Evaluation of treatment regimens to cure Helicobacter pylori infection—a meta-analysis


Laheij Department of Gastroenterology, University Hospital Nijmegen, PO Box 9101, NL 6500 HB Nijmegen, The Netherlands. E-mail: r.laheij@mie.kun.nl



: To assess effectiveness of treatment to cure Helicobacter pylori infection.

Data Synthesis

: Meta-analysis of 666 manuscripts (full papers, abstracts, letters to the editor) identified through Medline and a manual search (1986 to January 1998). Data were overviewed by regression analysis with weighted random effects models.


: 53 228 patients with H. pylori infection.


: Patients were treated with 132 different medication combinations.

Main outcome measure

: Cure of H. pylori infection per protocol and intention-to-treat basis at least 28 days after treatment.


The nationality of the patients and therapeutic regimen have a significant impact on the results, after correction for the heterogeneity in the precision of the cure rate caused by different study sizes and random effect for study. On the basis of the original sample size, cure rates of 80–85% were achieved using combinations of a proton-pump inhibitor or ranitidine bismuth citrate with two antibiotics including clarithromycin, amoxycillin and metronidazole or tinidazole. Comparable cure rates were also achieved using a combination of a proton-pump inhibitor or H2-receptor antagonist with bismuth subcitrate or tripotassium dicitrato bismuthate, metronidazole and tetracycline. The dose of clarithromycin influenced cure rates. Treatment duration did not influence the outcome.


Several therapeutic regimens are eligible to cure H. pylori infection. However, none of the medication combinations were able to cure H. pylori infection in more than 85% of the patients assessed by intention-to-treat. The countries in which the studies were performed also had a significant impact on eradication rates.


Helicobacter pylori frequently infects the human gastric mucosa. H. pylori causes peptic ulcer disease and is related to the development of gastric cancer. Since the rediscovery of the bacterium by Warren and Marshall1 an enormous number of studies have been performed in which an attempt was made to identify an optimal therapeutic regimen. Various antibiotics separately or in combination, with or without bismuth salts and gastric-acid-secretion inhibitors, have been evaluated. However, the data obtained show that treatment is difficult and despite all efforts, no single therapeutic regimen is optimal. Furthermore, the study results are difficult to compare because of differences in clinical features and methodological aspects which may explain the discordant cure rates reported. Meanwhile clinicians must decide for themselves which therapeutic regimen is the most effective and the least harmful method to cure H. pylori infection. Therefore many different regimens are used in clinical practice and a consensus on the most optimal regimen is lacking.

To promote comprehensibility, many systematic reviews and meta-analysis have appeared on treatment. Most have dealt with only selections of the literature, because of the immense amount of data available. In this article we tried to assess, systematically, the effectiveness of various treatment regimens to cure H. pylori infection. Unlike other reviews or meta-analyses, we tried to identify all relevant publications and did not exclude a priori letters to the editor, abstracts or symposium proceedings.


Identification and eligibility of publications

A computerized and manual literature search was performed in early 1998. Relevant publications were identified in Medline (1983–97) with the following Subject Heading terms: ‘Campylobacter or Helicobacter pylori’, eradicat*, ‘Human in TG’, or ‘Helicobacter-infections-drug-therapy in TI, AB, MeSH’, not restricted to any language. Furthermore, a manual search was conducted of the American Journal of Gastroenterology, Gastroenterology, Gut and the Scandinavian Journal of Gastroenterology published between 1986 and January 1998. Additional publications were retrieved by reviewing references in reviews and meta-analyses.

All retrieved publications were checked by two independent reviewers to ensure that they met the following inclusion criteria: (i) description of the complete treatment and acid suppressive treatment; (ii) biopsy-based methods (histology, culture or rapid urease test) or breath tests had been used to assess the H. pylori status before and after treatment (in the period immediately after the rediscovery of the bacteria in 1983 other, less reliable methods were used to evaluate the H. pylori status); and (iii) measurement of H. pylori status had been performed at least 4 weeks after cessation of treatment, as earlier measurement does not sufficiently discriminate between clearance and eradication. Exclusion criteria: the study was excluded if patients received acid suppressive medication in the 4 weeks after antibiotic therapy; studies with no information about the number of patients treated or cured were excluded.

As we intended to include all publications, attempts were made to exclude duplicates of studies or preliminary presentations. We did not exclude any medication a priori, to minimalize bias. Disagreements between the two observers were resolved by consensus, and the reasons for study exclusion were recorded.

Data analysis

From each study we calculated the percentage of patients successfully cured for each treatment protocol and whenever possible for the total patient sample. Both total patient sample and treatment protocol cure rates were used to explore possible differences between clinical features of the study populations and methodological aspects of the studies, both having a bearing on the outcome of the study. The trials were stratified into: report type (abstract, article and other); publication year (1983–97); whether or not the study population was a consecutive patient series; the nationality of the study population; the diagnosis of the patients (peptic ulcer disease, non-ulcer disease, other); the reference standard used to monitor treatment (biopsy-based, breath test or both biopsy-based and breath test); duration of follow-up period; whether or not treatment allocation was random; whether or not antibiotic susceptibility was measured; whether or not the study was multicentre; whether or not a total patient sample (intention-to-treat) cure rate could be calculated; and the therapeutic regimen applied (medication, dose and duration).

The medication was categorized as follows: proton-pump inhibitors (omeprazole, lansoprazole, pantoprazole and rabeprazole); H2-receptor antagonists (cimitidine, famotidine, nizatidine, ranitidine and roxatidine); ranitidine bismuth citrate; bismuth salts (bismuth subnitrate, bismuth subsalicylate, colloidal bismuth subcitrate and tripotassium dicitrato bismuthate); macrolides (azithromycin, clarithromycin, erythromycin and roxithromycin); tetracyclines (doxycyclin, oxytetracycline and tetracycline); penicillins (amoxycillin, ampicillin, bacampicillin, pencillin, pivampicillin and oxacillin); nitromidazoles (furazolidone, odinazole, ornidazole, metronidazole and tinidazole); cephalosporins (cefixin, cefuroximaxetil); fluoroquinolones (ciprofloxacin, norfloxacin and ofloxacin); antimycotium (itraconazole and rifampicin); lincomycin (clindamycin).

Statistical methods

To correct for outlying values and to stabilize the variance in cure rate, arcsin transformation was performed.2 We modelled the heterogeneity between the studies by means of an ordinary least square regression equation, in which all the clinical features and methodological aspects were included simultaneously. It is very likely that the cure rates of different therapeutic regimens were correlated within the same study. By introducing a random intercept for the study we modelled dependency between outcomes within the same study.3 Moreover, the imprecision of the cure rate varied per study. In order to correct for the heterogeneity in the precision of the cure rates caused by different study sizes, we also performed a weighted regression analysis with weights proportional to four times the size of the study population.2 For each regression model an overall F- (NDF, DDF) test was used to examine whether the hypothesis of no fixed effect should be rejected. NDF is the degree of freedom in the numerator and DDF the degree of freedom in the denominator of the F-test.

Difficulty of cure by country was estimated by the percentage of patients successfully cured per total patient sample (intention-to-treat) in more than one trial. The dose–response relation of therapeutic regimens with the highest cure rates were evaluated further. Only the intention-to-treat cure rate was used to explore possible differences between medication, dose–response and dose–duration relationships. We modelled the heterogeneity by means of a weighted regression analysis with a correction for random effects for study. Unfortunately, we could only model therapeutic regimens that were investigated in enough manuscripts because of convergence problems. In each model we corrected for features and methodological aspects that had a substantial bearing on the difference in cure rates between studies. The dose–response and dose–duration relationships were estimated for each drug separately. If the dosage of the drug or treatment duration was not of any significant importance the lowest investigated dose was described. All statistical analyses were performed with SAS software.


The computerized and manual searches revealed 2689 relevant manuscripts, of which 2507 were available in Dutch libraries; 666 manuscripts met the inclusion criteria. There was initial disagreement between the reviewers about the inclusion or exclusion of 332 manuscripts (kappa = 0.72). The remaining 2023 manuscripts were not included in the analysis for the following reasons: 363 were reviews or commentaries, 39 were not easily translatable, 347 provided insufficient data about the therapeutic regimen, 390 assessed H. pylori status within 4 weeks after treatment, 92 lacked information about the number of patients or cure rates, 85 did not use an adequate method to evaluate the H. pylori status (Table 1). In addition, 296 manuscripts were excluded because the same or preliminary results were reported in another publication.

Table 1.  . Eligibility of the identified manuscripts [reference] Thumbnail image of

In the 666 manuscripts included, 132 different medication combinations were used in 1295 trials. The average number of patients per trial was small ( Figure 1). The median was 30 patients with an interquartile range (IQR) of 19 and 50 patients. Overall, the median cure rate in the 1295 trials was 71% (IQR; 44–85) in the per protocol analysis and 67% (IQR; 42–82) in the intention-to-treat analysis reported in 868 out of the 1295 trials. Variation in eradication rates was related to differences in patient features and methodological aspects of the studies (Table 2). The results of studies that mentioned intention-to-treat cure rates besides per protocol cure rates differed significantly from studies that did not. Furthermore, the weighted regression analysis to correct for random effects between studies revealed that two more features were responsible for the observed difference in both per protocol and intention-to-treat cure rates between studies: the therapeutic regimen used and the nationality of the study population (Table 3).

Figure 1.

. Cure rates of H. pylori infection by study size.

Table 2.  . Clinical features and methodological aspects of the selected studies associated with differences in cure rate of H. pylori infection across studies Thumbnail image of
Table 3.  . Cure rates by country Thumbnail image of

Adjusted cure rates showed that no therapeutic regimen was optimal. The most frequently investigated therapeutic regimens that provided sufficient effectiveness were triple therapies: two antimicrobial drugs supplemented with a proton-pump inhibitor. Results showed that a combination of a proton-pump inhibitor and macrolide plus nitroimidazole cured 87.1% and 82.9% of the patients per protocol and intention-to-treat, respectively (Table 4). We found no differences between the proton-pump inhibitors (lansoprazole, omeprazole and panteprozole) or the type of nitroimidazole (metronidazole or tinidazole) used (Table 5). Of the two macrolides evaluated, clarithromycin was superior to azithromycin. Furthermore the dose of clarithromycin influenced the cure rate; 1.5 g of clarithromycin cured H. pylori in more patients than 500 mg, 600 mg, 750 mg or 1 g. The dosage of the other drugs did not influence the cure rate. The same antibiotics in combination with ranitidine bismuth subcitrate also performed very well. Although analysed in only seven studies, 90.8% and 78.4% of the patients were cured per protocol and on intention-to-treat basis, respectively.

Table 4.  . Adjusted cure rates of therapeutic regimens Thumbnail image of
Table 5.  . Evaluation of therapeutic regimens with dose response analysis Thumbnail image of

Amoxycillin in combination with a proton-pump inhibitor and clarithromycin, led to estimated per protocol and intention-to-treat cure rates of 86.2% and 80.0%, respectively. With this medication combination, again the clarithromycin dosage influenced the eradication rate. A dosage of 1.5 g clarithromycin performed better than 500 mg, 750 mg, 800 mg and 1 g. In the six studies performed, the results showed that the therapeutic regimen of ranitidine bismuth citrate and a macrolide and amoxycillin may be promising. Ranitidine bismuth citrate and clarithromycin cured H. pylori infection in 87.7% of the compliant patients (per protocol) and in 6.4% fewer when loss to follow-up was taken into account (intention-to-treat). The results from 26 studies that investigated combinations of a proton-pump inhibitor, bismuth salt, nitroimidazole and tetracycline showed that 90.6% of the compliant patients were cured. Exchanging the proton-pump inhibitor for an H2 receptor antagonist cured 89.0% of per protocol patients. Bismuth salts, nitroimidazole and tetracycline without a gastric-acid-secretion inhibitor cured 79.1% of the infected patients. When loss to follow-up was taken into account the cure rates decreased to 81.7%, 79.7% and 75.5%, respectively. In the only quadruple therapy we were able to model (omeprazole, colloidal bismuth subcitrate or tripotassium dicitrato bismuthate, metronidazole and tetracycline), none of the drugs or the dosages had any substantial influence on the cure rate. Additionally, the duration of treatment did not influence the cure rate of any of the above-mentioned therapy regimens.


By the beginning of 1998, at least 363 reviews, meta-analyses and commentaries on therapeutic regimens to eradicate H. pylori infection had published since the bacterium was rediscovered by Warren and Marshall in 1983.1 Our meta-analysis differs from these in several respects. Unlike other reviews, we tried to analyse all published data and did not a priori exclude the results of non-randomized studies, abstracts and letters to the editor. Moreover, instead of pooling treatment results, we attempted to explore the heterogeneity in features of the patient populations and in methodological aspects of the studies. The term heterogeneity characterizes the situation in which differences in study outcomes cannot readily be accounted for by sampling variation. We used two outcome measures: cure rates per protocol and intention-to-treat in our analysis to explore heterogeneity. The elimination of loss to follow-up from the analysis (per protocol analysis) overestimates the effect of the therapeutic regimen. The cure rates in a selection of patients with good compliance (per protocol) were substantially better than those in patients with poor compliance. After treatment has begun, many study-related events that occur are likely to be intercorrelated. By categorizing patients on the basis of therapy received, selection bias will appear. To prevent this bias, cure rates on the basis of the original population size (intention-to-treat) should be used.

Two features were responsible for the significant differences in cure rates between studies. That cure rates differ when different therapeutic regimens are used is to be expected. More striking however, is the other important feature responsible for the significant differences in cure rates between studies: the nationality of the study population. A number of reasons might explain the differences in treatment outcome found between countries. First, differences in food and tobacco consumption, therapy compliance and alcohol intake might be involved. In countries with a high alcohol intake, a therapeutic regimen that includes metronidazole might result in lower success rates because if the patients continue to drink they will stop treatment due to side-effects. Furthermore, several strains of H. pylori have been identified4. Epidemiological studies showed that the occurrence of different strains varies between geographical areas. Treatment response might be influenced by differences in genotypic strains. However, in general the most important reason for treatment failure with antimicrobial treatment was antibiotic resistance. Helicobacter pylori resistance against metronidazole and clarithromycin varies between countries. In countries such as Belgium, France and Spain the resistance of H. pylori for metronidazole is considerably higher than in other countries, such as the Netherlands and Norway. Resistance against metronidazole may vary considerable even in one country, especially between big cities and rural communities. For instance, resistance against metronidazole in the Amsterdam area is considerably higher than that in the eastern part of the Netherlands.

The other factor that markedly influenced the cure rates was the therapeutic regimen used. None of the therapeutic regimens cured H. pylori infection in 100% of the patients. To achieve an acceptable result, antimicrobial drugs combined with gastric-acid- secretion inhibitors or bismuth salts were necessary. At most, 85% of the patients who were fully treatment compliant (and not lost to follow-up) could be cured of the infection. Thus in routine clinical practice, eradication rates of H. pylori will be even lower. Several therapeutic regimens can be used to treat patients with an H. pylori infection. The most effective of them contained two antibiotics in combination with a gastric acid inhibitor. Antimicrobial agents that performed well were the macrolide clarithromycin, the nitroimidazole metronidazole and tinidazole, and amoxycillin. A combination of two drugs was more effective than one antibiotic alone because there seemed to be a synergistic response. The use of combinations of drugs with different actions and mechanisms against resistance should decrease the emergence of resistant organisms during treatment. The only therapeutic regimen in which metronidazole resistance did not seem to have any influence was quadruple therapy with a proton-pump inhibitor, bismuth salts, metronidazole and tetracycline. As a result of the changing patterns of antibiotic susceptibility it may be necessary to test sensitivity regularly before deciding which therapeutic regimen to use, because more therapeutic regimens are available to treat H. pylori. However, sometimes H. pylori will be sensitive to an antibiotic and sufficient concentrations will be achieved at the treatment site but the bacteria will be able to escape the effects of the drug.

Most of the previous reviews only considered randomized controlled trials to determine which therapeutic regimens resulted in the highest eradication rates. The analysis of non-randomized studies would increase the number of patients needed to detect small effects and may improve the generalizability of the study results. However, in studies on the effects of treatments, non-randomized comparisons can be affected by confounding.5 Therapeutic interventions are commonly prompted by an indication. Physicians selectively treat patients, leading to confounding by indication. We did not find any statistical difference in per protocol and intention-to-treat cure rates between randomized and non-randomized studies. Therefore, in our meta- analysis, the diversity in patient populations did not result in any substantial heterogeneity among the different trials with respect to confounding by indication. It seems that randomization of the patient population to compare therapeutic regimens to cure H. pylori infection is unnecessary.

We included more abstracts (n = 405) than full papers (n = 243) in our analysis. The diversity of manuscripts did not lead to any substantial heterogeneity among the different trials. A disadvantage of including abstracts in the analysis is the limited information about features of the patient population and methodological aspects of the study. Furthermore, the risk of including the (preliminary) results from multiple publications increases. Most of the abstracts we included were presented at meetings and had not (yet) been published as a full paper. Some of these studies will never be published as a full paper, because of the negative or reconfirming study results. The inclusion of abstracts may prevent or reduce bias. We further tried to exclude bias by not restricting our search to certain languages. We were able to evaluate studies published in English, German, French and Dutch. In manuscripts written in any other language, we searched for an abstract in English, German, French and Dutch. As a result only 39 studies had to be excluded because of language problems.

In conclusion, several therapeutic regimens can be used to treat patients with an H. pylori infection. Although, most studies report excellent cures rates, the success rate will frequently be below 80% in routine clinical practice. We found that the success rate of H. pylori eradication was dependent only on the country in which the study was performed and the therapeutic regimen.


We thank professor J.R.B.J. Brouwers, University Centre of Pharmacy, University of Groningen and Dr W.A de Boer, St. Anna Hospital, Oss for comments that improved this paper.