4.2 Referral guidance for endoscopy at presentation
4.2.4 Evidence review [update 2014]
The aim of this question was to identify patients who have had treatment for dyspepsia or GORD without previously having had anendoscopy, or who have not had an endoscopy in the past 12 months. This includes patients who are still symptomatic or have other newly onset signs and symptoms following lifestyle advice, and/or H pylori test and treat and/or empirical PPI treatments. This question was not looking at the use of endoscopy to assess the outcomes of interventions for dyspepsia or GORD.
A systematic search was conducted (see appendix C) which identified 5097 references. After removing duplicates the references were screened on their titles and abstracts and 58
references were obtained and reviewed against the inclusion and exclusion criteria (appendix C).
Overall, 56 studies were excluded as they did not meet the eligibility criteria, such as study design or relevant controls or interventions. A list of excluded studies and reasons for their exclusion is provided in appendix G.
The 2 remaining studies did meet the eligibility criteria and were included. Data was extracted into detailed evidence tables (see appendix D) and are summarised in table 11 below.
The overall quality of the 2 included studies was very poor quality and therefore with very low confidence in the effect estimates (predictors/risk factors). Both studies were retrospective cross-sectional studies.
Issues on study design
There were a number of methodological issues with the included studies that might contribute to substantial risk of bias, for example:
The 2 included studies were retrospective studies, which indicated that the predictive variables (risk factors/indicators) selected to be studied were driven by what data was routinely collected locally, rather than a set of pre-defined risk factors/predictors of interest to be investigated (that is, studies were data driven by local available data collection).
Update 2014
The characteristics of the study population of both included studies were unclear, which indicated that the results may not be generalisable to the UK’s ‘uninvestigated dyspepsia’
population.
Both included studies did not have long-term follow-up to investigate the downstream patient outcomes based on the endoscopic findings (for example, whether differential diagnosis has been confirmed; whether the treatment plan or strategy has been reviewed based on the endoscopic findings; whether there was symptomatic improvement).
As well as issues on study design, the included studies also suffered a number of limitations on statistical analysis. For example:
Both included studies used multivariate analyses (logistic regression) to analyse collected data. However, different predictive variables (risk factors/indicators) were included in different studies in the regression models. The two studies didn’t use the same set of risk factors/indicators in the regression model.
Some predictive variables (risk factors/indicators) have different thresholds and different references in different studies.
Only 1 of the 2 included studies carried out model diagnostics for the regression model.
For example (key diagnostics):
- Assumptions of normality and homoscedasticity were not tested.
- Multicollinearity was not assessed.
- Model fit (goodness-of-fit) was not assessed.
Due to all the above methodological and statistical issues, meta-analyses on individual predictors were not appropriate. However, the evidence was synthesized using a modified- GRADE approach to aid decision making. The criteria used in the modified-GRADE
approach were adapted from the Hayden et al. (2006) QUIPS checklist for prognostic study (please see appendix C, section C3 for the summary of the modified GRADE approach).
As the only 2 included studies have different predicted endpoints, it was considered
misleading to have presented the evidence by outcomes (predictors/risk factors), therefore the evidence was presented as individual study.
Update 2014
Table 11: Summary table of included studies Study
reference Population
Risk factors/ signs &
symptoms Control Follow-up Outcomes Author Conclusions
Lieberman (2004) ID: 758
Two distinct groups: (1) Reflux dyspepsia included patients with reflux symptoms, and (2) non-reflux dyspepsia included patients with upper abdominal pain or discomfort who did not have reported reflux symptoms, dysphagia, or known Barrett’s
esophagus, were identified.
Indications for examination included presence or absence of alarm symptoms, the potential alarm symptoms in patients with dyspepsia were defined as:
Weight loss
Vomiting
Evidence of GI bleeding (suspected upper GI bleed, hematemesis, melena, anaemia, or iron deficiency)
Reflux symptoms
Race and ethnicity (data only available in 85.0% of the procedures)
Gastric or duodenal ulcer at endoscopy were the endpoints of the logistic regression.
N/A Retrospective data between 2000 and 2002, no follow-up of patient’s outcomes post 2002.
Age, gender, race ethnicity (Black non- Hispanic, Hispanic), reflux symptoms, vomiting (with or without reflux), and bleeding were
significant predictors of gastric or duodenal ulcer for patients with
‘dyspepsia’ undergoing endoscopy. (for more details please see modified-GRADE profiles).
Although limited to patients with dyspepsia who receive endoscopy, these data provide an interesting profile of this group. These data cannot be generalisd to the general population of patients with dyspepsia symptoms, most of whom never have endoscopy.
The benefits of endoscopy in patients less than 50 years of age without alarm symptoms are uncertain and require further study.
Voutilainen (2003) ID: 1029
All patients with
‘dyspeptic symptoms’
sent for upper GI endoscopy in a hospital by GPs between 1 January and 31 December 1996. The study excluded:
Those had H pylori eradication therapy or oesophagogastric surgery
Those underwent endoscopy owing to sinister symptoms and signs suggestive of acute GI bleeding or for follow-up endoscopy.
Variables (signs, symptoms, risk factors, indicators) that were entered in the multivariate analyses were:
Age
Gender
H pylori infection
Alarm symptoms (anaemia, weight loss, dysphagia, vomiting)
High/low referral area
N/A Retrospective data in 1996, no follow-up on patient’s outcomes post 1996.
Gender and H pylori infection were
significant predictors of duodenal ulcer.
H pylori infection and alarm symptoms were significant predictors of gastric ulcer.
Age was significant predictor of gastric polyp, while gender and H pylori infection were significant predictors of not having gastric polyp.
(for more details please see modified-GRADE profiles).
This was a cross-sectional uncontrolled study with probable selection bias: GPs may have referred older patients for endoscopy more often than younger ones, the latter being treated
empirically. In conclusion, the present study revealed that alarm symptoms are strongly associated with significant endoscopic findings, such as gastric ulcer and cancer.
Update 2014
Table 12: Modified GRADE profiles: Predictors of gastric or duodenal ulcer for patients with ‘dyspepsia’ undergoing endoscopy Lieberman (2004)
Predict ors
Adjusted RR (95%CI)
Predict ors
Adjusted RR (95%CI)
Predictors Adjusted RR (95%CI)
Predictors Adjusted RR (95%CI)
Predictors Adjusted RR (95%CI)
Predictors Adjusted RR (95%CI) Age*
40–49 50–59 60–69
1.27 (1.08 to
1.50) 1.46 (1.25 to
1.71) 1.94 (1.66 to
2.28)
Gender
Male 1.14
(1.03 to 1.27)
Race/ethni city*
Black-NH Hispanic Asian/Pacifi c Islander- NH Native American- NH
1.20 (1.02 to
1.41) 1.26 (1.09 to
1.46) 1.15 (0.86 to
1.52) 1.01 (0.65 to
1.57)
Reflux symptoms Reflux symptoms
0.34 (0.31 to
0.39)
Vomiting- reflux interaction Vomiting-no reflux Vomiting- reflux
1.48 (1.24 to
1.77) 2.58 (1.83 to
3.65)
Bleeding cluster**- gender interaction Bleeding cluster-females Bleeding cluster-males
2.38 (1.97 to
2.88) 3.35 (2.80 to
4.00)
Footnote:
*Reference for Age = <40; reference for Race/ethnicity = White NH
**Bleeding cluster = defined as suspected upper GI bleeding, hematemesis, melena, anaemia, or iron deficiency NH = non-Hispanic
Modified GRADE
Risk of bias Serious1
Indirectness Serious2
Inconsistency N/A
Imprecision No serious
Other considerations Serious3
CONFIDENCE Very low
Footnote:
1 = Downgraded 1 level: retrospective study and did not control for potential confounding factors.
2 = Downgraded 1 level: unclear study population – not reported whether the study population was ‘uninvestigated dyspepsia’, not sure the study population is generalizable.
3 = Downgraded 1 level: no follow-up data that investigated the patient outcomes based on the endoscopic findings.
N/A = Not applicable (single study)
Update 2014
Table 13: Modified GRADE profiles: Predictors of duodenal ulcer, gastric ulcer and gastric polyp for patients with ‘dyspepsia’ undergoing endoscopy
Voutilainen (2003)
Predictors Duodenal ulcer Adjusted OR (95%CI)
Predictors Gastric ulcer
Adjusted OR (95%CI)
Predictors Gastric polyp
Adjusted OR (95%CI) Age (per decade)
Male
H pylori infection Alarm symptoms High referral rate
- 1.6 (1.1 to 2.2) 3.9 (2.7 to 5.5)
- -
Age (per decade) Male
H pylori infection Alarm symptoms High referral rate
- - 2.6 (1.9 to 3.5) 2.0 (1.4 to 2.7)
-
Age (per decade) Male
H pylori infection Alarm symptoms High referral rate
2.0 (1.1 to 3.5) 0.5 (0.3 to 0.9) 0.3 (0.1 to 0.6)
- 1.7 (1.0 to 2.8) Footnote:
High referral rate = ≥3.3/1000/year
Alarm symptoms = anaemia, weight loss, dysphagia, vomiting
Modified GRADE
Risk of bias Serious1
Indirectness Serious2
Inconsistency N/A
Imprecision Serious3
Other considerations Serious4
CONFIDENCE Very low
Footnote:
1 = Downgraded 1 level: retrospective study and did not control for potential confounding factors.
2 = Downgraded 1 level: unclear study population – not reported whether the study population was ‘uninvestigated dyspepsia’, not sure the study population is generalizable.
3 = Downgraded 1 level: no model diagnostics or validation.
4 = Downgraded 1 level: no follow-up data that investigated the patient outcomes based on the endoscopic findings.
N/A = Not applicable (single study)
Update 2014