not dif-ferentiate between adenomas and nonadenomatous polyps.

In such studies, colonoscopy is used to verify all lesions de-tected by virtual colonoscopy.

Our study was limited to the colonoscopic miss of adeno-matous and nonadenomatous polyps. There are many studiessuggesting that colorectal carcinomas are also missed. In arecent study, Roberston et al. investigated the short-term oc-currence of colorectal neoplasms in 2,915 patients who hadundergone a full colonoscopy with polypectomy of at leastone adenoma (26). Six patients (0.2%) were diagnosed witha neoplasm within 24 months of the initial polypectomy. Al-though there was no specific evidence (such as in a same-daytandem study), the authors suggest that polyps or small can-cers might have been missed during the initial colonoscopyin these patients.

Several important issues concerning the colonoscopic pro-cedure could not be addressed in this review. None of thestudies reported the identification of anatomical landmarksduring colonoscopy except for the csecal valve. Therefore, wewere not able to analyze the miss rate by site. It may be use-ful to report the location of polyps in eight colonic segments(rectum, sigmoid, ascending, flexura lienalis, transversum,flexura hepatica, descending, and cecal valve-cecum), whichleads to more accurate reporting on “problem areas” in whichthe miss rate might be higher.

In most studies the examinations were performed by highlyexperienced gastroenterologists, which is reflected by the sur-prisingly high proportion of patients with a complete exami-nation to the cecum of 97%. These results are not in line withother evidence. Bowles et al. (2005) reported a large studyincluding 9,223 colonoscopies in 68 endoscopy units in theU.K. in which the target of a complete examination to thececum in 90% of the patients was not achieved (19).

Another factor that may have influenced polyp detec-tion was suboptimal bowel preparation. Two studies reportedbowel preparation on a categorical scale (excellent, good, ad-equate or poor). None of the studies reported the use of seda-tives or patient acceptance of the colonoscopic procedure.

Several techniques to improve colonoscopic polyp detec-tion are already available or currently under investigation. Itis reported that chromoendoscopy improves the detection ofsmall adenomas and facilitates the visualization of mucosallesions but the technique is time-consuming which makesit less appropriate for widespread use (23, 24). New devel-opments such as narrow-band imaging, auto-fluorescencecolonoscopy, and immunoscopy are promising techniquesand might change the standard of reference in the future,but are currently in an experimental phase (25). Availablestudies on these topics were not included in our analysis be-cause none of them used a tandem-colonoscopy to determinemiss rate.

There is no proper reference standard for determining the“true” colonoscopic miss rate. In theory, although unethical,the ultimate reference standard would be total colectomy. In2002, Postic reported a study in which the miss rate of polypsand colonic cancer was assessed by comparing colonoscopywith partial colon resection (20). The very low miss rate oflarge polyps was comparable to our results: in the resected

Polyp Miss Rate Determined by Tandem Colonoscopy 349

segments of 156 patients, an endoscopist had overlooked one10-mm polyp in the ascending colon and two>10 mm cancerswithin one patient with cancer of the sigmoid. The two latterlesions, missed in the ascending colon, were discovered dur-ing surgery only, probably because colonoscopy could not becompleted to the proximal colon.

Another method to get closer to the “true” colonoscopicmiss rate is by virtual colonoscopy guided “second lookcolonoscopy.” Recent studies using virtual colonoscopy re-ported that 11–17% of the polyps 10 mm or larger weremissed by the initial colonoscopy (14, 21, 22). Although ev-idence is still limited, these data indicate that colonoscopicmiss rate of polyps 10 mm or larger might be significantlyhigher as compared with the results of our review. We believethat, if the main objective of a study is limited to estimating thecolonoscopic miss rate, virtual colonoscopy can be consid-ered as a helpful tool. However, within studies evaluating thesensitivity and specificity of virtual colonoscopy a problemoccurs when a composite reference standard (colonoscopyguided by virtual colonoscopy) is used to evaluate the testunder evaluation (virtual colonoscopy). By using the resultsof the test under evaluation (virtual colonoscopy) to enhancethe performance of the reference standard (colonoscopy), thesensitivities of the virtual colonoscopy might influence a phe-nomenon known as incorporation bias. This generally leadsto inflated estimates of sensitivity and specificity of the testunder evaluation (6). We therefore decided to perform a sys-tematic review using studies in which the colonoscopic missrate was determined independently of any other test. Subse-quently, our results can be used to get an insight into whetherobserved sensitivities and specificities might be inflated orunderestimated in all studies using colonoscopy as an inde-pendent reference standard (5). Especially for polyps smallerthan 10 mm, the colonoscopic miss rate might have a sub-stantial impact on the accuracy of the test under evaluation. Instudies evaluating virtual colonoscopy, we support the report-ing of the performance using the reference standard beforeunblinding alongside the results after unblinding, as done byPickhardt et al. (14).

Concluding RemarksFor colonoscopy, the miss rate of large polyps appears to beacceptably low, but more evidence on possible causes of missrate is warranted. The consequences of missing smaller sizedpolyps and finding the optimal method for their detection re-main matters for future research. We would like to point outthat, in the absence of a proper reference standard, an indepen-dent assessment of miss rate of any colonoscopic techniquewill be liable to the intrinsic limitations of the technique andits susceptibility to observer variability.

STUDY HIGHLIGHTS

What Is Current Knowledge� The polyp miss rate at colonoscopy is unclear.

What Is New Here� This literature review indicates one in five polyps aremissed at colonoscopy.� Larger polyps are less likely to be missed but some bigadenomas are overlooked (2 in a 100 that are 1 cm orlarger).� One in four small adenomas (1–5 mm) may be missed.� More data are needed as the number of studies on thistopic are remarkably few.

Reprint requests for correspondence: J.C. van Rijn, M.D., De-partment of Clinical Epidemiology & Biostatistics, Room J1B-210,Academic Medical Center, Meibergdreef 9, 1105 AZ, Amsterdam,The Netherlands.

Received July 25, 2005; accepted September 26, 2005.

REFERENCES

1. Jemal A, Tiwari RC, Murray T, et al. Cancer statistics, 2004.CA Cancer J Clin 2004;54:8–29.

2. Winawer SJ, Zauber AG, Ho MN, et al. Prevention of col-orectal cancer by colonoscopic polypectomy. The NationalPolyp Study Workgroup. N Engl J Med 1993;329(27):1977–81.

3. Winawer S, Fletcher R, Rex D, et al. Gastrointestinal Con-sortium Panel. Colorectal cancer screening and surveillance:Clinical guidelines and rationale-update based on new evi-dence. Gastroenterology 2003;124:544–60.

4. Mandel JS, Bond JH, Church TR, et al. Reducing mor-tality from colorectal cancer by screening for fecal occultblood. Minnesota Colon Cancer Control Study. N Engl JMed 1993;328:1365–71.

5. Pepe MS. The Statistical Evaluation of Medical Tests forClassification and Prediction. Oxford University Press NewYork, 2004;7.3:194–8.

6. Whiting P, Rutjes AW, Reitsma JB, et al. Sources of varia-tion and bias in studies of diagnostic accuracy: A systematicreview. Ann Intern Med 2004;140:189–202.

7. Brown LD, Cai TT, DasGupta A. Interval estimation for abinomial proportion. Stat Sci 2001;16:101–33.

8. Hixson LJ, Fennerty MB, Sampliner RE, Mc Gee D,Garewal H. Prospective study of the frequency and size dis-tribution of polyps missed by colonoscopy. J Natl CancerInst 1990;82:1769–72.

9. Hixson LJ, Fennerty MB, Sampliner RE, Garewal HS.Prospective blinded trial of the colonoscopic miss-rate oflarge polyps. Gastrointest Endosc 1991;37:125–7.

10. Rex DK, Cutler CS, Lemmel GT, et al. Colonoscopic missrates of adenomas determined by back-to-back colono-scopies. Gastroenterology 1997;112:24–8.

11. Matsushita M, Hajiro K, Okazaki K, et al. Efficacy of to-tal colonoscopy with a transparent cap in comparison withcolonoscopy without the cap. Endoscopy 1998;30:444–7.

12. Rex DK, Chadalawada V, Helper DJ. Wide anglecolonoscopy with a prototype instrument: Impact on missrates and efficiency as determined by back-to-back colono-scopies. Am J Gastroenterol 2003;98:2000–5.

13. Harrison M, Singh N, Rex DK. Impact of proximal colonretroflexion on adenoma miss rates. Am J Gastroenterol2004;99:519–22.

14. Pickhardt PJ, Choi JR, Hwang I, et al. Computed to-mographic virtual colonoscopy to screen for colorectal

350 van Rijn et al.

neoplasia in asymptomatic adults. N Engl J Med 2003;349:2191–200.

15. Schoen RE, Gerber LD, Margulies C. The pathologic mea-surement of polyp size is preferable to the endoscopic esti-mate. Gastrointest Endosc 1997;46:492–6.

16. Gopalswamy N, Shenoy VN, Choudhry U, et al. Is in vivomeasurement of size of polyps during colonoscopy accurate?Gastrointest Endosc 1997;46:497–502.

17. Bossuyt PM, Reitsma JB, Bruns DE, et al. Towards completeand accurate reporting of studies of diagnostic accuracy: TheSTARD initiative. Radiology 2003;226:24–8.

18. Jass JR. Hyperplastic polyps and colorectal cancer: Is therea link? Clin Gastroenterol Hepatol 2004;2:1–8.

19. Bowles CJ, Leicester R, Romaya C, et al. Prospective studyof colonoscopy practice in the UK today: Are we adequatelyprepared for national colorectal cancer screening tomorrow?Gut 2004;53:277–83.

20. Postic G, Lewin D, Bickerstaff C, Wallace MB. Colono-scopic miss rates determined by direct comparison ofcolonoscopy with colon resection specimens. Am J Gas-troenterol 2002;97:3182–5.

21. van Gelder RE, Florie J, Stoker J. Colorectal cancerscreening and surveillance with CT colonography: Currentcontroversies and obstacles. Abdom Imag 2005;30:5–12.

22. Van Gelder RE, Nio CY, Florie J, et al. Computed tomo-graphic colonography compared with colonoscopy in pa-tients at increased risk for colorectal cancer. Gastroenterol-ogy 2004;127:41–8.

23. Brooker JC, Saunders BP, Shah SG, et al. Total colonicdye-spray increases the detection of diminutive adenomasduring routine colonoscopy: A randomized controlled trial.Gastrointest Endosc 2002;56:333–8.

24. Kiesslich R, von Bergh M, Hahn M, et al. Chromoendoscopywith indigocarmine improves the detection of adenoma-tous and nonadenomatous lesions in the colon. Endoscopy2001;33:1001–6.

25. Dekker E, Fockens P. Advances in colonic imaging: newendoscopic imaging methods. Eur J Gastroenterol Hepatol2005;17:803–8.

26. Robertson DJ, Greenberg ER, Beach M, et al. Colorectalcancer in patients under close colonoscopic surveillance.Gastroenterology 2005;129:34–41.