Professional Documents
Culture Documents
Ford 2008
Ford 2008
com
Colorectal cancer
Colorectal cancer
adj5 ‘‘cancer’’, ‘‘colo$’’ adj5 ‘‘adenocarcinoma’’, ‘‘rectal’’ adj5 of diagnostic studies (Box 2).23 Studies were evaluated according
‘‘cancer’’ and ‘‘rectal’’ adj5 ‘‘adenocarcinoma’’ (all free text to whether assessors were blinded, cases were consecutive and
terms). These were combined using the set operator AND with whether sample size was adequate.
papers evaluating clinical features using the terms ‘‘medical
history taking’’, ‘‘anaemia’’, ‘‘iron deficiency anaemia’’, ‘‘diar-
Data synthesis and analysis
rhoea’’, ‘‘alarm’’, ‘‘weight’’ adj5 ‘‘loss’’, ‘‘altered’’ adj5 ‘‘bowel’’
Diagnoses established according to individual alarm features
and ‘‘rectal’’ adj5 ‘‘bleed$’’ (all free text terms). There were no
and statistical models were analysed separately. The primary
language restrictions, and papers published in abstract form
goal of the study was to describe the performance of the various
were eligible for inclusion in the review. Abstracts of the papers
alarm features and statistical models in distinguishing colorectal
identified by the initial search were evaluated for appropriate-
carcinoma from all other organic and functional lower GI
ness to the study question, and all potentially relevant papers
diseases. The sensitivity, specificity, positive likelihood ratio
were obtained and evaluated in detail. The bibliographies of
(LR), negative LR and 95% CI were calculated for each alarm
identified studies were used to perform a recursive search of the
literature. Authors were contacted where study data or feature using a Microsoft Excel spreadsheet (XP professional
methodology required further clarification. edition; Microsoft Corp, Redmond, Washington, USA) and
checked using StatsDirect version 2.4.4 (StatsDirect, Sale,
Cheshire, UK). Data were pooled using a random effects
Study selection
model,24 and StatsDirect was used to generate Forest plots of
Studies were required to report prospectively on unselected
pooled sensitivities, specificities, and positive and negative LRs.
cohorts of adult patients attending for investigation of lower GI
Where sufficient studies reported the utility of an individual
symptoms, and had to record symptoms prior to investigation.
alarm feature, the sensitivity and (1 – specificity) for each study
Those comparing accuracy of alarm features, a statistical model
were plotted graphically and a pooled summary receiver
or a combination of both with the results of lower GI
operating characteristics (ROC) curve constructed, and the area
investigation were eligible for inclusion (Box 1). More than
under the curve calculated, using Meta-DiSc version 1.4
90% of subjects were required to undergo lower GI investiga-
(Universidad Complutense, Madrid, Spain).
tion, defined as colonoscopy, barium enema or CT colography.
Heterogeneity between pooled studies for each alarm feature
Studies that used flexible sigmoidoscopy alone to evaluate
was assessed using the I2 statistic and x2. Where statistically
patients were only eligible if patients were followed-up for at
significant heterogeneity between studies existed (I2.25% or
least 1 year, or there was evidence that data on all potentially
p,0.10)25 for a particular alarm feature and a sufficient number
missed colorectal cancers were collected at study end. Case–
of studies were available, potential reasons for this were
control studies evaluating patients with cancer and comparing
explored informally via subgroup analyses according to study
them with patients without cancer were not included, as they
tend to bias results in favour of the diagnostic test being setting (primary care-based vs secondary care-based), number of
studied.22 Articles were independently assessed by three centres (single vs multicentre studies), type of lower GI
researchers according to the predefined eligibility criteria, with investigation used (colonoscopy vs other lower GI investiga-
disagreements resolved by consensus. tions), method of symptom data collection (questionnaire vs
clinical history vs method not reported) and study sample size
(,500 patients vs >500 patients). These are exploratory
Data extraction and quality assessment
analyses only, and the results should be interpreted with
Data were extracted by three reviewers on to predesigned forms,
caution. We compared LRs between these subgroups using the
and discrepancies in data extraction were resolved by consensus.
The quality of included studies was assessed using a meta-
analysis, which identified factors that influenced the outcome Box 2 Method used for assigning quality of evidence
Colorectal cancer
Cochrane Q statistic and, due to multiple analyses, a p value of 12 of the studies. The pooled positive and negative LRs for rectal
,0.01 was considered statistically significant. bleeding were poor, and there was statistically significant
heterogeneity between study results (table 2). The ROC curve
RESULTS indicated that rectal bleeding had a limited accuracy for diagnosing
The search strategy identified 11 169 studies of which 205 were lower GI malignancy, with an area under the curve of 0.60 (fig 2).
possibly relevant to the systematic review and retrieved (fig 1). Two studies collected more detailed information on the
Of these, 15 studies were eligible for inclusion, evaluating a total character of rectal blood loss in 4440 patients,17 18 with a pooled
of 19 443 patients,16–19 26–36 with a pooled prevalence of colorectal prevalence of colorectal cancer of 5% (95% CI 4% to 6%). The
carcinoma in all studies of 6% (95% CI 5% to 8%). The pooled prevalence of dark blood loss per rectum was 5% (95% CI
prevalence of colorectal carcinoma in individual studies varied 2% to 9%), as distinct from bright red rectal bleeding. The
from 3% to 14.6%. Polyp detection rates in individual studies, specificity of this symptom was high, which meant that the
where reported, varied between 7% and 50%. Complete colonic pooled positive LR was greater than for rectal bleeding of
imaging rates for lower GI investigations varied between 56% unspecified character (table 2).
and 98%. None of the included studies reported on operator
experience, in terms of number of previous procedures Change in bowel habit
performed, or withdrawal times for colonoscopy. Only three There were 11 studies evaluating a current change in bowel
studies reported the method of bowel preparation,28 32 35 and habit in 17 581 patients,17–19 27–30 32 34–36 with a pooled prevalence
only one study reported the quality of bowel preparation of colorectal cancer of 6% (95% CI% 5% to 8%). Only two
achieved,30 which was inadequate in 16% of patients. Other studies specified the duration of this symptom, and this was
study characteristics are provided in table 1. between 3 and 12 months.17 34 The pooled prevalence of a
Thirteen studies examined the accuracy of alarm features in change in bowel habit in the studies was 32% (95% CI 22% to
predicting colorectal carcinoma, one study reported on the 43%). The pooled positive and negative LRs for a change in
accuracy of a statistical model and one study reported on both. bowel habit were poor, with statistically significant hetero-
Symptom data were collected using a questionnaire in seven geneity between studies (table 2).
studies,17 18 26 27 29 31 34 via clinical history in three studies,16 32 33
and the method of collection was unclear in the remaining Anaemia
studies. One study used barium enema alone as the lower GI We identified seven studies reporting on the presence or absence
investigation of choice,29 and no study used CT colography of anaemia in 4404 patients,17 27 28 31 32 34 35 with a pooled
alone. Several studies included patients who required endo- prevalence of colorectal cancer of 8% (95% CI 6% to 11%). The
scopic visualisation of abnormalities detected at barium enema, pooled prevalence of anaemia was 11% (95% CI 8% to 15%).
or who were asymptomatic but undergoing follow-up or Pooled positive and negative LRs were again disappointing for
surveillance for previous colorectal carcinoma, polyps or anaemia in predicting underlying colorectal cancer (table 2).
inflammatory bowel disease.16 27 28 31 33 35 36 These groups of Four studies specifically stated that they examined the utility
patients were always excluded from our analyses, as they were of iron deficiency anaemia in predicting a diagnosis of colorectal
not relevant to the clinical question we were addressing. carcinoma in 1571 patients,28 31 32 34 though only one study
defined this precisely,34 with a pooled prevalence of cancer in
Rectal bleeding these studies of 9% (95% CI 5% to 13%). The prevalence of iron
We identified 14 studies evaluating 19 189 patients,16–19 27–36 with a deficiency anaemia was 14% (95% CI% 9% to 22%). However,
pooled prevalence of colorectal cancer of 7% (95% CI 5% to 8%). the pooled LRs did not improve over those for anaemia of
The pooled prevalence of rectal bleeding in the studies was 49% unspecified type (table 2).
(95% CI 38% to 59%), the character of which was not specified in
Weight loss
We identified five studies evaluating weight loss in 7418
patients,17 18 29 31 34 with a pooled prevalence of colorectal cancer
of 6% (95% CI% 4% to 8%). Overall, the pooled prevalence of
weight loss in the studies was 12% (95% CI% 6% to 20%).
Specificity of weight loss was generally high, meaning that the
pooled positive LR was almost 2, but the pooled negative LR
was poor (table 2).
Diarrhoea
There were five studies reporting on diarrhoea in 3904
patients,18 27 31 32 34 with a pooled prevalence of colorectal cancer
of 9% (95% CI 5% to 13%). The pooled prevalence of diarrhoea
in the studies was 20% (95% CI 6% to 38%). Both pooled
positive and negative LRs were poor, with a positive LR ,1,
suggesting that the presence of diarrhoea was actually a
negative predictor of colorectal cancer (table 2).
Abdominal mass
Two studies reported on the presence or absence of an
Figure 1 Flow diagram of assessment of studies identified in the abdominal mass on physical examination in 2465 patients,17 35
systematic review. with a pooled prevalence of cancer in the studies of 6% (95% CI
Colorectal cancer
5% to 7%). The pooled prevalence of a palpable abdominal mass statistical significance, and the positive and negative LRs
was 3% (95% CI 2% to 4%). Pooled specificity of an abdominal remained poor. The type of lower GI investigation used and
mass was high, but pooled sensitivity was very low, resulting in study sample size appeared to have little effect on the utility of
poor pooled positive and negative LRs (table 2). alarm features. Sensitivity was lower, and specificity higher, in
studies that used colonoscopy when change in bowel habit was
evaluated, but not for rectal bleeding or anaemia. The reason for
Statistical models
this is unclear, and is probably a chance finding.
We identified two studies reporting on statistical models in a
total of 2522 patients, with a pooled prevalence of colorectal
carcinoma of 4% (95% CI 3% to 5%).18 26 One of the models was DISCUSSION
designed specifically to differentiate between colorectal carci- To our knowledge this is the first systematic review and meta-
noma and other organic diseases of the colon and rectum, and analysis of the accuracy of alarm features or statistical models in
was applied prospectively,18 and the other was designed to predicting which individuals need urgent investigation to
differentiate between organic and functional lower GI disease, exclude colorectal cancer. The current study suggests that
and was applied retrospectively,26 but the reporting of data in symptoms and signs traditionally thought to have high
this study allowed us to examine the model’s accuracy in sensitivity for the diagnosis of colorectal cancer, when taken
predicting colorectal cancer compared with all other organic and individually, perform little better than chance alone in
differentiating between cancer and other organic and functional
functional lower GI disease. The models predicted a pooled
lower GI disease. Statistical models, validated in only two
prevalence of colorectal carcinoma of 44% (95% CI 36% to 51%).
studies, were also suboptimal. This is because these models were
Pooled sensitivity was high, meaning that the pooled positive
heavily weighted towards detecting all colorectal cancers, and
and negative LRs were better than for most of the individual
therefore demonstrated high sensitivity, at the expense of
symptom items (table 2).
specificity. This aspect of their design means that many
individuals would need to be investigated to detect one case
Subgroup analyses of colon cancer, and they are therefore not suitable for triaging
There were sufficient studies available to perform subgroup patients for urgent referral.
analyses for rectal bleeding, change in bowel habit and anaemia However, the current data also provide additional novel, and
of unspecified type (table 3). There were trends for a change in clinically more useful, information. The presence of a palpable
bowel habit to perform better as a predictor of colorectal cancer abdominal mass on examination and a report of dark red rectal
in secondary care studies compared with primary care studies, bleeding by the patient both had a specificity of .95% for the
and for anaemia of unspecified type to perform better in studies diagnosis of colorectal carcinoma, because the vast majority of
that collected symptom data via a questionnaire, compared individuals without cancer did not exhibit either of these
with those that did not report the method of data collection, clinical features. This means that their presence in a patient
and in multicentre studies. However, these did not achieve effectively rules in the diagnosis of colon cancer. These features
Colorectal cancer
Table 2 Sensitivity, specificity, and positive and negative likelihood ratios of individual symptom items and
statistical models
Symptom item and studies Positive likelihood Negative likelihood ratio
reporting Sensitivity (95% CI) Specificity (95% CI) ratio (95% CI) (95% CI)
Rectal bleeding
Tate 198832 64% (35% to 87%) 65.5% (56% to 74%) 1.86 (1.07 to 2.77) 0.55 (0.25 to 0.96)
Brenna 199033 96% (79% to 100%) 40% (35% to 46%) 1.61 (1.32 to 1.80) 0.10 (0.02 to 0.50)
Berkowitz 199328 79% (49% to 95%) 61% (55% to 67%) 2.03 (1.33 to 2.57) 0.35 (0.12 to 0.78)
Neugut 199336 77% (66% to 86%) 34% (31% to 37%) 1.17 (1.0 to 1.31) 0.67 (0.43 to 1.0)
Brewster 199430 33% (15% to 57%) 65.5% (61% to 70%) 0.97 (0.49 to 1.62) 1.02 (0.69 to 1.28)
Steine 199429 31.5% (19.5% to 45.5%) 85% (83% to 86%) 2.06 (1.33 to 2.98) 0.81 (0.65 to 0.94)
Cheong 200035 55% (31.5% to 77%) 72% (66% to 77%) 1.95 (1.18 to 2.82) 0.63 (0.36 to 0.93)
Morini 200127 70% (53.5% to 83%) 37% (31% to 42%) 1.10 (0.85 to 1.33) 0.82 (0.49 to 1.28)
de Bosset 200231 49% (35% to 63%) 71% (68% to 74%) 1.71 (1.23 to 2.24) 0.71 (0.53 to 0.90)
Selvachandran 200218 86% (78% to 92.5%) 31% (29% to 33%) 1.25 (1.12 to 1.34) 0.45 (0.26 to 0.72)
Panzuto 200334 44% (28.5% to 60%) 60% (53% to 66%) 1.09 (0.72 to 1.54) 0.94 (0.68 to 1.21)
Ahmed 200516 63% (45% to 78.5%) 25% (21% to 29.5%) 0.84 (0.62 to 1.04) 1.49 (0.90 to 2.24)
Bjerregaard 200717 68% (59% to 76%) 47% (45% to 49%) 1.28 (1.11 to 1.44) 0.68 (0.52 to 0.87)
Thompson 200719 71% (67% to 75%) 37% (36% to 38%) 1.13 (1.06 to 1.20) 0.78 (0.67 to 0.89)
Summary measures 64% (55% to 73%) 52% (42% to 63%) 1.32 (1.19 to 1.47) 0.76 (0.66 to 0.87)
Change in bowel habit
Tate 198832 64% (35% to 87%) 78% (70% to 85.5%) 2.98 (1.64 to 4.79) 0.46 (0.21 to 0.79)
Berkowitz 199328 7% (0.2% to 34%) 73% (67.5% to 78%) 0.26 (0.05 to 1.19) 1.27 (0.93 to 1.42)
Neugut 199336 12% (6% to 22%) 83% (80.5% to 86%) 0.72 (0.38 to 1.31) 1.06 (0.94 to 1.13)
Brewster 199430 38% (18% to 62%) 82% (77% to 85%) 0.98 (0.53 to 1.55) 1.01 (0.67 to 1.32)
Steine 199429 45% (32% to 69%) 61% (56% to 66%) 1.00 (0.72 to 1.29) 1.00 (0.76 to 1.24)
Cheong 200035 25% (9% to 49%) 81% (76% to 86%) 1.34 (0.58 to 2.66) 0.92 (0.65 to 1.11)
Morini 200127 12.5% (4% to 27%) 94% (90% to 96%) 1.95 (0.78 to 4.57) 0.93 (0.79 to 1.02)
Selvachandran 200218 86% (78% to 92.5%) 28% (26% to 29.5%) 1.19 (1.07 to 1.28) 0.50 (0.29 to 0.80)
Panzuto 200334 19.5% (9% to 35%) 79.5% (74% to 84%) 0.95 (0.48 to 1.76) 1.01 (0.82 to 1.16)
Bjerregaard 200717 63% (54% to 72%) 55% (53% to 57%) 1.40 (1.20 to 1.60) 0.67 (0.52 to 0.83)
Thompson 200719 77% (73% to 81%) 56% (55% to 57%) 1.76 (1.66 to 1.85) 0.41 (0.35 to 0.48)
Summary measures 41% (23% to 60%) 69% (58% to 78%) 1.29 (1.05 to 1.59) 0.82 (0.66 to 1.01)
Anaemia
Tate 198832 7% (0.2% to 34%) 91% (85% to 96%) 0.83 (0.14 to 4.18) 1.02 (0.75 to 1.12)
Berkowitz 199328 7% (0.2% to 34%) 84% (80% to 88%) 0.46 (0.08 to 2.09) 1.10 (0.81 to 1.20)
Cheong 200035 2% (0% to 20%) 95% (92% to 98%) 0.52 (0.05 to 4.53) 1.02 (0.85 to 1.07)
Morini 200127 15% (6% to 30%) 93% (90% to 96%) 2.24 (0.96 to 4.89) 0.91 (0.76 to 1.00)
de Bosset 200231 12% (4% to 24%) 90% (87% to 92%) 1.14 (0.52 to 2.35) 0.98 (0.85 to 1.06)
Panzuto 200334 68% (52% to 82%) 83% (77% to 87%) 3.98 (2.77 to 5.59) 0.38 (0.24 to 0.57)
Bjerregaard 200717 11% (6% to 18.5%) 90% (89% to 92%) 1.19 (0.71 to 1.95) 0.98 (0.90 to 1.03)
Summary measures 17% (5.5% to 33%) 90% (87% to 92%) 1.43 (0.75 to 2.74) 0.96 (0.87 to 1.06)
Weight loss
Steine 199429 31% (19% to 45%) 81% (79% to 83%) 1.65 (1.07 to 2.38) 0.85 (0.69 to 0.98)
de Bosset 200231 8% (2% to 19%) 97% (96% to 98%) 3.01 (1.10 to 7.85) 0.95 (0.84 to 1.00)
Selvachandran 200218 18% (11% to 27%) 92% (91% to 94%) 2.39 (1.50 to 3.67) 0.89 (0.79 to 0.96)
Panzuto 200334 37% (22% to 53%) 89% (84% to 92%) 3.24 (1.86 to 5.41) 0.71 (0.54 to 0.87)
Bjerregaard 200717 21% (14% to 30%) 79% (77% to 81%) 1.03 (0.71 to 1.43) 0.99 (0.89 to 1.08)
Summary measures 22% (14% to 31%) 89% (81% to 95%) 1.96 (1.25 to 3.08) 0.91 (0.84 to 0.98)
Diarrhoea
Tate 198832 14% (2% to 42%) 78% (70% to 86%) 0.66 (0.18 to 2.0) 1.09 (0.76 to 1.29)
Morini 200127 1% (0% to 11%) 95% (92% to 97%) 0.26 (0.03 to 2.37) 1.04 (0.94 to 1.07)
de Bosset 200231 2% (0% to 10%) 89% (87% to 91%) 0.18 (0.03 to 0.94) 1.10 (1.01 to 1.14)
Selvachandran 200218 69.5% (59% to 78.5%) 61% (59% to 63%) 1.78 (1.52 to 2.02) 0.50 (0.36 to 0.66)
Panzuto 200334 24% (12% to 40%) 69% (62% to 74%) 0.78 (0.43 to 1.31) 1.10 (0.87 to 1.30)
Summary measures 19% (1% to 54%) 80% (63% to 93%) 0.74 (0.34 to 1.62) 0.98 (0.85 to 1.13)
Iron deficiency anaemia
Tate 198832 7% (0% to 34%) 91% (85% to 96%) 0.83 (0.14 to 4.18) 1.02 (0.75 to 1.12)
Berkowitz 199328 7% (0% to 34%) 84% (80% to 88%) 0.46 (0.08 to 2.09) 1.10 (0.81 to 1.20)
de Bosset 200231 12% (4% to 24%) 90% (87% to 92%) 1.14 (0.52 to 2.35) 0.98 (0.85 to 1.06)
Panzuto 200334 68% (52% to 82%) 83% (77% to 87%) 3.98 (2.77 to 5.59) 0.38 (0.24 to 0.57)
Summary measures 23% (2%% to 57%) 87% (83% to 91%) 1.38 (0.48 to 3.94) 0.88 (0.68 to 1.13)
Abdominal mass
Cheong 200035 5% (0% to 29%) 96% (93% to 98%) 1.37 (0.23 to 7.27) 0.99 (0.79 to 1.04)
Bjerregaard 200717 4% (1% to 9%) 97% (96% to 98%) 1.50 (0.62 to 3.51) 0.99 (0.93 to 1.01)
Continued
Colorectal cancer
Table 2 Continued
Symptom item and studies Positive likelihood Negative likelihood ratio
reporting Sensitivity (95% CI) Specificity (95% CI) ratio (95% CI) (95% CI)
Summary measures 5% (2% to 9%) 97% (96% to 98%) 1.47 (0.68 to 3.19) 0.99 (0.95 to 1.02)
Dark red rectal bleeding
Selvachandran 200218 7% (3% to 15%) 97% (97% to 98%) 2.71 (1.28 to 5.57) 0.95 (0.88 to 0.99)
Bjerregaard 200717 24% (17% to 32%) 94% (93% to 95%) 4.27 (2.94 to 6.07) 0.81 (0.72 to 0.88)
Summary measures 15% (3% to 34%) 96% (93% to 98%) 3.83 (2.62 to 5.61) 0.88 (0.75 to 1.04)
Statistical models
Bellentani 199026 95% (65% to 100%) 54% (48% to 61%) 2.10 (1.47 to 2.46) 0.08 (0.01 to 0.59)
Selvachandran 200218 91% (83% to 96%) 62% (60% to 64%) 2.38 (2.16 to 2.57) 0.15 (0.08 to 0.28)
Summary measures 90% (84% to 95%) 59% (52% to 66%) 2.35 (2.16 to 2.55) 0.15 (0.08 to 0.26)
could therefore have potential value to prioritise access to negative LR of these four studies.34 This suggests that definite
urgent colonoscopy in the future. The caveat to this is that objective evidence of iron deficiency anaemia may provide
these features were reported only in secondary care-based greater accuracy in predicting a diagnosis of colorectal carci-
studies, and both the spectrum of disease and the doctor making noma, rather than anaemia of unspecified type. One study also
the diagnosis will be different from those encountered in quantified the degree of weight loss,17 though the positive and
primary care. negative LRs derived from these data remain disappointing.
There may be ways to improve the utility of other individual Another approach to improve the accuracy of alarm features
alarm features in predicting colorectal cancer. Examples include in predicting underlying colorectal carcinoma might be to
specifying the amount of weight lost (and over what period of combine them. Two of the studies included in this meta-
time); the degree of anaemia (and whether there is definite analysis used this approach.19 36 However, the sensitivity and
evidence that this is due to iron deficiency); and using more positive and negative LRs remained poor. As data reporting was
precise definitions of diarrhoea and change in bowel habit based inconsistent between the two studies that combined symp-
on either stool frequency, form or a combination of the two. Of toms, in terms of the individual symptoms that were selected
the seven studies that reported on the presence of anaemia in for combination, it was not possible to pool the results for the
this meta-analysis, only four stated that this was iron deficiency purpose of this meta-analysis in order to assess the value of this
anaemia,28 31 32 34 and only one of these reported the threshold approach further. Doctors often combine other items from the
used to define this.34 Excluding the three studies that did not clinical history, such as patient age and any relevant family
expand on the term anaemia from the meta-analysis,17 27 35 so history, with the patient’s symptoms as part of the diagnostic
that only studies examining the utility of iron deficiency process,37 and this may improve overall accuracy. This is
anaemia were pooled, had little effect on the sensitivity, essentially the approach taken by statistical models, but these
specificity, or positive and negative LR of anaemia in predicting are currently suboptimal as they have focused on achieving a
colorectal carcinoma. However, from table 2 it can be seen that high sensitivity, by predicting that between 40% and 50% of
the study providing a precise definition of what constituted iron patients presenting with lower GI symptoms will have color-
deficiency anaemia gave the best sensitivity, positive LR and ectal cancer. A health service with a limited budget would be
unable to cope with the demands placed upon it if all these
patients were referred for urgent investigation. Future models
should concentrate on maximising specificity in order to rule in
a diagnosis of colorectal cancer and prioritise urgent referral
more effectively.
This systematic review is limited by the quality of the studies
included. The majority of the studies identified, while large in
several cases, were of poor quality in terms of reporting whether
either patient recruitment was consecutive or assessors were
blinded. As studies did not routinely report quality assurance
data such as quality of bowel preparation, operator experience,
withdrawal time for colonoscopy and completion rates, the
accuracy of colonoscopy or barium enema as a diagnostic test
cannot be assessed as part of the current study. However, even
if the quality of the examinations is suboptimal in some of these
studies, this is unlikely to account for the disappointing
accuracy of most alarm features.
Subgroup analyses conducted according to study setting, type
of lower GI investigation used, method of symptom data
collection, number of study centres and study sample size failed
to have any significant impact on the LRs of individual
symptom items, where sufficient studies reported them, and
Figure 2 Receiver operating characteristics curve for accuracy of did not reveal any obvious cause for the observed heterogeneity
rectal bleeding in predicting colorectal carcinoma. AUC, area under between studies. The poor performance of alarm features in
the curve. predicting colorectal carcinoma may have been due, in part, to
Colorectal cancer
Table 3 Subgroup analyses for rectal bleeding, change in bowel habit and anaemia
Sensitivity Specificity Positive LR p Value for the Negative LR p Value for the
Symptom item and subgroup No of studies (95% CI) (95% CI) (95% CI) difference (95% CI) difference
Rectal bleeding
Study setting
Primary care-based 2 37% (26% to 49%) 73% (46% to 93%) 1.49 (0.80 to 2.77) 0.84 (0.72 to 0.98)
Secondary care-based 11 70% (61% to 77%) 51% (43% to 59%) 1.35 (1.21 to 1.50) 0.76 0.69 (0.59 to 0.81) 0.08
Lower GI investigation used
Colonoscopy 8 69% (58% to 80%) 51% (37% to 65%) 1.42 (1.16 to 1.74) 0.69 (0.51 to 0.93)
Other investigations 6 58% (43% to 78%) 55% (37% to 71%) 1.23 (1.11 to 1.36) 0.21 0.79 (0.68 to 0.91) 0.43
Symptom data collection
Via questionnaire 6 59% (42% to 75%) 56% (35% to 75%) 1.31 (1.16 to 1.49) 0.75 (0.64 to 0.87)
Not reported 5 65% (53% to 76%) 54% (40% to 67%) 1.33 (1.10 to 1.62) 0.76 (0.61 to 0.93)
Via clinical history 3 76% (49% to 94%) 43% (23% to 64%) 1.34 (0.82 to 2.17) 0.99 0.56 (0.18 to 1.78) 0.88
Number of centres
Single-centre studies 9 66% (52% to 78%) 54% (38% to 69%) 1.40 (1.19 to 1.64) 0.73 (0.58 to 0.92)
Multicentre studies 5 62% (50% to 74%) 50% (36% to 63%) 1.24 (1.10 to 1.39) 0.23 0.76 (0.66 to 0.87) 0.77
Study sample size
,500 patients 8 63% (49% to 77%) 53% (40% to 66%) 1.35 (1.07 to 1.72) 0.78 (0.57 to 1.05)
>500 patients 6 65% (52% to 77%) 51% (34% to 68%) 1.27 (1.15 to 1.41) 0.64 0.76 (0.66 to 0.82) 0.88
Change in bowel habit
Study setting
Primary care-based 2 32% (11% to 59%) 67% (41% to 89%) 0.99 (0.76 to 1.29) 1.01 (0.88 to 1.16)
Secondary care-based 9 43% (23% to 64%) 69% (56% to 80%) 1.38 (1.10 to 1.73) 0.06 0.77 (0.59 to 1.01) 0.08
Lower GI investigation used
Colonoscopy 5 22% (9% to 38%) 83% (75% to 89%) 1.26 (0.62 to 2.57) 1.00 (0.86 to 1.16)
Other investigations 6 57% (39% to 74%) 56% (45% to 66%) 1.26 (1.00 to 1.58) 0.99 0.73 (0.50 to 1.05) 0.12
Symptom data collection
Via questionnaire 5 46% (20% to 73%) 64% (44% to 81%) 1.23 (1.07 to 1.41) 0.85 (0.71 to 1.02)
Not reported 5 32% (4% to 69%) 71% (57% to 84%) 1.05 (0.63 to 1.75) 0.56 0.87 (0.57 to 1.34) 0.92
Number of centres
Single-centre studies 7 51% (32% to 70%) 62% (50% to 73%) 1.34 (1.02 to 1.76) 0.75 (0.50 to 1.12)
Multicentre studies 4 26% (5% to 56%) 79% (59% to 94%) 1.18 (0.82 to 1.70) 0.58 0.93 (0.80 to 1.08) 0.33
Study sample size
,500 patients 6 26% (14% to 41%) 79% (67% to 88%) 1.30 (0.77 to 2.20) 0.99 (0.85 to 1.14)
>500 patients 5 57% (33% to 80%) 56% (42% to 69%) 1.26 (0.98 to 1.60) 0.92 0.69 (0.44 to 1.07) 0.13
Anaemia
Lower GI investigation used
Colonoscopy 5 10% (5% to 17%) 91% (87% to 94%) 1.24 (0.74 to 2.08) 1.00 (0.96 to 1.06)
Other investigations 2 37% (0% to 91%) 87% (79% to 94%) 2.21 (0.68 to 7.19) 0.38 0.63 (0.25 to 1.58) 0.33
Symptom data collection
Via questionnaire 4 25% (6% to 50%) 90% (87% to 92%) 1.91 (0.94 to 3.91) 0.88 (0.75 to 1.03)
Not reported 2 4% (0% to 15%) 91% (77% to 98%) 0.48 (0.13 to 1.79) 0.07 1.06 (0.96 to 1.17) 0.05
Number of centres
Single-centre studies 3 5% (1% to 13%) 91% (82% to 97%) 0.59 (0.21 to 1.67) 1.05 (0.96 to 1.15)
Multicentre studies 4 25% (6% to 50%) 90% (87% to 92%) 1.91 (0.94 to 3.91) 0.07 0.88 (0.75 to 1.03) 0.08
Study sample size
,500 patients 5 17% (1% to 47%) 90% (84% to 94%) 1.57 (0.67 to 3.72) 0.91 (0.75 to 1.10)
>500 patients 2 12% (8% to 17%) 90% (89% to 91%) 1.18 (0.78 to 1.79) 0.56 0.98 (0.93 to 1.04) 0.47
LR, likelihood ratio.
the fact that the majority of included studies were based in may reflect a degree of selection bias. Eleven of the studies
secondary care. This could have created a referral bias, whereby stated that recruitment had been prospective, nine that they
primary care doctors preferentially referred only those indivi- had included consecutive patients, only one study specified that
duals in whom it was not possible to reach a clear diagnosis, investigators were blinded to patient data,29 and in four studies
after a comprehensive history and physical examination, to the clinical data were collected by the doctor performing the
studies. However, this is unlikely to be a major explanation for investigative procedure,27 31–33 though the latter would be
the findings of the current study, as the performance of these expected to bias studies towards overestimating the accuracy
individual symptoms was no better when only primary care- of alarm features in diagnosing colorectal cancer.
based studies were included in the meta-analysis. The pre- The findings of the current study demonstrate that there is
valence of colorectal cancer varied between studies, but in the little evidence to support current NICE guidelines on rapid
majority of cases was between 4% and 8%. Three studies referral for suspected colorectal cancer.14 15 The presence of a
reported a prevalence in excess of 10%,27 32 34 higher than would palpable abdominal mass and dark red rectal bleeding appeared
be expected in a Western population; two of these studies were to be the most useful individual clinical features to prioritise
conducted in Italy.27 34 The reasons for this are speculative, but access to urgent colonoscopy, due to their high specificity. The
Colorectal cancer
latter item is not utilised in current guidelines, but this is an 16. Ahmed S, Leslie A, Thaha MA, et al. Lower gastrointestinal symptoms are not
predictive of colorectal neoplasia in a faecal occult blood screen-positive population.
important observation and needs further validation in future Br J Surg 2005;92:478–81.
studies. Finally, models assessing a combination of features 17. Bjerregaard NC, Tottrup A, Sorensen HT, et al. Diagnostic value of self-reported
should be developed to focus on specificity rather than symptoms in Danish outpatients referred with symptoms consistent with colorectal
cancer. Colorectal Dis 2007;9:443–51.
sensitivity. These recommendations for research may enable 18. Selvachandran SN, Hodder RJ, Ballal MS, et al. Prediction of colorectal cancer by a
future referral criteria to better identify patients with high patient consultation questionnaire and scoring system: a prospective study. Lancet
likelihood of lower GI malignancy who warrant urgent 2002;360:278–83.
19. Thompson MR, Perera A, Senapati A, et al. Predictive value of common symptom
investigation. combinations in diagnosing colorectal cancer. Br J Surg 2007;94:1260–5.
20. Haynes RB, Sackett DL, Guyatt GH, et al. Evaluating diagnostic tests. In: Clinical
Competing interests: None. epidemiology: how to do clinical practice research. Philadelphia, PA: Lippincott,
Williams, and Wilkins, 2006:273–322.
21. The Cochrane Collaboration. The Cochrane Methods Group on systematic review
REFERENCES of screening and diagnostic tests: Recommended methods. Available at: http://www.
1. American Cancer Society. Cancer facts & figures 2007. http://www.cancer.org/ Cochrane.org/docs/sadtdoc1.2007 (accessed 13 Aug 2008).
downloads/STT/CAFF2007PWSecured.pdf (accessed 13 Aug 2008). 22. Deeks JJ. Systematic reviews of evaluations of diagnostic and screening tests. Br
2. Ferlay J, Autier P, Boniol M, et al. Estimates of the cancer incidence and mortality in Med J 2001;323:157–62.
Europe in 2006. Ann Oncol 2007;18:581–92. 23. Scherer K, Bedlack RS, Simel DL. Does this patient have myasthenia gravis? JAMA
3. Majumdar SR, Fletcher RH, Evans AT. How does colorectal cancer present? 2005;293:1906–1914.
Symptoms, duration, and clues to location. Am J Gastroenterol 1999;94:3039–45. 24. DerSimonian R, Laird N. Meta-analysis in clinical trials. Control Clin Trials
4. Keddie N, Hargreaves A. Symptoms of carcinoma of the colon and rectum. Lancet 1986;7:177–88.
1968;292:749–50. 25. Higgins JPT, Thompson SG, Deeks JJ, et al. Measuring inconsistency in meta-
5. Agreus L, Svardsudd K, Talley NJ, et al. Natural history of gastroesophageal analyses. Br Med J 2003;327:557–60.
reflux disease and functional abdominal disorders. Am J Gastroenterol 26. Bellentani S, Baldoni P, Petrella S, et al. A simple score for the identification of
2001;96:2905–14. patients at high risk of organic diseases of the colon in the family doctor consulting
6. Boyce PM, Koloski NA, Talley NJ. Irritable bowel syndrome according to varying room. Fam Pract 1990;7:307–12.
diagnostic criteria: are the new Rome II criteria unnecessarily restrictive for research 27. Morini S, Hassan C, Meucci G, et al. Diagnostic yield of open access colonoscopy
and practice. Am J Gastroenterol 2000;95:3176–83. according to appropriateness. Gastrointest Endosc 2001;54:175–9.
7. Crosland A, Jones R. Rectal bleeding: prevalence and consultation behaviour. Br 28. Berkowitz A, Kaplan M. Indications for colonoscopy. An analysis based on
Med J 1995;311:486–8. indications and diagnostic yield. S Afr Med J 1993;83:245–8.
8. Ford AC, Forman D, Bailey AG, et al. Irritable bowel syndrome: a 10-year natural 29. Steine M, Stordahl A, Laerum F, et al. Referrals for double-contrast barium
history of symptoms, and factors that influence consultation behavior. examination. Factors influencing the probability of finding polyps or cancer.
Am J Gastroenterol 2008;103:1229–39. Scand J Gastroenterol 1994;29:260–4.
30. Brewster NT, Grieve DC, Saunders JH. Double-contrast barium enema and flexible
9. Hillila MT, Farkkila MA. Prevalence of irritable bowel syndrome according to different
sigmoidoscopy for routine colonic investigation. Br J Surg 1994;81:445–7.
diagnostic criteria in a non-selected adult population. Aliment Pharmacol Ther
31. de Bosset V, Froehlich F, Rey JP, et al. Do explicit appropriateness criteria enhance
2004;20:339–45.
the diagnostic yield of colonoscopy? Endoscopy 2002;34:360–8.
10. Hungin APS, Whorwell PJ, Tack J, et al. The prevalence, patterns and impact of 32. Tate JJT, Royle GT. Open access colonoscopy for suspected colonic neoplasia. Gut
irritable bowel syndrome: an international survey of 40 000 subjects. Aliment 1988;29:1322–5.
Pharmacol Ther 2003;17:643–50. 33. Brenna E, Skreden K, Waldum HL, et al. The benefit of colonoscopy.
11. Jones R, Lydeard S. Irritable bowel syndrome in the general population. Br Med J Scand J Gastroenterol 1990;25:81–8.
1992;304:87–90. 34. Panzuto F, Chiriatti A, Bevilacqua S, et al. Symptom-based approach to colorectal
12. Talley NJ, Jones M. Self-reported rectal bleeding in a United States community: cancer: survey of primary care physicians in Italy. Dig Liver Dis 2003;35:869–75.
prevalence, risk factors, and health care seeking. Am J Gastroenterol 1998;93:2179–83. 35. Cheong KL, Roohi S, Jarmin R, et al. The yield for colorectal cancer and adenoma by
13. Wilson S, Roberts L, Roalfe A, et al. Prevalence of irritable bowel syndrome: a indication at colonoscopy. Med J Malaysia 2000;55:464–6.
community survey. Br J Gen Pract 2004;54:495–502. 36. Neugut AI, Garbowski GC, Waye JD, et al. Diagnostic yield of colorectal neoplasia
14. NICE. Referral guidelines for suspected cancer in adults and children. www.nice.org. with colonoscopy for abdominal pain, change in bowel habits, and rectal bleeding.
uk/nicemedia/pdf/CG027fullguideline.pdf (accessed 14 Aug 2008). Am J Gastroenterol 1993;88:1179–83.
15. NICE. Guidance on cancer services. Improving outcomes in colorectal cancers. 37. Thompson MR, Heath I, Ellis BG, et al. Identifying and managing patients at low risk
www.nice.org.uk/nicemedia/pdf/CSGCCfullguidance.pdf (accessed 14 Aug 2008). of bowel cancer in general practice. Br Med J 2003;327:263–5.
These include:
References This article cites 32 articles, 9 of which can be accessed free at:
http://gut.bmj.com/content/57/11/1545.full.html#ref-list-1
Email alerting Receive free email alerts when new articles cite this article. Sign up in
service the box at the top right corner of the online article.
Notes