Bayesian network meta-analysis suggests a similar effectiveness between a monovalent and a pentavalent rotavirus vaccine

Abstract

To assess the comparative effectiveness of a monovalent and a pentavalent rotavirus vaccine (RV1 and RV5), a Bayesian network meta-analysis was conducted. Data of randomized trials from the Cochrane Review in 2012 were extracted and synthesized. For the prevention of severe rotavirus disease up to 2 years, no statistical difference was found in the effectiveness between the 2 types of vaccine (odds ratio: 2.23, 95% credible interval: 0.71–5.20). Similarly, the comparative effectiveness of RV1 and RV5 appeared equivalent for other rotavirus-associated outcome measures, such as prevention of severe disease up to 1 year and all severity of rotavirus infections for up to both 1- and 2-year follow-ups. These results indicates that, overall, RV1 and RV5 offer similar benefits to prevent rotavirus diseases; nonetheless, credible intervals are generally wide, highlighting the necessity of further meta-analyses including updated information or, ideally, controlled trials comparing both vaccines directly.

Keywords: :

• monovalent rotavirus vaccine
• pentavalent rotavirus vaccine
• network meta-analysis
• comparative effectiveness research
• Cochrane Database Systematic Review

Introduction

Rotavirus causes a substantial burden among infants and young children in both developed and developing countries.¹ An example of this burden is in the United States, where rotavirus gastroenteritis was related to 55 000–70 000 hospitalizations or 20–60 deaths in 2006.² Globally, in 2008, diarrhea attributable to rotavirus infection resulted in an estimated 453 000 deaths in children aged younger than 5 y.³

Rotavirus gastroenteritis is a highly vaccine-preventable disease. Currently, 2 forms of this vaccine are available: a monovalent vaccine (RV1: Rotarix [GlaxoSmithKline Biologicals]) and a pentavalent vaccine (RV5: RotaTeq [Merck and Co., Inc.]).^4,5 Both vaccines are reported to be highly effective in preventing rotavirus diseases. For example, the Cochrane Database Systematic Review (CDSR) found that both RV1 and RV5 are related to 90% of the reduction of severe rotavirus infection compared with placebo.⁶ A large amount of evidence has been reported regarding the effectiveness of rotavirus vaccines, all of which support the high effectiveness of vaccines.⁷ However, to date, all of such studies evaluated vaccine effectiveness vs. placebo and no head-to-head trials have compared the effectiveness of RV1 and RV5. Therefore, it is yet be answered whether one type of vaccine is superior to another.

When there exist no trials comparing 2 or more interventions directly, traditional or pairwise meta-analysis is of limited value to address the clinical question of which treatment option is better.⁸ Recently, network meta-analysis has gained popularity. This type of analysis allows indirect comparison of treatments in the absence of head-to-head studies.^9,10

The rationale of the current study was to evaluate the comparative effectiveness of RV1 and RV5 by synthesizing researches comparing “RV1 vs placebo” with “RV5 vs placebo” in the 2012 CDSR.

Materials and Methods

Data were extracted from the CDSR in 2012, which summarized 41 rotavirus vaccine trials up to May in 2012.⁶ All of the studies in the CDSR were randomized controlled trials (RCTs) comparing placebo with rotavirus vaccine (RV1 or RV5) and none of the trials compared RV1 with RV5 directly. Although the CDSR has extensively reported outcomes for vaccine effectiveness including all-cause diarrhea or all-cause death, in the present study, analyses were limited to 4 major rotavirus-related outcomes: severe disease up to 1 y and 2 y, and rotavirus diarrhea of all severities up to 1 y and 2 y. The definition for “severe disease” was the same as that used by investigators in the original study. For evaluating these outcomes, a total of 31 trials out of 41 in the original systematic review were used. The list of included studies is available in the Tables S1–3.

Extracted data were synthesized in a Bayesian framework.¹¹ A random-effects model that can incorporate the potential underlying heterogeneity among studies,¹² such as geographical differences or quality of trials, was used.⁶ Model fitness was evaluated by residual deviance statistics; obtained results were sufficiently small for the number of trial arms, ensuring that the random-effects model was suitable for analysis of the present data set.¹² The programming code for network meta-analyses, which was developed by The National Institute for Health and Clinical Excellence,^12,13 was adopted. WinBUGS version 1.43 was used for statistical inferences.¹⁴ The comparative effectiveness of vaccines was simulated using the placebo as a common comparator and is presented as odds ratios (ORs) with 95% credible intervals (CrIs), which is an analog to 95% confidence intervals in traditional meta-analyses. In a Bayesian context, the 95% Crl directly shows that the true OR lies within the range of Crls with a probability of 95%.¹² There is no statistical testing or statistical measures, such as P values, in Bayesian statistics; judgment of significance relies on 95% Crls. Simulated results were obtained based on 50 000 samples from 3 chains, after 25 000 burn-in samples. The convergence was checked graphically and confirmed through the window that WinBUGS output (e.g., “history” plot) after the burn-in phase.¹²

Previous reports have suggested heterogeneity among studies, attributed to differences in the level of mortality in the country (low- vs high- mortality) or definitions of severe illnesses.^15,16 To examine whether these factors could affect the study results, subgroup analyses were conducted, when appropriate.

Results

Characteristics of included studies

Among 31 trials in the meta-analysis, 17 trials were conducted in low-mortality countries and 14 were in high-mortality countries, as defined by WHO mortality stratum (Tables S1–3). Fifteen RV1 trials comprised 10 clinical trials in low-mortality countries and 5 trials in high-mortally countries. In the 16 RV5 studies, 7 studies were performed in low-mortality nations and 9 were in high-mortality nations.

Vaccine effectiveness for the prevention of severe rotavirus disease was evaluated in a total of 20 trials (Tables S1 and S2). To define severe rotavirus illnesses, the Vesikari 20-point scale, the Clark 24-point scale, medical attentions requiring hospitalization/rehydration therapy and the combination of Vesikari scale and medical attentions were used in 12, 3, 2, and 3 trials, respectively. In one study investigating primary safety and immunogenicity of RV1, there was no clear definition of severe rotavirus diseases. A cut-off point of 16 was consistently used to define severe cases in all 3 studies using the Clark scale. However, among 15 trials with the Vesikari scale, the cut-off point varied, ranging from 8 to 13.

Network meta-analysis

The summary statistics of comparative effectiveness of 2 types of vaccine are shown in Table 1. Overall, the limits of Crls for all primary rotavirus-associated outcomes do not exclude 1, indicating that there are no statistical differences in the effectiveness between RV1 and RV5.

Table 1. Comparative effectiveness between RV1 and RV5

Outcome	Odds ratio^1 (95% Crl)	No. of studies (RV1, RV5)	No. of vaccines		No. of events
			RV1	RV5	RV1	RV5
Severe disease (up to 1 y)	1.75 (0.44–4.52)	16 (9, 7)	26 232	4130	126	40
Severe disease (up to 2 y)	2.23 (0.72–5.20)	17 (10, 7)	19 857	4568	144	119
All rotavirus diseases (up to 1 y)	1.32 (0.57–2.67)	13 (6, 7)	7749	6315	297	178
All rotavirus diseases (up to 2 y)	1.62 (0.77–2.92)	12 (6, 6)	5861	4542	148	314

¹ Odds ratio > 1 favors RV1 over RV5.

The network meta-analysis also identified, as expected, that RV1 and RV5 have high odds for preventing rotavirus disease compared with placebo (Table 2).

Table 2. Effectiveness of RV1 and RV5 compared with placebo

Outcome	RV1 (OR with 95% Crl)	RV5 (OR with 95% Crl)
Severe disease (up to 1 y)	5.52 (2.96–12.7)	3.60 (1.72–11.0)
Severe disease (up to 2 y)	5.68 (3.21–12.0)	2.83 (1.51–6.83)
All rotavirus diseases (up to 1 y)	3.50 (2.11–6.54)	2.89 (1.77–5.11)
All rotavirus diseases (up to 2 y)	3.11 (1.96–5.39)	2.06 (1.41–3.36)

Odds ratio > 1 favors RV1 or RV5 over placebo.

When subgroup analyses were conducted to examine whether the comparative efficacy of RV1 and RV5 differed between low- and high-mortality countries, the results were similar to the results presented above and one type of vaccine was not superior to other (Table 3). Because of a large variation in the definition of severe gastroenteritis, subgroup analyses stratified by severity scale could not be performed.

Table 3. Comparative effectiveness between RV1 and Rv5 stratified by the level of mortality in countries

	Country Level
Outcome	Low-Mortality Countries	High-Mortality Countries
Severe Disease (up to 1 y)	1.54 (0.01–5.46)	2.04 (0.38 –4.38)
Severe Disease (up to 2 y)	0.63 (0.04–2.32)	1.18 (0.40–2.42)
All Rotavirus Disease (up to 1 y)	1.91 (0.06–4.78)	1.20 (0.28–4.94)
All Rotavirus Disease (up to 2 y)	2.04 (0.38–4.64)	1.33 (0.10–14.3)

Presented in odds ratio with 95%Crls; odds ratio > 1 favors RV1 over RV5.

Discussion

This indirect meta-analysis, reanalyzing data from the CDSR, observed that the comparative effectiveness of RV1 and RV5 was equivalent for preventing rotavirus-associated outcomes. To date, this is the first report to evaluate vaccine effectiveness by network meta-analysis, a recently developed meta-analytic technique.

The significance of this study lies in the current context of comparative effectiveness research (CER).^17,18 By definition, CER is the conduct and synthesis of research comparing the benefits of different interventions in “real world” settings.¹⁸ CER can help in the decision making of healthcare providers, including decisions on vaccination policy. At a clinician-patient level, CER may provide an answer for which vaccine works better among available vaccines. CER can also address the question of which vaccine is more cost effective (e.g., at a country level). As illustrated by examples, CER is essential for those who are involved in healthcare systems, from physicians to stake holders. However, unfortunately, direct evidence from head-to-head trials is often lacking in RCTs,⁹ as exampled by rotavirus vaccine research. Similarly, standard meta-analysis that can compare only 2 interventions at a time is not feasible when head-to-head trials are lacking.^8,10 Therefore, comparative effectiveness of available treatment options remains largely unknown, even in the era of evidence-based medicine.^17,18 For these issues, network meta-analysis, which enables assessment of relative effectiveness in the absence of direct comparison, is now increasingly recognized as a useful tool for CER. Hopefully this study, which is the first network meta-analysis, is an initial step for CER in vaccine studies.

The findings in this report are consistent with results from 2 recent observational studies that reported similar efficacy between RV1 and RV5.^19,20 Cortese et al.¹⁹ reported that, in 2013, RV1 had 91% efficacy in reducing rotavirus-associated emergency care using a cohort of children with gastroenteritis. They also found that RV5 had 92% efficacy using the same cohort. The efficacy of RV1 and RV5 was also similar when the control cohort captured by an electronic immunization information system was used for calculation. Another study by Payne et al.²⁰ in 2013 reported that there were no statistical differences between RV1 and RV5 against medically attended rotavirus gastroenteritis, indicating that both vaccines share a similar efficacy. These observational studies and the present network meta-analysis supplement the absence of direct comparison between RV1 and RV5 in RCTs.

This study is preliminary because the results presented here were obtained by re-analysis of data from an existing systematic review,⁶ using a different statistical methodology. The process of conducting a meta-analysis comprises 2 crucial steps: comprehensive literature retrieval and data synthesis. A de novo literature search would waste valuable resources if good systematic reviews with comprehensive searches already exist.²¹ Because of this reason, as well as limited human resources for a literature search, existing data from the CDSR that identified relevant clinical trials of rotavirus vaccines were used. Using these existing data also allowed challenge of complex data synthesis, by minimizing the effort of literature review. The next logical step is to undertake an updated network meta-analysis on the same topic, integrating RCTs published since 2012.

There are 3 potential limitations in this study. First, there were relatively wide Crls across 1 (Table 1), which may indicate that this study does not provide conclusive results. The wide Crls can be explained by 2 factors, both of which arose from the nature of the included studies.²² The first factor is the relatively small number of trials in each intervention arm. The second factor is that the number of participants with unfavorable outcomes (e.g., severe rotavirus infection) was small because both vaccines were highly effective. For more precise estimation of comparative effectiveness of RV1 and RV5, further RCTs may be required. Another limitation is that large variation in the definition of severe illnesses did not allow subgroup analyses to be conducted. Therefore, the comparativeness effectiveness to prevent severe rotavirus gastroenteritis could be biased, arising from the inconsistency of the definitions of severe cases. However, the present study provides robust data as for comparative effectiveness for preventing rotavirus disease of all severities. Additionally, it is possible that random-effects model incorporated with this heterogeneity, to some extent, as well as for effectiveness for the prevention for severe rotavirus infections.¹² Therefore, the author believes that the presented results were not significantly different, despite the variation of definition of severe cases. The final limitation is that the number of vaccinees in RV1 and RV5 was unbalanced (Table 1), which could occur in the presence of a publication bias.²³ Therefore, study results might be affected by reporting bias, if any.

In summary, this network meta-analysis finds that the effectiveness of RV1 and RV5 is similar, though further researches are needed for more precise estimation of effectiveness. Currently, direct comparison of both vaccines is lacking. The result of this study could support clinical decision making for all healthcare providers.

Abbreviations:
RV1	=	monovalent rotavirus vaccine
RV5	=	pentavalent rotavirus vaccine
CDSR	=	Cochrane Database Systematic Review
RCT	=	randomized controlled trial
CER	=	comparative effectiveness research

Disclosure of Potential Conflicts of Interest

No potential conflicts of interest were disclosed.

Acknowledgments

The author thanks Dr Mayumi Hangai and Dr Yoichiro Oda (Chigasaki Municipal Hospital) for critical review of the first draft of this manuscript. There is no specific funding relevant to this study.

10.4161/hv.28284