The reviewed studies (Tables 3–6) varied widely in their design approaches and included examples of the three overarching study types described in Table 2: pregnancy registry studies, electronic data sources (i.e., secondary data analysis), and passive surveillance studies. In this review, secondary data analysis studies were used for both vaccines recommended for use in pregnancy and not broadly recommended for use in pregnancy. Passive surveillance studies were particularly common for vaccines that were not broadly recommended to be administered during pregnancy. Among the studies of vaccines not broadly recommended for use in pregnancy, the passive surveillance studies tended to have larger sample sizes (range: 306–2,942) relative to the other designs (range: 2–207). Additionally, as anticipated, actual and target sample sizes (reported for newer studies without publicly available results at the time of the search) tended to be higher among studies for vaccines recommended to be administered during pregnancy, as compared to those not broadly recommended to be administered during pregnancy.

The FDA guidance refers to registries being based on the voluntary participation of individuals who have been exposed to a specific drug or biological product (including vaccines) during pregnancy and unexposed individuals who enroll into the comparator cohort. For products that are anticipated to be used rarely during pregnancy (e.g., treatment of advanced cancer), the FDA guidance mentions that the sponsor “can consider a pregnancy surveillance program (i.e., a structured approach for data collection with targeted questionnaires to obtain follow-up information on all exposed pregnancies of which sponsors become aware).” These pregnancy surveillance programs usually need worldwide safety data collection “to identify a sufficient number of exposed pregnancies for clinical safety assessment.” This guidance document also stresses that while pregnancy registries are important, their design usually makes them insufficient for use in isolation, and therefore, the FDA recommends a complementary approach of using a registry and another method (e.g., an electronic database study).

The EMA guidance documents note that existing databases and/or existing disease-specific registries are recommended whenever possible, however do not recommend a specific study design (including for non-vaccine studies) nor offer vaccine-specific recommendations. Features of different study designs are described and references to the GVP Module VIII (European Medicines Agency, 2017) and CHMP Guideline on the Exposure to Medicinal Products During Pregnancy: Need for Post-authorisation Data (European Medicines Agency, 2005) are made.

The current FDA guidance section “Pregnancy Registries/Overview” does not include vaccine-specific recommendations which address the challenges unique to vaccine studies. Additional guidance on the criteria for determining the most appropriate design for vaccine studies is desirable. For vaccines recommended during pregnancy, in which the number of exposures are generally expected to be large, establishing a registry with active enrolment seems appropriate. For vaccines not recommended during pregnancy, mainly inadvertent exposures (particularly before pregnancy or during the first pregnancy trimester) are expected. For these vaccines, the choice of the most appropriate design is less straightforward. In some cases, the number of exposed pregnant individuals may not be negligible if the vaccine is widely administered in a population of childbearing potential. For other vaccines, the number of exposed pregnant individuals may be expected to be low, and thus setting up a registry with active recruitment may face feasibility challenges.

Additionally, the current guidance lacks clarity on specific factors that determine when pregnancy registries conducted via passive surveillance (e.g., passive surveillance studies through spontaneous case reporting in PV database) are appropriate as compared to active pregnancy registries with active patient recruitment and data collection. There is a need for guidance to define criteria for selecting passive surveillance studies versus active pregnancy registries based on rarity of exposure, availability of clinical trial data, vaccine type (e.g., yearly influenza strain updates), vaccine coverage, existing data robustness, pregnancy recommendations, contraindications, and the feasibility of active surveillance implementation.

Of the reviewed example studies (Tables 3–6), those designed as passive surveillance pregnancy studies based on spontaneous reporting all lacked comparison groups (for vaccines with or without recommendation for use in pregnancy). In contrast, most of the reviewed cohort studies with active data collection, whether prospective or retrospective, included a comparison group. The comparators consisted of individuals who either received other vaccines during pregnancy or were exposed to the vaccine under study before becoming pregnant. In the studies that did not include internal comparison groups, the estimates of the exposed pregnant people were compared to background population estimates.

Studies using secondary data analysis for vaccines not widely recommended during pregnancy seemed to face more challenges in identifying internal comparator groups. For instance, one study on the safety of Boostrix-Tdap (Tseng et al., 2022) selected a historical unexposed control cohort rather than a concurrent one, as the vaccine uptake among pregnant people exceeded 80%, which impaired the feasibility of 1:1 matching and could introduce confounding by indication.

Both the FDA and EMA guidelines encourage the use of multiple comparator groups in pregnancy studies. Specifically, the FDA guidance recommends the use of a concurrent internal comparison group of pregnant people who are unexposed to the treatment under study. While the FDA guidance acknowledges that in certain situations, background rates from another pregnancy registry or external database may be the only available comparator, it cautions that this approach has limitations that should be considered when interpreting results.

Both the FDA and EMA guidelines provide general recommendations for comparators in pregnancy safety studies and lack specific guidance for vaccine pregnancy studies, particularly for studies utilizing secondary data analysis. While these guidelines highlight the advantages of disease-specific pregnancy registries over product-specific ones, as the former include built-in comparator groups and can mitigate the risk of confounding by indication, these registries have limited utility in preventive vaccine studies. The FDA guidance emphasizes the importance of selecting appropriate comparators—warning that “comparing dissimilar populations could bias the study results, indicate a risk when none exists, or mask an increased risk that exists”—however, detailed recommendations relevant for vaccine studies are notably absent.

The comparator group could consist of unexposed pregnant people, although this approach must carefully address potential unmeasured confounding as the decision to receive or decline vaccination may correlate with health behaviors, socioeconomic factors, or underlying health conditions that affect pregnancy outcomes. Background rates in the target population can serve as comparators. These rates can be either concurrent with vaccine exposures or from periods before vaccination campaigns began, though each approach has limitations. Concurrent rates may be “contaminated” by vaccine effects if a large proportion of the population is vaccinated, while historical rates might be influenced by temporal changes in diagnostic practices and reporting systems. Furthermore, the availability and quality of pre-vaccination historical data often presents a significant challenge, potentially limiting the reliability of historical comparisons. Alternatively, comparisons may include pregnant people who received another vaccine during pregnancy, though this approach is increasingly challenging due to the growing number of recommended multi-component vaccines during pregnancy. More comprehensive recommendations on the selection of preferred and appropriate comparators for vaccine safety studies would be beneficial to stakeholders.

We observed varying approaches to defining exposure periods (Tables 3–6). Ten studies, regardless of data collection methodology or pregnancy use indication, considered vaccine exposure during a brief period (typically 1 month) before pregnancy start and throughout the entire pregnancy. For studies that differed from this approach, vaccine exposure was defined in several ways: administration of a vaccine dose at any time during pregnancy (Adacel-TdaP, Spikevax-Covid-19 studies); administration of a vaccine dose within a defined period during pregnancy, either related to approved gestational ages (Abrysvo-RSV) or gestational ages as recommended by the Advisory Committee on Immunization Practices (ACIP) in the US (Cervarix-HPV/secondary data; Table 4). In general, the studies defined pregnancy start at the last menstrual period (LMP) date or conception date.

Only three studies explicitly described their approach to handling multiple vaccine doses (Cervarix-HPV study/primary data, Trumenba-MenB, Varivax-Varicella). The Cervarix-HPV study (primary data; Table 3) classified multiple exposures during pregnancy by the earliest trimester of exposure, and when exposure occurred both before and after conception, classification was based on the post-conception dose. In two other studies (Trumenba-MenB, Varivax-Varicella), multiple doses were possible during the exposure period, with pregnant individuals considered exposed at the time of the first vaccine dose. Of note, a multi-dose immunization schedule is not recommended for all vaccines, and for single dose regimens, multiple vaccine doses may indicate administration errors.

Eight studies considered strata of timing of exposure in the analysis, such as a month before LMP and/or trimesters of gestation (Varivax-Varicella, Cervarix-HPV/primary data Table 3, Trumemba-MenB, Menveo- MenACWY, Fluarix IIV4/FluLaval IIV4-Influenza, Comirnaty-Covid-19, Spikevax-Covid-19, Abrysvo-RSV).

The EMA Guidance emphasizes the importance of accurate exposure measurement during embryo-fetal development (precise timing for both pregnancy start and medicine use) and gestational age at time of exposure. It describes susceptible periods for adverse pregnancy outcomes based on the gestational age at time of exposure. These periods include (1) the first 2 weeks after conception, when interference may result in early pregnancy loss; (2) gestational weeks 4–16, which are critical to study the association of exposure and major birth defects, though each congenital abnormality has its specific critical period (e.g., neural tube defect between the gestational days 29 and 42 or between days 15 and 28 post-conception); (3) week 16 until delivery, when interference may cause minor anomalies, growth impacts, or functional defects; (4) late pregnancy ³ and delivery, which carry risks of physiological impacts on the neonate. Finally, throughout the entire pregnancy, exposure to environmental agents, including medicines, as well as exposure to infections or worsening of existing chronic diseases, can result in pregnancy loss or stillbirth.

The FDA Guidance advises collecting detailed information on medication use (including start/stop dates, dose, frequency, duration, and indication), incorporating exposure information from the period just before pregnancy (especially for products with long half-lives), and conducting stratified analyses by gestational timing of exposure, with a separate analysis of first-trimester exposures for MCMs. For complementary (e.g., secondary database) studies, the FDA Guidance emphasizes accurate gestational age estimation, suggesting various methods for identification (e.g., birth certificates, ICD codes, electronic health records, ultrasound), and recommends obtaining informed consent for medical record review in population-based case-control studies to confirm diagnoses and collect additional exposure data.

Both EMA and FDA guidance recommends recording exposure to any other medical product during pregnancy as well as the medical history and other potential confounders.

A notable gap exists in both FDA and EMA guideline documents, which lack recommendations for designing pregnancy studies specific to vaccines. Current recommendations primarily focus on drug use parameters such as start/stop dates, frequency, and duration, which are not fully applicable to characterize vaccine exposure. Critical aspects of vaccine exposure documentation include brand name, batch number, and dose number when it is part of an immunization series against the same disease (DeSilva et al., 2016). This set of data is essential for comprehensive exposure assessment but remains unaddressed in current regulatory guidance. Additional guidance on the critical elements to characterize vaccine exposure is required.

It is widely recognized that the first trimester is the most critical period for exposure during pregnancy with regards to subsequent effects on fetal development (teratogenicity). Both FDA and EMA guidelines emphasize the importance of considering specific timing of exposure when evaluating outcomes of MCMs. The EMA guidance provides insights on the relationship between exposure timing and gestational age regarding the risk of MCMs and other adverse outcomes, detailing susceptible periods throughout pregnancy. However, DeSilva et al. (2016) offer more precise recommendations that could enhance regulatory guidance; these authors suggest including maternal vaccination from 30 days prior to conception up to 20 weeks gestational age. The rationale for this time frame is to account for wider windows of teratogen exposure and potential errors in assigning conception dates and gestational age.

It is worth noting that different exposure windows may be relevant for certain outcomes and should be considered accordingly. Additional regulatory guidance on defining appropriate exposure periods for vaccines would be valuable for researchers conducting vaccine safety studies during pregnancy.

Across all 15 studies identified, we describe in Table 7 the alignment of the prespecified outcomes in the study protocol with the EMA (European Medicines Agency, 2019; The European Network of Centres for Pharmacoepidemiology and Pharmacovigilance, 2010; European Medicines Agency, 2005), FDA (United States Food and Drug Administration, 2019), and Brighton Collaboration guidelines (Munoz et al., 2015; Jones C. E. et al., 2016) for both primary and secondary data designs. Many of the primary data collection studies were passive surveillance studies, making it challenging to identify target outcomes due to the lack of an a priori list. In these cases, only outcomes specifically described as outcomes of interest were summarized.

Almost all studies included outcomes for fetal death/stillbirth (N = 14), major congenital malformations (MCMs; N = 14), and spontaneous abortion/miscarriage (N = 13). Fewer, but still a majority, included an outcome for preterm delivery (N = 12), elective termination/induced abortion for any reason (N = 10), low birth weight (N = 8), and small for gestational age (N = 7). Few studies evaluated maternal death (N = 2), ectopic pregnancies (N = 4), and molar pregnancies (N = 1). Developmental milestones were captured in two studies, but only up to a maximum of 1 year. Other endpoints not specifically recommended by health authorities (e.g., preeclampsia, premature rupture of membranes, gestational diabetes, postpartum hemorrhage, gestational hypertension, cesarean delivery, placenta previa, and large for gestational age) were collected inconsistently across studies. Additional adverse events (AEs) and serious adverse events (SAEs) may have been collected and reported in accordance with routine PV practices, in addition to the prespecified outcomes, as established within the primary data collection studies.

Overall, there was wide variability in identified outcomes. All nine primary data collection studies collected information on congenital anomalies; however, only three reported that these anomalies were reviewed by an independent expert, with two utilizing a dysmorphologist (Cervarix-HPV; Multiple manufacturers-COVID-19 vaccines). Only one study (Comirnaty-COVID-19) explicitly adhered to the FDA draft guidance, where an expert dysmorphologist, serving as a co-investigator for the OTIS Pregnancy Registry studies, reviewed all medical records and reports of MCMs. This review process was blinded to exposure status and consistently applied across both exposed and comparator cohorts, aligning with FDA recommendations.

Among the six secondary data analysis studies, five studies assessed congenital malformations and, of those, three documented varying validation approaches. Two studies implemented expert review: the Menveo-MenACWY study utilized a birth defect evaluator with genetics expertise serving on an independent Scientific Advisory Committee and applying the Centers for Disease Control and Prevention (CDC’s) Metropolitan Atlanta Congenital Defects Program (MACDP) definitions, while the Cervarix-HPV study employed adjudication by two independent external teratology experts who remained blinded to exposure status. The Boostrix-Tdap study relied on chart confirmation of AEs. These approaches reflect varying adherence to FDA guidance, which recommends outcome validation through clinical expert review, linkage to birth defect registries, or previously validated high-performing algorithms in the same (or similar) database planned for the current analysis.

The Brighton Collaboration (Munoz et al., 2015) has defined a consensus list of terms and concept definitions of key events for monitoring immunization in pregnancy through a rigorous process involving systematic literature review, worldwide stakeholder survey, and taskforce meetings. Priority outcomes were defined as those that are key for safety assessment of vaccine studies in pregnancy (summarized in Table 7). No priority outcomes were provided for childhood outcomes, as maternal vaccination effects on older children were deemed unlikely. Of note, guidance for assessing neurodevelopmental delay was later added to the Brighton Global Alignment of Immunization Safety Assessment in pregnancy [GAIA] Case Definitions (Villagomez et al., 2019).

The FDA guidance (summarized in Table 1) does not provide a specific list of outcomes but includes recommended data collection elements in Appendix A under different subsections (General, Maternal information, Neonatal information) which can be used to deduce relevant outcomes. While the FDA guidance is mostly tailored to a primary data collection study design, it is noted that the list of outcomes may also be applicable to complementary studies (e.g., studies with a secondary collection design). Furthermore, the FDA provides guidance on outcome ascertainment and validation in study designs relying on secondary data (including electronic administrative claims data and/or electronic health record databases), recommending validation using a gold standard method such as via medical record review from clinical experts or via linkage to birth defect registries and/or birth certificate data, or through a high-performing algorithm validated for the specific outcome in the same or similar database.

EMA guidance documents (summarized in Table 1), indicate that all “relevant outcomes throughout the human developmental cycle […] with child follow up for a long enough period to capture relevant information […]” should be evaluated and offer a summary of commonly studied outcomes. The EMA guidance also suggests monitoring longer-term child outcomes such as neurodevelopmental disorders but provides no guidance for outcome validation.

Other initiatives, such as the effort coordinated by St George’s, University of London, have leveraged the comprehensive work of harmonized case definitions for key obstetric and neonatal outcomes, done by the GAIA project and the Brighton Collaboration network to provide comprehensive guidance in harmonizing the data collected in case report forms (CRFs) for clinical trials (Jones C. E. et al., 2016). However, there are still gaps. The CRF data collection guidance is best suited for primary data collection in clinical trials. While it can be adapted for use in post-approval secondary data studies, the limitations of secondary use of data preclude full implementation. Furthermore, while several of the harmonized priority data collection elements recommended in the guidance can be mapped to the EMA and FDA guidance outcomes, there are a few EMA/FDA recommended outcomes that cannot be directly mapped to the CRF data elements (Jones C. et al., 2016). For example, the data collection guidance does not specifically indicate that elective termination/induced abortion for any reason, and termination of pregnancy due to fetal anomaly (both of which are recommended for collection in the EMA and FDA guidance documents) should be collected. Although it recommends that information on “any new medical conditions, surgical procedures or hospitalizations” during pregnancy be collected. These open-ended questions lead to challenges in harmonization.

While both regulatory agencies emphasize that studies should capture both live and non-live birth pregnancy outcomes, and recommended or commonly studied outcomes are listed, neither regulatory guidance document outline a specific subset of outcomes that all studies must evaluate. Additional guidance that provides (1) a comprehensive list of a subset of outcomes that all studies must evaluate across indications and, within this list of required outcomes, (2) a prioritization of which outcomes are most important and thus should drive calculations of sample size and study design, would be greatly beneficial to ensure consistency across similar studies, allow for meaningful comparisons, and clearly communicate expectations. For studies requiring additional outcomes beyond the prioritized list, a decision framework for selecting the appropriate outcomes is needed.

Additional guidance is needed regarding when and how long follow-up is needed, and the appropriate strategies for assessing childhood outcomes (e.g., growth and development). As this is likely to be dependent on factors such as the vaccine, disease, and setting, a decision framework would be useful in determining relevance for a given study.

Validation is crucial for ensuring data reliability but should be undertaken judiciously, guided by clear criteria delineating when it is necessary. In cases where existing validated algorithms have been successfully employed in analogous or identical data sources, it is prudent to utilize these established methodologies. This approach obviates the need for redundant efforts in developing new algorithms, advocating instead for the consistent application of proven processes. Such consistency would enhance the strength and reliability of data and conclusions. Nonetheless, it is acknowledged that these algorithms may require updates and adaptations to specific studies. While the FDA has outlined expectations regarding validation of study outcomes, achieving harmonization in validation expectations across regulatory authorities remains an essential goal.

Electronic databases are increasingly utilized for pregnancy safety studies. In these data sources, researchers often rely on algorithms to identify cases. The integration of innovative machine learning techniques for improving algorithms may offer opportunities to refine case definitions. Continued methodological research in this area may further strengthen the validity and reliability of case definitions. Guidance on the use of machine learning techniques to refine case definitions would be of value.

An important practical hurdle in pregnancy research is achieving adequate sample sizes as vaccine exposure may be uncommon, and some maternal, fetal or neonatal outcomes of interest are rare. Sample sizes in identified studies (Tables 3–6) varied widely: from 2 to 2,942 exposures for vaccines not recommended in pregnancy, and from 239 to about 16,606 exposures for vaccines recommended for use in pregnancy.

The FDA guidance mentions that for pregnancy registries, the “determination of an adequate sample size depends on the objective(s) and design of the registry and the background rate of the outcome in the study population. If more than one pregnancy outcome is considered, sample size determination should be based on the outcome with the lowest background rate (e.g., MCM).” It also notes that a single teratogen is unlikely to increase the risk of all MCMs and acknowledges that pregnancy registries often lack power to study specific MCM risks unless the effect size is substantial.

The EMA guidance suggests pooling data from multiple databases to address sample size concerns. The sample size should be adequate to evaluate single events of birth defects rather than all MCMs aggregated.

The sample size calculation is also impacted by the exposure period. Indeed, as the window of exposure gets smaller, fewer pregnancies meet the exposure criteria, thus reducing the sample size in the exposed cohort. The FDA guidance mentions that “typically, a specific defect or pattern of defects is associated with a specific teratogenic exposure during a critical period”, which generally occurs during the first trimester. In earlier FDA guidance (United States Food and Drug Administration, 2005) this is further described, stating that “each part, tissue, and organ of an embryo has a critical period during which its development may be disrupted” and that “evaluating the timing of exposure is also important when assessing the power of a study.”

Frequent adverse outcomes include preterm birth (10% of US infants in 2022) (CDC, 2025a), and small for gestational age (11.1% of US live births) (Jensen et al., 2019). Major structural or genetic birth defects affect 3% of births in the US (Centers for Disease Control and Prevention, 2008); prevalence estimates ranged from 0.63 per 10,000 live births for common truncus to 18.65 for clubfoot (Stallings et al., 2024) during 2016–2020. Other outcomes of interest are rare (e.g., stillbirth, 5.7 per 1,000 births (CDC, 2025b)). For each outcome, the minimal sample size to be reached can be determined, based on the prevalence in the comparison group, the desired minimal detectable risk, and the statistical power. A pregnancy safety study assessing multiple outcomes of interest will have a variable minimal detectable risk per outcome for constant power and sample size.

Additional guidance on minimal detectable risk and statistical power for priority outcomes is essential to ensure that studies are being designed to meet the targets expected by regulators. The requirements may differ based on the state of knowledge (no specific risk identified versus specific safety concern), which will inform the selection of the study objectives, and based on the expected level and timing of exposure, impacting study feasibility.

We observed varying durations for data collection across studies (Tables 3–6), with primary data collection studies spanning three to 19 years and secondary data studies spanning one to 5 years. These differences were driven by sample size targets and differences in vaccine exposure among pregnant persons (e.g., indications, recommendations, and uptake rates in participants of childbearing potential and pregnant persons). Many studies, particularly passive surveillance studies, did not have pre-specified enrolment targets. These studies are typically established as post-approval requirements or commitments with regulatory agencies, requiring periodic reports (e.g., annually for 5 years) and a final report upon study completion. In some cases (e.g., FDA Fluarix (US Food and Drug Administration, 2012)), regulators may require continued enrollment until they approve study closure, leading to extended durations. In the case of varicella zoster virus (VZV)-containing vaccines ⁴ , two factors limited the study’s feasability: the low exposure rate among varicella-susceptible women of childbearing age and the rarity of congenital varicella syndrome. These made it unlikely to collect robust data within a reasonable timeframe (Marin et al., 2014). As a result, the FDA supported closing the registry and revising product label information. The registry, initiated in March 1995, ceased patient enrollment in October 2013.

Regulatory guidelines do not address the duration of the inclusion period, which is tied to target sample size, the feasibility of achieving it within reasonable timelines, and participant retention. The FDA recommends pregnancy registries continue until one or more of the following occurs: “(1) sufficient information has accumulated to meet the scientific objectives of the registry, (2) the feasibility of collecting sufficient information diminishes to unacceptable levels due to low exposure rates, poor enrollment, or loss to follow-up, and (3) other methods of gathering appropriate information become achievable or are deemed preferable.”

The study duration is also determined by the outcomes of interest. Longer-term outcomes, such as child health, growth, and neurodevelopment, require extended follow-up duration. Some cardiac, renal, and intestinal malformations may not be diagnosed immediately postpartum, and their incidence is significantly influenced by the duration of follow-up and diagnostic tests availability. Therefore, long-term follow-up is recommended when possible and appropriate. According to the EMA GVP guidelines, the decision to pursue long-term follow-up for childhood outcomes should be based on biological plausibility and supporting evidence from available data such as non-clinical findings, clinical outcomes, pharmacological properties, and identified signals.

Better estimation of the duration of pregnancy safety studies is essential for planning. Defining a target sample size upfront helps determine feasible study design approaches (e.g., active prospective registry, passive surveillance studies, linked electronic healthcare claims and/or EMR data). Clear sample size targets for priority outcomes (refer to Section 4.5) and appropriate follow-up duration is essential to better estimate study duration and optimize planning efforts.

Estimating pregnancy exposure for new products might be challenging, especially in case of inadvertent exposure. Extending a study with insufficient recruitment or unmet objectives may not be appropriate. If recruitment goals are not met despite best efforts, adjustments may be necessary in consultation with regulatory authorities. It is therefore important for MAHs to discuss study progress with regulators at regular intervals, such as during the submission of annual study progress reports, to revise study plans when recruitment and retention fall below initial expectations. Stopping rules could be defined in the protocol for cases where the number of outcomes occurring in a pre-defined period is below the pre-defined threshold (in the context of adequate recruitment and follow-up).

Of the example studies reviewed, only one is a collaborative study spanning multiple MAHs (Multiple manufacturers-COVID-19 vaccines; Table 5). This study is a primary data collection study of COVID-19 vaccines, covering more than 20 countries with a target sample size of at least 6,000 individuals (at least 500 per vaccine brand). Besides this study, the geographical representation of pregnancy registries was notably limited and most passive pregnancy surveillance studies predominantly consisted of cases from the US. The multi-product COVID-19 vaccine pregnancy registry stands out for its global recruitment efforts and innovative approach of engaging pregnant people through social media platforms, representing an advancement beyond traditional recruitment methods. The absence of multi-stakeholder collaborative studies among secondary data analysis studies likely reflects that, while studies may utilize common real-world data sources, it may be currently more efficient for each MAH to develop separate study protocols and analytic plans that generate product-specific results rather than collaborating in a pre-competitive scenario.

In current guidance and recent workshops held to advance optimal approaches to generate high quality, timely safety data about product use during pregnancy, the FDA recommends collaborations across MAHs and within therapeutic areas. These collaborations would drive efficiency across stakeholders (i.e., patients/providers/MAHs/regulatory authorities), increase quality, and promote trust. The anti-retroviral therapy pregnancy registry (APR), a multi-sponsor study that has been ongoing for over 30 years to monitor antiretroviral (ARV) drug use during pregnancy for early signs of potential teratogenicity, is frequently cites as an exemplar of a successful cross-industry collaboration (Albano et al., 2024).

Recent EMA guidance does not specifically recommend cross-industry collaboration but it highlights two key advantages: (1) methodological benefits of evaluating the impact of product use during pregnancy through multi-product or disease-based registries, including the ability to construct internal comparators and better control of potential confounding due to indication, and (2) efficiency gains to overcome relatively low pregnancy exposure and outcomes prevalence through well-designed, large, multi-national/multi-database approaches that may utilize common data models or data pooling (e.g., Nordic registers).

Outside of exceptional examples like the APR and the North American Antiepileptic Drug (AED) Pregnancy Registry, which were set up with successful partnerships across MAHs with common post-marketing commitments/requirements to study drugs with expected intended use during pregnancy, there have been barriers for MAHs to collaborate on a single study or shared platform. These barriers to collaboration include differences across MAHs on timelines of approval/milestones, labeling of indication and treatment guidelines, variation in uptake and footprint, feedback from regulatory agencies that does not support a harmonized approach, lack of common data/operational infrastructure, funding structures, and confidentiality concerns about sensitive information.

Establishing an independent, common registry or RWD platform with a core protocol that is hosted by a third party, and from which MAHs with similar needs could partner for study conduct of secondary database studies, is a potential solution to overcome many of the current barriers to pre-competitive collaborative approaches. A key enabler for driving scientifically rigorous and efficient evidence generation would be the creation of a minimum core set of safety outcomes aligned across health authorities (Table 7) and vaccine MAH.

To facilitate and incentivize the creation of multi-sponsor registries that are fit for purpose to fulfil post-approval safety and/or effectiveness requirements, more guidance and harmonization across regulatory agencies is needed, including flexibility for registry platforms to be owned by third parties (e.g., contract research organizations). Compared to traditional, single product registries, the benefits of multi-sponsor registries can include efficiency in financial and human resource requirements (MAH, regulatory agency, investigators), time (from study start to delivery of actionable evidence), and participants number (simplified enrolment reducing burden, re-use of comparator/un-exposed groups). Through collective alignment on critical design and analytic specifications, multi-sponsor registries can drive scientific rigor to minimize bias and enhance reliability of results. Finally, their larger and more representative sample sizes will allow for reduced uncertainty and improved confidence in the study results. For vaccines without an expected recommendation for use during pregnancy, these benefits would be even more acute, since lower uptake rate would likely render traditional registry approaches unsuccessful.

In addition to the traditional cohort study designs that have been described thus far, other analysis techniques may also be used to generate safety data in pregnancy. For example, one study (Abrysvo-RSV; described in Table 6) utilizes both a rapid cycle analysis component and a subsequent cohort study analysis. Rapid cycle analysis is a technique to rapidly identify potential signals associated with vaccine safety (Klein et al., 2021; Yih et al., 2011). The technique uses the maximized sequential probability ratio test (maxSPRT) to conduct repeated statistical tests to determine whether there is a significant association between an exposure and an outcome. To account for the repeated testing procedure and reduce the rate of false positives, the proportion of total alpha that is used at each test is spent according to the proportion of total information that has been received at the time of the test (and by the predetermined shape of the alpha spending function that is selected). In contrast to a traditional cohort study in which sufficient sample size must be achieved before statistical testing can be initiated, the rapid cycle analysis allows for testing to begin as soon as data begins to accrue. While this technique can help to identify potential safety signals, confounding control is typically limited given small available sample size, and thus is best followed up with a well-designed and large cohort study when feasible to assess causality. The increasing use of rapid cycle analysis prior to conducting a well-designed cohort study allows for early identification of vaccine safety information, which can inform appropriate risk management.

There was no use of a rapid cycle analysis in the primary data collection studies identified.

Despite an increasing need to generate faster safety evidence, no guidance was identified from the FDA or EMA regarding the use of rapid cycle analyses in post-approval pregnancy safety studies.

Additional guidance on the role of rapid cycle analyses for the generation of safety evidence is needed from regulatory authorities. This includes information on when a rapid cycle analysis might be most effective (e.g., for vaccines recommended vs. not recommended in pregnancy, outcomes) and additional information regarding basic requirements for this method to be implemented (e.g., recommended data sources).