Protocol for a sequential, prospective metaanalysis to describe coronavirus disease 2019 (COVID-19) in the pregnancy and postpartum periods

We urgently need answers to basic epidemiological questions regarding SARS-CoV-2 infection in pregnant and postpartum women and its effect on their newborns. While many national registries, health facilities, and research groups are collecting relevant data, we need a collaborative and methodologically rigorous approach to better combine these data and address knowledge gaps, especially those related to rare outcomes. We propose that using a sequential, prospective meta-analysis (PMA) is the best approach to generate data for policy- and practice-oriented guidelines. As the pandemic evolves, additional studies identified retrospectively by the steering committee or through living systematic reviews will be invited to participate in this PMA. Investigators can contribute to the PMA by either submitting individual patient data or running standardized code to generate aggregate data estimates. For the primary analysis, we will pool data using two-stage meta-analysis methods. The meta-analyses will be updated as additional data accrue in each contributing study and as additional studies meet study-specific time or data accrual thresholds for sharing. At the time of publication, investigators of 25 studies, including more than 76,000 pregnancies, in 41 countries had agreed to share data for this analysis. Among the included studies, 12 have a contemporaneous comparison group of pregnancies without COVID-19, and four studies include a comparison group of non-pregnant women of reproductive age with COVID-19. Protocols and updates will be maintained publicly. Results will be shared with key stakeholders, including the World Health Organization (WHO) Maternal, Newborn, Child, and Adolescent Health (MNCAH) Research Working Group. Data contributors will share results with local stakeholders. Scientific publications will be published in open-access journals on an ongoing basis. The protocol for this serial prospective meta-analysis was registered with PROSPERO (ID: 188955) on May 28, 2020. A prospective meta-analysis identifies studies that will contribute data to the meta-analysis, as well as establishes the analysis plans, before the results of the individual studies are known [19]. This approach is similar to a multi-site registry or cohort in the sense that studies work to harmonize collection of key outcomes, but differs from multi-site studies in that each site implements a study design and local protocol that is appropriate for their context [19]. Our goal is to provide a maximally flexible, robust, collaborative, and inclusive analysis framework that is methodologically appropriate. All individual study sites will obtain appropriate ethical approval for their independent data collection activities. As this project is a meta-analysis using aggregate study estimates or analyses of secondary, de-identified individual patient data, it is not classified as human research and thus is exempt from IRB approval. Not all those who are pregnant or give birth identify as women. We use the terms “pregnant women” and “pregnant people” throughout. This accurately reflects both study designs (e.g. a comparison group including those who self-report as women between the ages to 15 and 45) and the language used by the studies contributing data. The study aims to answer basic epidemiological questions about COVID-19 and its impact on maternal and newborn health by pooling data from independent studies using harmonized data definitions and an individual patient data meta-analytic framework to minimize data variability. Analyses will be performed in three groups. First, we will provide descriptive statistics among pregnant women with COVID-19. To understand the relative risks associated with COVID-19 in pregnancy, we will compare pregnant people with COVID-19 to i) other pregnant people without COVID-19 and ii) non-pregnant women of reproductive age (15–45 years) with confirmed or suspected COVID-19. Specific objectives regarding COVID-19 among pregnant or postpartum women include: i) describe the natural history of COVID-19; ii) estimate the incidence of and risk factors for hospitalization, admission to intensive care unit, receipt of critical care, and use of invasive ventilation (for COVID-19); and iii) estimate the infection and case fatality rate. For each of these outcomes, we will also compare the risk for pregnant women with COVID-19 to non-pregnant women of reproductive age with confirmed or suspected COVID-19. Specific objectives regarding maternal health among pregnant or postpartum women with COVID-19 are to estimate the incidence or prevalence of: i) maternal morbidities including hypertensive disorders of pregnancy, abnormal placentation, preterm prelabor rupture of membranes, hemorrhage, embolic disease, etc.; ii) maternal and pregnancy-related death; iii) adverse pregnancy outcomes including stillbirth, preterm birth, small-for-gestational age birth, and low birthweight. For each of these outcomes, we will also compare the risk for pregnant women with COVID-19 to other pregnant women without COVID-19. Specific objectives regarding newborn health among newborns born to patients with COVID-19 are to estimate the incidence and timing of and risk factors for: i) congenital anomalies; ii) perinatal and (early) neonatal mortality; and iii) intrauterine, intrapartum, or early peripartum transmission of SARS-CoV-2 from mother to child. When appropriate, we will compare these risks to those of newborns born to women without COVID-19 during pregnancy. Specific objectives regarding SARS-CoV-2 in biospecimens (e.g. nasopharyngeal, vaginal, or rectal swab, maternal blood, pregnancy tissue, placenta, amniotic fluid, cord blood, breast milk) include estimating: i) the proportion of biospecimens with detectable SARS-CoV-2 virus and median viral load; ii) the association between virus or viral load in biospecimen and key outcomes described above. We originally recruited study sites to join the proposed prospective meta-analysis first via professional research networks, and subsequently via key stakeholder networks. Stakeholders at the National Institute of Child Health and Human Development (NICHD) at the U.S. National Institutes of Health (NIH) supported recruitment of NIH-funded maternal and child health network groups and other U.S. government funded projects. Stakeholders at the World Health Organization (WHO) in the Departments of Maternal, Newborn, Child, and Adolescent Health (MNCAH) and of Sexual and Reproductive Health and Research (SRH) supported recruitment of researchers engaged in the COVID-19 MNCAH research network and the SRH pregnancy cohorts working group [20]. Stakeholders from the International Federation of Gynecology and Obstetrics (FIGO) supported recruitment by issuing an invitation through their international network. Studies were invited to participate based solely on study design. Eligible study designs included: i) registries enrolling all suspected or confirmed cases in pregnancy or postpartum period, ii) cohorts enrolling all pregnant women, or iii) case-control studies enrolling pregnant or postpartum women with suspected or confirmed COVID-19. There were no a priori sample size limitations due to the dynamic epidemiology of the pandemic. Study investigators confirmed their intent to contribute to the PMA via a letter of intent or signing a collaboration agreement. Building on concepts laid out in the Framework for Adaptive Meta-analyses (FAME) [21], we will also collaborate with the PregCOV-19 Living Systematic Review Consortium to identify studies that might be eligible for post-publication inclusion into the proposed meta-analysis. The search strategy for the LSR was previously published [8, 22]. We will screen all published studies included in the living systematic review for potential inclusion in the PMA using the following criteria: i) the study conforms to the study designs outlined above; ii) there is a defined catchment area (e.g., certain hospitals, states, etc.); and iii) the sample size included at least 25 pregnant or postpartum people with confirmed or suspected COVID-19 who were consecutively recruited or identified through surveillance. The purpose of including only studies with a defined catchment is to allow us to understand the underlying population from which participants are sampled and thus assess potential risk of bias and to support our efforts to ensure there are no overlapping participants. Confirmed cases of COVID-19 will be defined as those with laboratory-confirmed SARS-CoV-2 infection via a nucleic acid amplification test, regardless of clinical signs or symptoms. The original protocol was expanded to include COVID-19 cases confirmed via antigen tests as they became validated and widely used over the course of the pandemic. Suspected cases will be defined according to the WHO August 7, 2020 case definition based on either a) clinical (acute onset fever and cough, or acute onset of three or more pre-identified signs and symptoms) and epidemiological criteria (residing, working, or traveling in an area with high risk of transmission or community transmission in the past 14 days or working in any health care setting) or the severe acute respiratory illness (SARI) case definition [23]. Probable COVID-19 infections will also be defined according to the WHO August 7, 2020 case definition (clinical criteria and contact with a case; a suspected case with chest imaging, recent anosmia or ageusia onset without other cause, respiratory disease preceding death and contact with a case) [23]. Outcomes related to women’s health will include: Neonatal outcomes of interest include: We developed the draft data modules and questions in April 2020 based on a proposed set of questions from the Pregnancy CoRonavIrus Outcomes RegIsTrY (PRIORITY) study (ClinicalTrials.gov Identifier: {“type”:”clinical-trial”,”attrs”:{“text”:”NCT04323839″,”term_id”:”NCT04323839″}}NCT04323839). We also reviewed and included questions from the case report and data collection forms developed by the Department of Sexual and Reproductive Health (SRH), HRP, WHO and the WHE, WHO prospective cohort study, and the U.S. Centers for Disease Control and Prevention (CDC) Surveillance for Emerging Threats to Mothers and Babies Network (SET-NET) study [26, 27]. We requested two rounds of feedback via a survey and by email from the >50 participants of the bi-monthly informal meeting established at the beginning of the pandemic called the “Perinatal COVID-19 Global Gathering”. The current data modules reflect feedback and general consensus among survey respondents. The final draft of the data modules and core variables was finalized and shared broadly on June 2, 2020 (S1 File). We updated the data modules in September 2020 to reflect evolving understanding of SARS-CoV-2 infection in newborns and to reflect the updated SRH, HRP, WHE UNITY generic protocol developed by WHO for COVID-19-related pregnancy cohort studies [26] (S2 File). We will develop a codebook and statistical codes for each contributing study to map original study variables to the PMA core variables. Investigators can join the PMA either by submitting individual patient data or running standardized code to generate aggregate data estimates that can be included in the meta-analysis. The same data quality and consistency checks will be performed for each study, and any issues will be resolved with study investigators. Studies will be eligible to contribute data to the PMA when they have accrued at least 25 confirmed cases with completed follow up. Study-specific estimates for the two-stage meta-analysis will be produced using standardized analytical codes, and aggregated measures will be exported into a database for the meta-analyses. Individual studies are expected to proceed with their own study publications and can use any analyses generated by the PMA for publication or policy making in their study context. For each updated meta-analysis, we will review included studies for potentially overlapping participants to avoid including individual participants in the estimates for more than one study. We will choose a single study to contribute cases from given locations (e.g. health facilities, cities, states, or countries). The study with the most complete data will be selected to contribute, or in the case of similar data between studies, we will include data from the study that was the first to join the PMA. The number of studies and the number of cases per study is therefore likely to change with each updated meta-analysis, given this robust analytical strategy. We will use a modified version of the Newcastle Ottawa Scale to assess the risk of bias of individual studies. Ideally, all individual-level data would be combined for one-stage meta-analysis. However, we anticipate that the ability to quickly share data and the degree of willingness to share individual patient data may vary by country and across collaborators and institutions. Thus, we will plan for a step-wise statistical analysis plan where the most feasible and simple analyses (that contribute directly to our research questions) are prioritized, and more advanced statistical modeling will be conducted subsequently. For the first stage of analysis, prevalence, and incidence data for the respective outcomes, overall and by risk factor, will be pooled using the conventional DerSimonian-Laird random-effects model [28]. For analyses where only proportions or crude incidence rates are used, the Arcsine method may be applied to stabilize the statistics and ensure approximate asymptotic normality [29]. For analyses comparing pregnant COVID-19 cases to a) COVID-19 negative pregnancies or b) non-pregnant women with COVID, we will calculate the relative risk or odds ratio (for case-control studies) for each outcome; effect estimates will be pooled using the conventional DerSimonian-Laird random-effects model [28]. We will assess forest plots visually for heterogeneity. When at least ten studies are being pooled, we will also quantify heterogeneity by the I2 statistic [30]. As the meta-analysis is designed to be sequential and will be expanded and updated over time, we will revisit the proposed data synthesis methods for subsequent stages of analysis to ensure we are using the most appropriate methods given available data as the project evolves. Where appropriate and as sample size allows, we will consider meta-regression or subgroup analyses by the following study level characteristics: study design and sampling strategy, proportion of confirmed COVID-19 cases (out of suspected and confirmed cases), national maternal mortality ratio, national neonatal mortality rate, and geographic region. We will also consider subgroup analysis by the following individual patient characteristics to identify risk factors or effect modifiers for specific outcomes: confirmed versus suspected COVID-19 case status; gestational age at COVID-19 onset (by week or by trimester), COVID-19 severity, pre-pregnancy health conditions or comorbidities [diabetes, hypertension, cardiovascular disease, obesity (Body Mass Index ≥30 kg/m2), underweight (Body Mass Index <18.5 kg/m2), tuberculosis, malaria, HIV/AIDS, syphilis, anemia (hemoglobin < 11 g/dL)], maternal morbidities (also described above as outcomes), maternal vaccination status (influenza, COVID-19), calendar time or time since first COVID-19 diagnosis in the study area, parity, maternal age, race or ethnicity, and maternal education. Some of these risk factors may be considered both outcomes of having COVID-19 in pregnancy or factors that exacerbate SARS-CoV-2 infection in pregnancy; we will attempt to disambiguate between the two based on the timing of diagnoses whenever possible. Consistent with the GRADE event-based approach, we will avoid “very low precision” by requiring a minimum number of three sites or a total of 100 events for each outcome to run the first analysis [31]. As data accumulate, we will evaluate the robustness of the inferences and whether the answer can be reasonably inferred via a conservative sample size. Assuming at least 50% heterogeneity between sites, the need for 90% power and 5% type 1 error to detect a difference, a conservative 1% prevalence/incidence and a risk ratio of 1.5 for the considered subgroup, a minimum of 30,000 participants are required to demonstrate an effect for each outcome and the proposed subgroups (risk factors). Alternatively, a total of 400 events or more may be used as a threshold for ‘sufficient evidence’ in accordance with recommendation of the GRADEPROfiler [31, 32]. When these thresholds are met, the steering committee must decide whether additional, updated meta-analyses will be performed. The steering committee will consist of at least one member from each participating site, key stakeholders (S3 File), and the technical coordinating team at the George Washington University. When a formal vote is needed, the following teams will each cast one official vote: each group of investigators linked to participating studies, each key stakeholder organization, and the technical coordinating team. The steering committee will prioritize research questions and agree on common elements of data collection. They will disseminate results, including rapid reports to key stakeholders, webinars, and submission of manuscripts to preprint servers and scientific journals. The technical coordinating team will develop protocols for data transfer and ensure data quality; write the statistical analysis plans; and conduct meta-analyses. At the time of publication, 25 studies (in 41 countries) were actively participating in the ongoing PMA (Table 1). Among these studies, 11 have a contemporaneous comparison group of pregnancies without COVID-19; two studies include a comparison group of non-pregnant women of reproductive age. The median anticipated sample size of participating studies is 1,500. More than 76,000 pregnancies in total are expected to contribute to the completed meta-analyses (Table 1). These studies include data from 41 countries: Argentina, Australia, Bangladesh, Belgium, Brazil, Canada, Chile, China, Colombia, Democratic Republic of the Congo, Egypt, France, French Guyana, The Gambia, Germany, Ghana, Guatemala, Hong Kong, India, Israel, Italy, Kenya, Lebanon, Mali, Mexico, Mozambique, Netherlands, New Zealand, Nigeria, Pakistan, Peru, Portugal, Puerto Rico, Rwanda, South Africa, Spain, Sweden, Switzerland, Turkey, Uganda, United Kingdom, United States, and Zambia (Fig 1). Data from additional countries is expected from the WHO prospective cohort study and other studies that will be identified via the published literature. A description of each study, including the study investigators and institutional affiliations, study design description, recruitment methods, anticipated sample sizes, recruitment timeline, and primary outcomes are presented in supplementary tables (S3 File). 1 Democratic Republic of the Congo, Ghana, Kenya, Nigeria, South Africa, and Uganda. 2 USA, Canada, Mexico, Colombia, French Guyana, Peru, Brazil, Argentina, Chile, Australia, Rwanda, Lebanon, Israel, China, Portugal, Spain, France, Italy, United Kingdom, Switzerland, Germany, Belgium, Netherlands, Ireland, and Egypt. 3 India, Pakistan, Bangladesh, Kenya, DRC, Zambia, and Guatemala. 4 United Kingdom, Albania, Argentina, Austria, Chile, China, Czech Republic, Egypt, Greece, Hong Kong, Indonesia, India, Italy. 5 The Gambia, Mozambique, and Kenya. 6 Uganda, Malawi. 7 Universal testing prior to labor & delivery admission was not uniform across all participating clinical sites, but occurred in the majority by May of 2020. The remaining hospitals initiated universal testing for scheduled delivery admissions only. 8 The generic protocol for the WHO prospective cohort study is available for countries at: https://www.who.int/publications/m/item/a-prospective-cohort-study-investigating-maternal-pregnancy-and-neonatal-outcomes-for-women-and-neonates-infected-with-sars-cov-2. The final list of participating countries that will contribute to the meta-analysis is being confirmed.