Meta-analysis: the diagnostic value of alarm symptoms for upper gastrointestinal malignancy


Dr G. A. J. Fransen, Department of Gastroenterology and Hepatology, University Medical Center St Radboud Nijmegen, Internal code: 547, PO box 9101, 6500HB Nijmegen, the Netherlands.


Background : With the advent of empirical treatment strategies for patients with dyspeptic symptoms, it becomes increasingly important to select patients with a high risk of having cancer for immediate endoscopy. Usually alarming symptoms are used for this matter, but their diagnostic value is by no means clear.

Aim : To investigate the diagnostic value of alarm symptoms for upper gastrointestinal malignancy.

Methods : Meta-analysis of studies describing prevalence of alarm symptoms in patients with and without endoscopically verified upper gastrointestinal malignancy were identified through a Medline search. The prevalence, pooled sensitivity, specificity, positive and negative predictive values were calculated.

Results : About 17 case studies and nine cohort studies were selected. The mean prevalence of gastrointestinal malignancies in the cohort studies was 2.8% of 16 161 patients. Five cohort studies indicated that 25% of the patients diagnosed with upper gastrointestinal malignancy had no alarm symptoms. The pooled sensitivities of individual alarm symptoms varied from 9 to 41%, the pooled positive predictive value ranged from 4.6 to 7.9%, and was 5.9% for ‘having any alarm symptom’. The pooled negative predictive value was 99.4% for ‘having any alarm symptom’.

Conclusion : The risk of upper gastrointestinal malignancy in any individual without alarm symptoms is very low, but approximately one in four patients with upper gastrointestinal cancer have no alarm symptoms at the time of diagnosis.


With the advent of empirical treatment strategies (like Helicobacter pylori test-and-treat or acid suppressive therapy) for patients with dyspeptic symptoms, fewer patients have upper gastrointestinal (GI) endoscopy.1–3 However, it is important not to delay diagnosis in patients with underlying malignant disease. Therefore, it becomes increasingly important to select patients with a high risk of having cancer for immediate endoscopy in order to safely treat the remaining patients empirically. On the contrary, this selection should be specific in order to prevent large number of people having unnecessary endoscopy, which is costly and troublesome to the patients. Usually alarm symptoms such as weight loss, dysphagia, nausea, vomiting and anaemia or bleeding are used for this matter.4–6 However, although generally accepted, the diagnostic value of alarm symptoms is by no means clear, in spite of the amount of research conducted in this area.

Recently Boldys et al.7 concluded that weight loss and melaena are certainly indices of more advanced cancer, and in order to diagnose gastric cancer at an early stage the occurrence of these symptoms should not be awaited. On the contrary, Meineche-Schmidt and Jørgensen8 did not find a higher risk of cancer in patients with alarm symptoms in a large cohort comprising more than 600 patients.

A difficulty in the identification of factors related to diagnosis of upper GI malignancies is the low incidence of these cancers. As a result, large cohorts need to be studied, which often is not feasible. This problem may be overcome by performing meta-analysis. The aim of this study was to evaluate the diagnostic value of alarm symptoms for upper GI malignancy by means of a meta-analysis.


Identification and selection of publications

A literature search was performed in August 2003. Relevant publications were identified in Medline using key words such as ‘alarm’, ‘sinister’, ‘dyspepsia’ and ‘gastrointestinal malignancy’. Furthermore, additional publications were retrieved by reviewing references of selected studies. In order to be selected, publications had to fulfil the following criteria: (i) the presence or absence of upper GI malignancy was established endoscopically, and (ii) the number of patients and prevalence of alarm symptoms in patients with and without upper GI malignancy were described or could be calculated. To minimize selection bias all abstracts were assessed by two independent reviewers; in case of a difference of opinion a third independent reviewer was consulted.

Data analysis and statistical methods

Literature search showed two types of studies eligible for inclusion: cohort studies and case studies. Cohort studies included patients with and without upper GI malignancy and measured the alarm symptoms prospectively (i.e. before endoscopic examination), while case studies included only patients with confirmed gastric or oesophageal malignancy.

Sensitivity was calculated for cohort studies and case studies separately, by pooling the results of the individual studies using an approximation to the inverse variance approach.9 The estimate of the overall sensitivity is the sum of all true positives divided by the sum of all patients with upper GI cancer in the pooled studies. Additionally, for the cohort studies, overall specificity, and overall positive/negative predictive values (PPV/NPV) were calculated similarly. Analyses were performed for individual alarm symptoms separately (dysphagia, weight loss, bleeding, anaemia, nausea/vomiting) and, whenever possible, for having ‘any alarm symptom’ (vs. no alarm symptoms at all). Missing values were excluded from analyses.


Literature search identified 305 papers. After initial review, 39 papers were retrieved to be studied in detail. Of these papers, 13 were excluded because either the prevalence of alarm symptoms or malignancy was not clearly described or because not all patients underwent upper GI endoscopy. The remaining 26 publications, 17 case studies and nine cohort studies, were included in analyses.

The case studies described 1552 patients with upper GI malignancy (Table 1a), mostly identified through retrospective review of hospital records. The pooled sensitivities for the individual alarm symptoms (dysphagia, anaemia/bleeding, nausea/vomiting and weight loss) were respectively 30%, 27%, 21%, and 41% (Table 2). When looking at having ‘any alarm symptom’ the pooled sensitivity was 95%. Overall 10 case studies described 63 (6%) patients diagnosed with upper GI malignancy without any alarm symptom.

Table 1.  Characteristics of selected publications: (a) case studies; (b) cohort studies
AuthorYear of patient selectionLand of patient selectionPopulation NMean age (range)
(a) Case studies
Canga and Vakil201990–2000USAIn teaching hospital identified from electronic database, all ADV34168 (22–94)
Christie et al.211986–1992UKOpen-access endoscopy unit in district hospital25? all < 55
Sue-Ling et al.221970–1990UKPatients operated for EGC in university hospital reviewed4669* (38–86)
Gillen and McColl231989–1993UKHospital patient records, patients identified by means of West of Scotland Cancer Registry7650* (31–54)
Gillen and McColl237351* (37–54)
Eckardt et al.241977–1986GermanyRecords of out-patients who had an endoscopy resulting in a histological diagnosis of gastric cancer241EGC 62*; ADV 59*
Houghton et al.25Since 1965UKUniversity hospital records of patients with EGC3567* (44–79)
Carter et al.261974–1982USAPatients operated for gastric cancer in university hospital reviewed31EGC 44 (26–61); ADV 67 (37–83)
Ballantyne et al.271978–1985UKPatients operated for EGC in university hospital reviewed2074* (51–87)
Oleagoita et al.281975–1985SpainPatients operated for EGC in university hospital reviewed14255 (19–82)
Moreaux and Bougaran291965–1995FranceHospital patients operated for EGC10161 (28–82)
Gozzetti et al.301974–1985ItalyPatients operated for EGC in university hospital reviewed4959 (29–90)
Pinto et al.311977–1989ItalyPatients admitted for EGC in university hospital reviewed14264 (34–84)
Thong-Ngam et al.321994–1998ThailandHospital patient records11960 (22–91)
Ribichini et al.331971–1983ItalyPatients operated for EGC reviewed6063 (34–84)
Saubier et al.341970–1980FranceHospital patient records3060 (31–84)
Traynor et al.35 ? 10-year periodIrelandPatients operated for EGC reviewed1157 (35–70)
Heughan et al.361972–1980CanadaHospital patient records10? (43–76)
AuthorYear of patient selectionLand of patient selectionPopulationCancer prevalence (N cases/N)Mean age (range)
  1. EGC, early gastric cancer; ADV, advanced gastric cancer; GP, General Practitioner; GI, gastrointestine.

  2. * Median.

  3. † Gillen and McColl23 describe 76 patients with oesophageal cancer and 73 patients with gastric cancer, which are analysed separately.

(b) Cohort studies
Melleney and Willoughby101999–2000UKGP referral and selection by specialist (one-stop-dyspepsia clinic)7.1 (6/84)59* (15–84)
Manes et al.371998ItalyAll patients referred for endoscopy0.8 (6/706)47 (15–86)
Sung et al.381998–1999Hong KongGP referral for dyspepsia, endoscopy in regional hospital0.8 (23/2627)48
Numans et al.391986–1988The Nether- landsGP referral for recurrent and/or long-lasting episodes and/or alarming symptoms2.4 (21/861)?
Boldys et al.71983–1993PolandPatients with symptoms suggesting cancer were referred to secondary referral unit9.7 (83/860)44* (IQR: 35–54)
Adang et al.401989–1990The Nether- landsAll patients referred for endoscopy by GP and hospital specialist2.4 (70/2900)56 (3–96)
Wallace et al.19? (average of 5 years)USAPatients referred to open-access endoscopy unit: four units were academic and primary care referral centres, two were military centres2.1 (81/3815)47 (s.d.: 18)
Johannessen et al.411985–1987NorwayOut-patients referred to open-access endoscopy unit, mostly by GP1.0 (?/930)50 (s.d.: 14)
Voutilainen et al.421996FinlandGP referral for upper GI endoscopy0.5 (17/3378)58 (IQR: 25)
Table 2.  Sensitivity of alarm symptoms in case studies
StudyNumber of patientsTrue positiveFalse negativeSensitivity (%)
  1. * Gillen and McColl23 describe 76 patients with oesophageal cancer and 73 patients with gastric cancer, which are analysed separately.

 Canga and Vakil2034115518645
 Christie et al.212581732
 Gillen and McColl23*76185824
 Gillen and McColl23*73621185
 Thong-Ngam et al.3211981117
634Pooled sensitivity (95% CI)40 (36–43)
Weight loss
 Canga and Vakil2034114519643
 Christie et al.2125141156
 Gillen and McColl23*76472962
 Gillen and McColl23*73462763
 Sue-Ling et al.2246103622
 Eckardt et al.242418615536
 Houghton et al.2535221363
 Carter et al.2631183137
 Moreaux and Bougaran291014974
 Gozzetti et al.3049133627
 Pinto et al.31142588441
 Ribichini et al.3360273345
 Thong-Ngam et al.32119645554
 Traynor et al.35114736
 Heughan et al.36101910
1350Pooled sensitivity (95% CI)41 (37–44)
Bleeding and/or anaemia
 Canga and Vakil203418725426
 Christie et al.212571828
 Houghton et al.2535152043
 Heughan et al.36104640
401Pooled sensitivity (95% CI)27 (24–31)
 Gillen and McColl23*7661430
 Gillen and McColl23*7332810
 Sue-Ling et al.2246132285
 Eckardt et al.24241146218
 Carter et al.26312713
 Ballantyne et al.272054410
 Oleagoita et al.281425965
 Moreaux and Bougaran2910171355
 Gozzetti et al.304954111
 Pinto et al.316075312
 Saubier et al.343042613
 Thong-Ngam et al.32119348529
988Pooled sensitivity (95% CI)11 (8–13)
 Gillen and McColl23*7631715
 Gillen and McColl23*73175922
 Ballantyne et al.27204695
 Moreaux and  Bougaran2910111001
 Saubier et al.34301293
300Pooled sensitivity (95% CI)9 (6–11)
 Canga and Vakil203415328816
 Gillen and McColl23*76212546
 Gillen and McColl23*733820316
 Sue-Ling et al.224692626
 Eckardt et al.24241274936
 Houghton et al.2535264736
 Carter et al.263151525
 Ballantyne et al.272091336
 Oleagoita et al.281423710526
 Gozzetti et al.3049112035
 Pinto et al.31142212843
1196Pooled sensitivity (95% CI)21 (18–25)
Any alarm symptom
 Canga and Vakil203413241795
 Christie et al.212524196
 Gillen and McColl23*7671593
 Gillen and McColl23*73730100
 Sue-Ling et al.224642491
 Oleagoita et al.28142136696
 Moreaux and Bougaran2910195694
 Pinto et al.31142137596
 Saubier et al.343026487
 Ribichini et al.3360491182
 Traynor et al.35117464
1047Pooled sensitivity (95% CI)94 (92–96)

In the cohort studies, a total of 16 161 patients were described with a mean prevalence of upper GI malignancy of 2.8% (Table 1b). Most of the studies described patients referred for endoscopy by the general practitioner, the reasons for referral and way of patient selection varied between the studies. For example, the studies by Melleney and Willoughby10 and Boldys et al.7 included patients with a high suspicion of upper GI malignancy, which may explain the relatively high prevalence of upper GI malignancy. The pooled sensitivities for dysphagia, anaemia/bleeding, nausea/vomiting, weight loss, and ‘any alarm symptom’ were respectively 25%, 17%, 27%, 24% and 75% (Table 3). The pooled PPVs were respectively 6.6%, 4.6%, 7.5%, 7.9% and 5.9%.

Table 3.  Sensitivity, specificity and predictive value of alarm symptoms in cohort studies
 AuthorNTPFPFNTNSensitivity (%)Specificity (%)PPV (%)NPV (%)
  1. TP, true positive; FP, false positive; FN, false negative; TN, true negative; PPV, positive predictive value; NPV, negative predictive value.

DysphagiaSung et al.382627128222576499399.2
Numans et al.398611319986156276698.7
Adang et al.402900141695626112094897.9
 7058Pooled estimate25946.698.8
  95% Confidence interval17–2393–944.2–9.098.5–99.1
Weight lossSung et al.38262731620258813991699.2
Numans et al.398611419576456777798.9
Adang et al.4029001712653265424951298.0
Johannessen et al.41930724426837826399.7
Boldys et al.786091747761199.99091.3
 8178Pooled estimate24937.997.9
  95% Confidence interval18–3092–935.8–10.097.6–98.2
Bleeding and/or anaemiaSung et al.382627179222525497199.1
Numans et al.39861362167321692597.9
Boldys et al.786015606871718922091.3
Adang et al.402900155115522692182397.6
 7248Pooled estimate17904.697.5
  95% Confidence interval12–2389–903.0–6.097.1–97.5
Nausea/vomitingBoldys et al.7860201456363224811290.9
Numans et al.398618199136143876497.9
 1721Pooled estimate27787.594.3
  95% Confidence interval18–3576–804.9–10.293.0–95.5
Any alarm symptomMelleney and Willoughby108453714183531297.6
Manes et al.37706434266667951199.7
Boldys et al.7860613342244373571595.3
Voutilainen et al.423378121092522697168199.8
Sung et al.382627191094249583961599.8
 7655Pooled estimate75795.999.4
  95% Confidence interval67–8278–804.8–7.099.2–99.6
Age > 45, male, anaemia or bleedingWallace et al.19381574286678689177399.2

The pooled NPV for the individual alarm symptoms (dysphagia, anaemia/bleeding, nausea/vomiting and weight loss) were 98.8%, 97.9%, 97.5%, and 94.3 respectively, for ‘having any alarm symptom’ it was 99.4%. This means that there is approximately a 1:175 chance of missing malignancy in patients without alarm symptoms.

Overall, 16 studies (2238 patients) differentiated between early and advanced stages of upper GI malignancy: 710 patients had early stage cancer and 743 had advanced stage cancer.


Alarm symptoms are generally accepted as an indication for direct endoscopy. This is reflected in several guidelines for treatment of patients with dyspeptic symptoms1–4 and in the exclusion criteria for patient selection of most trials.11, 12 As a result, the usefulness of alarm symptoms for selection of patients with a higher risk of having upper GI malignancy is rarely questioned. However, the existing evidence does not consistently support the usefulness of alarm symptoms for this matter.

Therefore, we performed a meta-analysis to study the diagnostic value of alarm symptoms. Our results show that the sensitivity of symptoms indicating malignancy is rather disappointing, both for individual alarm symptoms and for having ‘any alarm symptom’. In addition to this, the pooled PPV is relatively low, indicating that only a small proportion of the patients with alarm symptoms does have cancer. On the contrary, the high NPV for having ‘any alarm symptom’ shows that there is little chance of missing a malignancy in patients without alarm symptoms, although this partially caused by the low prevalence of upper GI malignancy.

Moreover, it should be pointed out that the patients studied in the papers included show some selection. This is obvious for the case studies, but the cohort studies also show a degree of selection. This is reflected by the relatively high prevalence of upper GI cancer, with a mean of almost 3%, ranging from 0.5 to 9.7%. However, in most western countries the prevalence of upper GI malignancy is below 0.5%. Therefore, in the general population, both the sensitivity and the PPV of alarm symptoms will be worse, although the NPV will remain high.

In addition to this, the prevalence of alarm symptoms is high in patients consulting their general doctors for dyspeptic symptoms. This is illustrated by Meineche-Schmidt and Jørgensen,13 who showed that 10% of primary care patients with dyspeptic symptoms had one or more alarm symptoms, whereas only 3% of investigated patients with alarm symptoms were found to have cancer.

Combining the high prevalence of alarm symptoms with their relatively poor sensitivity and PPV leads to the following conclusion: if guidelines are strictly followed, a large number of patients will have endoscopy, whereas a relatively low number of cancers will be identified, and, although the absolute chance of missing malignancy because of the absence of alarm symptoms is small, relatively many malignancies (up to 25%) will be missed. This is supported by two studies in the United Kingdom that compared short-term endoscopies for high-risk patients with open-access endoscopies.14, 15 The results from both studies showed that relatively few patients with cancer were identified in the ‘high-risk’ group while most patients with cancer were found in the open-access group.14, 15

Furthermore, several studies showed that in patients with upper GI cancer the occurrence of alarm symptoms usually indicates more advanced cancer with worse prognosis.7, 14–17 Therefore, it is questionable whether prompt endoscopy improves prognosis for patients with alarm symptoms. This is confirmed by Martin et al.18 who did not find a relationship between delay in diagnosis and prognosis for patients with gastric cancer. Unfortunately, most studies included in this meta-analysis did not provide sufficient information about the stage of the cancers detected in relation to the presence of alarm symptoms. Therefore, it was not possible to perform a proper subgroup analysis.

Possibly, a better diagnostic tool might be created by combining a selection of alarm symptoms with patient characteristics. One of the cohort studies showed a sensitivity of 91% and a specificity of 77% when using a combination of age over 45 years, male gender and anaemia or bleeding.19 This result indicates that the combined presence of alarm symptoms, especially in combination with other factors-like age, gender or smoking might be a better tool for selection of high-risk patients. More research is necessary to define such a tool.

In conclusion, although alarm symptoms are generally accepted as an indication for prompt endoscopy, our results show that the diagnostic value of alarm symptoms is not optimal. Other factors such as age and family history should be taken into account when estimating a patient's risk of having upper GI cancer.