The international Perinatal Outcomes in the Pandemic (iPOP) study: Protocol

Preterm birth is the leading cause of infant death worldwide, but the causes of preterm birth are largely unknown. During the early COVID-19 lockdowns, dramatic reductions in preterm birth were reported; however, these trends may be offset by increases in stillbirth rates. It is important to study these trends globally as the pandemic continues, and to understand the underlying cause(s). Lockdowns have dramatically impacted maternal workload, access to healthcare, hygiene practices, and air pollution – all of which could impact perinatal outcomes and might affect pregnant women differently in different regions of the world. In the international Perinatal Outcomes in the Pandemic (iPOP) Study, we will seize the unique opportunity offered by the COVID-19 pandemic to answer urgent questions about perinatal health. In the first two study phases, we will use population-based aggregate data and standardized outcome definitions to: 1) Determine rates of preterm birth, low birth weight, and stillbirth and describe changes during lockdowns; and assess if these changes are consistent globally, or differ by region and income setting, 2) Determine if the magnitude of changes in adverse perinatal outcomes during lockdown are modified by regional differences in COVID-19 infection rates, lockdown stringency, adherence to lockdown measures, air quality, or other social and economic markers, obtained from publicly available datasets. We will undertake an interrupted time series analysis covering births from January 2015 through July 2020. The iPOP Study will involve at least 121 researchers in 37 countries, including obstetricians, neonatologists, epidemiologists, public health researchers, environmental scientists, and policymakers. We will leverage the most disruptive and widespread ‘natural experiment’ of our lifetime to make rapid discoveries about preterm birth. Whether the COVID-19 pandemic is worsening or unexpectedly improving perinatal outcomes, our research will provide critical new information to shape prenatal care strategies throughout (and well beyond) the pandemic.

Our aim is to capture, at a minimum, data on all births (live and stillbirth) from 28 +0 to 44 +6 weeks gestation inclusive; or above ≥1000g birth weight. We also aim to capture additional data on all births (live and stillbirth) from 22 +0 to 27 +6 weeks gestation, or between 500g and 999g. These data will be included in enhanced analyses. The main analysis study period is January 1, 2015 to July 31, 2020, covering the first lockdown period (in 2020) and the previous five calendar years. We will include data from January 1, 2018 to July 31, 2020 in investigative analysis if earlier data is not available. We will request the most recent data available to allow enhanced analyses covering a wider time period. The primary exposure will be a binary variable for lockdown based on the stringency index. We will use the stringency index from the Oxford COVID-19 Government Response Tracker. The Oxford COVID-19 Government Response Tracker provides a systematic cross-national, cross-temporal measure to understand how government responses have evolved over the full period of the disease’s spread. It collects information on different policies and interventions that governments have instituted in response to the COVID-19 pandemic and using standardized series of indicators creates a suite of composites indices to measure the extent of these responses. The indicators cover information on containment and closure policies (e.g. school closures and restrictions in movement) (C1-C8); economic policies (e.g. income support to citizens or provision of foreign aid) (E1-E4); and record health system policies (e.g. COVID-19 testing regimes or emergency investments into healthcare) (H1-H5). The lockdown stringency index is calculated using only the policy indicators C1-C8 and H1. The value of the index on any given day is the average of nine sub-indices pertaining to the individual policy indicators, each taking a value between 0 and 100. If the most stringent policy is only present in a limited area or region, a binary flag variable denotes limited scope. The codebook for the stringency index is publicly available. We will define lockdown as a score of ≥50 on the Oxford COVID-19 Government Response Tracker stringency index. The decision on this arbitrary cut off has been influenced by scoping of stringency index data in high income settings and comparison of stringency indexes in settings which have and have not implemented lockdown measures. For example, Sweden (which has not had a ‘lockdown’) never implemented measures during the study period that added up to higher than 50 on the stringency index, compared to neighbouring Denmark, which scored above 50 throughout the study period in 2020. We will record timing of reaching a score ≥50 separately for each country/region. Our primary analysis will focus on the start date of pandemic lockdown defined as the first date when a country/region’s stringency exceeded 49 (i.e. as a stringency score of ≥50). Subsequent analyses may include the: •    Time period of pandemic lockdown: defined as a continuous calendar period during which a country/region has a stringency score of ≥50 •    Total duration of pandemic lockdown: defined as the sum of all calendar periods during which a country/region has a stringency score of ≥50 Note: The beginning and length of lockdown may vary by country/region Births during the 2020 lockdown periods will be compared with births occurring before the first date when a country/region’s stringency exceeded 49 (i.e. as a stringency score of ≥50), defined by lockdown stringency index in each country/region. The exact comparator time period may vary by country/region. Primary outcome •    Preterm birth rate_m (main analysis: any birth 28 +0- 36 +6 weeks gestation; denominator total births ≥28 +0 weeks). •    Preterm birth rate_e (enhanced analysis: any birth 22 +0- 36 +6 weeks gestation; denominator total births ≥22 +0 weeks). Secondary outcomes •    Early preterm birth rate_m (main analysis: any birth 28 +0 – 31 +6 weeks gestation; denominator total births ≥28 +0 weeks). •    Early preterm birth rate_e (enhanced analysis: any birth 22 +0 – 31 +6 weeks gestation; denominator total births ≥22 +0 weeks). •    Extreme preterm birth rate_e (enhanced analysis: any birth 22 +0 – 27 +6 weeks gestation; denominator total births ≥22 +0 weeks). •    Spontaneous preterm birth rate_e (enhanced analysis: any birth 28 +0- 36 +6 weeks gestation which is preceded by spontaneous contractions or preterm prelabour rupture of membranes [PPROM]; denominator total births ≥28 +0 weeks). •    Spontaneous preterm birth rate_e (enhanced analysis: any birth 22 +0- 36 +6 weeks gestation which is preceded by spontaneous contractions or preterm prelabour rupture of membranes [PPROM]; denominator total births ≥22 +0 weeks). •    Post term birth rate_m (main analysis: any birth ≥42 +0 weeks gestation; denominator total births ≥28 +0 weeks). •    Stillbirth rate_m (main analysis: any stillbirth ≥28 +0 weeks gestation (or ≥1000g if gestation not available); denominator total births ≥28 +0 weeks (or ≥1000g if gestation not available). •    Stillbirth rate_e (enhanced analysis: any stillbirth ≥22 +0 weeks gestation (or ≥500g if gestation not available); denominator total births ≥22 +0 weeks (or ≥500g if gestation not available). •    Low birth weight rate_m (main analysis: any birth 1000–2500g; denominator live births ≥1000g). •    Low birth weight rate_e (enhanced analysis: any birth 500–2500g; denominator live births ≥500g). •    Very low birth weight rate_m (main analysis: any birth 1000 – 1500g; denominator live births ≥1000g). •    Very low birth weight rate_e (enhanced analysis: any birth 500 – 1500g; denominator live births ≥500g). •    Extremely low birth weight rate_e (enhanced analysis: any birth 500g – 1000g; denominator live births ≥500g). Potential confounders/effect modifiers for the entire iPOP study are represented in a DAG ( Figure 2). We recognise that i) many of the variables in the DAG (e.g. maternal age distribution) are unlikely to have significantly changed within the timeframe of the analysis and thus unlikely to be confounders, and ii) our initial analysis strategy is to compare changes in association with lockdown within datasets; thus these variables are less relevant. To allow expedient provision and analysis of data we propose using aggregate data for WP1 and WP2; with more complex analysis enabled with provision of individual participant data and provider level data in subsequent WPs. National/regional level societal characteristics that we are interested in exploring include mediating and moderating factors obtained from publicly available datasets as described in the section below. Country classification by income as defined by the World Bank (LIC, LMIC, UMIC, HIC) as a proxy for wider social security and healthcare system. We have extended invitations for national, regional and institutional data custodians of birth data to participate through formal and informal networks, social media, lay and scientific media. Participating countries as of December 1 st 2020 are shown in Figure 4. We will request aggregate data from each data provider using an excel spreadsheet template, which includes details on levels of missing data. We will classify data provided to iPOP as Standard, Enhanced, or Investigative, based on the characteristics described in Table 1. We will also ask for completion of a questionnaire regarding the source of data including, country of origin, region(s) covered and size of population covered. To assist data providers on which template to use to capture their data, we have constructed a data flow diagram ( Figure 5). For WP2 we will use the following publicly available data sources: •     Lockdown stringency: Using the stringency index (see section Exposures) and COVID-19: Containment and Health Index defined as a continuous (0–100) or categorical measures. •     Socioeconomic status: Measured by Organisation for Economic Co-operation and Development (OECD) better life index. •     Ambient air quality: Estimated using the Data Integration Model for Air Quality (DIMAQ) 24 , which uses input data from a variety of public sources including: Open Air Quality, NASA Modern-Era Retrospective analysis for Research and Applications version 2 ( MERRA-2) global modelling initiative, satellite imagery data from the Multiangle Implementation of Atmospheric Correction ( MAIAC), and global population density from the NASA/Columbia University Socioeconomic Data and Applications Center. •     Adherence to lockdown indicated by traffic and movement trends: Obtained from publicly available Google mobility data. •     COVID-19 rates: Nationally available via John Hopkins COVID-19 infection rates •     Parental leave policy: Measured by World Bank Data (yes/no; length of paid maternity leave). •     Other country-level characteristics: Measured by World Bank Data (including variables such as world region, GDP, income expenditure, hospital beds, maternal education), The Global Gender Gap Index, The Global Hunger Index and Political stability index. We will use the Secure Anonymised Information Linkage (SAIL) Databank, Swansea Wales, to store all data provided to iPOP. Upon completion of a Data Contribution Agreement between each iPOP data provider and the SAIL Databank, each data providers will either: Data will be transferred into SAIL using the “Split-file” process with the support of the Informatics Service, National Health Services (NHS) of Wales. Person-level demographics are translated to an Anonymous Linking Field (ALF). Additional information on the SAIL File Structure & Data Transfer processes can be found here. iPOP Team Members (analysis team) will access data stored within SAIL via a remote access and conduct data analyses remotely on the International COVID-19 Data Alliance (ICODA) Workbench, via a federated approach. ICODA is a new data platform that allows scientists and researchers across the globe to discover, access and analyse multi-dimensional datasets in a confidential and secure environment. More information can be found on the HDR UK website. To ensure outputs are confidential and safe, all statistical outputs will be checked using Statistical Disclosure Control (SDC) procedures before being exported out of the virtual environment. We will use SDC guiding principles from the Handbook on SDC for Outputs by the UK Data Service. This will prevent the identity of a birth from being revealed or inferred from outputs. A catalogue on the data variables captured will be recorded alongside relevant metadata. These high-level summaries will be made publicly available. All analyses will be fully specified in a comprehensive Statistical Analysis Plan. We will adhere to relevant reporting guidance for example the Strengthening the Reporting of Observational studies in Epidemiology ( STROBE). Descriptive analysis We will use summary statistics and data visualisations to describe, explore and compare the national/regional data to describe the study outcomes and other perinatal characteristics, including: •    All births •    Live and stillbirths •    Preterm and post term births •    Low birth weight •    Spontaneous preterm births In WP2 we will use summary statistics and data visualisations (e.g. choropleth maps) to describe, explore and compare the national/regional data. Statistical modelling We will undertake population-based ITSA for main analyses of primary and secondary outcomes. We will use time-series techniques to capture any underlying temporal trends and seasonality in the data before the implementation of lockdown measures. We will consider both linear and more flexible trends. We will use these time-series regression models to forecast (or predict) the expected trends and will compare these to the observed trends seen after the lockdown measures. This will capture both immediate (i.e. step) changes and gradual (i.e. slope) changes in the outcome in relation to implementation of lockdown measures in our models. All analyses will be prespecified in a Statistical Analyses Plan before analysis. Meta-analysis We will undertake a meta-analysis of national/regional results, on the step-change and the difference between the forecast and observed outcomes at different time points after the implementation of lockdown measures. We will also stratify by country income setting as a dichotomous variable (LIC+LMIC vs UMIC+HIC), since existing data suggests differing effects in these groups. Statistical heterogeneity will be assessed using I 2 test. For WP2, we will use these pooled estimates from WP1 in meta-regression analyses. These will incorporate the moderator/mediator variables as potential mechanisms at a national/regional level. This will measure the influence of these mechanisms on the association between lockdown measures and adverse perinatal outcomes. Sensitivity analyses Where enhanced datasets are available for an outcome, we will perform similar modelling techniques to those described above with these enhanced data as sensitivity analyses to test the robustness of the main analyses in different populations. These analyses will be specified further in a comprehensive Statistical Analysis Plan. Predefined examples include: •    Sensitivity analyses restricting the denominator for our main outcomes of interest (excluding outcomes on spontaneous preterm birth) from all births to only live births. These analyses will be informative for the appropriateness of using datasets which only include information on live births. •    Sensitivity analyses with varying cut-off points for our lockdown definition (i.e. above and below 50) from the stringency index to test the robustness of assigning ≥50 as the primary cut-off point. These analyses will also allow inclusion of countries with less strict lockdown measures, such as Sweden, and inform whether/to what extent the observed associations might vary by lockdown stringency. We will conduct supplementary analyses in investigative datasets. Output confidentiality All outputs will be checked for any potential disclosure and confidentiality breaches, using guidance from the Handbook on SDC for Outputs by the UK Data Service. Public and patient involvement early in study design and development ensures research studies are responsive to input, guidance and advice, and can help identify and mitigate potential challenges early in the research process 25 . Further, public and patient involvement helps to identify research outcomes that are meaningful and pragmatic to knowledge users. The iPOP team has engaged parents as patient partners early in the study design and have built a working group to capture and integrate patient involvement in the iPOP study as it moves forward. Meeting monthly, patient partners will be involved in developing effective and meaningful knowledge translation and communication strategies for disseminating iPOP findings. Specific to WP2, patient partners will work with researchers to examine mechanistic effects of the pandemic lockdown on perinatal outcomes. Patient partners will also work with researchers to develop knowledge translation strategies to ensure effective and meaningful dissemination of findings to knowledge users. To ensure transparent, equitable, and meaningful engagement, we have developed Guiding Principles that outline the terms of agreement for study leads and collaborators who are involved in the iPOP Study. Each member of the iPOP Study must read and sign the guiding principles document in order to collaborate on the study. While not legally binding, this document provides guidance and parameters around authorship, roles and responsibilities, research integrity, communication and Team Science guidelines. The iPOP Study ensures confidentiality and security of the processing of data for electronic files. Data will be safeguarded by an appropriate level of security, technical and organisational measures to prevent unauthorized disclosure or access, accidental or unlawful destruction, accidental loss or alteration, and unlawful forms of processing. WP1 and WP2 will be based on de-identified aggregate data only. It will be assumed that any Team member sharing data within the iPOP Study does so in accordance with relevant and applicable legal and regulatory standards and obligations including but not limited to, confidentiality, data protection and intellectual property, and access governance agreements. iPOP collaborators must adhere to these policies and processes. All collaborators must respect the iPOP principles of data protection and processing, which include the following: All contributed to iPOP data must be •    Processed fairly and lawfully •    Collected for specified and legitimate purposes •    Accurate •    Absent of personal identifiers (names, addresses, etc.) •    Stored not longer than necessary •    Processed under the responsibility and liability of the data Controller for the provided data set •    Handled according to the EU GDPR rules (when hosted in the UK)

N/A

• Dima Health