Primary outcomes: The primary aim of this systematic review and meta-analysis was to assess the total number of adverse events occurring following treatment for H pylori with bismuth compounds, either alone, or in combination with antibiotics and/or acid suppression therapy, compared to treatment with antibiotics alone, acid suppression therapy alone, a combination of the two, or no treatment/placebo.

Secondary outcomes: The secondary aims were to evaluate the number of specific individual adverse events occurring and the number of withdrawals of therapy due to adverse events, and to assess the effect of long-term (defined as 1 mo or more) therapy on number of adverse events (both total number and by specific category) and withdrawals due to adverse events.

Types of studies: In order to best estimate adverse events that were directly attributable to the use of bismuth, studies were only eligible for inclusion in this systematic review if they were RCTs that compared bismuth monotherapy with either acid suppression therapy alone, placebo, or no treatment, or compared bismuth compounds in combination with either antibiotics, or antibiotics and acid suppression therapy as part of a recognised efficacious eradication regimen with an identical dose and duration of antibiotics either alone or in combination with acid suppression therapy. We defined an efficacious bismuth-containing eradication regimen as any one of: bismuth triple therapy (bismuth in combination with two antibiotics); bismuth quadruple therapy (as for triple therapy, but with the addition of acid suppression therapy); or ranitidine bismuth citrate dual (with one antibiotic) or triple (with two antibiotics) therapy.

Types of participants: Patients were required to be H pylori-positive adults (over the age of 16 years) taking any bismuth compound for more than 1 d with a comparison group of H pylori-positive patients who were not taking bismuth.

Types of assessment: Bismuth toxicity had to be assessed and recorded using one or more of the following methods: medical databases; face-to-face interviews; telephone interviews; symptom diaries; or questionnaire in order for studies to be eligible for inclusion. The questionnaire used was not required to be previously validated but, if there were sufficient studies using questionnaires, we aimed to assess the impact of this in a sensitivity analysis.

Types of outcome measures: The proportion of patients that reported any adverse event and the proportion experiencing specific individual adverse events were assessed wherever trial reporting allowed this.

Search strategy: Two authors performed searches of the medical literature to identify articles from MEDLINE (from 1966 up to October 2007), EMBASE (from 1988 up to October 2007), and the Cochrane Library and Current Contents electronic databases. RCTs using bismuth salts were identified using the medical subject heading term “bismuth”. These studies were combined using the set operator and with papers that used a variety of free text terms including “Denol”, “Pepto-Bismol”, “bismuth”, “subsalicylate”, “tripotassium dicitrato bismuthate”, “subnitrate”, “subgallate”, “ranitidine bismuth citrate”, “pylorid”, “quadruple therapy”, “pylera”, and “bismuth subcitrate potassium”. There were no language restrictions, and papers published in abstract form only were also eligible for inclusion in the review. The abstracts of all papers identified by the initial search were evaluated for appropriateness to the study question, and all potentially relevant studies were retrieved and examined in greater detail to determine whether or not they met all eligibility criteria. The bibliographies of identified studies were then used to perform a recursive search of the literature to identify other potentially eligible studies. In addition, Digestive Disease Week, United European Gastroenterology Week, and European H pylori Study Group conference abstract books between 2000 and 2007 were hand-searched.

Selection of studies: Two reviewers screened all titles and abstracts of trials that were identified by the search strategy as being potentially eligible for inclusion in the systematic review to confirm or refute eligibility. This was performed using pre-designed eligibility forms. A third reviewer adjudicated where any disagreements arose, and a consensus view was taken.

Assessment of study quality: The quality of studies was assessed according to the following pre-defined criteria: method of assessment of occurrence of adverse events (interview, diary, and questionnaire), generation of randomisation schedule, method of allocation of concealment, and blinding of assessor as to patient allocation to therapy.

Data concerning total number of adverse events and number of specific individual adverse events were extracted on to specially developed forms by two reviewers and all data extraction was checked by a third reviewer. These verified data were then entered onto a Microsoft Excel spreadsheet (XP professional edition; Microsoft Corp, Redmond, WA, USA), and again this was double-checked by a third reviewer. Trial characteristics including setting (population-based, primary care, secondary care), country of origin, number of centres involved, duration of bismuth therapy and dosage schedule, type of bismuth compound, mean age of included patients, and proportion of male patients were recorded to allow exploration of potential reasons for any heterogeneity detected between trial results.

Data were extracted as dichotomous outcomes and pooled using a random effects model[13], where sufficient data were available. The impact of bismuth therapy on the incidence of total and specific individual adverse effects versus comparison regimen was expressed as a combined relative risk (RR) with a 95% confidence interval (CI). The number needed to harm with bismuth therapy to cause one adverse event, and a 95% CI, were calculated as the reciprocal of the risk difference from the meta-analysis, and where this was statistically significant the results were reported.

Due to differences in methodology, patient populations, and outcome measures between eligible trials, the results of individual studies can be very diverse and therefore when they are included in the same meta-analysis this may affect the accuracy of the overall result. This inconsistency within a single meta-analysis can be quantified with a statistical test of heterogeneity, to assess whether the variation across trials is due to true heterogeneity, or chance. This quantity is termed I², and its value ranges from 0 to 100 percent, with 0 percent representing no observed heterogeneity, and larger values indicating increasing heterogeneity. A value below 25 percent is arbitrarily chosen to represent low levels of heterogeneity[14]. Where the degree of statistical heterogeneity is greater than this, clinical reasons within individual trials that may account for some of this inconsistency can be explored. Wherever statistically significant heterogeneity existed between trial results in this systematic review, possible explanations were investigated informally using sensitivity analyses. These are exploratory only, and may explain some of the observed variability, but the results should be interpreted with caution.

All statistical analyses were performed using Stats Direct version 2.2.4 (Stats Direct Ltd, Sale, Cheshire, UK), which was used to generate Forest plots of pooled relative risks for total adverse event rates and specific individual adverse event rates by category, as well as funnel plots to assess for evidence of publication bias.

The search strategy identified 1800 studies, of which 549 were possibly eligible. After reviewing the abstracts of these it became clear that 209 were RCTs of bismuth, and these were retrieved for further assessment. Of these, 35 were eligible for inclusion in the meta-analysis[7,15-48], reporting on 4763 H pylori-positive patients, 2435 of whom received bismuth or bismuth-based regimen, and 2328 received a comparison regimen (Figure 1). Thirty-three of the trials were found in fully published form, and two were only published as abstracts[29,44]. Seven of the RCTs used more than one bismuth-containing regimen[7,20,31,34,35,41,47].

Detailed trial characteristics are provided in Table 1. Nineteen of the trials were conducted in Europe[15-18,20,23-25,27-30,37,40,42-45,47], eight in the Far East[21,22,32,35,36,38,46,48], four in the USA[31,33,34,41], one in the Middle East[26], one in South America[19], one in Australia[7], and one was a multi-national study[39]. Eleven of the studies were multi-centre RCTs[15,21,25,29,31,32,34,39,41,44,45]. Duration of bismuth therapy ranged from 7 to 56 d, with a total daily dose of between 400 mg and 2100 mg. Nineteen studies used ranitidine bismuth citrate[15,17,20,22,24,28-32,34,36,39-41,43,45,46,48], ten studies colloidal bismuth subcitrate[7,16,18,21,26,27,37,38,42,44], two studies tripotassium dictrato bismuthate[23,35], two studies bismuth subsalicylate[25,33], one study bismuth subnitrate[19], and one study both bismuth subnitrate and colloidal bismuth subcitrate[47]. Comparison regimens were proton pump inhibitor or H₂-receptor antagonist (H₂-RA)-based eradication therapy in 23 studies[15,17-24,26-30,32,35,36,39,40,43,45,46,48], antibiotics alone in four studies[16,38,44,47], antibiotics or placebo in three studies[31,34,41], H₂-RA alone in two studies[25,37], placebo alone in two studies[33,42], and H₂-RA in combination with either one antibiotic or placebo in one study[7]. The mean age of individuals in included studies ranged from 36.7 years to 50.5 years, and the proportion of male patients varied between 32 percent and 78 percent. The number of participants in each RCT ranged from 20 to 530 individuals.

Thirteen of the trials were double-blind randomised studies[7,15,20,22,31,33,34,37-39,41,42,47], the remainder being either single-blind or open. Five of the single-blind trials specifically stated that assessors were blinded to treatment allocation[21,24,45,46,48]. Ten of the studies reported the method of generation of the randomization schedule[15,18,20,26,30,32,37,39,47,48], but only four the method of concealment of allocation[27,32,47,48]. Four of the studies recorded adverse events using a questionnaire[20,24,28,40], but only two of these stated that the questionnaire was validated[24,28]. Seven studies collected information concerning adverse events using a diary or diary cards[21,27,32,34-36,48], two via face-to-face interview[43,45], and one via telephone interview[46]. The remainder of trials did not state how they collected adverse events data.

There were no serious adverse events such as death or neurotoxicity in either arm of any of the included RCTs. Twenty-five trials reported the total number of individuals experiencing any adverse event with bismuth or bismuth-containing regimens versus comparison regimen[15-20,23-27,30,32,33,35-40,42,44,45,47,48]. Three of these studies utilized more than one regimen[20,35,47], allowing 28 comparisons to be made. The relative risk of an adverse event with bismuth or bismuth-containing regimens versus comparison regimen was 1.01 (95% CI: 0.87 to 1.16) (Figure 2 and Table 2). There was statistically significant heterogeneity detected between trial results (heterogeneity test: I²= 30.3%). The Egger test did not suggest any trend for funnel plot asymmetry (P = 0.16). Sensitivity analysis according to trial setting, country of origin, dose of bismuth salt used, type of bismuth salt used, mean age of patients included in the study, and proportion of males included in the study failed to reveal any obvious explanation for the observed heterogeneity.

Abdominal pain: Thirteen trials reported the total number of individuals experiencing abdominal pain with bismuth or bismuth-containing regimens versus comparison regimen[17,18,20,21,24,26,28,30,34,39,40,46,47]. Three of these studies utilized more than one regimen[20,34,47], allowing 16 comparisons to be made. The relative risk of abdominal pain with bismuth or bismuth-containing regimens versus comparison regimen was 1.06 (95% CI: 0.64 to 1.74) (Table 2). There was no statistically significant heterogeneity detected between trial results (heterogeneity test: I²= 22.0%), and the Egger test did not suggest any trend for funnel plot asymmetry (P = 0.15).

Dark stools: Four trials reported the total number of individuals experiencing dark stools with bismuth or bismuth-containing regimens versus comparison regimen[17,42,46,48]. The relative risk of dark stools with bismuth or bismuth-containing regimens versus comparison regimen was 5.06 (95% CI: 1.59 to 16.12) (Figure 3 and Table 2). There was marginal statistically significant heterogeneity detected between trial results (heterogeneity test: I²= 25.2%), but no obvious causes were found, and the Egger test did not suggest any trend for funnel plot asymmetry (P = 0.28). The number of patients needed to harm with bismuth or bismuth-containing regimen versus comparison regimen to cause one case of dark stools was 7.5 (95% CI: 4 to 71).

Diarrhoea: Twenty-two trials reported the total number of individuals experiencing diarrhoea with bismuth or bismuth-containing regimens versus comparison regimen[7,17,18,20,24-28,30-32,34,36,39-42,45-48]. Six of these studies utilized more than one regimen[7,20,31,34,41,47], allowing 28 comparisons to be made. The relative risk of diarrhoea with bismuth or bismuth-containing regimens versus comparison regimen was 1.01 (95% CI: 0.72 to 1.42) (Table 2). There was marginal statistically significant heterogeneity detected between trial results (heterogeneity test: I²= 26.2%), but no obvious causes were found, and the Egger test did not suggest any trend for funnel plot asymmetry (P = 0.75).

Dizziness: Eight trials reported the total number of individuals experiencing dizziness with bismuth or bismuth-containing regimens versus comparison regimen[21,26,31,32,35,41,46,48]. Three of these studies utilized more than one regimen[31,35,41], allowing 11 comparisons to be made. The relative risk of dizziness with bismuth or bismuth-containing regimens versus comparison regimen was 1.18 (95% CI: 0.81 to 1.72) (Table 2). There was no statistically significant heterogeneity detected between trial results (heterogeneity test: I²= 0%), and the Egger test did not suggest any trend for funnel plot asymmetry (P = 0.20).

Headache: Fourteen trials reported the total number of individuals experiencing headache with bismuth or bismuth-containing regimens versus comparison regimen[17,18,20,24-26,30,31,34,39-41,46,47]. Five of these studies utilized more than one regimen[20,31,34,41,47], allowing 19 comparisons to be made. The relative risk of headache with bismuth or bismuth-containing regimens versus comparison regimen was 1.31 (95% CI: 0.81 to 2.11) (Table 2). There was no statistically significant heterogeneity detected between trial results (heterogeneity test: I²= 0%), and the Egger test did not suggest any trend for funnel plot asymmetry (P = 0.83).

Metallic taste: Fourteen trials reported the total number of individuals experiencing metallic taste with bismuth or bismuth-containing regimens versus comparison regimen[17,20,24,27,30,34,35,39-42,45,46,48]. Four of these studies utilized more than one regimen[20,34,35,41], allowing 18 comparisons to be made. The relative risk of metallic taste with bismuth or bismuth-containing regimens versus comparison regimen was 1.02 (95% CI: 0.81 to 1.28) (Table 2). There was no statistically significant heterogeneity detected between trial results (heterogeneity test: I²= 0%), though the Egger test suggested there was funnel plot asymmetry (P = 0.01).

Nausea and/or vomiting: Twenty trials reported the total number of individuals experiencing nausea and/or vomiting with bismuth or bismuth-containing regimens versus comparison regimen[17,18,20,21,24-28,30-32,34,35,39-42,46,47]. Six of these studies utilized more than one regimen[20,31,34,35,41,47], allowing 26 comparisons to be made. The relative risk of nausea and/or vomiting with bismuth or bismuth-containing regimens versus comparison regimen was 1.16 (95% CI: 0.89 to 1.52) (Table 2). There was no statistically significant heterogeneity detected between trial results (heterogeneity test: I²= 0%), and the Egger test did not suggest that there was any evidence of funnel plot asymmetry (P = 0.85).

Twenty-eight trials reported the total number of individuals who terminated therapy due to experiencing adverse events with bismuth or bismuth-containing regimens versus comparison regimen[16-18,20,22-32,34-37,39-43,45-48]. Six of these studies utilized more than one regimen[20,31,34,35,41,47], allowing 34 comparisons to be made. The relative risk of withdrawal of therapy due to adverse events with bismuth or bismuth-containing regimens versus comparison regimen was 0.86 (95% CI: 0.54 to 1.37) (Table 2). There was no statistically significant heterogeneity detected between trial results (heterogeneity test: I²= 0%), though the Egger test suggested that there was evidence of funnel plot asymmetry (P = 0.05).

The duration of bismuth therapy was one month or more in eleven of the included studies[7,18,25,31,34,37-39,41,42,47]. There were sufficient trials to pool data to examine the effect of duration of therapy on total number of adverse events, some of the specific individual adverse events (including diarrhoea, headache, and nausea and/or vomiting), and withdrawal of therapy due to adverse events.

Total number of adverse events: Seven trials provided data on total number of adverse events in 945 individuals (467 of whom were assigned to bismuth)[18,25,37-39,42,47], and one study utilized more than one regimen allowing eight comparisons to be made[47]. The relative risk of an adverse event with bismuth or bismuth-containing regimens used for one month or more versus comparison regimen was 1.20 (95% CI:1.00 to 1.44), with no statistically significant heterogeneity detected between trial results (heterogeneity test: I²= 0%).

Diarrhoea: Nine studies provided data on the incidence of diarrhoea with one month or more of bismuth in 1601 patients (859 of whom were assigned to bismuth)[7,18,25,31,34,39,41,42,47], with five of the studies utilizing more than one regimen allowing fourteen comparisons to be made[7,31,34,41,47]. The relative risk of diarrhoea with bismuth or bismuth-containing regimens used for one month or more versus comparison regimen was 1.72 (95% CI: 1.14 to 2.60), with no statistically significant heterogeneity detected between trial results (heterogeneity test: I²= 0%).

Headache: Seven studies provided data on the incidence of headache with one month or more of bismuth in 1435 patients (778 of whom were allocated to bismuth)[18,25,31,34,39,41,47], with four of the studies utilizing more than one regimen allowing eleven comparisons to be made[31,34,41,47]. The relative risk of headache with bismuth or bismuth-containing regimens used for one month or more versus comparison regimen was 1.39 (95% CI: 0.76 to 2.53), with no statistically significant heterogeneity detected between trial results (heterogeneity test: I²= 0%).

Nausea and/or vomiting: Eight studies provided data on the incidence of nausea and/or vomiting with one month or more of bismuth in 1501 patients (810 of whom were allocated to bismuth)[18,25,31,34,39,41,42,47], with four of the studies utilizing more than one regimen allowing twelve comparisons to be made[31,34,41,47]. The relative risk of nausea and/or vomiting with bismuth or bismuth-containing regimens used for one month or more versus comparison regimen was 1.47 (95% CI: 0.87 to 2.48), with no statistically significant heterogeneity detected between trial results (heterogeneity test: I²= 0%).

Withdrawal of therapy due to adverse events: Nine studies provided data on the incidence of withdrawal of therapy due to adverse events with one month or more of bismuth in 1554 patients (837 of whom were allocated to bismuth)[18,25,31,34,37,39,41,42,47], with four of the studies utilizing more than one regimen allowing thirteen comparisons to be made[31,34,41,47]. The relative risk of withdrawal of therapy due to adverse events with bismuth or bismuth-containing regimens used for one month or more versus comparison regimen was 0.86 (95% CI: 0.47 to 1.57), with no statistically significant heterogeneity detected between trial results (heterogeneity test: I²= 0%).

Bismuth compounds are often used as part of eradication therapy for Helicobacter pylori (H pylori). There are concerns about toxicity of these compounds in some countries, particularly as a result of their potential neurological sequelae.

Data concerning toxic effects of bismuth compounds are mainly derived from studies that have used these compounds at a high dose for a prolonged period of time. We conducted a meta-analysis of adverse events resulting from a 1 to 2-wk course of bismuth based H pylori eradication therapy.

The current study demonstrated that bismuth compounds, when used short-term for 1 to 2 wk in H pylori eradication therapy, are safe. The only adverse event that occurred more frequently in patients receiving bismuth was dark stools.

Potential adverse events from bismuth compounds in a 1 to 2-wk course of H pylori eradication therapy are now quantified. Gastroenterologists can be assured that these compounds are safe to use.

The number needed to harm is the number of patients that would need to be treated with bismuth compounds for one patient to experience an adverse event.

There are now problems in obtaining satisfactory eradication rates of H pylori with PPI-based triple therapies, so the use of bismuth containing regimens has been recommended as a potential first line therapy in the Maastricht guidelines. Furthermore, there are now new bismuth combinations commercially available. For these reasons it is important to be sure of the safety of bismuth compounds.