Search strategy and screening

For the literature search, we first considered the studies published in three review articles published in 1999 [20], 2001 [21] and 2004 [22]. This work was supplemented by a systematic search for literature in the MEDLINE and EMBASE databases using appropriate keywords - "Helicobacter pylori" combined with "occupational risk, endoscopy, gastroenterologist, healthcare worker" - for the years 1999 to 2010. We also searched through the reference lists of the chosen studies and included appropriate publications in our work.

The criteria for inclusion related to the following:

Study design: cohort study or cross-sectional study

Study population: the study investigates gastroenterologists and/or their assistants (nurses)

Exposure: the study investigates occupation as a risk factor

Languages: German and English

Study quality

The methodological quality of the studies was assessed as moderate or good. A study was categorised as moderate if it did not take into account the potential confounding effect of age and socioeconomic status or did not adequately describe the control group's origin in order to check for the potential of confounding. Very small studies with fewer than 30 subjects per group also counted as methodologically moderate. Three authors carried out the literature screening and quality evaluation independently from one other and then compared their findings. Where they disagreed, a consensus was reached by means of discussion.

Statistical analysis

The studies are differentiated into prospective incidence studies and retrospective prevalence studies. For the purposes of statistical analysis, we used the information on the number of gastroenterological personnel and the control group as a whole, and the proportion in each case that had tested positive for H. pylori. This data was used to calculate prevalence ratios as effect estimates in the case of retrospective original studies. These are described as relative risks (RR). 95% confidence intervals (95%CI) were generated to serve as statistical tests. For the purpose of meta-analysis, a combined effect estimate was calculated using the Mantel-Haenszel method for dichotomous outcomes. Stratification enabled us to conduct further differentiated analyses relating to individual occupational groups, the kind of controls, the study region and the time of publication.

Heterogeneity

We carried out a chi-square test (χ²) in order to examine the statistical heterogeneity between studies. If there was statistically significant heterogeneity (P < 0.05) the random effect model was used to calculate the combined effect estimate, otherwise the fixed effect method was used.

Sensitivity analysis

The impact of the studies on the combined effect estimate finding was tested by excluding individual studies from the analysis and examining the estimate stability [23]. The quality of the studies was also compared in this context. Where there is a clearly recognisable difference, the moderate studies can be considered separately or excluded from the analysis.

Publication bias

We first showed a possible publication bias graphically using a funnel plot. The effect estimate was plotted versus the precision of the estimate (defined as the inverse of standard error (1/SE_i)). An asymmetry of this funnel plot indicates publication bias. The funnel plot asymmetry was additionally measured following a linear regression approach on the natural logarithmic scale of the effect estimate. In this method, the standard normal deviate, defined as the effect estimate divided by its standard error, is regressed against the precision. The intercept provides a measure of asymmetry. The greater the deviation from zero, the stronger the evidence of asymmetry [24].

The analyses were carried out using Review Manager (RevMan 5) and Microsoft Excel.

Studies identified and assessment of study quality

When study quality was assessed on the basis of the abovementioned criteria, 15 studies were categorised as good [25–36, 46–48], and nine as methodologically moderate [37–45].

Meta-analysis

Pooled analysis of all 24 studies included showed that gastroenterological staff (RR 1.34; 95%CI 1.14-1.58) exhibited a significantly increased risk of H. pylori infection. Pooling of all methodologically good studies confirmed this increased risk (RR 1.52; 95%CI 1.27-1.81), whereas the methodologically moderate studies provided no evidence of higher risk.

A differentiated analysis of the studies, which drew a difference between the information on gastroenterologists, their nurses/assistants and gastroenterological personnel in general, produced a similar picture. While the studies as a whole and the methodologically good studies (Figure 1) showed statistically significant results, the moderate studies showed no difference in H. pylori prevalence between gastroenterological personnel and the controls (Table 3).

The studies chose to use different control groups, recruiting them either directly from the population at large or non-medical fields ( = without contact with patients, Table 1), or from hospital staff from other departments ( = in contact with patients, Table 2). When the studies are stratified by non-medical controls, pooled analysis showed statistically significant risks for all personnel (RR 1.74; 95%CI 1.23-2.48), for doctors (RR 1.39; 95%CI 1.09-1.77) and for nurses/assistants (RR 1.37; 95%CI 1.08-1.74). In contrast, comparison with medical controls revealed no statistically significant risks.

When the studies were differentiated by location in Europe, Asia and America/Australia, studies conducted in Asia showed a marked 50% increase in risk for all occupational groups in gastroenterological departments. Assistants were the only group observed in the European studies to be at significantly higher risk statistically, whereas doctors were at significantly higher risk in America/Australia. If only methodologically good studies are considered, there is no change in the Asian studies, whereas in Europe a statistically significant increase in risk for all personnel can be seen.

Stratification by method of diagnosis revealed no striking differences between breath test diagnosis and diagnosis by serological examination. Stratification by date of publication showed significantly higher risks for the period 1990-1999 in all groups statistically, while no increase in risk could be seen prior to 1990. For studies published in 2000 or later, statistically significant risks could be seen for endoscopy personnel in general, and for nurses/assistants.

Heterogeneity and sensitivity analysis

Testing for homogeneity revealed clear evidence of heterogeneity among most studies included in this paper. Accordingly, the random effect model was used to calculate the pooled effect estimate. A sensitivity analysis was also carried out, excluding individual studies from the meta-analysis one by one so as to show their influence on the pooled effect estimate. The relative risks thus calculated ranged from 1.26 to 1.40 and were all statistically significant.

Publication Bias

The funnel plot showed no evidence of a marked publication bias (funnel plot not shown), nor did linear regression reveal any significant funnel plot asymmetry (intercept 0.59; 95%CI -1.95-1.83).

Study area

Diagnostic method

In epidemiological studies, a H. pylori infection is mainly diagnosed by non-invasive methods such as serology or breath tests. More rarely, a stool antigen test is used. In the present studies this method was used to ascertain prevalence in only one study. In contrast, 16 studies used serological tests and seven used breath tests. After stratification by diagnostic method, the meta-analysis showed no differences in risk. The main advantage of serology is that it is a quick, low-cost method for use with large study populations. Nonetheless, this method has various limitations. These relate in part to defining a cut-off value to differentiate between positive and negative results and to the sensitivity of the test to changes in reagents and laboratory conditions, which is particularly relevant in longitudinal studies and repeat examinations. The breath test (C13-UBT) is somewhat more cost-intensive, but its sensitivity and specificity are in the order of 90-95%, and this method is just as suitable as the stool test, particularly when examining children [49, 50].

Comparison with other studies

Since the H. pylori bacterium was discovered, numerous reviews have been published on the subject. Some dealt with the occupational risk of infection run by medical personnel in general, while some examined gastroenterological personnel in particular. Matysiak-Budnik [51] showed an association between occupational exposure and an increased risk of infection. Williams [20], too, stated that there were increased occupational risks for endoscopy personnel. However, the evidence in this review appears contradictory, since the findings varied between no risk and a five times greater risk. De Schryver et al. [21, 22] were able to show in their reviews increased risks for gastroenterologists and endoscopy personnel. Magalhaes Queiroz [52] found controversial data on the occupational risk, but they considered only some of the studies also included here for gastroenterological personnel.

Strength and limitations of the review

To our knowledge, this is the only meta-analysis to date to focus on the prevalence of H. pylori infection among endoscopy personnel. The strengths of this work are those of a meta-analysis. A meta-analysis can give a comprehensive overview of the state of research. Pooling different studies on a topic increases statistical power and validity in comparison with smaller individual studies and increases the likelihood of being able to identify actual differences that exist between groups [53]. Nonetheless, this form of study also has limitations, and the result of the effect estimate can only be considered and interpreted in relation to the underlying data. One problem in this paper is the statistical heterogeneity among studies. This was taken into account first by applying the random effect model. In addition, sub-group analyses were carried out so that only studies with identical characteristics were pooled and analysed [23].

The search strategy to listed papers might introduce publication bias, because inconclusive studies might be less likely published in listed journals. However, we controlled for publication bias and found no evidence for it.

An adequate control for confounding is crucial in occupational infectious disease epidemiology, especially when transmission of infection mainly occurs in childhood. Therefore, for a study to be classified as good, it needed to give the confounding nature of age and social status due consideration. However, the control for significant confounders via selection of the control group or by adjustment could not be assessed for effectiveness.

Although the assessment of study quality was taken into account, further criteria must be heeded when pooling and calculating pooled estimates - differences in risk factors, in the observance or adjustment of confounders, and in the study population (different regions, different survey dates, different study approaches/control groups). A lack of information about age, gender, ethnicity and work-related factors, such as the use of gloves and masks or the frequency of gastroscopies performed, precludes more differentiated analyses.