Changes in preterm birth and stillbirth during COVID-19 lockdowns in 26 countries

Preterm birth (PTB) is the leading cause of infant mortality worldwide. Changes in PTB rates, ranging from −90% to +30%, were reported in many countries following early COVID-19 pandemic response measures (‘lockdowns’). It is unclear whether this variation reflects real differences in lockdown impacts, or perhaps differences in stillbirth rates and/or study designs. Here we present interrupted time series and meta-analyses using harmonized data from 52 million births in 26 countries, 18 of which had representative population-based data, with overall PTB rates ranging from 6% to 12% and stillbirth ranging from 2.5 to 10.5 per 1,000 births. We show small reductions in PTB in the first (odds ratio 0.96, 95% confidence interval 0.95–0.98, P value <0.0001), second (0.96, 0.92–0.99, 0.03) and third (0.97, 0.94–1.00, 0.09) months of lockdown, but not in the fourth month of lockdown (0.99, 0.96–1.01, 0.34), although there were some between-country differences after the first month. For high-income countries in this study, we did not observe an association between lockdown and stillbirths in the second (1.00, 0.88–1.14, 0.98), third (0.99, 0.88–1.12, 0.89) and fourth (1.01, 0.87–1.18, 0.86) months of lockdown, although we have imprecise estimates due to stillbirths being a relatively rare event. We did, however, find evidence of increased risk of stillbirth in the first month of lockdown in high-income countries (1.14, 1.02–1.29, 0.02) and, in Brazil, we found evidence for an association between lockdown and stillbirth in the second (1.09, 1.03–1.15, 0.002), third (1.10, 1.03–1.17, 0.003) and fourth (1.12, 1.05–1.19, <0.001) months of lockdown. With an estimated 14.8 million PTB annually worldwide, the modest reductions observed during early pandemic lockdowns translate into large numbers of PTB averted globally and warrant further research into causal pathways.


Data preparation and management
Suppressed data: For the few datasets where suppressed values were present due to small numbers (<5), these were inputted using two different methods depending on how data providers suppressed the data: (1) for datasets where the total number of births did not equal the total number of non-suppressed cells, we divided the number of non-allocated births by the number of suppressed cells and (2) for datasets where the total number of births equalled the total number of non-suppressed cells, we inputted the suppressed cell as the midpoint between 1 and the threshold for suppression (usually <5 births) and recalculated the total number of births.
Missing and outlier data: The distribution of the number of births with missing information on gestational age was investigated to determine if these data were missing at random with respect to lockdown. If there was no evidence to suggest that data missing was not at random and if the percentage of births missing information on gestational age did not change between the lockdown and pre-lockdown periods, then we assumed that these were missing gestational age completely at random and re-allocated these births proportionally across the gestational age groups. Where data on births were completely missing for a given month, linear interpolation of the outcome rates was performed using data from the 6 nearest surrounding time-points for the population-based data. For the non-population-based data, where there were higher levels of missing data for consecutive months in some of the datasets, we did not input these values and only modelled using the observed data. We graphed the preterm and stillbirth rates for each month for each dataset to check that the fell within plausible ranges; all plots were reviewed by the statistical analysis team (including clinicians, statisticians and epidemiologists) and where implausible rates were identified, we followed-up with the data provider to check if there had been a data entry error. Where the rates could not be corrected, the implausible data points were treated as missing for analysis.
Bias in capture of births in lockdown: Given the early stage of the pandemic, we would not expect to see any changes in the number of births being observed in our data sources compared to pre-lockdown unless driven by a bias in which women were giving birth in different locations and not being recorded, or due to changes in recording practices. To assess this, we forecasted the expected total numbers of births using a Poisson time series, based on pre-lockdown seasonal and yearly trends, and compared the observed number of births to expected number of births. We calculated the percentage change in the total number of births in the lockdown period by dividing the observed total number of births by the expected number of births. Any population-based datasets where there was a relative change of 10% or more in the number of observed compared to expected births following lockdown were excluded from the population-based analysis, and analyzed as a non-population-based dataset.
Data Management: Data were stored and analyzed in the UK Secure Anonymized Information Linkage (SAIL) Databank 1,2 , Swansea Wales, in compliance with the European General Data Protection Regulation guidelines, adhering to the global gold standard of data governance. All data contributors completed a Data Contribution Agreement (DCA) between their institution and SAIL and were provided with a secure link to upload data directly to the SAIL repository.
To ensure outputs were confidential and safe, all statistical outputs were checked using Statistical Disclosure Control (SDC) procedures before being exported out of the virtual environment. We used SDC guiding principles from the Handbook on SDC for Outputs by the UK Data Service 3 . This prevented the identity of a birth from being revealed or inferred from outputs.

Patient Partner Interpretation
Behind every statistic, there is a story of a baby and a family. Patient organizations from around the globe were raising awareness about inequalities in the area of maternal and newborn health long before the COVID-19 pandemic. Disparities have existed between countries in the delivery of prenatal care for many years; however, the lack of robust data collection strategies and standardized birth registries have hampered efforts to understand these disparities and gain insight towards the underlying causes of preterm birth. As a patient community, we were optimistic that the iPOP Study findings might help us identify reasons why rates of prematurity and stillbirths may have declined in some countries early in the pandemic and that these 'reasons' might be leveraged to help reduce the global preterm birth and stillbirth rates. We perceive two major learnings from the iPOP Study: one related to the study results and another related to the challenges faced by the researchers.
The iPOP Study results revealed small differences in preterm and stillbirth rates during the COVID-19 pandemic, and while the scope of this paper did not identify a reason, we feel it may be due to the impact on access to care. The experience of patient organizations working with families who experience preterm birth indicate that because of pandemic enforced changes to maternal and neonatal care, the patient experience has been dramatically altered 55 .
With access to existing care pathways and evidence-based family-centered care severely disrupted, patient organizations have reported increasing numbers of families seeking alternative sources of support and resources 4 . Our experience leads us to believe that the iPOP Study results are likely related to the significant shift in maternal and newborn care pathways around the globe.
The iPOP Study researchers faced many challenges related to data collection and quality. They had access to limited numbers of globally distributed data sets and obtaining comparable data, especially from LMICs, proved very difficult. These challenges lead us to conclude that maternal and newborn health is still not prioritized as a topic warranting immediate and urgent attention in numerous health systems around the world. GLANCE, the Global Alliance for Newborn Care was launched in 2019 by the European Foundation for the Care of Newborn Infants (EFCNI). Patient organizations from 15 countries contributed towards a Call to Action, advocating for the development of initiatives aimed at improving newborn and maternal health worldwide. Up-to-date, reliable data gathered through standardized methodologies is the cornerstone upon which future research and quality care initiatives must be built and as a collective voice. As such, we are calling for researchers and health providers to learn from the iPOP Study and the pandemic as a whole, to address the deficit in reliable and consistent global maternal and newborn health data.

Supplementary Figures
Supplementary Figure 1   Data started from January 2018, and in most facilities there was a data quality exercise conducted in early 2019 inflating preterm birth rates during this period, making it impossible to draw inferences about impact of lockdown in 2020 (see Supplementary Figure 1).

Facility 5 & 6
No available data on gestational age, only birth weight. These are peripheral facilities which generally do not collect data on gestational age.

Facility 7
While this is the largest private facility conducting private deliveries in the state, there were relatively small number of births (<50 per month).

Facility 2
No data available from April 2020 onwards.

Uganda
Facility 1 Seven months of data missing in 2019.

Sensitivity analysis: comparison of change in association between lockdown and preterm births rates in the meta-analysis when removing large countries
Supplementary