INTRODUCTION

There is a very long tradition of registering pregnant persons and births. Collecting data during pregnancy and childbirth is feasible as most pregnant persons attend prenatal care and deliver with skilled birth attendants. UNICEF estimated that between 2015 and 2020 87% of pregnant persons in the world had at least one prenatal visit, and 83% of births were attended by skilled attendants.^1,2 This is the case in most low- and middle-income countries (LMICs), such as the Democratic Republic of the Congo (DRC) where, in 2018, 82% had at least one prenatal visit, 85% of births were attended by a skilled birth attendant and 82% of deliveries took place in a health facility.^1,2

Prenatal care and delivery room registries have been used in paper form for many years. Evidence suggests that maternity registries are a reliable data source because of their consistency with medical records. In 2000, we found that in three departments in Guatemala the date of delivery coincided in 96.2% of cases where it was recorded in both delivery room registries and medical records; the person who attended the delivery coincided in 97.8% of cases; and cesarean sections, maternal complications, and neonatal deaths coincided in 100% of cases.³ Data from sentinel prenatal clinics are often used to generate population-based estimates of key indicators, such as HIV/AIDS seroprevalence.⁴ Perinatal data collection systems have evolved from registries to more detailed and structured perinatal records. Perinatal databases were among the first ones to be digitized, with electronic perinatal databases appearing in the early 1980s.⁵

Our objectives are to review perinatal information systems and health indicators. We define the perinatal period as including pregnancy, and/or delivery, and/or the neonatal period (0–28 days). Some systems may include follow-up data beyond the neonatal period. We will also summarize data analysis methods, with an emphasis on the choice of comparison groups and potential biases.

PERINATAL INFORMATION SYSTEMS

We performed a landscape analysis to identify perinatal information systems, which could inform maternal immunization active safety surveillance in LMICs. The landscape analysis included a scoping review, expert consultation, and interviews of key informants.⁶ The scoping review found eight eligible information systems that included research networks (Global Network’s Maternal Newborn Health Registry (GN-MNHR)^7,8 and International Network for the Demographic Evaluation of Populations and their Health (INDEPTH)),^9,10,11 perinatal electronic medical records (Perinatal Informatic System (SIP),^12,13,14,15 Pregnancy Exposure Registry & Birth Defects Surveillance (PER/BDS)),¹⁶ and SmartCare,^17,18 or custom software platforms for health information systems (District Health Information Software 2 (DHIS 2),^19,20,21,22 Open Medical Record System (OpenMRS),^23,24,25,26 and Open Smart Register Platform (OpenSRP))^27,28,29 (Table 1). These are active and existing systems being used in LMICs in Africa, Asia, and Latin America that can collect individual-level perinatal data, at the facility and community levels, and have the capacity to inform background rates on adverse perinatal outcomes.

These systems are supported by different types of institutions³⁰ such as private foundations (e.g., Bill & Melinda Gates Foundation), governmental agencies in high-income countries (e.g., United States Agency for International Development (USAID), Centers for Disease Control and Prevention, NORAD), global health initiatives (e.g., The Global Alliance for Vaccines and Immunization), research institutes (e.g., National Institute of Child Health and Human Development), non-governmental organizations (e.g., Comic Relief), international organizations (e.g., World Health Organization (WHO)/Pan American Health Organization), universities (e.g., University of Oslo), private sector organizations (e.g., GlaxoSmithKline), and LMIC governments (e.g., South Africa National Department of Health). Some organizations were also responsible for the development and implementation of the systems and for its optimal operability, such as the University of Oslo in the case of DHIS 2.

Each information system uses its own procedures to collect data, follow documented manuals, and certify the training of users and administrators. DHIS 2, GN-MNHR, and INDEPTH were selected as the top three systems after experts reviewed the governance, system management, system design, data source, data management, and data quality domains of each system in an expert consultation. The selected systems demonstrate features that support the protection and privacy of collected information through anonymization of data, implementation of passwords before access, or external security (cybersecurity), and internal and external quality control procedures. The three systems can collect patient data and longitudinally track pregnant persons’ progress and their babies over the prenatal and postnatal periods. However, the timing of capturing information from antenatal visits is different among the systems. GN-MNHR and INDEPTH collect antenatal care data retrospectively, GN-MNHR collects their data at enrollment and delivery, and INDEPTH collects self-reported data retrospectively from household visits. It is important to note that INDEPTH has been adapted to collect data prospectively in Burkina Faso, Kenya, and Mozambique to evaluate artemisinin exposure and monitoring in pregnant people.³¹

PERINATAL HEALTH INDICATORS

The Global Alignment of Immunization Safety Assessment in pregnancy (GAIA) project³² and the WHO Global vaccine safety in pregnancy multi-country collaborative study (MCC) have started to define and pilot case definitions for priority perinatal outcomes. The GAIA partners developed the first set of over 29 standardized case definitions of prioritized obstetric and neonatal outcomes based on the standard Brighton Collaboration process.

The feasibility to implement GAIA variables in the field remains an open question.^32,33 One challenge is to define what is the minimum acceptable set of variables that are needed to have quality information about perinatal outcomes and to inform maternal immunization active surveillance systems during pregnancy. As part of our landscape analysis, experts were invited to select the minimum set of variables, or maternal and neonatal outcomes, that should be collected in routine perinatal systems in LMICs to monitor COVID-19 vaccine post-marketing surveillance during pregnancy (Table 2).³⁴

This panel’s prioritization of maternal, neonatal, and vaccine-related variables can guide the necessary adaptation of existing maternal and neonatal data-collection systems in LMICs for maternal vaccine active safety surveillance. Among the eight perinatal information systems identified by our landscape analysis, we can see that all systems collect vital data, such as maternal and neonatal deaths (Tables 3 and 4). The most frequently recorded perinatal outcomes are fetal distress, postpartum hemorrhage, antenatal bleeding, dysfunctional labor, spontaneous abortion, congenital anomalies, neonatal infections, preterm birth, stillbirth, low birth weight, small for gestational age, and respiratory distress. Some outcomes are not currently recorded by any of the selected systems (i.e., premature preterm rupture of membranes, preterm labor, insufficient cervix, and neurodevelopmental delay).

Variability in case definitions and diagnostic criteria across data sources and among different cultures and languages is a challenge. The lack of harmonization of case confirmation/classification among systems is also a defined problem. These eight systems are not currently using the harmonized case definitions developed by the Brighton Collaboration GAIA project,³⁵ however, some of them (e.g., GN-MNHR) follow the WHO perinatal definitions and others (i.e., DHIS 2, INDEPTH, OpenMRS, SIP, and PER/BDS) are utilizing the tenth revision of the International Classification of Diseases (ICD10) for coding outcomes and conditions.

GN-MNHR and INDEPTH coordinate the data collection and validation across their sites. In contrast, DHIS 2 offers different modules on an open-source platform that each site can customize, modify, and adapt for use with total autonomy. However, all of them have the flexibility to add new variables.

DATA ANALYSIS

Comparison groups

The choice of a comparison group is a key step in the analysis of vaccine outcomes. The analyses compare vaccinated ('exposed') to non-vaccinated ('non-exposed') pregnant persons. Epidemiologists classify subjects enrolled in cohort studies as 'exposed' and 'non-exposed' to avoid confusion with 'cases' and 'controls' enrolled in retrospective case-control studies that compare persons with and without a health condition.³⁶

It is important to have high-quality data for both the vaccinated and non-vaccinated persons. Several variables, often listed as potential adverse events, are not rare and are expected to occur frequently among non-exposed. For example, the INTERGROWTH-21st study found 5.5% of all newborns in urban populations in eight countries were low birthweight (LBW) newborns.³⁷ Thus, it is not sufficient to collect high-quality data among the non-vaccinated. Comparisons to vaccinated persons can be affected by measurement biases if the quality of the data is not homogenous. For this reason, it may be easier to interpret perinatal databases covering all pregnant individuals than active safety surveillance studies only collecting data among vaccinated persons.

Vaccine studies often compare frequencies of health outcomes among the exposed to background rates in the general population. In practice, background rates generally differ between different populations and different data-collection systems. For frequent events, such as LBWs, data are often only available in LMICs from surveys, such as the Demographic and Health Surveys.³⁸ Data might not be available for less frequent adverse events in LMICs. Thus, comparisons to background rates are often imprecise and can only detect very large differences. Analyses comparing vaccinated and non-vaccinated pregnant persons from the same perinatal database are much more likely to provide precise estimates than comparisons to background rates from the general population. As seen above, the long-standing tradition of systematically collecting data during pregnancy provides a unique opportunity to compare vaccinated and non-vaccinated persons.

Biases

The fundamental characteristic of perinatal data, the length of gestation, needs to be considered. The likelihood to be vaccinated increases with the length of gestation (Figure 1). Crude data will thus give the impression that vaccines 'protect' against preterm delivery and other adverse birth outcomes. Influenza vaccine studies have shown that not adjusting for length of gestation can bias estimates in favor of vaccination by 10–26%.³⁹ The analysis should also consider the gestational age at which the vaccination is performed. Epidemiologists see the time before vaccination as 'immortal time' because death occurring during that time would have made vaccination impossible.³⁹ Death and other adverse events (e.g., preterm delivery) occurring during the 'immortal time' would thus erroneously be attributed to the absence of the vaccine. Availability of vaccines can also vary with time, which could create additional biases with persons delivering early having a smaller likelihood of still being pregnant when vaccinations become available.³⁹ The key point is to take time into account.

Potential confounding variables also need to be taken into account. There might be a healthy vaccine bias, as a higher socio-economic status is associated with both vaccination status and better health outcomes. On the other hand, persons with comorbidities might be vaccinated more often, which would create a non-causal association between vaccination and poor health outcomes. One approach to adjust for both time and potential confounders is to perform multivariable analyses. A Cox regression model is often used because it gives an adjusted hazard ratio, which can be interpreted as a relative risk. One example, shown in Table 5, from Zerbo and collaborators, demonstrates the misleading effect of crude data, which gives the impression that the influenza vaccine protects against preterm birth and LBW.⁴⁰ Crude risk ratios were statistically significantly lower than 1. This effect disappears after performing a Cox regression, with adjusted hazard ratios not statistically significantly different from 1.

CONCLUSION

A significant number of perinatal information systems are active in LMICs. They provide a unique opportunity to collect data in both vaccinated and non-vaccinated populations. Data are limited to minimum basic data sets, but systems often allow additional variables to be added. The data analysis needs to take length of gestation and other potential confounders into account.

PRACTICE RECOMMENDATIONS

ACKNOWLEDGEMENTS

This work was supported, in part, by the Bill & Melinda Gates Foundation [INV008443].

We thank Erin Goucher for her editorial assistance.