Antiemetics in Children With Acute Gastroenteritis: A Meta-analysis

This systematic review was registered in the PROSPERO International Prospective Register of Systematic Reviews (CRD42016035236). This article complies with the recommendations of the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) extension for NMA.¹¹ 

We searched Medline (Ovid), Embase (Ovid), Cochrane Central Register of Controlled Trials, Cumulative Index to Nursing and Allied Health Literature, and Latin America and the Caribbean Literature on Health Sciences from the inception to December 31, 2018. Search strategies were developed in liaison with an experienced librarian (Supplemental Information). We used validated filters for identifying pediatric articles and RCTs.^7,12  No language or publication status limits were used. We performed gray literature searches through trial registries (www.clinicaltrials.gov and World Health Organization Clinical Trials Registry Platform).

We included RCTs and quasi RCTs in which authors evaluated antiemetics used for controlling vomiting in children with ADG. Our interventions of interest were metoclopramide, ondansetron, domperidone, dexamethasone, dimenhydrinate, alizapride, and granisetron at any dose and presentation in children with ADG and vomiting. Researchers had to compare any of the interventions against them, a placebo, conventional treatment with ORT, or different doses or administration routes of the same intervention and had to report at least 1 of the outcomes of interest.

Our primary outcomes were cessation of vomiting and hospitalization. The secondary outcomes included the need for intravenous rehydration (measured as the number of participants who required intravenous rehydration during the ED stay and up to 3 days after discharge); revisit to the ED (measured as the number of participants that revisited the ED up to 72 hours after discharge); number of vomiting (measured as the mean number of vomiting episodes during the observation period); and side effects.

Regarding side effects, as a post hoc analysis, we analyzed side effects (any side effect reported by the authors) and diarrhea separately. We separated the outcomes because diarrhea was reported as a dichotomous (presence or absence of diarrhea in the observation) and continuous variable (mean number of diarrheal stools), with the latter being the most commonly reported. The presence of diarrhea (dichotomous) was analyzed along with the rest of the side effects, but it was not possible to combine with the continuous data. Lastly, the worsening of diarrhea has been described as a major concern for the use of ondansetron in previous systematic reviews^9,10  and CPGs.^4,7  Thus, we wanted to determine the specific effect on the number of stools beside the incidence of side effects to better inform further clinical decision-making.

Two reviewers (L.F.N.-S. and J.A.-R.) performed independently and in duplicate the screening of available titles and abstracts to assess their eligibility. Studies were retrieved in full text if either one of the reviewers considered them eligible. Potentially eligible studies were reviewed, and studies were included if both reviewers agreed on their eligibility. In case of disagreement, a third reviewer (I.D.F.) resolved it. We tried to contact authors of primary studies during data extraction for missing information.

We used a prespecified and piloted form to extract the data. Among the extracted data were study characteristics (design, year, duration of follow-up, sample size, setting); patient characteristics (age, impatient or outpatient, days of disease, hydration status); intervention details (doses, administration forms); and outcome results (number of events, mean and SD or SEs, per arm) at the longest duration of follow-up. Two reviewers (L.F.N.-S. and J.A.-R.) independently and in duplicate conducted data extraction. When consensus was not reached, a third reviewer was involved (I.D.F).

We assessed, independently and in duplicate, all included studies for their risk of bias (RoB) using a modified version of the Cochrane RoB tool¹³  on the basis of the following criteria: sequence generation, allocation concealment, blinding of participants, personal and outcome assessors, completeness of follow-up, selective outcome reporting, and other biases. For each criterion, an RoB score was assigned as “definitely low,” “probably low,” “probably high,” or “definitely high” risk.¹⁴  Disagreements were resolved by consensus, and a third reviewer was involved (I.D.F) when consensus was not reached.

We performed a pairwise random-effects meta-analysis of each available direct comparison. Treatment effects were estimated using odds ratios (ORs) for dichotomous outcomes and mean differences (MDs) for continuous outcomes, along with their 95% credible intervals (CIs). We used vague priors for all model parameters and a common half-normal prior distribution for the between-study SD (τ∼N [0,1], τ > 0) across all treatment comparisons per outcome, given that many treatment comparisons were informed by single studies (20). Heterogeneity in pairwise meta-analysis for all the direct comparisons was quantified with the statistic for heterogeneity in direct comparisons (I²) expressed as a percentage of variability that is due to true differences between studies rather than sampling error.¹⁵  All analyses were performed by using the Markov chain Monte Carlo method. A geometry plot was used to present all the available direct comparisons per outcome, in which each node represents one intervention.

We performed an NMA to analyze all the potential comparisons among interventions for each outcome. An NMA, also known as multiple-treatment comparisons or multiple-treatment meta-analysis, is a special statistical technique that provides a methodology to address the issue of having available many interventions for the same condition under study, mostly compared against a placebo but less or not compared against each other.¹⁶  NMA takes advantage of 2 statistical approaches. First, it takes advantage of the use of indirect comparisons: we can estimate the effect of intervention A versus intervention B, indirectly if both A and B have been compared against an intervention C (usually a placebo). Second, the combination of direct and indirect comparisons allows researchers to obtain more precise estimates (ie, narrower confidence intervals or credible intervals in the results).¹⁶  With an NMA, we can obtain an effect estimate to determine the differences between any pair of interventions, even if they have not been directly compared, and summarize all the available evidence in one single study.

For each outcome and a connected network of studies, we performed a Bayesian random-effects NMA if the assumptions of between-study homogeneity, transitivity, and incoherence across treatment comparisons were judged to be justifiable. Transitivity¹⁷  is the assumption that an indirect comparison is a valid method to compare 2 treatments because the studies are sufficiently similar in important clinical and methodological characteristics, or in other words, they are similar in their distributions of effect modifiers.¹⁸  Incoherence (also called inconsistency) is defined as the statistical difference between direct and indirect treatment effects.¹⁹ 

In the absence of direct evidence for a given comparison, we indirectly estimated treatment effectiveness and safety. In the presence of both direct and indirect evidence, the NMA provided a combined effect estimate.²⁰  A Bayesian hierarchical model with vague priors adjusting for correlation of multiarm trials was fitted. After discarding the first 10 000 iterations, series of 100 000 burn-in simulations with thinning of 10 values were used to allow convergence. The model convergence was checked by visual inspection of the evaluation of the mixing of 2 chains. The analysis was performed in OpenBUGs (version 3.2.3).²¹ 

Variables used for the assessment of the transitivity assumption included the mean number of vomiting before recruitment (fewer or >4 episodes per hour), the follow-up time of outcome measurement (less than and >12 hours), and the route of administration (intravenous or oral). The statistical incoherence between the direct and indirect estimates was assessed with both a global χ² test by using the random-effects design-by-treatment interaction model²²  and a local z test by using the loop-specific approach calculating the ratio of OR.²³ 

We conducted meta-regression, sensitivity, and subgroup analyses to explore the potential sources of heterogeneity and incoherence. Meta-regression was performed by using the number of vomiting episodes before the recruitment as the independent variable. Three sensitivity analyses were conducted on the basis of the RoB, excluding studies with (1) overall high RoB, (2) high RoB because of allocation concealment, and (3) high RoB because of incomplete outcome. Lastly, we performed three subgroup analyses for route of administration, vomiting number before recruitment, and time of follow-up for outcome measurement. Our hypotheses were as follows: the effect of the interventions might be inferior with oral medication (versus intravenous) in children with >4 episodes of vomiting per hour (versus <4 episodes) or when outcomes are measured >12 hours after the recruitment (versus <12 hours).

When 10 or more studies were available for an outcome, we assessed small-study effects and publication bias using the comparison-adjusted funnel plot,²⁴  which was used to inform the Grading of Recommendations, Assessment, Development, and Evaluation (GRADE) assessment (see Rating the Confidence in the Effect Estimates section below). We calculated the surface under the cumulative ranking (SUCRA) curve values to rank the available treatments according to their efficacy, and we captured the uncertainty in the parameter values that informed treatment rankings calculating their corresponding 95% CI.^25-27  We graphically depicted the SUCRA curve values for all outcomes in a rank-heat plot.²⁸ 

Reviewers (L.F.N.-S. and J.A.-R.), in pairs and independently, assessed the quality of evidence for each reported outcome according to the GRADE approach.¹⁹  Any disagreement was resolved by a third reviewer (I.D.F.). We rated confidence as high, moderate, low, or very low. The direct comparisons assessment was based in 5 categories: study limitations (RoB),²⁹  imprecision,³⁰  inconsistency,³¹  indirectness,³²  and publication bias.³³  For NMA, the approaches by Puhan et al¹⁹  and Brignardello-Petersen et al³⁴  were applied. These consider, in addition, the assessment of intransitivity and incoherence criteria.

With the aim of optimizing results interpretation and clinical applicability, we present a summary using a novel approach that has been previously described to summarize results from NMA.³⁵  We grouped the interventions according to the magnitude of the effect in comparison to a placebo and the quality of evidence (according to the GRADE approach). The different categories (marked by different colors) are displayed in Fig 1. Dark colors represent interventions with moderate- to high-quality evidence (high certainty on the results). Light colors represent interventions with very low– to low-quality evidence (low certainty on the results).

We identified 3196 titles from databases and 4 additional records through other sources. After removing duplicates, 1840 titles and abstracts were screened. Sixty-six studies were identified for full-text screening. We excluded 42 studies because of reasons presented in Supplemental Table 4 and included 24 RCTs^36–59  enrolling 3482 children. In Supplemental Table 5, we describe the characteristics of included studies. The flow diagram of the study selection is shown in Fig 2. The eligible studies were conducted in 16 countries from 5 continents. The mean number of vomiting episodes before recruiting was 7.09 (SD = 4.28) and the age of children across the studies was 35.1 months (SD = 24.3; range: 5.2–120.6). The interventions studied were metoclopramide, ondansetron, domperidone, dexamethasone, dimenhydrinate, and granisetron, mostly compared against a placebo. The search did not retrieve studies on alizapride. The network geometry plots with the available direct comparisons for the 6 outcomes are shown in Fig 3.

Among the included studies, 6 (25%) revealed concerns for high RoB due to allocation concealment and blinding of participants and/or outcome assessors. Five studies had high RoB because of incomplete outcome reporting and 5 studies because of inadequate sequence generation. The RoB assessment is described in Supplemental Table 6.

We conducted pairwise meta-analyses in 10 direct comparisons (2627 patients) for cessation of vomiting. According to the I² result, heterogeneity was low to medium in all comparisons except for the placebo-dimenhydrinate comparison (I² = 53.5%). Ondansetron was found to be better than metoclopramide, dexamethasone, and placebo in direct meta-analyses. The remaining treatment comparisons revealed no statistical differences (Supplemental Table 7).

In the NMA, we obtained 21 paired effect estimates. Ondansetron revealed the largest effect in comparison to the placebo (placebo versus ondansetron; OR = 0.28 [CI = 0.16 to 0.46]) with a high quality of evidence. Ondansetron was also better than metoclopramide (ondansetron versus metoclopramide; OR = 3.27 [CI = 1.20 to 9.19]) (Supplemental Table 7). No incoherence was found with the global assessment (P = .95). The loop-specific approach also revealed no statistically significant incoherence, but the high ratio of OR may suggest that some degree of incoherence exists (Supplemental Fig 6). Ondansetron was the best intervention measured with the SUCRA values (SUCRA = 1.0) (Supplemental Table 8). Forest plots and funnel plots are displayed in Supplemental Figs 7 and 8, respectively.

In the meta-regression analysis using the number of vomiting before recruitment as a covariate, a marginally significant coefficient was obtained (β = 0.24; CI = 0.01 to 0.48; on log OR scale) (Supplemental Table 9). In the subgroup analysis by route of administration, in comparison to the placebo, both oral (placebo versus ondansetron; OR = 0.34 [CI = 0.17 to 0.67]) and intravenous ondansetron (placebo versus ondansetron; OR = 0.21 [CI = 0.07 to 0.53]) were found to be effective (Supplemental Table 10). In the subgroup analyses by severity of the episodes, only ondansetron was better than the placebo (placebo versus ondansetron; OR = 0.32 [CI = 0.18 to 0.56]), than domperidone (domperidone versus ondansetron; OR = 0.34 [CI = 0.14 to 0.90]), and better than metoclopramide (metoclopramide versus ondansetron: OR = 0.31 [CI = 0.1 to 0.83]) in the subgroup of <4 episodes per hour. We found no differences in the subgroup of >4 episodes per hour. In the subgroup analysis by time of follow-up, ondansetron was similarly effective, although the effect was larger when follow-up was <12 hours than when follow-up was >12 hours (Supplemental Table 10). Three sensitivity analyses based on the RoB were performed; ondansetron maintained its effect across all the analyses (Supplemental Table 11).

Thirteen studies provided information on hospitalization rates (2008 patients). In the pairwise meta-analyses, ondansetron was better than domperidone (domperidone versus ondansetron; OR = 2.72 [CI = 1.56 to 5.89]) and better than the placebo (placebo versus ondansetron; OR = 3.63 [CI = 1.16 to 21.3]). Heterogeneity was low in all the comparisons except for the domperidone-ondansetron comparison (I² = 65.9%). In the NMA, ondansetron was better than the placebo (placebo versus ondansetron; OR = 2.93 [CI = 1.69 to 6.18]) and also better than domperidone (domperidone versus ondansetron; OR = 3.31 [CI = 1.21 to 15.8]). The remaining comparisons revealed no differences (Supplemental Table 12). We found no incoherence with the global test (P = .21) or the loop-specific approaches (Supplemental Fig 9). The SUCRA values, forest plots, and funnel plot are displayed in Supplemental Table 13 and Supplemental Figs 10 and 11, respectively.

In the meta-regression analysis using the number of vomiting episodes before recruitment as a covariate, we found no association (β = −0.04 [CI = −0.24 to 0.15]; on log OR scale) (Supplemental Table 14). In the subgroup analysis by route of administration, the placebo compared to oral ondansetron was found effective (placebo versus ondansetron; OR = 3.29 [CI = 1.65 to 8.56]) in contrast to intravenous ondansetron (placebo versus ondansetron; OR = 2.19 [CI = 0.44 to 14.5]) (Supplemental Table 15). In the subgroup of <4 episodes of vomiting per hour, only ondansetron was more effective than the placebo (placebo versus ondansetron; OR = 2.06 [CI = 1.18 to 3.8]). We found no differences when vomiting frequency was >4 episodes per hour. Lastly, ondansetron revealed to be effective when follow-up was >12 hours after the intervention (placebo versus ondansetron; OR = 2.27 [CI = 1.08 to 5.55]) (Supplemental Table 15). Ondansetron maintained its effect across all the sensitivity analyses on the basis of different RoB criteria (Supplemental Table 16). In Fig 4, we display the league table with all the NMA effect estimates for both primary outcomes.

We analyzed 10 comparisons (1544 patients) that measured the need for intravenous rehydration. In the pairwise meta-analysis, ondansetron was better than metoclopramide (ondansetron versus metoclopramide; OR = 0.03 [CI = 0.00 to 0.46]) and better than the placebo (placebo versus ondansetron; OR = 3.22 [CI = 2.02 to 5.43]) (Supplemental Fig 12). In the NMA, ondansetron revealed the greatest effect in comparison to the placebo (placebo versus ondansetron; OR = 3.0 [CI = 1.9 to 5.1]; moderate quality) (Supplemental Fig 13). Ondansetron was also more effective than metoclopramide (ondansetron versus metoclopramide; OR 0.02 [CI = 0.00 to 0.48]; moderate quality) (Supplemental Table 17). The incoherence, SUCRA values, and funnel plot for the outcome need for intravenous rehydration are displayed in Supplemental Fig 14, Supplemental Table 18, and, Supplemental Fig 15, respectively.

Seventeen studies reported the number of vomiting episodes (2504 patients). In the direct meta-analysis, ondansetron was better than the placebo (placebo versus ondansetron; MD = 1.46 [CI = 0.74 to 2.63]) and granisetron was more effective than the placebo (placebo versus granisetron; MD = 0.60 [CI = 0.11 to 1.09]) (Supplemental Fig 16). Heterogeneity was low except for the comparisons of the placebo versus ondansetron (I² = 89.1%), placebo versus domperidone (I² = 80.6%), and placebo versus dimenhydrinate (I² = 43.4%) (Supplemental Table 19). In the NMA, only ondansetron was more effective than the placebo (placebo versus ondansetron; MD = 1.48 [CI = 0.81 to 2.62]; very low quality). No incoherence was found with the global or the local assessment (P = .99) (Supplemental Fig 17). The SUCRA values, forest plots of NMA, and funnel plot for number of vomiting episodes are displayed in Supplemental Table 20 and Supplemental Figs 18 and 19, respectively.

Twelve studies had information about the revisiting (1763 patients). In the pairwise meta-analysis, the placebo was better than granisetron (placebo versus granisetron; OR = 0.31 [CI = 0.09 to 0.87]) (Supplemental Fig 20). In the NMA, none of the interventions revealed differences to the placebo (Supplemental Table 21; Supplemental Fig 21). The incoherence, SUCRA values, and funnel plot for revisiting are displayed in Supplemental Fig 22, Supplemental Table 22, and Supplemental Fig 23, respectively.

In 12 studies, the authors reported side effects (1816 children, 5 treatments). In 4 studies, the authors reported that no side effects were found. In the NMA, dimenhydrinate was the only intervention that revealed significantly more side effects than the placebo, including somnolence, sleepiness, sedation, and drowsiness (placebo versus dimenhydrinate; OR = 0.14 [CI = 0.01 to 0.7]; very low quality) (Supplemental Table 23; Supplemental Figs 24 and 25). The incoherence analyses, SUCRA values, and funnel plot for side effects are displayed in Supplemental Fig 26, Supplemental Table 24, and Supplemental Fig 27, respectively.

Regarding diarrhea, domperidone revealed a reduction of the number of stools in comparison to ondansetron in the pairwise meta-analysis (domperidone versus ondansetron; MD = −1.25 [CI = −1.34 to −1.15]) (Supplemental Fig 28) and in the NMA (dimenhydrinate versus ondansetron; MD = −2.04 [CI = −3.9 to −0.05]; very low quality) (Supplemental Fig 29; Supplemental Table 25). The incoherence, SUCRA values, and funnel plot for diarrhea are displayed in Supplemental Fig 30, Supplemental Table 26, and Supplemental Fig 31, respectively. A rank-heat plot summarizing the ranking statistic across all interventions and outcomes is displayed in Supplemental Fig 32.

In Fig 5, we summarize the results on the basis of the effect estimates and quality of the evidence. With high certainty, we found that for cessation of vomiting, hospitalization prevention and the need for intravenous rehydration, ondansetron was the only intervention categorized as the “best intervention.” The best intervention category means that, with high certainty, ondansetron was better than placebo and also better than at least one of the other interventions. For cessation of vomiting and the need for intravenous rehydration, ondansetron was also better than metoclopramide (Fig 4 and Supplemental Table 17, respectively); for hospitalization, it was also better than domperidone (Fig 4).

For the same outcomes, the remaining interventions were categorized as “similar to placebo” or as “may be similar to placebo.” The similar to placebo category means that, with high certainty, metoclopramide, domperidone, dexamethasone, granisetron, and dimenhydrinate were no different from the placebo. The category may be similar to placebo means that these interventions, with low certainty, seem to be similar to the placebo. Lastly, ondansetron was considered similar to the placebo, with high certainty, for side effects and as may be similar to placebo, with low certainty, for causing diarrhea (Fig 5)

In this systematic review and NMA, we evaluated available antiemetics for controlling vomiting in children with ADG. Moderate to high quality of evidence indicates that ondansetron is the best intervention for cessation of vomiting, preventing hospitalization, and the need for intravenous rehydration. There is no evidence to support the use of domperidone, dimenhydrinate, metoclopramide, alizapride, or granisetron for the cessation of vomiting and preventing hospitalizations because they were classified as similar or may be similar to placebo in all the effectiveness outcomes.

Interestingly, in our subgroup analyses, we found that the effect of ondansetron seems to be larger when used orally (rather than intravenously) in children with a low number of vomiting episodes (<4 episodes per hour) and when the hospitalization outcome was measured after 12 hours. Meta-regression revealed a significant coefficient in cessation of vomiting, meaning that the larger the number of vomiting episodes before recruitment, the lower the effects of antiemetics. Although these results suggest that ondansetron should be administered orally and that its effect seems to be lower when vomiting is severe, these findings should be studied further.

Our study is the first NMA that includes all the currently available antiemetics used in children with ADG. A previous NMA included only 11 studies and studied only ondansetron, metoclopramide, granisetron, and dexamethasone.¹⁰  In contrast, we included 24 studies with additional evidence from dimenhydrinate and ondansetron. We also applied advanced statistical techniques to explore causes of heterogeneity and incoherence, and we assessed the quality of the evidence with GRADE. A previous Cochrane review⁹  also summarized all the available evidence but only performed direct comparisons and therefore did not study differences among interventions. Despite the differences between our work and previous reviews, we all concluded that ondansetron is most likely the best intervention. Nevertheless, we can now be more confident in these results because we provide updated evidence, include more studies and interventions, and provide more precise effect estimates (narrower CIs), all of which reduce uncertainty around the results.

We found that dimenhydrinate was the only intervention inferior to the placebo in terms of safety. Regarding diarrhea, all the interventions revealed no statistically significant differences against the placebo. Ondansetron has been previously associated with an increase in diarrheal episodes. In our results, we found that ondansetron increases diarrhea in comparison to dimenhydrinate but not in comparison to the placebo. Ondansetron may have a slight impact on the number of stools that require monitoring in some children, but in most of the cases, it will not be clinically significant.

Antiemetics use in children with ADG has been controversial. CPGs from organizations such as the World Health Organization,⁵  the National Institute for Health and Care Excellence,⁷  and the American Academy of Pediatrics⁴  do not recommend their use. Conversely, the European Society for Pediatric Gastroenterology, Hepatology, and Nutrition CPGs⁸  recommend ondansetron. This disagreement may be due to the lack of updated CPGs in the former organizations that did not consider recent evidence. Therefore, our results may be crucial for future CPG updates.

Our study has several strengths. We conducted a comprehensive systematic review including all the available evidence regardless of the language and publication status of the studies. Our review is based on statistical analyses that considered both NMA effect estimates and probability rankings, including subgroup, sensitivity, and meta-regression analyses. This allowed us to explore possible effect modifiers and prove the robustness of our results. We used the GRADE to appraise the quality of evidence and provided a straightforward presentation of our findings (Fig 5), which summarizes in a single resource the relative performance of each intervention per outcome, categorized by the certainty on the evidence. Lastly, we followed Cochrane¹³  and International Society for Pharmacoeconomics and Outcomes Research recommendations for developing a rigorous NMA.⁶⁰ 

However, our study is not free of limitations. The evidence in most treatment comparisons is of low or very low quality. This is the result of the presence of a significant RoB and imprecise estimates. The latter is explained by the lack of enough evidence from direct comparisons among the interventions given that most of the evidence comes from comparisons against placebo.

Our results may be beneficial for clinicians, researchers, guideline developers, and decision-makers. Clinicians count with updated evidence to support the use of ondansetron, which is the only intervention with moderate- to high-quality evidence that supports its use in children with ADG and vomiting. Considering the amount of evidence and its quality, when designing new trials, researchers should consider comparing new alternatives against ondansetron rather than against a placebo. Guideline developers may use our results for future CPG updates. CPGs that have not recommended antiemetics may want to consider ondansetron as an alternative in children at risk for failure of oral rehydration and to prevent hospitalization. Finally, decision-makers may consider ondansetron as the standard of therapy and facilitate the decision-making process about what interventions should be funded in health benefit plans or for reimbursement policies.

Ondansetron is the only intervention that, with moderate to high certainty, showed an effect on cessation of vomiting, hospitalization prevention, and the need for intravenous rehydration. There is no evidence to support the use of metoclopramide, dimenhydrinate, domperidone, alizapride, and dexamethasone in these patients. Ondansetron was found to be a safe intervention, whereas dimenhydrinate was the only intervention that produced more side effects than the placebo.

We thank Jesenia Avendaño, a health sciences librarian, for her assistance in the design of the search strategy.

ADG

acute diarrhea and gastroenteritis

CI

credible interval

CPG

clinical practice guideline

ED

emergency department

GRADE

Grading of Recommendations, Assessment, Development, and Evaluation

I²

statistic for heterogeneity in direct comparisons

MD

mean difference

NMA

network meta-analysis

OR

odds ratio

ORT

oral rehydration therapy

RCT

randomized clinical trial

RoB

risk of bias

SUCRA

surface under the cumulative ranking