Removing the Age Restrictions for Rotavirus Vaccination: A Benefit-Risk Modeling Analysis

Summary

Background:
To minimize potential risk of intussusception, the World Health Organization (WHO) recommended in 2009 that rotavirus immunization should be initiated by age 15 weeks and completed before 32 weeks. These restrictions could adversely impact vaccination coverage and thereby its health impact, particularly in developing countries where delays in vaccination often occur.

Methods and Findings:
We conducted a modeling study to estimate the number of rotavirus deaths prevented and the number of intussusception deaths caused by vaccination when administered on the restricted schedule versus an unrestricted schedule whereby rotavirus vaccine would be administered with DTP vaccine up to age 3 years. Countries were grouped on the basis of child mortality rates, using WHO data. Inputs were estimates of WHO rotavirus mortality by week of age from a recent study, intussusception mortality based on a literature review, predicted vaccination rates by week of age from USAID Demographic and Health Surveys, the United Nations Children's Fund (UNICEF) Multiple Indicator Cluster Surveys (MICS), and WHO-UNICEF 2010 country-specific coverage estimates, and published estimates of vaccine efficacy and vaccine-associated intussusception risk. On the basis of the error estimates and distributions for model inputs, we conducted 2,000 simulations to obtain median estimates of deaths averted and caused as well as the uncertainty ranges, defined as the 5th–95th percentile, to provide an indication of the uncertainty in the estimates.

We estimated that in low and low-middle income countries a restricted schedule would prevent 155,800 rotavirus deaths (5th–95th centiles, 83,300–217,700) while causing potentially 253 intussusception deaths (76–689). In contrast, vaccination without age restrictions would prevent 203,000 rotavirus deaths (102,000–281,500) while potentially causing 547 intussusception deaths (237–1,160). Thus, removing the age restrictions would avert an additional 47,200 rotavirus deaths (18,700–63,700) and cause an additional 294 (161–471) intussusception deaths, for an incremental benefit-risk ratio of 154 deaths averted for every death caused by vaccine. These extra deaths prevented under an unrestricted schedule reflect vaccination of an additional 21%–25% children, beyond the 63%–73% of the children who would be vaccinated under the restricted schedule. Importantly, these estimates err on the side of safety in that they assume high vaccine-associated risk of intussusception and do not account for potential herd immunity or non-fatal outcomes.

Conclusions:
Our analysis suggests that in low- and middle-income countries the additional lives saved by removing age restrictions for rotavirus vaccination would far outnumber the potential excess vaccine-associated intussusception deaths.



Please see later in the article for the Editors' Summary

Introduction

Rotavirus infection is the leading cause of fatal diarrhea among children younger than 5 y, accounting for 453,000 deaths in the year 2008 based on recently published World Health Organization (WHO) estimates [1]. To curb this large toll of severe rotavirus disease, in 2006, the WHO recommended two rotavirus vaccines—Rotarix (GSK Biologicals) and RotaTeq (Merck & Co.)—for use in Europe and the Americas, and in 2009, they expanded this recommendation to all children worldwide [2]. These recommendations reflected the growing availability of evidence of the good efficacy profile of rotavirus vaccines—first from clinical trials in high- and middle-income countries in the Americas and Europe in 2006 and then also from low-income settings in Africa and Asia in 2009 [3]–[6].

Because a previous rotavirus vaccine (RotaShield) was found to be associated with intussusception, a rare form of bowel obstruction [7], the pivotal pre-licensure trials in the Americas and Europe for both currently available rotavirus vaccines were conducted in over 60,000 infants each to exclude this risk; these trials did not show an increase in risk of vaccine-associated intussusception similar to that found with Rotashield [3],[4]. However, recent data on the postlicensure safety of rotavirus vaccines generated from these countries has suggested a possible low level risk of intussusception (∼one to two excess cases per 100,000 vaccinated infants) in some countries but not in others [8],[9]. On the basis of considerations that this low level risk is greatly exceeded by the observed health benefits of vaccination, national and international policy and regulatory bodies have continued to support recommendations for use of rotavirus vaccine [8],[9].

In 2009, WHO recommended that rotavirus vaccines should not be initiated for infants aged 15 wk or older, with all doses being completed by 32 wk [2]. These age restrictions were driven by concerns about intussusception risk. Natural intussusception rarely occurs before 3 mo of age and the incidence increases 10-fold between 3 and 6 mo of age [10]. Therefore, a constant vaccine-associated relative risk (RR) of intussusception, particularly with the first vaccine dose that has been primarily associated with risk, would translate to more excess cases if infants were vaccinated late, beyond 3 mo of age. Similar findings were observed in the United States after use of RotaShield, prompting a debate about whether restriction of RotaShield to infants younger than 3 mo of age would have averted withdrawal of the vaccine [10]–[12]. A consequence of these strict age restrictions in countries with vaccination delays is that those arriving late for immunization would potentially not have access to the benefits of rotavirus vaccination [13],[14].

To facilitate decision making, we previously undertook a scenario analysis assessing the benefits and risks of a rotavirus vaccination strategy with and without an age restriction [15]. Since this analysis, new evidence has been published on several key parameters for the scenario analysis, including data on efficacy of rotavirus vaccines in Africa and Asia [5],[6], the effect of rotavirus vaccines on diarrhea deaths [16],[17], postlicensure data on risk of intussusception with current rotavirus vaccines [8],[9],[18], the release of updated estimates of rotavirus mortality by WHO [1] and age distribution of rotavirus disease by week of age [19], and updated data on timeliness of vaccination coverage in low- and middle-income countries [20]. The availability of these new data and the imminent introduction of rotavirus vaccines in many developing countries in Africa during the next 2 y prompted us to revise our previous analysis to provide policy makers with the most up-to-date evidence to inform decisions of best approaches to global implementation of rotavirus vaccines.

Methods

We focused this analysis exclusively on the benefits of rotavirus mortality reduction and potential risk of fatal intussusception in children <5 y of age in 158 low- and middle-income countries with a birth cohort of 123.6 million where 99.9% of the global rotavirus mortality occurs. To explore the effect of age restriction in different parts of the world, we grouped these countries on the basis of child mortality rates, according to WHO mortality strata [21], and assigned to one of four groups: group B and C (countries with low child mortality), group D-Americas (countries in the Americas with high child mortality), group D-Asia (countries in Asia with high child mortality), and group D & E-Africa (countries in Africa with high child mortality). Because group A countries with very low child mortality (i.e., high-income) represent <0.1% of the global rotavirus deaths, they were excluded from this analysis.

Vaccination Strategies and Coverage Estimates

For both immunization strategies, restricted and unrestricted, we assumed that rotavirus vaccine would be given at the same time as the diphtheria-tetanus-pertussis (DTP) vaccine and that vaccine coverage in the individual countries would be equal to the proportion of infants receiving each of the three DTP doses by week of age (i.e., proportion vaccinated, ρ_v) during the first 3 y of life. Under the restricted schedule, if infants received their first DTP dose by ≤14 wk of age, we assumed they would receive all doses up to 32 wk of age, but if they first appeared after 14 wk, they would remain unvaccinated. On the unrestricted schedule, vaccine would be administered according to the age-specific coverage rates for each of the DTP dose up to 3 y of age.

Our DTP coverage estimates are based on vaccination data from household USAID Demographic and Health Surveys (DHSs) [22] and the United Nations Children's Fund (UNICEF) Multiple Indicator Cluster Surveys (MICS) [23] that were administered in 48 countries between 1996 and 2009. To estimate coverage for countries without DHS or MICS data, overall WHO-UNICEF 2010 country-specific coverage estimates were converted into age-specific coverage rates using regression coefficients to predict lognormal curves of timeliness. These were derived from the available DHS/MICS survey data and extrapolated to countries without a survey within a WHO region and mortality stratum. Timeliness was determined by WHO sub-region and adjusted for trends between the DHS/MICS survey year and 2010 using the WHO-UNICEF 2010 best estimates for DTP coverage data, drop-out rate between DTP1 and DTP3, the target age recommended in the country schedule, and the gross domestic product per capita [24]. This process was done separately for DTP1 and DTP3. DTP2 timeliness assumed the average of the regression coefficients used for DTP1 and DTP3.

Our analysis does not allow catch-up immunization and assumes no improvement in timeliness with the introduction of rotavirus vaccine.

Assessment of Benefits—Base Scenario

Estimated numbers of country-specific rotavirus deaths (λ_rv) were obtained from WHO, using the 95% CIs to define the triangular distributions around the point estimate (Table 1) [1]. On the basis of a WHO-sponsored review of published and unpublished studies on age distribution of diarrhea mortality and rotavirus-associated hospitalizations by week of age, we predicted 1-wk gamma age distributions for the first year of life and 4-wk age categories thereafter for countries in different WHO regions [19].

Rotavirus vaccine efficacy (ε_rv) against fatal rotavirus disease was estimated from clinical trials or vaccine effectiveness studies in each WHO region (Tables 1–2) [3],[5],[6],[25]–[29]. Because efficacy against rotavirus mortality could not be directly measured in the trials, we applied efficacy estimates against the most severe rotavirus disease outcome reported in the study [3],[5],[6],[25]–[29]. This approach was reasonable given that three nationwide studies from Latin America have documented reductions in diarrhea deaths after vaccine introduction that has approximated reductions based on the efficacy of these vaccines against severe rotavirus disease [16],[17],[30]. Because both rotavirus vaccines have performed similarly in clinical trials, we assumed the same overall efficacy for the two-dose Rotarix and the three-dose RotaTeq vaccine. The efficacy parameters were age-stratified (<1 y and >1 y of age) because studies have documented lower efficacy among children older than 1 y of age [5],[25],[27]. Efficacy of partial vaccination (first dose) was also available from one country in the B & C region [27], and one country in the D-Americas region [25], but not for D-Asia and D & E-Africa. We therefore reduced the point estimates for full vaccine efficacy for Asia and Africa by the same proportion as the relative difference in efficacy between the full and partial series in D: Americas region. We used 95% CIs from the respective studies to define the beta distribution around the vaccine efficacy point estimates.

The number of rotavirus deaths prevented was obtained from λ_rvε_rvρ_v, where λ_rv is the number of rotavirus deaths by week of age, ε_rv is the vaccine efficacy, and ρ_v is the proportion vaccinated by week of age.

Assessment of Risk—Base Scenario

Risk of intussusception has been documented after postlicensure use of Rotarix and RotaTeq in four different studies [8],[9],[31],[32]. Each of these studies identified an approximate 4- to 6-fold increase in risk relative to background during the first week after dose 1 (Table 3), a magnitude of risk that would not have been detected in the clinical trials. No effect modification of risk with age at vaccination was reported in these studies, but the first dose of vaccine was largely administered before 15 wk. In two additional countries, no risk of intussusception was identified after the first vaccine dose [9],[18]. Risk of intussusception was not identified after the first dose in Brazil (RR = 1.1; 95% CI = 0.3–3.3) or the United States (RR = 1.2; 95% CI = 0.03–6.8). However in view of the wide CIs, particularly in the United States, a risk of small magnitude similar to that detected in the other four studies cannot be excluded [9],[18]. In Brazil, a statistically significant 2-fold risk was also identified in the first week after dose 2.

We obtained dose-specific pooled estimates of RR from each of the regions where some increase in RR of intussusception was identified (Table 3). To err on the side of risk, we excluded the US safety data from the pooled analysis because no risk was identified. For pooled estimates of vaccine-associated intussusception risk, we used the weighted average of the logarithm of the RR, ∑log(RR_i)ω_i/∑ω_i, where weight (ω_i) for each study [8],[9],31,32 is the inverse of the variance computed from the reported 95% CIs [33]. The variance of the weighted average log RR is the inverse of the sum of the each weight (1/∑ω_i) and was used to compute the 95% CIs for the pooled risk estimate. For the uncertainty analyses, we used the 95% CIs to define the gamma distribution around the RR estimates.

The average annual incidence of natural intussusception by week of age ((λ_is) was estimated from published studies. Because natural intussusception is a very rare disease, we restricted our review to studies reporting either national incidence of intussusception or incidence of intussusception from a minimum of five hospitals with known catchment population, stratified by age [34]–[51]. While intussusception incidence in this review ranged from 18–88 per 100,000 infants, the age distribution of intussusception was similar between the different studies. Thus, to obtain intussusception incidence by week of age (λ_is), we applied the global intussusception incidence among infants and fit a gamma curve to intussusception surveillance data from the United States [45], the only country where intussusception incidence was available by week of age. For uncertainty analysis, parameters of the gamma curve for λ_is were sampled from a normal distribution, assuming standard deviation is equal to 5% of the mid-parameter values.

Death caused by intussusception is uncommon in industrialized countries, occurring in <1% of the cases [52]. In a recently conducted national study from 16 hospitals in Mexico and 43 hospitals in Brazil (WHO group B & C), case fatality for intussusception was 1% and 5%, respectively [9]. One large study from nine countries across Africa indicated an average case fatality of about 12% [53]. No reliable estimates of case fatality were available for countries in D-Americas and D-Asia. Thus, we conservatively estimated the case fatality (δ_is) to be 5% for B & C countries, 10% for D-Americas, 25% for D-Asia, and 25% for D & E-Africa. We sampled from a beta distribution, assuming standard deviation is equal to 5% of the mid parameter values to specify the upper and lower limits of δ_is in uncertainty analyses.

The number of intussusception deaths associated with vaccination, during the first week after dose 1 and 2, was obtained from Bρ_v[(λ_isRR_i) − λ_is]δ_is, where B is the number of births, ρ_v is the proportion vaccinated by week of age, λ_is is the intussusception incidence by week of age, RR_i is the RR during the week after each dose, and δ_is the proportion of intussusception events that lead to death.

Sensitivity Analysis

We conducted a one-way sensitivity analysis to determine the impact on the benefit-risk ratios when assuming four conservative scenarios that would favor risk and one that would favor vaccine: (1) We assumed a relative increase of 20% in incidence and case fatality of intussusception. (2) We explored the impact of effect modification of risk by age at vaccination, by doubling estimates of RR of intussusception when dose 1 of rotavirus vaccine was administered to infants older than 14 wk of age. (3) We assumed a scenario of low vaccine efficacy by inputting the lower confidence limit for each of the efficacy estimates. (4) We explored the effect of a “pessimistic” situation combining all of the preceding three scenarios. (5) We also assessed the effect of an “optimistic” scenario of high vaccine efficacy related to factors such as possible indirect benefits or higher efficacy among children vaccinated at older ages with lesser interference of vaccine take from circulating transplacental antibodies.

Uncertainty Analysis

The above analyses yielded estimates of rotavirus deaths averted and intussusception deaths caused under age-restricted and -unrestricted vaccination strategies. We conducted a probabilistic uncertainty analysis to assess the potential impact of simultaneous variation of each of the model inputs (λ_rv, ε_rv, ρ_v, λ_is, RR) on the precision of the benefit-risk estimates. We shifted the lognormal timeliness curves and gamma rotavirus and intussusception age curves by simultaneously sampling new shape, shift, and scale parameters for each run, with each parameter being sampled from a normal distribution with standard deviation equal to 5% of the original parameter value. On the basis of the error estimates and error distributions described for each of the model inputs, we conducted 2,000 simulations to obtain the median estimates of deaths averted and caused as well as the uncertainty ranges, defined as the 5th–95th percentile, to provide an indication of the uncertainty in the estimates. All analyses were done with Microsoft EXCEL (Microsoft Corp, 2007).

Results

Approximately 453,000 rotavirus-associated deaths are estimated among children younger than 5 y annually without a rotavirus vaccination program (Figure 1). We project that a rotavirus vaccination program under the current age-restricted schedule would prevent almost 33% or 155,800 of these deaths (5th–95th centiles, 83,300–217,700) if delivered at the same ages at which the DTP vaccine is currently being delivered in these countries (Table 4). Without the age restrictions, a program would prevent 45% or 203,000 deaths of all rotavirus deaths (102,000–281,500), which would represent 47,200 more deaths prevented (18,700–63,700) than with an age-restricted schedule. These additional deaths prevented under an unrestricted vaccination schedule reflect an additional 18%, 21%, 25%, and 22% of the children receiving DTP1 in the WHO B & C, D-Americas, D-Asia, and D-Africa countries, respectively, compared to the age-restricted schedule in these countries (Figure 2).

From the perspective of risk, a rotavirus vaccination program limiting vaccination to children <15 wk of age would cause about 253 intussusception deaths (76–689) (Table 4). In contrast, a program without age restrictions would cause nearly 547 intussusception deaths (237–1,160). Thus, a vaccination policy without any age restrictions for use of rotavirus vaccines in low- and middle-income WHO countries would avert an additional 47,200 rotavirus-associated deaths and cause an additional 294 intussusception-associated deaths, compared to the current age-restricted strategy (Table 5). The median incremental benefit-risk ratio in all mortality strata was nearly 154 deaths averted for every death caused, ranging from 55–318 deaths averted for every death caused across the different mortality strata (Figures 3 and 4).

Under the scenarios of effect modification of risk with age at vaccination and increased incidence and case fatality of intussusception, an unrestricted schedule would cause 603 (174–946) and 423 (232–678) excess deaths, respectively, while averting about 47,200 rotavirus deaths (18,700–63,700) (Table 5). A scenario where efficacy approximated the lower confidence limit in the clinical trials would avert an additional 20,400 rotavirus deaths (8,500–34,300) under an unrestricted schedule. With pessimistic assumptions of high intussusception incidence and case fatality, high risk, and low efficacy, a vaccination program without age restrictions would cause 868 intussusception deaths (506–1,362) while preventing 20,400 rotavirus deaths (8,500–34,300), for a benefit-risk ratio of 24. In contrast, the benefit-risk ratio would approximate 220 (116–407) under an optimistic scenario of high vaccine efficacy.

Discussion

Our analysis demonstrates that if first dose of rotavirus vaccine is restricted to children 14 wk of age or younger, rotavirus vaccines would prevent about 155,800 of the 453,000 rotavirus deaths occurring in children <5 y of age annually worldwide while resulting in 253 intussusception deaths. While most of the gap in preventable rotavirus deaths is due to the moderate efficacy of the vaccines in high mortality settings, the current age restrictions on rotavirus vaccination also contribute by potentially excluding nearly 21%–25% of the world's children, those with the highest risk of rotavirus mortality, from receiving these vaccines. Lifting the age restriction for the first dose of rotavirus vaccination would save an additional 47,200 lives yearly and would result in an additional 294 intussusception deaths, for an incremental benefit of saving 154 lives for each excess intussusception death caused.

In the past 5 y, with the introduction of rotavirus vaccines in nearly 30 countries worldwide, substantial experience has been gained with regard to the safety and effectiveness of these vaccines in the real-world setting, including against deaths [8],[9],[16]–[18],[25],[27],[54],[55]. Moreover, clinical trials for these vaccines have documented their efficacy in target populations of Asia and Africa, where majority of the rotavirus deaths occur. Given these encouraging data, the ability of the vaccines to reach children with the highest mortality will be a major determinant of their life-saving impact.

Our base estimates are conservative, erring on the side of overestimating vaccine risk for four reasons. First, over 45 publications have documented remarkable declines in severe diarrhea and rotavirus disease, including deaths, since their introduction in national immunization programs worldwide [55]. Many of these studies from different locations have demonstrated significant declines in unvaccinated members of the community, indicating indirect benefits of vaccination that we did not account for in our analysis [56]–[59]. Second, because of interference from circulating transplacental antibodies during the first several months of life, immune response to vaccine and thus efficacy is likely to be higher when children are vaccinated at older ages. For example, anti-rotavirus IgA geometric mean titers for Vietnamese infants vaccinated against rotavirus at 9 and 13 wk were lower (77 U/ml) compared to infants vaccinated at 9 and 17 wk of life (176 U/ml) [60]. Third, we assumed that some risk of intussusception exists following each of the first two doses of rotavirus vaccine in all countries worldwide; however, risk of intussusception has varied by setting, and robust studies in two large countries have not identified risk after dose 1 [9],[18]. Fourth, even in our base scenario, we assumed high rates of intussusception case fatality in all WHO regions, about 2-fold higher than those reported in the literature.

On the other hand, the benefit-risk ratios might be inflated due to several factors. First, our base scenario assumes that the risk of intussusception relative to background does not increase with age. After the withdrawal of RotaShield, a debate persisted with regard to whether the RR of intussusception might have been higher for infants vaccinated beyond 14 wk of age [11],[12]. While limited data from an evaluation in Mexico does not suggest effect modification of risk by age for current vaccines [9], we incorporated a scenario of increased risk with age at vaccination that indicated that vaccination would avert 75 rotavirus deaths for each excess intussusception death. Second, our model might have overestimated vaccine coverage among children at the highest risk of dying from rotavirus as these might be the hardest to reach, thus inflating the mortality benefits of vaccination relative to the risks in our model. However, data from Mexico and Brazil, where substantial reductions in diarrhea deaths have occurred in all regions of both countries after the introduction of vaccine [16],[17], provides some reassurance that vaccine is reaching those at the highest risk of dying.

While the numerical benefits of relaxing the age restriction on rotavirus vaccination exceed the risks, other factors are relevant for policy considerations. First, the age restrictions for rotavirus vaccines potentially offer an incentive to improve timeliness of vaccination, which would potentially have far reaching benefits beyond just prevention of rotavirus disease. However, reasons for delays in vaccination in developing countries are complex and it is not known if a policy of restricting the first dose of rotavirus vaccines alone would be a sufficient motivational factor for parents and countries to improve timeliness of vaccination. Indeed, some delays may be due to unavoidable factors, such as contraindications. Second, while the unrestricted vaccination scenario allows for vaccination at any age during the first 3 y of life, few children arrive for vaccination beyond 1 y of life. It is important to note that delays in vaccination particularly beyond 1 y of life will reduce benefits substantially because of increasing probability of acquiring natural immunity from wild-type rotavirus infection. Third, a death caused by an intervention may be perceived worse than a death caused by a failure to intervene [61]–[63]. However, some evidence suggests that individuals may regret disease resulting from withholding vaccine as much as side effects from vaccination [63]. Furthermore, after the RotaShield experience, ethicists argued equal culpability for deaths caused by withholding the vaccine as for deaths resulting from the vaccine [64]. Finally, our analysis did not address high income countries where mortality from both rotavirus disease and from intussusception is uncommon, and thus the benefit-risk considerations will differ. Furthermore, vaccination is more timely in these settings (e.g., in the United States, 93% of the DTP1 is given by 15 wk of age [65]), and thus decisions will likely have to be made at a country level based on evaluation of local data.

In summary, using emerging, real-world data on rotavirus and intussusception mortality and rotavirus vaccine efficacy, safety, and coverage, we estimate that removing the age restrictions on rotavirus vaccination would avert 47,200 additional rotavirus deaths in low- and middle-income countries. In April 2012, WHO's Strategic Advisory Group of Experts reviewed the evidence presented in this paper and recognized that the 15-wk and 32-wk age restrictions for rotavirus vaccines are preventing vaccination of many vulnerable children [66]. SAGE encourages timely vaccination, but no longer universally recommends the age restrictions, supporting their removal in seetings where mortality benefits outweigh the risk so that many thousands more deaths would be averted and immunization programs are able to immunize children who are currently excluded from the benefits of rotavirus vaccines. Age restriction policies will ultimately be decided at country level, but this analysis has shown a clear case for a change in policy that will be particularly instrumental for saving lives in settings where mortality from rotavirus is high and delays in timing of vaccination are common.