Global Excess Burden of Pregnancy-Related Adverse Outcomes Among Women of Reproductive Age During the COVID-19 Pandemic Authors: Chenyuan Qin, Jian Xu, Xingang Sang, Yansheng Ding, Jue Liu Published: 2026 ABSTRACT PurposeInfectious disease emergencies may affect pregnancy-related health through biological vulnerability, disrupted obstetric and reproductive care, and health-system shocks. We estimated excess incidence of pregnancy-related adverse outcomes during 2020–2021 relative to a no-pandemic counterfactual.MethodsWe conducted a counterfactual time-series modelling study using data from the Global Burden of Disease Study 2021. Incidence data for pregnancy-related adverse outcomes from 1990 to 2021 were extracted. Five forecasting approaches were trained on 1990–2016 data and validated against 2017–2019 data. For each outcome and location, the best-performing model was selected primarily based on validation mean absolute error (MAE), with root mean square error (RMSE) used as a secondary criterion, to prioritize average out-of-sample prediction accuracy. The best-performing model was used to estimate expected age-standardised incidence rates (ASIRs) for 2020–2021 under a no-pandemic scenario. Observed and expected ASIRs were compared using population-weighted absolute rate differences and relative percentage changes.ResultsGlobally, observed ASIRs for most pregnancy-related adverse outcomes exceeded counterfactual expectations, except for ectopic pregnancy. The largest absolute excesses were observed for maternal sepsis and other maternal infections (32.21 per 100,000 women; +3.40%) and abortion and miscarriage (30.92; +1.58%), followed by maternal hypertensive disorders (3.79; +0.41%), maternal hemorrhage (2.61; +0.36%), and obstructed labor and uterine rupture (1.06; +0.15%). Excess in abortion and miscarriage was most pronounced among women aged 15–19 and 25–44 years, whereas excess maternal sepsis and other maternal infections was greatest among women aged 20–24 years. Significant national-level excess increases were observed in 26 countries and territories for maternal sepsis and other maternal infections, 42 for maternal hemorrhage, 41 for maternal hypertensive disorders, 35 for obstructed labor and uterine rupture, 56 for abortion and miscarriage, and 11 for ectopic pregnancy.ConclusionSeveral pregnancy-related adverse outcomes exceeded expected no-pandemic levels during 2020–2021, with substantial variation by age and country. These findings underscore the need to safeguard essential obstetric and reproductive health services and strengthen equity-oriented surveillance and health-system resilience for future infectious disease emergencies, including Disease X. Introduction Maternal health remains a central priority in global health.1,2 Although substantial progress has been made in reducing preventable maternal morbidity and mortality over recent decades, this progress has been uneven across regions and remains vulnerable to health-system disruption, socioeconomic instability, and interruptions in essential reproductive and obstetric care.3–5 Pregnancy-related health outcomes can be particularly sensitive to infectious disease emergencies because biological vulnerability, health-care dependence, and social vulnerability may converge to increase the risk of adverse outcomes.6–10 Understanding how major infectious disease shocks affect pregnancy-related health is therefore essential for strengthening health-system resilience and protecting progress towards global maternal health targets. The coronavirus disease 2019 (COVID-19) pandemic caused substantial disruption to health systems worldwide.11 Beyond the direct effects of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection, pandemic-related containment measures, reduced service availability, delayed care-seeking, workforce shortages, and disruptions to referral systems may have affected access to antenatal care, emergency obstetric care, safe delivery services, and reproductive health services.7–9 Previous studies have reported increases in selected adverse maternal outcomes during the pandemic, including maternal morbidity, pregnancy complications, and reduced use of essential maternal health services.9,12–14 However, much of the available evidence has been derived from hospital-based or country-specific studies, and many analyses have relied on short-term before-and-after comparisons. Although such studies provide important individual-level clinical evidence, they are less suited to distinguishing pandemic-period deviations from pre-existing secular trends or to making standardised comparisons across countries and age groups. GBD 2021 addresses this complementary need by providing standardised, age-specific estimates across 204 countries and territories over a long historical time series, enabling consistent temporal and cross-national comparisons within a common analytical framework.15 Despite the growing evidence on maternal health during COVID-19, it therefore remains unclear whether specific pregnancy-related adverse outcomes during 2020–2021 deviated from their expected pre-pandemic trajectories and how such deviations varied across maternal outcomes, reproductive-age groups, and geographical settings. Although the acute phase of the COVID-19 pandemic has passed in many settings, the continued transmission and evolution of SARS-CoV-2 and the possibility of future high-impact infectious disease events underscore the need to understand how maternal health systems respond to large-scale systemic shocks.16–18 A global counterfactual assessment is therefore needed to determine whether observed pregnancy-related adverse outcomes during the COVID-19 pandemic deviated from the trajectories expected in the absence of the pandemic. By using pre-pandemic historical trends to estimate expected outcomes under a no-pandemic scenario, counterfactual time-series modelling allows pandemic-period rates to be evaluated against expected trajectories rather than through simple before-and-after comparisons.19,20 Chen et al applied GBD 2021 time-series modelling across a broad range of causes of disease burden, demonstrating the utility of this framework for macro-level assessment of pandemic-period deviations.19 However, that study was designed to characterize changes across a wide spectrum of diseases rather than to provide a dedicated assessment of maternal disorders. Building on this broader framework, the present study focuses specifically on six pregnancy-related adverse outcomes among women of reproductive age, allowing outcome-specific and age-specific patterns to be examined within a coherent maternal-health context. In this study, using standardised GBD 2021 estimates, we developed counterfactual prediction models based on age-standardised incidence rates (ASIRs) to quantify absolute and relative excess changes in six pregnancy-related adverse outcomes among women of reproductive age during 2020–2021 and to characterize their variation at global, regional, national, and age-specific levels. By treating COVID-19 as a real-world stress test of maternal health systems, this study also provides evidence relevant to preparedness for future infectious disease emergencies, including potential outbreaks caused by Disease X. Material and Methods Study Design and Data Sources We conducted a counterfactual time-series modelling study to estimate excess changes in pregnancy-related adverse outcomes during the COVID-19 pandemic. Data were obtained from the GBD 2021 database, which provides standardised and comparable estimates of disease burden by location, year, age, sex, and cause.15 The pre-COVID-19 period was defined as 1990–2019, and the COVID-19 pandemic period was defined as 2020–2021. Annual ASIRs from 1990 to 2019 were used to model expected pre-pandemic trajectories of pregnancy-related adverse outcomes, and counterfactual predictions for 2020–2021 were generated under a no-pandemic scenario. Observed rates during 2020–2021 were then compared with counterfactual predictions to estimate absolute rate difference (ARD) and relative percentage change (RPC). Detailed outcome definitions and model specifications are provided in the Supplementary Methods. As this study exclusively used publicly available, aggregated, and de-identified data from the GBD database and did not involve individual-level participant data, the requirement for ethical approval was waived by the Ethics Committee of Peking University. Data Extraction We extracted annual incidence rates and 95% uncertainty intervals for pregnancy-related adverse outcomes among women aged 15–49 years from GBD 2021 for 1990–2021 at the global level, across 21 GBD regions, and in 204 countries and territories. GBD regions are defined according to geographical proximity, epidemiological similarity, and patterns of cause-specific mortality.15 Age-specific estimates were extracted in 5-year age groups. Maternal disorders were analyzed as six individual causes: maternal sepsis and other maternal infections, maternal hemorrhage, maternal hypertensive disorders, obstructed labor and uterine rupture, abortion and miscarriage, and ectopic pregnancy. The incidence rates used in this study were population-based, with women aged 15–49 years as the population denominator; thus, the ASIRs represent incidence per 100,000 women of reproductive age rather than risk per pregnancy or per live birth. We also extracted the Socio-demographic index (SDI), the standard population of women aged 15–49 years by 5-year age group, and the number of live births by maternal age group. SDI is a composite measure of development ranging from 0 to 1, with higher values indicating higher income per capita, higher educational attainment, and lower fertility.15 Fertility rate was defined as the number of live births per 1000 women aged 15–49 years. SDI and fertility rate were included as exogenous covariates in the prediction models to account for long-term socioeconomic and reproductive trends. Counterfactual Modelling Framework For each location and outcome, the annual ASIR series was divided into a training period (1990–2016), validation period (2017–2019), and prediction period (2020–2021). Because the no-pandemic counterfactual is inherently unobservable, these predictions represent model-based estimates under the assumption that pre-pandemic temporal relationships would have continued through 2020–2021 in the absence of the pandemic. To ensure that the exogenous covariates were consistent with the no-pandemic counterfactual scenario, SDI and fertility rate for 2020–2021 were extrapolated from their pre-pandemic trajectories using data from 1990 to 2019 rather than using their observed pandemic-period values. These covariates were therefore not held constant at pre-pandemic levels but were allowed to evolve according to their pre-pandemic temporal patterns. Observed values for 2020–2021 were not used because they could themselves have incorporated pandemic-related socioeconomic or reproductive changes into the no-pandemic baseline. Autoregressive integrated moving average models (ARIMA) were fitted separately for each covariate, with model parameters selected automatically using the auto.arima function. The resulting counterfactual covariate estimates for 2020–2021 were then used as exogenous inputs in the outcome prediction models. Five forecasting approaches were evaluated: ARIMAX, Prophet, XGBoost, ARIMAX+XGBoost, and Prophet+XGBoost. These candidate models were selected to represent complementary forecasting structures. ARIMAX was included to capture temporal dependence while incorporating exogenous covariates,21 whereas Prophet provides a flexible framework for modelling long-term trends and changes in trend structure.22 XGBoost was included to capture nonlinear relationships and interactions among temporal, lagged, and exogenous predictors.23 The hybrid models further combined statistical time-series modelling with nonlinear residual correction by XGBoost, allowing the two model classes to capture complementary components of the time series. Rather than prespecifying a single model for all series, candidate models were compared using temporal validation, and the best-performing model was selected separately for each location and outcome. These models were used to estimate ASIRs of pregnancy-related adverse outcomes per 100,000 women aged 15–49 years in 2020–2021 under a no-pandemic counterfactual scenario. The hybrid models were specified to combine the ability of time-series models to capture temporal dependence with the ability of machine-learning models to capture non-linear patterns. Details of model training, hyperparameter tuning, validation, and testing are provided in the Supplementary Methods. Residual autocorrelation was assessed using the Ljung–Box test, with a non-significant result interpreted as no evidence of residual autocorrelation. Model performance was evaluated during the validation period using mean absolute error (MAE) and root mean square error (RMSE).24,25 Because the outcomes were annual incidence-rate time series and the primary validation period comprised three held-out years (2017–2019), our primary objective was to minimize the typical magnitude of annual forecast error when reconstructing the expected no-pandemic trajectory. MAE was therefore selected as the primary criterion because it directly measures the average absolute forecast error on the original incidence-rate scale and applies a linear penalty to the magnitude of prediction errors, whereas RMSE is more sensitive to relatively large prediction errors because errors are squared before averaging.24,25 Because candidate models were compared within the same location–outcome series, the scale dependence of MAE did not affect within-series model ranking. RMSE was retained as a secondary criterion to account for relatively large prediction errors. When candidate models had similar MAE values after rounding to one decimal place, the model with the lowest RMSE was selected. The selected model was then refitted using data from 1990 to 2019 and used to generate counterfactual predictions for 2020–2021. MAE and RMSE were calculated as follows: \documentclass[12pt]{minimal} \usepackage{wasysym} \usepackage[substack]{amsmath} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage[mathscr]{eucal} \usepackage{mathrsfs} \DeclareFontFamily{T1}{linotext}{} \DeclareFontShape{T1}{linotext}{m}{n} {linotext }{} \DeclareSymbolFont{linotext}{T1}{linotext}{m}{n} \DeclareSymbolFontAlphabet{\mathLINOTEXT}{linotext} \begin{document}$$RMSE = \sqrt {{1 \over n}\mathop \sum \limits_{i = 1}^n {{\left({\widehat {{y_i}} - {y_i}} \right)}^2}} $$\end{document} \documentclass[12pt]{minimal} \usepackage{wasysym} \usepackage[substack]{amsmath} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage[mathscr]{eucal} \usepackage{mathrsfs} \DeclareFontFamily{T1}{linotext}{} \DeclareFontShape{T1}{linotext}{m}{n} {linotext }{} \DeclareSymbolFont{linotext}{T1}{linotext}{m}{n} \DeclareSymbolFontAlphabet{\mathLINOTEXT}{linotext} \begin{document}$$ MAE = {1 \over n}\mathop \sum \limits_{i = 1}^n \left| {\widehat {{y_i}} - {y_i}} \right|$$\end{document} where \documentclass[12pt]{minimal} \usepackage{wasysym} \usepackage[substack]{amsmath} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage[mathscr]{eucal} \usepackage{mathrsfs} \DeclareFontFamily{T1}{linotext}{} \DeclareFontShape{T1}{linotext}{m}{n} {linotext }{} \DeclareSymbolFont{linotext}{T1}{linotext}{m}{n} \DeclareSymbolFontAlphabet{\mathLINOTEXT}{linotext} \begin{document}${y_i}$\end{document} denotes the observed value at time point i, \documentclass[12pt]{minimal} \usepackage{wasysym} \usepackage[substack]{amsmath} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage[mathscr]{eucal} \usepackage{mathrsfs} \DeclareFontFamily{T1}{linotext}{} \DeclareFontShape{T1}{linotext}{m}{n} {linotext }{} \DeclareSymbolFont{linotext}{T1}{linotext}{m}{n} \DeclareSymbolFontAlphabet{\mathLINOTEXT}{linotext} \begin{document}$\widehat {{y_i}}$\end{document} denotes the expected value, and n denotes the number of observations. After selection of the optimal model for each location and outcome, counterfactual incidence rates and corresponding 95% uncertainty intervals for 2020–2021 were generated under the no-COVID-19 scenario. Population-weighted ARD and RPC between observed pandemic-period rates and counterfactual estimates were calculated to quantify excess changes in pregnancy-related adverse outcomes among women of reproductive age. The overall counterfactual modelling workflow is presented in Figure 1. Detailed procedures for model training, hyperparameter tuning, validation, selection, and testing are provided in the Supplementary Methods. Figure 1Counterfactual modelling framework for estimating excess changes in pregnancy-related adverse outcomes during the COVID-19 pandemic.The flowchart outlines the process of modeling age-standardized incidence rates (ASIRs) of six pregnancy-related adverse outcomes among women of reproductive age from 1990 to 2021. It begins with GDB 2021, leading to ASIRs analysis across global regions and countries. Two exogenous covariates are used: SDI and fertility rate. The modeling approaches include ARIMAX, Prophet, XGBoost, ARIMAX plus XGBoost and Prophet plus XGBoost. Validation prediction for 2017 to 2019 combines these predictions. The validation set is from 2017 to 2019, followed by MAE and RMSE calculations to select the optimal model. Counterfactual predictions for ASIRs are made for 2020 to 2021, compared with observed values. The final step involves calculating absolute rate difference (ARD) and relative percentage change (RPC).A flowchart of ASIRs modeling for pregnancy-related outcomes from 1990 to 2021 using various models.Abbreviation: ARD, absolute rate difference; ARIMAX, autoregressive integrated moving average model with exogenous variables; GBD, global burden of disease; MAE, mean absolute error; RMSE, root mean square error; RPC, relative percentage change; SDI, socio-demographic index; XGBoost, extreme gradient boosting. Counterfactual modelling framework for estimating excess changes in pregnancy-related adverse outcomes during the COVID-19 pandemic. Sensitivity Analysis To assess the robustness of the counterfactual estimates, we conducted three sensitivity analyses. We first compared the main 9:1 training–validation split (1990–2016 and 2017–2019) with 8:2 (1990–2013 and 2014–2019) and 7:3 (1990–2010 and 2011–2019) splits. We then repeated model selection using RMSE as the primary criterion and MAE as the secondary criterion. Finally, we refitted the prediction models without exogenous covariates to examine the sensitivity of the estimated excess changes to model specification. Statistical Analysis For each outcome, counterfactual ASIRs and age-specific incidence rates for 2020–2021 were estimated at the global, regional, and national levels. ASIRs (per 100,000 women of reproductive age) were calculated using the global standard population: \documentclass[12pt]{minimal} \usepackage{wasysym} \usepackage[substack]{amsmath} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage[mathscr]{eucal} \usepackage{mathrsfs} \DeclareFontFamily{T1}{linotext}{} \DeclareFontShape{T1}{linotext}{m}{n} {linotext }{} \DeclareSymbolFont{linotext}{T1}{linotext}{m}{n} \DeclareSymbolFontAlphabet{\mathLINOTEXT}{linotext} \begin{document}$$ASIR = {{\mathop \sum \nolimits_{i = 1}^n {a_i}{w_i}} \over {\mathop \sum \nolimits_{i = 1}^n {w_i}}}$$\end{document} where \documentclass[12pt]{minimal} \usepackage{wasysym} \usepackage[substack]{amsmath} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage[mathscr]{eucal} \usepackage{mathrsfs} \DeclareFontFamily{T1}{linotext}{} \DeclareFontShape{T1}{linotext}{m}{n} {linotext }{} \DeclareSymbolFont{linotext}{T1}{linotext}{m}{n} \DeclareSymbolFontAlphabet{\mathLINOTEXT}{linotext} \begin{document}${a_i}$\end{document} denotes the age-specific incidence rate in age group i, and \documentclass[12pt]{minimal} \usepackage{wasysym} \usepackage[substack]{amsmath} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage[mathscr]{eucal} \usepackage{mathrsfs} \DeclareFontFamily{T1}{linotext}{} \DeclareFontShape{T1}{linotext}{m}{n} {linotext }{} \DeclareSymbolFont{linotext}{T1}{linotext}{m}{n} \DeclareSymbolFontAlphabet{\mathLINOTEXT}{linotext} \begin{document}${w_i}$\end{document} denotes the corresponding standard population weight. Because the underlying population denominator comprised all women aged 15–49 years rather than pregnancies or live births, the ASIRs quantify population-level incidence and should not be interpreted as the probability of an adverse outcome conditional on pregnancy. To quantify pandemic-period deviations from expected trends, population-weighted absolute rate differences and relative percentage changes were calculated by comparing observed incidence rates with counterfactual estimates for 2020–2021. The ARD and RPC were defined as follows: \documentclass[12pt]{minimal} \usepackage{wasysym} \usepackage[substack]{amsmath} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage[mathscr]{eucal} \usepackage{mathrsfs} \DeclareFontFamily{T1}{linotext}{} \DeclareFontShape{T1}{linotext}{m}{n} {linotext }{} \DeclareSymbolFont{linotext}{T1}{linotext}{m}{n} \DeclareSymbolFontAlphabet{\mathLINOTEXT}{linotext} \begin{document}$${\mathrm{ARD}} = {R_{{\mathrm{obs}}}} - {R_{{\mathrm{pred}}}}$$\end{document} where \documentclass[12pt]{minimal} \usepackage{wasysym} \usepackage[substack]{amsmath} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage[mathscr]{eucal} \usepackage{mathrsfs} \DeclareFontFamily{T1}{linotext}{} \DeclareFontShape{T1}{linotext}{m}{n} {linotext }{} \DeclareSymbolFont{linotext}{T1}{linotext}{m}{n} \DeclareSymbolFontAlphabet{\mathLINOTEXT}{linotext} \begin{document}${R_{obs}}$\end{document} denotes the population-weighted observed incidence rate and \documentclass[12pt]{minimal} \usepackage{wasysym} \usepackage[substack]{amsmath} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage[mathscr]{eucal} \usepackage{mathrsfs} \DeclareFontFamily{T1}{linotext}{} \DeclareFontShape{T1}{linotext}{m}{n} {linotext }{} \DeclareSymbolFont{linotext}{T1}{linotext}{m}{n} \DeclareSymbolFontAlphabet{\mathLINOTEXT}{linotext} \begin{document}${R_{{\mathrm{pred}}}}$\end{document} denotes the population-weighted expected incidence rate. All analyses were conducted using R software (version 4.4.2). Data visualization was performed using GraphPad Prism (version 10.4.1), and spatial maps were generated using ArcGIS (version 10.8). All statistical tests were two-sided, and P values less than 0.05 were considered statistically significant. Results Global and Age-Specific Excess Burden of Pregnancy-Related Adverse Outcomes At the global level, observed ASIRs exceeded the no-pandemic counterfactual estimates for five of the six pregnancy-related adverse outcomes, with ectopic pregnancy being the exception (Figure 2, Figure S1 and Table S1). The largest excess was observed for maternal sepsis and other maternal infections, with an absolute increase of 32.21 (95% CI: 31.90, 32.52) per 100,000 women of reproductive age and a relative increase of 3.40% (95% CI: 3.37, 3.43), followed by abortion and miscarriage, with an absolute increase of 30.92 (95% CI: 30.48, 31.36) and a relative increase of 1.58% (95% CI: 1.55, 1.60). Smaller excesses were observed for maternal hypertensive disorders, maternal hemorrhage, and obstructed labor and uterine rupture. These global estimates are presented in detail in Table S1. Figure 2Global and age-specific excess changes of pregnancy-related adverse outcomes among women of reproductive age during the COVID-19 pandemic, 2020–2021.The image shows tables of pregnancy-related adverse outcomes by age during COVID-19. Each table lists: Rank, Outcomes, Absolute Rate Difference (ARD) with 95 percent Confidence Interval (CI) per 100,000 women and Relative Percent Change (RPC) with 95 percent CI. Among women aged 15–19 and 25–44 years, excess abortion and miscarriage was more pronounced, with the greatest excess observed among those aged 25–34 years.Tables rank pregnancy outcomes by age during COVID-19, showing absolute and relative changes.Notes: *P<0.05 indicates statistical significance.Abbreviation: ARD, absolute rate difference; CI, confidence interval; COVID-19, coronavirus disease 2019; RPC, relative percentage change. Global and age-specific excess changes of pregnancy-related adverse outcomes among women of reproductive age during the COVID-19 pandemic, 2020–2021. Age-specific patterns were heterogeneous (Figure 2 and Table S2). Excess abortion and miscarriage was more pronounced among women aged 15–19 years and 25–44 years, particularly those aged 25–34 years. Maternal sepsis and other maternal infections showed the largest excess among women aged 20–24 years, whereas maternal hypertensive disorders were relatively more prominent among women aged 45–49 years. Among women aged 15–19 years, obstructed labor and uterine rupture showed the largest relative increase. Complete age-specific estimates are provided in Table S2. Regional Heterogeneity in Excess Pregnancy-Related Adverse Outcomes Regional excess burden varied markedly by maternal outcome, with the most widespread increases observed for maternal hemorrhage and abortion and miscarriage (12 regions each), followed by maternal sepsis and other maternal infections (8 regions) and ectopic pregnancy (5 regions), while the largest excesses in maternal hypertensive disorders were observed in Western Sub-Saharan Africa and High-income North America and obstructed labor and uterine rupture showed large but geographically localized excesses (Figure 3 and Tables S3–S8). Figure 3Excess changes in pregnancy-related adverse outcomes among women of reproductive age across 21 GBD regions during the COVID-19 pandemic, 2020–2021. (A) Absolute rate differences; (B) Relative percentage changes.The bubble charts illustrate rate differences and percentage changes in maternal health outcomes across various GBD regions. In Image A, the x-axis represents regions like Global, High income Asia Pacific and others, while the y-axis lists outcomes such as maternal sepsis and other maternal infections, obstructed labor and uterine rupture, and more. A triangle indicates a decrease and a circle indicates an increase. The size and shading of markers show the rate change per 100,000 women. Image B presents relative percentage changes with the same axes. Here, markers indicate percentage changes. Both charts highlight regional variations in maternal health outcomes, with High-income North America showed excess increases across multiple outcomes.Bubble charts show regional spikes in pregnancy-related issues, notably abortion and miscarriage.Notes: *P<0.05 indicates statistical significance.Abbreviation: COVID-19, coronavirus disease 2019; GBD, global burden of disease. Excess changes in pregnancy-related adverse outcomes among women of reproductive age across 21 GBD regions during the COVID-19 pandemic, 2020–2021. (A) Absolute rate differences; (B) Relative percentage changes. High-income North America showed excess increases across multiple outcomes, with particularly pronounced deviations in maternal sepsis and other maternal infections (91.91 per 100,000 women of reproductive age; 13.95%), obstructed labor and uterine rupture (144.57 per 100,000; 23.93%), and abortion and miscarriage (114.09 per 100,000; 19.08%). Western Sub-Saharan Africa showed the largest absolute excess in maternal hypertensive disorders (177.31 per 100,000), whereas Eastern Europe had the largest absolute excess in abortion and miscarriage (506.39 per 100,000; 17.07%). Australasia showed prominent excesses in maternal hemorrhage and ectopic pregnancy, with the largest absolute regional increases for these two outcomes. Complete estimates across all 21 GBD regions are provided in Tables S3–S8. National and Territorial Patterns of Excess Pregnancy-Related Adverse Outcomes National and territorial patterns were highly heterogeneous across outcomes (Figure 4, Figure S2 and Tables S9–S14). Significant excess increases were observed in 26 countries and territories for maternal sepsis and other maternal infections, 42 for maternal hemorrhage, 41 for maternal hypertensive disorders, 35 for obstructed labor and uterine rupture, 56 for abortion and miscarriage, and 11 for ectopic pregnancy. Figure 4Absolute rate differences in pregnancy-related adverse outcomes among women of reproductive age across 204 countries and territories, 2020–2021. (A) Maternal sepsis and other maternal infections; (B) Maternal hemorrhage; (C) Maternal hypertensive disorders; (D) Obstructed labor and uterine rupture; (E) Abortion and miscarriage; (F) Ectopic pregnancy.The text describes six thematic world choropleth maps showing absolute rate differences (per 100,000 women of reproductive age) for various maternal health issues. Each map includes a scale bar (0, 3,000, 6,000 Km) and a legend with five classes. Image A focuses on maternal sepsis and infections, highlighting countries like the USA (89.54) and Lithuania (52.60). Image B covers maternal hemorrhage, with notable rates in Switzerland (149.94) and Israel (96.30) . Image C addresses maternal hypertensive disorders, featuring Ghana (1561.54) and Nigeria(134.52). Image D illustrates obstructed labor and uterine rupture, with the USA (166.42) and Cyprus (108.01) among the highlighted countries. Image E shows data on abortion and miscarriage, with high rates in Estonia (1535.42) . Finally, Image F presents ectopic pregnancy rates, noting Australia (23.85)and South Africa (19.43).Six thematic world maps show absolute rate differences for six pregnancy-related outcomes, 2020 to 2021.Abbreviation: COVID-19, coronavirus disease 2019. Absolute rate differences in pregnancy-related adverse outcomes among women of reproductive age across 204 countries and territories, 2020–2021. (A) Maternal sepsis and other maternal infections; (B) Maternal hemorrhage; (C) Maternal hypertensive disorders; (D) Obstructed labor and uterine rupture; (E) Abortion and miscarriage; (F) Ectopic pregnancy. Several particularly pronounced national deviations were evident. The United States showed the largest absolute excesses in maternal sepsis and other maternal infections (89.55 per 100,000 women of reproductive age) and obstructed labor and uterine rupture (166.42 per 100,000). Ghana showed an exceptionally large excess in maternal hypertensive disorders, with an absolute increase of 1561.54 per 100,000 women of reproductive age and a relative increase of 348.89%. This large observed–expected difference occurred against a distinct pre-pandemic trajectory: the ASIR of maternal hypertensive disorders declined rapidly during 2015–2019, whereas the observed ASIR increased during 2020–2021 relative to the continuing downward counterfactual trajectory. For abortion and miscarriage, Estonia and Russia showed the largest national absolute excesses, with increases of 1535.42 and 722.34 per 100,000 women of reproductive age, respectively, consistent with the pronounced regional deviation observed in Eastern Europe. Maternal hemorrhage and ectopic pregnancy also showed marked national heterogeneity, with particularly prominent deviations observed in Switzerland and Australia, respectively. Comprehensive country- and territory-specific estimates for all six outcomes are provided in Tables S9–S14. Sensitivity Analysis Sensitivity analyses supported the stability of the main findings (Figures S3–S5). The primary 9:1 training–validation split showed the lowest overall prediction error. Using RMSE rather than MAE as the primary model-selection criterion produced similar estimates for five of the six outcomes. Although the selected model for obstructed labor and uterine rupture changed under the alternative model-selection criterion, the direction and overall pattern of the estimated excess remained stable Estimates were also broadly consistent when models were refitted without exogenous covariates. Discussion This counterfactual modelling study provides a comprehensive global assessment of pandemic-period deviations in six pregnancy-related adverse outcomes among women of reproductive age relative to expected no-pandemic trajectories. During 2020–2021, observed incidence rates exceeded counterfactual expectations for five of the six outcomes, with substantial heterogeneity across age groups and geographical settings. Importantly, these observed–expected differences should be interpreted as departures from modelled no-pandemic trajectories rather than as causal effects attributable solely to the COVID-19 pandemic. The counterfactual estimates identify when, where, and among which age groups pregnancy-related outcomes deviated from pre-pandemic expectations, but they do not establish why these deviations occurred. Nevertheless, the geographically heterogeneous patterns highlight settings in which pregnancy-related health burden diverged most markedly from historical expectations and underscore the importance of maintaining essential maternal and reproductive health services during future large-scale health-system disruptions. Several pathways documented in previous studies may provide context for interpreting these pandemic-period deviations, although they cannot be identified as causal mechanisms in the present analysis. Two broad and potentially overlapping pathways are particularly relevant. First, SARS-CoV-2 infection during pregnancy has been associated with increased risks of severe maternal morbidity and selected obstetric complications.6,26 Second, the pandemic was accompanied by substantial changes in antenatal, obstetric, and reproductive health-service provision and utilization.9,27 These direct biological and indirect health-system pathways may have operated concurrently; however, our population-level counterfactual analysis did not include individual infection status, health-care utilization, or other relevant exposures and therefore cannot estimate their relative contributions to the observed–expected differences. Changes in fertility during 2020–2021 are also relevant to the interpretation of these findings. Because the ASIR denominator comprised all women aged 15–49 years rather than pregnancies or live births, the estimated excesses represent changes in population-level burden among women of reproductive age rather than changes in the probability of an adverse outcome conditional on pregnancy. Against the background of declining fertility during the pandemic period, these population-standardised rates should therefore not be interpreted as per-pregnancy risks.28 Age-specific analyses showed that excess abortion and miscarriage was more pronounced among women aged 15–19 years and 25–44 years, particularly those aged 25–34 years. These patterns may be considered in the context of age-related differences in reproductive circumstances and vulnerability to pandemic-related disruptions. Previous studies have reported that school closures and interruptions to sexual and reproductive health services were associated with increased unintended pregnancy among adolescents,29,30 while SARS-CoV-2 infection and reduced access to pregnancy-related care may also have affected pregnancy outcomes among women of reproductive age.27,31 However, these factors were not directly measured in our analysis, and the GBD category combines induced abortion and miscarriage, which may have different determinants. The observed age-specific heterogeneity should therefore not be interpreted as evidence for any specific causal mechanism and warrants further investigation using individual-level pregnancy data. High-income North America showed excess increases across multiple pregnancy-related adverse outcomes, with particularly pronounced deviations for maternal sepsis and other maternal infections, obstructed labor and uterine rupture, and abortion and miscarriage. Evidence from the USA provides relevant clinical context for a potential direct biological pathway. A multicenter study across 17 US hospitals found that SARS-CoV-2 infection during pregnancy was associated with increased serious maternal morbidity related to obstetric complications,26 while national delivery-hospitalization data also documented adverse maternal outcomes among women with COVID-19.32 At the same time, evidence from North America indicates that the pandemic substantially altered the delivery and utilization of prenatal care, although these changes were not uniform across populations or settings. In the USA, national data showed that changes in prenatal-care utilization differed according to Medicaid expansion status and race and ethnicity, suggesting heterogeneous effects on access to care.33 A population-based study from Ontario, Canada, similarly identified temporary disruptions in first-trimester prenatal visits and ultrasound use during the early pandemic, followed by recovery toward previous trends.34 These findings suggest that, even in high-resource settings, health-system resilience may depend not only on the overall availability of medical resources but also on maintaining equitable access and continuity of maternity care during major system-wide disruptions. Nevertheless, these external findings provide contextual evidence only; our population-level analysis cannot determine whether, or to what extent, direct infection-related risks or changes in health-care delivery accounted for the regional deviations observed in High-income North America. Ghana showed an exceptionally large observed–expected difference in maternal hypertensive disorders. This finding should first be interpreted in the context of the pre-pandemic trajectory used to construct the counterfactual. The ASIR declined particularly rapidly between 2015 and 2019, and the model consequently projected continuation of this downward trajectory into 2020–2021. Because the observed ASIR subsequently increased, a large deviation from the counterfactual trajectory resulted, and the magnitude of this excess is therefore partly sensitive to the assumption that the rapid pre-pandemic decline would otherwise have continued. At the same time, pandemic-related disruption of maternal health services may provide additional context for this reversal. National and facility-based studies from Ghana documented reductions in antenatal-care utilization during the pandemic, including substantial losses in ANC visits and changes in maternal-health service delivery.35 Because antenatal care is central to repeated blood-pressure measurement and screening for hypertensive disorders during pregnancy, reduced or disrupted access to such care could plausibly have affected the timely detection and management of hypertensive complications.36 However, these external data do not establish that health-service disruption caused the excess observed in our analysis. The Ghana finding should therefore be interpreted as a marked departure from the modelled pre-pandemic trajectory occurring in a context in which maternal-health service utilization was also disrupted. Eastern Europe showed the largest regional excess in the combined category of abortion and miscarriage, with particularly marked deviations in Estonia and Russia. Interpretation of this pattern requires caution because the GBD category combines induced abortion and miscarriage, which have distinct determinants and may respond differently to biological, behavioural, and health-system changes. Reproductive-health service responses during the pandemic also varied substantially across Europe. A multicountry assessment of abortion care found that some settings experienced reduced availability or additional barriers, whereas others introduced adaptations to maintain access; in Estonia, for example, fewer clinic visits, remote consultations, and home use of misoprostol were encouraged during the pandemic.37 Such evidence suggests that the pandemic altered the organization and delivery of reproductive health services, but not necessarily through a uniform reduction in access. These changes may provide relevant context for the regional heterogeneity observed in our analysis. However, they cannot establish the mechanisms underlying the specific deviations in Estonia or Russia, particularly because our aggregated data cannot distinguish whether the excess in the combined outcome reflected changes in induced abortion, miscarriage, or both. Further country-specific studies using individual-level pregnancy and health-service data are therefore needed to clarify the processes underlying these deviations. The standardised global coverage of GBD 2021 and the long pre-pandemic time series enabled consistent comparisons across countries, regions, and age groups and provided a robust historical basis for constructing counterfactual trajectories. In addition, the comparison of complementary statistical and machine-learning forecasting approaches, together with temporal validation and multiple sensitivity analyses, enhanced the robustness of model selection and the reliability of the estimated pandemic-period deviations. The multilevel assessment at global, regional, national, and age-specific levels further allowed heterogeneity in pregnancy-related health patterns to be characterized within a common analytical framework. This study has several limitations. First, pregnancy-related adverse outcomes were estimated at the population level among women of reproductive age rather than from individual-level pregnancy records; thus, the findings reflect population-level burden rather than per-pregnancy risks. Changes in fertility and pregnancy frequency during 2020–2021 may therefore have influenced these population-standardised rates. Second, the no-pandemic counterfactual was inherently unobservable and was constructed under the assumption that pre-pandemic temporal relationships, including the extrapolated trajectories of SDI and fertility rate, would have continued through 2020–2021 in the absence of COVID-19. Concurrent changes in socioeconomic conditions, fertility, health policies, health-care access, diagnostic practices, or surveillance could also have contributed to the observed–expected differences. Moreover, the aggregated GBD data did not include individual-level information on SARS-CoV-2 infection, health-care utilization, or other relevant exposures. Consequently, the observed deviations cannot be interpreted as causal effects of the pandemic or used to identify the mechanisms responsible for individual regional or national patterns. Temporal validation and sensitivity analyses assess the robustness of model prediction and specification but cannot validate the unobserved counterfactual or establish causal attribution. Third, variations in data quality and completeness across countries may have affected the estimated excess burden. Finally, the models lacked independent external validation, and their performance beyond the GBD framework remains uncertain. Nevertheless, the standardised global coverage and consistent counterfactual framework enable systematic comparison of pandemic-period deviations across populations and settings. Conclusions In conclusion, pregnancy-related adverse outcomes deviated from expected pre-pandemic trajectories during 2020–2021, with substantial heterogeneity across age groups and countries. These patterns underscore the need to safeguard continuity of essential obstetric and reproductive health services during infectious disease emergencies. Strengthening maternal health-system resilience, equity-oriented surveillance, and the capacity to maintain essential care during large-scale disruptions should be key priorities in future emergency preparedness.