Vaccines for Adults

Since the prior 2014 report on vaccine safety, a number of new vaccines have become available for adults, including recombinant zoster vaccine (RZV), hepatitis B vaccine with a novel immunostimulatory adjuvant, and adjuvanted inactivated influenza vaccine (aIIV), of which many are indicated primarily or exclusively for older adults.

Vaccines for Children

Among children, there continues to be a reasonably robust body of evidence related to vaccine safety.

Vaccines for Pregnant Women and Their Fetuses/Infants

This report also identified increasing evidence assessing the safety of vaccines among pregnant women. For this update, we identified no studies in pregnant women that assessed the effects of hepatitis B vaccines or RIV, which is an area that could be targeted for further research. We identified one study of quadrivalent IIV, but it did not assess any of the key adverse events pre-specified for this report, nor any other serious or severe adverse events. While we did not identify any studies of quadrivalent vaccines in pregnant women, our 2014 report uncovered no safety concerns related to trivalent IIV or monovalent H1N1 influenza vaccine, neither of which are currently in use. Although this report reviewed only vaccines in current use, we note that a recent case-control study¹⁰⁹³ of data collected over three influenza seasons (2012–13, 2013–14, 2014–15) found no increased risk of spontaneous abortions among pregnant women who received IIV (almost all trivalent vaccine).

We identified a number of new studies that examined the safety of Tdap for pregnant women and their newborn infants, including four RCTs and 12 cohort studies. This enabled us to make many new evidence statements, of which several had moderate SoE (maternal cardiovascular events, maternal death, maternal diabetes, eclampsia/pre-eclampsia, preterm labor, maternal reproductive system events, stillbirth, cardiovascular events in infants, death in infants, encephalitis/encephalopathy in infants, seizures in infants) and others had low SoE (maternal encephalitis/encephalopathy, autism in infants, birth defects in infants, febrile seizures in infants). We note that women who received Tdap had significantly lower risk of preterm labor (which included preterm delivery if preterm labor was not reported) than did comparison groups (moderate SoE). The mechanism for this protective effect is not clear, and our systematic review was not designed to assess the effectiveness of vaccines in reducing harms. Among five large cohort studies, two identified a slightly increased risk for chorioamnionitis,^{91, 154} while others reported no difference in risk.^{67, 121, 170} This potential slightly increased risk could be investigated further.

In terms of longer-term effects, we found a significantly decreased risk of autism (low SoE) among infants whose mothers received Tdap; this was based on one study,⁶⁷ and again the mechanism for this effect is not clear. Other studied outcomes for which no difference was found in infants exposed to Tdap in utero include neonatal microcephaly.

Strength of the Review

Our review of the literature was extensive and designed to maximally capture evidence on the presence or absence of adverse events associated with vaccines in studies with a comparator evaluating vaccines in human participants. To identify data, we relied not only on research databases, but also on publicly accessible grey literature. We searched Clinicaltrials.gov for information on unpublished trials with results posted in the trial record, as well as all Advisory Committee on Immunization Practices (ACIP) statements and vaccine package inserts. While we did not use FDA regulatory documents as a primary source, we drew upon such documents whenever any signal was identified for potentially increased risk (e.g., HEPLISAV-B and myocardial infarction).

After searching the literature, we employed a transparent, established protocol to minimize the risk of missing relevant studies. The review team followed inclusive decision rules and ordered full text copies of any publications reporting on interventions or observational studies of the vaccines of interest—even if ostensibly reporting only on clinical effectiveness—in order to identify studies that reported on harms only in the full publication.

We reviewed adverse events reported in comparative vaccine studies regardless of whether or not they might be attributed definitively to the intervention. We systematically identified evidence of the absence of specific adverse events (i.e., events that were assessed in research studies but that did not occur if reported as such).

We allowed for complex comparisons where a vaccine was evaluated in the presence of a second vaccine (e.g., HPV9 and Tdap) and compared to the second vaccine alone (e.g., Tdap). For some vaccines, many of the studies included an active vaccine comparator to ensure that that the intervention is compared to the current standard of care. Thus, we also allowed for comparisons between a vaccine and the vaccine it was replacing (e.g., quadrivalent IIV and trivalent IIV) or to the newest similar vaccine (e.g., MenACWY-TT to MenACWY-CRM or MenACWY-D). A comparison between a vaccine and an active comparator may underestimate rate of adverse effects relative to a comparison between that same vaccine and placebo. Given this, we clearly note when an active vaccine comparator was used throughout the text of this report. To address the difference in studies, we report first the risk estimates for the vaccine of interest compared to either placebo or a base treatment that the intervention group also receives (e.g., routine vaccines), and then we report the risk estimate when combined with studies that included an active vaccine comparator. We also allowed for combining of studies where populations were different (e.g., one study of RZV given to healthy adults and another of RZV given to adults with stem cell transplants); including studies of non-healthy populations enabled us to assess the risk of adverse events in populations that might be at higher risk for such events. While our inclusive approach added to the complexity of the review, this approach has the benefit of capturing the fullest evidence base possible for the range of vaccines and potential harms.

Limitations of the Review

While our literature search procedures were extensive, some unpublished trial results may not have been identified. As noted above, we were able to mitigate this possibility by searching trial databases directly and reference mining relevant publications, including ACIP recommendations. We also may have missed studies due to the challenging nature of assessing harms as contrasted with assessing effectiveness, since many publications focus on the clinical effectiveness of an intervention with either no, sparse, incomplete, or non-systematic assessment and/or reporting of safety data. While search filters exist for effectiveness studies, filters to address harms are not as successful in identifying relevant studies. Thus, as noted above, we intentionally screened the full text of all vaccine intervention studies to minimize missing safety data. We also note that the current review builds on the prior 2014 report, which itself built upon the 2011 Institute of Medicine (IOM) report. The prior 2014 report did not search for or include studies on vaccines that were covered in the IOM report and published prior to 2011. In the current report, only for vaccines for which there were new indications or for new vaccines did we perform targeted searches for research published prior to 2014. Wherever possible, we used data that could be combined in meta-analyses in this update from the prior report to estimate the relative risk based on all available research studies. In addition, we narratively synthesized research findings across both the prior 2014 report and the update to assess the SoE.

This report reviews currently recommended vaccines for routine use, and does not include new vaccines in development or under emergency use authorization, such as vaccines for the 2019 coronavirus disease (COVID-19) pandemic. We also excluded studies of vaccines not currently in use in the United States. For example, while HPV2 and HPV4 are no longer in use in the United States, studies of the safety of these vaccines (still in use in other countries) would likely inform discussions of the safety of HPV9. We acknowledge that studies of other widely used vaccines could be useful but were required to limit the scope to a focus on the United States. We also excluded non-English language studies. Although we were considering only vaccines approved for use in the United States, it is possible relevant epidemiological studies have been published in non-English journals.

While we did identify sub-group analyses within studies that met our inclusion criteria, there may be additional reports of studies that assessed specific risk factors but that would not have been included if there were no appropriate comparator (e.g., a study comparing HPV9 in men and women, where all participants receive the vaccine). Despite the large number of studies included, indirect analyses across studies to assess the effect of participant or administration variables were possible only for selected characteristics and selected outcomes. This was because studies varied widely in the strategy used to assess the safety of the vaccines and used different datasets (e.g., some used mining datasets that included data for many different vaccines). In addition, some studies reported adverse events of interest at a system or group level (e.g., rate of all cardiovascular events), but others did not. For studies that did not report system-level rates of system-level adverse events, we had to instead choose the most common adverse event within the relevant system (e.g., unstable angina) to use available data to the extent possible. This resulted in some estimates that are based on combined systems and event-level rates, but in cases where such analyses indicated more events in the vaccinated group, we also performed sensitivity analyses to assess the robustness of the findings and reviewed the nature of the events.

As was the case for the prior Agency for Healthcare Research and Quality 2014 report and the IOM report, the current report focused on the association of specific vaccines with particular adverse events. One of our inclusion criteria was that studies must include both an intervention vaccine and a comparator (either placebo, no vaccine, or the comparator vaccine closest to what the new vaccine is replacing). Because of this approach, drawing conclusions about the safety of the immunization schedule taken as a whole, multiple vaccines together, and/or certain adjuvants and preservatives is not generally possible from this report’s findings.

A further limitation results from the fact that not all immunizations and thus adverse events may have been captured, particularly if they occurred outside of the medical home and/or if the state did not have a vaccine registry. Statewide registries for adults are much less common than they are for children. Adults are also more likely to get their vaccines at alternate locations such as their workplace or at pharmacies. Given the nature of the studies that were included, all of which had to have some comparator, loss of information from this limitation is less likely for this report.

Finally, we note that many of the harms we were assessing as our key adverse events (e.g., acute disseminated encephalomyelitis, Guillain-Barré syndrome, transverse myelitis, anaphylaxis) are quite rare and the number of studies that reported on the presence or absence of events for a vaccine was sometimes small. As a result, despite our extensive searches for data that could be combined across studies, our confidence intervals are often wide, which make the risk estimates imprecise. Risk estimates had wide confidence intervals in individual studies given that often only one or two events occurred in the entire sample, but also in estimates combined across studies. Although statistical pooling is a data aggregation method, we used random effects meta-analyses, which do not average the results of the identified studies, but set out to estimate the true risk by treating the studies as samples from an underlying distribution. When only few studies are available, the effect estimates differ, and/or the event rate is rare, the confidence interval of the effect estimate will be very wide as the true effect is estimated from a very small sample of potential studies. With this review being no exception, rare events present considerable methodological challenges.⁴⁹ However, we investigated all instances of more observed events in the vaccinated group further regardless of the precision of the risk estimate and transparently present all rates. In addition, we considered all available data regardless of whether they could be combined in a pooled estimate, including post-marketing surveillance and other observational studies such as self-controlled case series and case-control studies, in the grading the SoE.

Strength of the Evidence Base

We identified a large number of studies reporting on the presence and absence of adverse events, and in particular a number of new studies on the use of Tdap in pregnant women. The studies reported on a group of patients vaccinated with the vaccine of interest and a control group, an important feature as many of the adverse events of interest are not unique to vaccines and we had to establish whether the event was more likely in the vaccine group. We included experimental and observational studies and many studies included in the current report are RCTs, which by design should minimize differences between intervention and control groups.

We extracted data both from published journal articles and clinical trial registries. The importance of trial registries has increased dramatically since reporting of results has become mandatory. Clinicaltrials.gov is set up to capture results that can be used in systematic reviews and meta-analyses, including data on severe adverse events, serious adverse events, and mortality. In general, the harms data in Clinicaltrials.gov have been found to be more complete than in the corresponding publications,^{1094, 1095} although we note that the database tends to better capture the presence of adverse events than the absence of such events (that is, when no adverse event occurs in either the intervention or comparator group). As a result, our findings may well underestimate the safety of vaccines, as we did not count the absence of adverse events in the study record as evidence of absence of effects; we only considered studies that explicitly reported the presence or the absence of the adverse events of interest.

Other innovative methodologic approaches have also improved the analysis of adverse events, particularly for post-marketing studies. For example, in the United States, the CDC’s Vaccine Safety Datalink uses data obtained through such systems at eight large healthcare organizations, enabling high-quality studies using methodologies such as self-controlled risk intervals analyses. Other nations with single payer healthcare can often leverage their extensive electronic registries, which allow for epidemiological studies of entire populations.

Limitations of the Evidence Base

An important limitation common to systematic reviews in general is the quality of the original studies included. We critically appraised included studies in detail. Studies that reported timing and severity and defined adverse events using standard, precise definitions were rated higher than those that did not.

Limitations of studies vary according to their design. Controlled trials often have insufficient sample sizes to identify very rare adverse events as discussed above, and may not follow participants long enough to identify long-term sequelae. Even in studies with generous follow-up times, the timing of events is not always optimally reported. Except where explicitly noted in the text, controlled trials of vaccines tend to be conducted in healthy patients. Thus, persons who may be more susceptible to adverse events may be excluded from trials yet eligible to receive a vaccine after it is licensed.

Given the above limitations of controlled trials, comprehensive reviews of vaccine safety must include post-licensure studies, but these are not without their own limitations. People who avoid vaccinations whether deliberately or not, may differ from those who receive vaccinations in terms of race, sex, age, socioeconomic status, and preexisting medical conditions, and these differences may in turn be associated with health outcomes. Observational studies typically attempt to control for such potential confounders by using matched cohorts or multivariate regression analysis. However, some factors such as environmental exposures may be unmeasured or challenging to control for adequately. The self-controlled case series was developed specifically to assess the safety of vaccines; this method eliminates confounding by all time-independent variables by using cases as their own controls and predefined “time windows” before and/or after vaccination. While self-controlled case series control for patient characteristics that do not change over time (i.e., gender, race, ethnicity, genetics), they cannot capture factors that are time related. This is especially important when studying vaccinations of infants and toddlers. Another limitation of such studies is that the data provided typically do not allow for pooling across studies (e.g., combining a self-controlled case series study of rotavirus vaccine that provides adjusted relative risk with an RCT of rotavirus vaccine that provides counts of events).

Studies using passive surveillance such as the Vaccine Adverse Event Reporting System (VAERS) are crucial in identifying signals regarding adverse events post-licensure. By definition, they do not consider the rates of such events in non-vaccinated populations, and thus are not designed to assess a statistical association between a vaccine and an adverse event; thus, these studies were excluded from this project. VAERS data might contain important adverse event signals that are not identified in this report and that warrant future research.

Post-licensure epidemiological studies are conducted to investigate possible associations between vaccines and adverse events reported in passive surveillance or multiple case reports. Such studies often do not limit their investigation to a particular brand or formulation. It is difficult to assess the utility of studies that do not report specific details about vaccines, or lump vaccines against a specific disease together. For example, a study might investigate the effect of “seasonal influenza vaccines” in general, even when the vaccines used may be quite different (e.g., live attenuated influenza vaccine and IIV). We excluded such studies that reported on a category of mixed vaccine types, if there were distinct vaccines in use at the time of the study (e.g., “meningococcal vaccines” when both MenACWY and MenB were in use during the same period).