Safety of Vonoprazan-Amoxicillin Dual Therapy Versus Concomitant Quadruple Therapy for H. Pylori Eradication

H. pylori Dual vs Concomitant Therapy - Protocol (English version) 3 nov 2026 · @Manuela INTRODUCTION

1. Background and rationale Helicobacter pylori (HP) infection is the most frequent bacterial infection in humans and a major global health burden. Although in gradual decline, its approximate global and regional prevalence rates are 44% and 57%, respectively.¹,² In Uruguay, the only available records reported 46% in 1997 and 70% in an African population (2010).³,⁴ Its morbidity lies in the non-negligible risk of malignant disease - non-hereditary gastric adenocarcinoma (HP being a group 1 carcinogen and its main risk factor), lymphoma and premalignant lesions (gastric atrophy and intestinal metaplasia) - and of benign conditions affecting quality of life, such as peptic ulcer or dyspepsia. A recent Uruguayan study of a random population undergoing endoscopy observed an HP prevalence of 42%, with gastric atrophy in 38.2% and intestinal metaplasia in 31.4%.⁵ The Maastricht VI/Florence consensus establishes the indications for testing and eradication: uninvestigated dyspepsia with alarm features, peptic ulcer, marginal zone (MALT) lymphoma, household contacts of infected patients, prolonged use of non-steroidal anti-inflammatory drugs, iron-deficiency anemia, idiopathic thrombocytopenia, a personal or first-degree family history of gastric cancer, a personal history of premalignant gastric lesions, increased gastric cancer risk (an ethnic background with higher infection prevalence, migration from an endemic area, certain hereditary carcinogenic syndromes), autoimmune gastritis, and gastric adenomatous or hyperplastic polyps.⁶ Regimens are adapted to regional resistance rates and drug availability, and as a rule combine a proton pump inhibitor (PPI) - omeprazole being the most widely available - with one or more antibiotics of synergistic mechanism. Where clarithromycin resistance exceeds 15% or is unknown, as in Uruguay, the consensus indicates quadruple therapy, preferably bismuth-containing. Since bismuth and tetracyclines are not readily accessible, our setting uses non-bismuth quadruple (concomitant) therapy - omeprazole, clarithromycin, amoxicillin and metronidazole (OCAM) - for 14 days, on the basis of an eradication rate above 90% and adequate tolerance.⁷ Low tolerance and adherence, attributable to adverse events and to a cumbersome dosing schedule, and above all antibiotic resistance with its selective pressure, prompt the search for evidence supporting fewer antibiotics. In Chile, a country with a very high prevalence of HP-related gastric adenocarcinoma, resistance to clarithromycin, metronidazole and amoxicillin has been reported at 12%, 53% and 4%, respectively.⁸ Locally, a 2006 study of 50 infected patients observed resistance rates of 36% to metronidazole and 12% to clarithromycin.⁹ No subsequent studies are available, and similar or higher resistance is presumed at present.

   Potassium-competitive acid blockers (pCAB) have demonstrated equal or greater potency for HP eradication and, since 2024, have been incorporated into first-line regimens.¹⁰,¹¹ One regimen under study combines vonoprazan (a pCAB recently available in Uruguay) with high-dose amoxicillin - the dual regimen - promising for its reduced antibiotic burden and consequent resistance, and for greater adherence through a simplified schedule and fewer adverse events.

   A 2025 meta-analysis found higher eradication rates with vonoprazan-based than with PPI-based dual therapy.¹² A United States trial of 1,046 patients demonstrated superiority of vonoprazan-based regimens, both dual (77.2%) and triple with clarithromycin (80.8%), over a lansoprazole-based regimen (68.5%), with no significant differences in adverse events.¹⁵ Adverse events deserve particular attention. Japanese guidelines from 2016 recommend caution in treatment selection in patients with reduced hepatic and/or renal function, to avoid unwanted interactions and toxicity.¹⁶ With concomitant therapy they are estimated at 40% (with reports of up to 90%) and usually result in a discontinuation rate of 2-10%.⁷,¹⁷ The greatest symptom burden is gastrointestinal: nausea, vomiting, dyspepsia, diarrhea and dysgeusia, the last above all with metronidazole. A Chinese randomized trial of 400 patients observed adverse events in 19% with dual therapy (vonoprazan and amoxicillin) against 53% with bismuth quadruple therapy (rabeprazole, amoxicillin, clarithromycin and bismuth citrate), with discontinuation for intolerance in 4% and 6.5%, respectively; dysgeusia predominated in the comparator arm. Self-limited skin rash, attributable to amoxicillin hypersensitivity, is the event most characteristically associated with the dual regimen.¹⁸,¹⁹ A Japanese prospective study found a smaller reduction in gut microbiota diversity and relative abundance with the dual method.²⁰ Administration independently of meals, afforded by the stable bioavailability of vonoprazan, could further improve adherence.²¹ Amoxicillin is a time-dependent antibiotic, and H. pylori eradication requires a sustained gastric pH above 4. Within the dual regimen, evidence is available for amoxicillin 750 mg every 6 hours and 1 g every 8 hours. The former ensures a sustained acidic gastric pH and lower interindividual variability; the latter, through a simplified schedule and the rapid and sustained pH reduction achieved by vonoprazan, favors adherence and reproducibility without compromise of efficacy, and was selected here on that basis and for logistical and technical reasons.¹⁹,²² The gap is specific: neither in Uruguay nor in the region are there data contrasting a pCAB-based dual regimen with non-bismuth concomitant quadruple therapy, the regimen effectively used here and the only feasible one given the limited accessibility of bismuth and tetracyclines. An ongoing pilot study evaluates a dual regimen (esomeprazole-amoxicillin) at Hospital de Clínicas, but no national data make that contrast. All evidence on vonoprazan-based dual therapy comes from Asian, North American and European populations, with resistance profiles, CYP2C19 polymorphisms and prescription patterns that cannot be extrapolated to the Uruguayan population, and in those studies the comparator was bismuth quadruple or clarithromycin triple therapy, never concomitant therapy.

   The expected contribution is threefold: the first national estimate of the comparative safety profile of both regimens, the main determinant of adherence and therefore of eradication success in practice; local evidence on a regimen that reduces the antibiotics administered from three to one, with the consequent impact on selective pressure in a country with no resistance data after 2006; and national eradication and adherence data as a basis for adequately powered non-inferiority studies and for updating local recommendations.

2. Objectives: general and specific General objective. To compare the safety profile and the effectiveness of dual therapy with vonoprazan and amoxicillin versus non-bismuth concomitant quadruple therapy with omeprazole, clarithromycin, amoxicillin and metronidazole for H. pylori eradication in the Uruguayan adult population.

Specific objectives.

1. To compare the frequency and the severity of adverse events between both treatment groups (primary objective).
2. To compare the effectiveness of both regimens, estimated as the H. pylori eradication rate by intention to treat and per protocol.
3. To compare treatment adherence between both groups.
4. To identify the variables associated with the occurrence and severity of adverse events, with control for baseline characteristics and recruitment center.

Hypotheses. Null (H₀): there is no difference in the distribution of adverse event severity between the two regimens. Alternative (H₁): the dual regimen is associated with a lower frequency and severity of adverse events, as a result of the reduction in antibiotic burden from three agents to one. Secondary: the dual regimen achieves eradication rates comparable to those of the concomitant regimen and superior adherence, attributable to the simplified dosing schedule and to the independence of administration from meals. It is expressly noted that the study was not powered to formally test non-inferiority in eradication; this hypothesis will be evaluated in an exploratory manner.

METHODS

1. Literature search strategy The background review was carried out through a structured search in PubMed, Cochrane Library and LILACS at the international level, and in Timbó and SciELO at the national level, between November 2025 and February 2026, with no language restriction. MeSH and free-text terms were combined: "Helicobacter pylori", "vonoprazan", "potassium-competitive acid blocker", "dual therapy", "concomitant therapy", "eradication", "adverse effects" and "antimicrobial resistance". Randomized clinical trials, systematic reviews with meta-analysis and consensus documents published from 2010 onward were prioritized. Reference management was performed with Mendeley.
2. PICO question In adult patients aged 18 to 75 years with confirmed Helicobacter pylori infection and an indication for eradication according to the Maastricht VI/Florence consensus, treated at four tertiary care centers in Uruguay (P), is dual therapy with vonoprazan 20 mg every 12 hours and amoxicillin 1 g every 8 hours for 14 days (I), compared with non-bismuth concomitant quadruple therapy - omeprazole, clarithromycin, amoxicillin and metronidazole every 12 hours for 14 days - (C), associated with a lower frequency and severity of adverse events (primary O) and with comparable eradication rates and greater adherence (secondary O)?
3. Patient participation Patients with confirmed H. pylori infection and an indication for its eradication at the participating centers, who meet the eligibility criteria, will take part. They will be referred either from Digestive Endoscopy, on the basis of histological presence or a positive urease test, or from the Gastroenterology outpatient clinic, on the basis of stool antigen (Ag), serology or breath test.
4. Design An experimental, prospective, longitudinal and multicenter study will be conducted, in the form of a randomized clinical trial with triple blinding (treating physician, research team and statistician), comparing the dual strategy with vonoprazan and amoxicillin against conventional treatment with omeprazole, clarithromycin, amoxicillin and metronidazole for H. pylori eradication. Results will be reported in accordance with the CONSORT statement.
5. Setting Four tertiary level hospital centers, public and private, caring for adults from across the country: Hospital de Clínicas "Dr. Manuel Quintela" (Montevideo), Asistencial Médica-Sanatorio Cantegril (Punta del Este), Hospital Policial (Montevideo) and COMEPA (Paysandú).

6. Eligibility Patients aged 18 to 75 years, users of the research centers, with a positive result and an indication for H. pylori eradication according to the Maastricht VI/Florence consensus, established by histological presence, urease test, stool antigen, serology or breath test, will be included.

   Exclusion criteria: known allergy to any study drug; previous H. pylori treatment; antibiotics in the past month, or anti-H2 drugs, PPIs or vonoprazan in the previous 2 weeks; pregnancy; previous gastric resection; and unwillingness to participate.

7. Randomization, intervention and data collection All patients diagnosed with H. pylori infection and treated during 2026 will be invited to participate. They will be informed of the study objectives and, should they agree, verbal and signed informed consent will be requested, with a copy handed to them (Appendix 1).

   Once the patient is eligible, the treating gastroenterologist will complete a form recording the H. pylori diagnostic method and the patient's demographic data (age, sex, race, obesity, active or past smoking, family history of gastric cancer). A person external to the research - the only person with access to these records, as a masking measure - will then inform the treating gastroenterologist of the patient's identifier number (1-102) and of which envelope to hand over (A or B), without disclosure of the composition of either regimen, which remains unknown to the treating physician and to the research team. At no point will the investigator have access to the spreadsheet recording the patient number and assigned regimen.

   The treating gastroenterologist will hand the patient, in an opaque envelope, the complete course of one regimen, together with the self-administered Treatment Adherence and Adverse Events questionnaire (Appendix 4), to be completed at the end of treatment and returned to the physician, and the request for the eradication test, to be performed at least 45 days after treatment initiation.

   Both treatments last 14 days. The dual therapy group takes 1 tablet of vonoprazan 20 mg every 12 hours and 2 tablets of amoxicillin 500 mg (1 g) every 8 hours. The conventional treatment group takes 1 tablet of omeprazole 20 mg, 1 of clarithromycin 500 mg, 1 of metronidazole 500 mg and 2 of amoxicillin 500 mg, all every 12 hours; in this group the morning intake must occur 30 minutes before the first meal of the day, to optimize omeprazole bioavailability.

   One month after completion of treatment (45 days after initiation), the eradication test will be performed by stool antigen, biopsy, breath test or urease test, as defined by the physician according to the patient's characteristics, at each treating center. The patient will be asked to discontinue PPI and vonoprazan until the test is performed. In the event of treatment failure, and to guarantee the duty of care, a second-line treatment will be offered.

8. Variables Demographic and diagnostic variables All measured at the time of the patient's inclusion in the study and recorded in the database under the categories listed.

   Variable Definition Type and scale Categories Sex Sex assigned at birth Dichotomous qualitative Male, female Age The patient's biological age Discrete quantitative, ratio scale 18-75 Race The patient's typical trait Nominal qualitative White, black, other Smoking Active or past consumption of tobacco in any of its vehicles Dichotomous qualitative Yes, no Obesity Body mass index ≥ 30 Dichotomous qualitative Yes, no First-degree family history of gastric cancer Gastric cancer in a parent, sibling or child Dichotomous qualitative Yes, no H. pylori diagnostic method The test confirming infection: gastric biopsies taken to look for H. pylori or the Sydney protocol, endoscopic urease test, stool antigen, positive serology or positive breath test Nominal qualitative Stool Ag, histology, serology, urease test, breath test Therapeutic variables

   1. Adverse events - an unwanted effect of the treatment plan, during treatment or after its completion, defined in accordance with the CTCAE criteria.²³ Recorded events are nausea, vomiting, diarrhea, abdominal pain, dysgeusia, dyspepsia, reflux, anorexia, asthenia, skin rash and others. For each, absence or presence is recorded and, if present, its intensity - mild (requires no treatment), moderate (requires treatment), severe (requires hospital consultation) - and the number of days with that event. Ordinal qualitative variable, measured after completion of eradication treatment.
   2. H. pylori eradication (eradication rate) - absence of infection confirmed by a method validated for post-treatment testing, performed at least 45 days after initiation of the regimen, equivalent to at least one month after its completion. Any negative result by stool antigen, histology, urease test or breath test, according to the method available at each center, is recorded as eradication achieved; any positive result, as not achieved. Serology is excluded as a test of cure, since the persistence of antibodies precludes any distinction between active and past infection. The method used in each participant is recorded. Dichotomous qualitative variable, measured at the visit subsequent to the eradication test. The eradication rate is not a variable measured in the individual but a derived statistic: the proportion with eradication achieved over the total in each arm, as a percentage with its 95% confidence interval, estimated by intention to treat and per protocol.
   3. Adherence - compliance with the prescribed dose, frequency and duration, recorded by questionnaire and established according to the number of treatment days completed with correct dose and frequency. Good adherence is defined as 80% or more of the treatment (11 days or more), poor adherence as less than 80%. Dichotomous qualitative variable, measured after completion of eradication treatment.
   4. Treatment discontinuation - discontinuation of the assigned regimen before the planned 14 days, whatever the cause, recorded from the treating physician's handwritten note in the participant's file and classified at data entry into one of five mutually exclusive categories: did not discontinue; discontinued because of an adverse event; discontinued by participant decision not attributed to an adverse event; discontinued because of forgetfulness or difficulty with compliance; reason not recorded. The treatment day of discontinuation is also recorded. Nominal qualitative variable, measured at the visit subsequent to the eradication test, from the record made during the treatment period.

9. Harms Collection instrument. The Adherence and Adverse Events Questionnaire (Appendix 4), recording for each symptom its presence, intensity and duration in days, will be administered systematically to all participants. Intensity is defined operationally by the intervention it requires - mild, moderate and severe as defined in the variables section - in correspondence with the grading logic of the CTCAE v5.0. The questionnaire records neither the assignment arm nor the identity of the drugs received: the instruction sheets (Appendices 2 and 3) identify the tablets by color alone. The participant completes it at the end of treatment and returns it at the visit subsequent to the eradication test.

   Clinical management of an adverse event. Management is symptomatic and does not require knowledge of the assignment arm, so blinding does not condition the quality of care. For a mild or moderate event, the treating physician will prescribe symptomatic treatment and decide on continuation according to clinical judgment. The management adopted is not collected as an independent variable: since severity is defined operationally by the intervention required, the assigned category already captures it.

   Treatment discontinuation. The decision to discontinue because of intolerance does not depend on knowledge of the assigned regimen. The treating physician will record the reason and the treatment day. At data entry, the non-investigator responsible for the anonymized database will classify that reason into the pre-established categories, without knowledge of the treatment group; notes that do not allow an unequivocal assignment will be coded as reason not recorded. Discontinuation attributable to an adverse event is reported as a secondary outcome and is not integrated into the severity variable: a participant may discontinue because of a symptom classified as mild, and incorporating it into the severity scale would distort the primary variable. Reasons will be reported disaggregated by group.

   Unblinding. Individual unblinding is permitted when knowledge of the assignment is indispensable for participant safety; a serious adverse event or the need for a concomitant treatment whose interaction depends on the regimen received are sufficient grounds, and the treating physician may request it at any time. The person external to the research who holds the assignment spreadsheet will disclose the arm for that participant alone and leave a written record. Unblinding does not entail exclusion: the participant continues to be analyzed in the assigned arm. The research team and the statistician will not be informed of individual unblinding events until database closure.

   Serious adverse events. The instruction sheets include a contact telephone number, external to the assignment of regimens, for queries and for the reporting of serious adverse events, with explicit mention of the alarm signs that must prompt immediate consultation - shortness of breath, chest pain or severe swelling of the tongue. Appropriate care will be provided, the event documented, and the Research Ethics Committee of Hospital de Clínicas notified within 72 hours of the event becoming known. The research team will assess in each case the need to modify or discontinue the study.

   Declared limitation of blinding. Blinding of the treating physician is structurally imperfect: the two regimens differ in number of tablets and frequency of administration, so arm membership could be inferred from the instruction sheet handed to the participant. Mitigating measures are identification of the drugs by color alone on both sheets, opaque envelopes of identical characteristics sealed until delivery, and the non-participation of the treating physician in apportioning or packaging, carried out at the Academic Unit of Gastroenterology. Blinding of the research team and of the statistician is complete until database closure, since neither accesses the assignment spreadsheet. This limitation will be reported explicitly in the publication of the results, per the CONSORT statement.

10. Timeline Duration from January 2026 (months 1-16). Activities, in order: literature search; definition of objectives; review of forms; organization of execution; data collection; data processing; data analysis; writing of results.

11. Sample size and sampling The sample size was estimated for the comparison of two independent proportions corresponding to the primary safety outcome - the presence of at least one adverse event during treatment - assuming a two-sided significance level of 5% and 80% power.

    The expected proportion in the dual arm was set at 0.19, taken from the randomized trial by Jiang et al., whose experimental arm used vonoprazan 20 mg every 12 hours and amoxicillin 1 g every 8 hours for 14 days - identical to the experimental arm here - and in which the overall incidence of adverse events was 19% among 200 participants.

    No point estimate was adopted for the comparator arm. The control arm of that trial received rabeprazole, amoxicillin, clarithromycin and bismuth citrate - a bismuth quadruple regimen with two antibiotics, with a 53% incidence of adverse events - which is not the comparator here, since the present study uses non-bismuth concomitant quadruple therapy with three antibiotics, metronidazole included. Published estimates for concomitant therapy span a wide range, around 40% and with reports reaching 90%, which argues against a single expected value as the basis for the calculation. The sample size is therefore justified by declaration of the minimum detectable magnitude, computed with the standard formula for the comparison of two independent proportions.

    With 50 participants per arm (100 in total), a two-sided significance level of 5% and p₁ = 0.19, the study achieves 80% power to detect an absolute difference of 26 percentage points or more in the frequency of adverse events, corresponding to an incidence of 45% or more in the comparator arm; for a difference of 30 points, power rises to 90%. It is expressly noted that differences below 26 percentage points cannot be ruled out by this trial, and that if the comparator arm frequency were at the lower end of the published range - around 40% - power would fall to approximately 64%. This limitation will be reported explicitly in the publication of the results. The calculation dichotomizes the outcome into presence or absence of adverse events, whereas the primary analysis uses the four-category ordinal variable in an ordinal logistic regression; insofar as the proportional odds assumption holds, the ordinal model uses the severity information that dichotomization discards, so the estimate above is conservative with respect to the effective power of the primary analysis.

    Non-probabilistic consecutive sampling will be used: at each center all patients meeting the eligibility criteria will be invited, in the order in which they are diagnosed, until the quota assigned to that institution is complete. The distribution is equitable by design, with 25 participants per center; two reserve numbers are assigned to Hospital de Clínicas as coordinating center to cover any replacements, hence the numbering range of 1 to 102. It does not derive from the documented clinical volume of each center but from the decision to guarantee balanced representation of the four institutions, assuming comparable outpatient flows, and it allows, together with permuted block randomization stratified by center, simultaneous balance of the arms within each institution.

    Should any institution fail to complete its quota within the period foreseen in the timeline, the remaining participants may be redistributed among centers with recruitment capacity available; such a decision will be documented, communicated to the Research Ethics Committee and reflected in the CONSORT flow diagram. The final distribution by center will be reported, and center will be entered as a covariate into the analysis model.

12. Recruitment A treating gastroenterologist will invite all patients with a positive H. pylori result at the participating centers who meet the eligibility criteria, and will request their signature on the informed consent form after clarifying any questions (Appendix 1). That gastroenterologist will begin completion of the patient form and finalize it after the intervention (Appendix 5).

13. Allocation of interventions Once consent has been signed and the Treating Physician Form started, one of the two regimens will be provided, in opaque envelopes of identical characteristics assembled beforehand, each containing all the necessary tablets, a standardized treatment instruction sheet (Appendices 2 and 3), the patient information sheet and informed consent form, and a telephone number for queries.

    The allocation will be communicated by a person external to the research, following a random sequence previously established in an Excel spreadsheet to which, as a masking measure, only he or she has access. That spreadsheet was constructed with permuted block randomization stratified by the 4 participating centers, with a non-reproducible cryptographic seed. At no time will the research team or the treating physician know or access the allocation received, so as to maintain blinding of both the sequence and the treatment.

    The assigned treatment will be apportioned into the exact number of tablets required, with partial retention of its original packaging (blister format and apportioned tablets). Envelopes remain sealed until the visit at which treatment is handed over; beforehand they are stored under strict control at the Academic Unit of Gastroenterology, without participation of the Pharmacy department of Hospital de Clínicas, and then distributed to each participating center. Apportioning and packaging are performed by physicians and non-physician health personnel (nursing team).

14. Data collection and management At least 1 month after completion of treatment, the physician will hold a new visit to assess the eradication test result, collect the Adherence and Adverse Events questionnaire (Appendix 4) completed by the patient, and complete the patient form (Appendix 5). All forms will be kept in a folder per patient, under strict control and accessed only by the treating gastroenterologist during the intervention; once the intervention stage is complete, the folders will be collected and a non-investigator will complete the anonymized database.

    General coordination will be the responsibility of Dr. Arocena, principal investigator, who will coordinate data entry prior to treatment, store the informed consent forms and collect the questionnaires. A data management system will be created for form entry and for patient control and follow-up. All forms will be entered without identifiers, and a password for the exclusive use of the investigators will be required to access the database.

15. Statistical analysis The analysis will be carried out with Stata version 18.0. No interim analyses are planned.

Analysis populations. The safety population, used for the primary outcome, comprises all randomized participants who received at least one dose and returned the adverse events questionnaire. The intention-to-treat (ITT) population comprises all randomized participants, analyzed in the assigned arm irrespective of treatment received or completed; in the eradication analysis, any participant without an available eradication test is counted as a failure. The per-protocol (PP) population comprises participants who completed at least 80% of the regimen (11 days or more), received no antibiotics, PPIs or vonoprazan between completion of treatment and the test, and have a valid test performed at least 45 days after initiation. ITT is primary for eradication; PP is reported as a sensitivity analysis.

Losses to follow-up. Any participant who does not attend the post-regimen test visit is a loss to follow-up. Losses will be described by arm with their reasons and represented in the CONSORT flow diagram. Baseline characteristics of those lost will be compared with those retained, to assess the plausibility of the missing-at-random assumption. An overall loss above 20% will be reported explicitly as a limitation.

Handling of missing data. The proportion of missing data by variable and arm will be reported. Below 5% in the primary outcome, a complete-case analysis will be performed; at or above 5%, multiple imputation by chained equations (20 imputed sets) under the missing-at-random assumption, with baseline variables and assignment arm in the imputation model and estimates combined by Rubin's rules. As a sensitivity analysis, a worst-case imputation will assign the most unfavorable outcome to participants lost from the experimental arm and the most favorable to those lost from the comparator arm.

Descriptive analysis. Qualitative variables will be described by absolute and relative frequencies; quantitative variables by mean and standard deviation or median and interquartile range, according to the distribution assessed by graphical inspection and the Shapiro-Wilk test. A table of baseline characteristics by arm will be presented without hypothesis testing between groups, per the CONSORT recommendations.

Primary outcome and effect measures. The primary outcome is the maximum severity reached by each participant across the symptoms collected, as a four-category ordinal variable (absent, mild, moderate, severe). Bivariate comparison between arms will use the chi-square test or Fisher exact test according to the expected frequencies. Effect measures will be reported with 95% confidence intervals in all cases: for the presence of at least one adverse event, the risk difference with a Newcombe interval and the relative risk; for ordinal severity, the proportional odds ratio from the ordinal logistic regression model; for eradication rate and adherence, the risk difference and relative risk. Effect magnitude and precision will be favored over the dichotomy of statistical significance.

Multivariable model and covariates. An ordinal logistic regression model will be estimated with adverse event severity as the dependent variable. The proportional odds assumption will be evaluated by the Brant test; should it not hold, a partial proportional odds model will be used or, failing that, multinomial logistic regression. Covariates are specified in advance: treatment arm, recruitment center, age, sex and smoking. The selection is deliberately pa