Maternal Risk Prediction Models in Africa: A Scoping Review of Approaches, Variables, and Contextual Gaps

Maternal Risk Prediction Models in Africa: A Scoping Review of Approaches, Variables, and Contextual Gaps | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Systematic Review Maternal Risk Prediction Models in Africa: A Scoping Review of Approaches, Variables, and Contextual Gaps Eric Kwasi Elliason This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9408936/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Background Maternal mortality remains unacceptably high in many low- and middle-income countries, particularly in sub-Saharan Africa, despite global commitments to its reduction. In recent years, artificial intelligence and machine learning–based predictive models have been increasingly applied to maternal health, with the aim of identifying women at high risk of adverse outcomes. However, the scope, design, and contextual relevance of these models remain unclear. Objective This scoping review aimed to map existing evidence on maternal risk prediction models, examine the types of outcomes and predictors used, identify methodological trends and gaps, and propose a conceptual framework to guide the development of more context-sensitive predictive approaches. Methods A scoping review was conducted following established methodological guidance. Peer-reviewed studies and relevant reports focusing on maternal risk prediction, maternal mortality, severe maternal morbidity, and the application of artificial intelligence or machine learning in maternal health were included. Data were charted on study characteristics, data sources, predictors, modeling approaches, predicted outcomes, and performance measures. Findings were synthesized narratively. Results The review identified a growing body of literature applying machine learning techniques to predict maternal mortality, near-miss events, and pregnancy-related complications such as postpartum hemorrhage, preeclampsia, and sepsis. Most models relied predominantly on clinical and obstetric variables and were developed using retrospective, facility-based datasets. Social determinants of health and health system factors were inconsistently incorporated, and external validation across diverse contexts was limited. High-performing models often lacked interpretability, and evidence on real-world implementation was scarce. These gaps highlighted a disconnect between predictive accuracy and practical applicability in maternal health settings. Conclusion Existing maternal risk prediction models demonstrated technical promise but remained limited in scope, contextual sensitivity, and equity orientation. This review highlighted the need for integrated predictive frameworks that incorporate clinical, social, and health system determinants of maternal risk. The proposed conceptual framework provides a foundation for developing more context-aware and actionable predictive models to support timely intervention and reduce preventable maternal deaths. Maternal & Fetal Medicine Health Policy Health Economics and Outcomes Research Health Law Maternal mortality Maternal risk prediction Artificial intelligence Machine learning Scoping review Health systems Social determinants of health Figures Figure 1 Figure 2 1.0 Introduction Maternal mortality remains one of the most persistent and troubling public health challenges in sub-Saharan Africa. Despite decades of global commitment, including the Safe Motherhood Initiative and the Sustainable Development Goals, the region continues to account for nearly two thirds of global maternal deaths (World Health Organization [WHO], 2023). In several countries, progress has stalled or reversed, raising concerns about the effectiveness of current strategies for identifying and managing maternal risk. Traditionally, maternal mortality reduction efforts have focused on improving access to antenatal care, skilled birth attendance, and emergency obstetric services. While these interventions are essential, they often rely on late recognition of complications, when preventive opportunities have already been lost (Knight et al., 2019). Many maternal deaths occur not because risks were entirely absent, but because warning signs were either underestimated, poorly monitored over time, or disconnected from timely action within the health system (Thaddeus & Maine, 1994). Risk identification in maternal health has largely been guided by static clinical checklists and categorical classifications. Factors such as maternal age, parity, anemia, hypertension, and obstetric history are commonly used to label women as low or high risk during pregnancy. However, evidence increasingly suggests that these approaches have limited predictive value, particularly in low resource settings where social vulnerability and health system constraints play a decisive role in outcomes (Souza et al., 2014). Many women classified as low risk at antenatal visits still experience severe maternal morbidity or death, while others labeled high risk may never develop complications. In response to these limitations, predictive models and risk scoring tools have been developed to support earlier detection of maternal danger. These range from simple rule based algorithms to more recent machine learning approaches designed to estimate the probability of adverse maternal outcomes. However, most existing models have been developed using hospital based data from high income settings, focus narrowly on biomedical variables, and treat risk as a single time point event rather than a dynamic process that evolves across pregnancy, childbirth, and the postpartum period (Graham et al., 2016; Tunçalp et al., 2015). Their applicability to African contexts, where delays in care, social determinants, and facility readiness strongly shape outcomes, remains uncertain. Furthermore, maternal risk in African settings is deeply embedded within broader structural and contextual factors. Poverty, nutritional insecurity, limited autonomy in health decision making, geographic barriers to care, and fragmented referral systems interact with clinical conditions to amplify vulnerability (Kruk et al., 2018). Predictive approaches that do not account for these layered realities risk oversimplifying maternal danger and may fail to support meaningful prevention. Despite growing interest in predictive analytics for maternal health, there is limited clarity on what types of maternal risk prediction models currently exist, what variables they incorporate, what outcomes they aim to predict, and how well they reflect the African context. A comprehensive mapping of this evidence is needed to understand the conceptual assumptions, methodological approaches, and practical limitations of existing models. This scoping review therefore aims to systematically map maternal risk prediction models applied in African and comparable low- and middle-income settings. Specifically, it seeks to examine the types of predictive approaches used, the domains of risk variables included, the maternal outcomes predicted, and the extent to which social and health system factors are incorporated. By identifying gaps in current models, this review intends to inform the development of more context sensitive, dynamic, and actionable frameworks for predicting maternal risk and preventing avoidable maternal deaths. 2.0 Methods 2.1 Study Design This scoping review was conducted to map existing evidence on maternal risk prediction models, with particular attention to their conceptual foundations, methodological approaches, and relevance to African and other low- and middle-income settings. The review followed the methodological framework proposed by Arksey and O’Malley (2005), further refined by Levac et al. (2010). Reporting was guided by the Preferred Reporting Items for Systematic Reviews and Meta-Analyses Extension for Scoping Reviews (PRISMA-ScR) to ensure transparency and methodological rigor. 2.2 Review Questions The review was guided by the following questions: What types of maternal risk prediction models have been developed or applied in African and comparable low- and middle-income contexts? What maternal outcomes are these models designed to predict? What categories of risk variables are incorporated into existing models? To what extent do these models account for social, contextual, and health system factors? 2.3 Eligibility Criteria Eligibility criteria were defined using the Population–Concept–Context (PCC) framework. The population included pregnant women, women in labor, and postpartum women. Studies focusing exclusively on neonatal outcomes without maternal risk prediction were excluded. The concept of interest was maternal risk prediction, including statistical models, risk scoring tools, algorithms, and machine learning approaches developed to predict maternal mortality, severe maternal morbidity, or pregnancy-related complications. The context included African countries and other low- and middle-income settings. Studies conducted exclusively in high-income settings were excluded unless they explicitly addressed applicability to low-resource contexts. Only peer-reviewed articles were included. Commentaries, editorials, conference abstracts, and opinion pieces without empirical or methodological content were excluded. 2.4 Information Sources and Search Strategy A systematic search was conducted across multiple electronic databases, including PubMed, Scopus, Web of Science, and CINAHL. Additional records were identified through Google Scholar and manual screening of reference lists of included studies. The search strategy combined controlled vocabulary terms and free-text keywords related to maternal health, risk prediction, and modeling approaches. Key terms included “maternal mortality,” “maternal morbidity,” “risk prediction,” “prediction model,” “algorithm,” “machine learning,” and “artificial intelligence,” combined with geographic terms such as “Africa” and “low- and middle-income countries.” The final search was completed in [insert month and year], and all retrieved records were exported into reference management software for screening and deduplication. 2.5 Study Selection The study selection process followed PRISMA-ScR guidelines. A total of 186 records were identified through database searching and additional sources. After removal of 36 duplicates , 150 records remained for title and abstract screening. During the screening stage, 118 records were excluded based on lack of relevance to maternal risk prediction, absence of predictive modeling components, exclusive focus on neonatal outcomes, or non-empirical study design. A total of 32 full-text articles were assessed for eligibility. Of these, 22 studies were excluded for the following reasons: lack of a predictive modeling component ( n = 9 ), exclusive focus on neonatal outcomes ( n = 6 ), and limited relevance to maternal health decision-making in low- and middle-income contexts ( n = 7 ). A final total of 10 studies met the inclusion criteria and were included in the review. The study selection process is presented in Fig. 1 . 2.6 Data Charting A data charting form was developed and iteratively refined to extract key information from the included studies. Extracted data included author and year of publication, country or region of study, study design, type of predictive model, data sources, maternal outcomes predicted, categories of risk variables included, and reported limitations. Particular attention was given to whether studies incorporated non-clinical variables such as socioeconomic status, geographic access, and health system characteristics. 2.7 Data Synthesis Extracted data were synthesized descriptively and thematically. Studies were grouped according to the type of predictive approach used, including statistical models, machine learning methods, and hybrid or ensemble approaches. Patterns in outcome focus, variable selection, and contextual sensitivity were examined to identify key trends and gaps in the literature. Consistent with scoping review methodology, no formal assessment of study quality or risk of bias was conducted. 3.0 Results 3.1 Study Selection The database search and supplementary screening identified a total of 186 records, including studies retrieved from PubMed, Scopus, Web of Science, and CINAHL, as well as additional sources such as Google Scholar and reference list screening. After removal of 36 duplicates, 150 records remained for title and abstract screening. During the screening phase, 118 records were excluded due to lack of relevance to maternal risk prediction, absence of predictive modeling components, exclusive focus on neonatal outcomes, or non-empirical study designs. A total of 32 full-text articles were assessed for eligibility. Following full-text review, 22 articles were excluded for the following reasons: lack of a predictive modeling component (n = 9), exclusive focus on neonatal outcomes (n = 6), and limited relevance to maternal health decision-making in low- and middle-income contexts (n = 7). A final total of 10 studies met the inclusion criteria and were included in the scoping review. The study selection process followed PRISMA-ScR guidance and is illustrated in Fig. 1 . Flow diagram illustrating the process of study identification, screening, eligibility assessment, and inclusion for the scoping review, conducted in accordance with PRISMA-ScR guidelines. A total of 186 records were identified, 150 were screened after removal of duplicates, 32 full-text articles were assessed for eligibility, and 10 studies were included in the final review. 3.2 Characteristics of Included Studies The characteristics of the included studies are summarized in Table 1 . The studies were published between 2020 and 2025, reflecting increasing interest in predictive and artificial intelligence–based approaches to maternal health. The evidence base comprised both primary empirical studies and review-based studies, indicating that the field remains in a phase of methodological development and consolidation. Geographically, the included studies represented a mix of global analyses, low- and middle-income country–focused research, and context-specific applications. Some studies were conducted in African settings, including Kenya and Afghanistan (Dawodi et al., 2020 ; Shah et al., 2023 ), while others adopted global or multi-context perspectives through systematic or narrative reviews (Silva Rocha et al., 2022 ; Khan et al., 2022 ; Lin et al., 2024 ). In terms of study design, review-based studies were prominent, particularly those synthesizing evidence on artificial intelligence and machine learning applications in maternal health (Silva Rocha et al., 2022 ; Tzimourta et al., 2025 ; Lin et al., 2024 ). Empirical studies most commonly employed retrospective cohort designs using hospital records or electronic health data. Other studies conducted secondary analyses of existing maternal health datasets to develop and evaluate predictive models (Westcott et al., 2020 ; Togunwa et al., 2023 ; Nirmala & Kambili, 2023 ). Across studies, the primary populations of interest were pregnant women receiving antenatal or intrapartum care, with limited explicit attention to the postpartum period. Data sources were predominantly facility-based, including hospital records and electronic health systems, with fewer studies incorporating community-level data or longitudinal follow-up. Table 1 Characteristics of Included Studies on Maternal Risk Prediction and Artificial Intelligence Author(s), Year Country / Region Study Design Data Source Study Population Type of Predictive Model Key Risk Variable Domains Maternal Outcome Predicted Silva Rocha et al., 2022 Global Systematic review Published literature Pregnant and postpartum women Machine learning models (various) Clinical, demographic Maternal mortality during and post-pregnancy Tzimourta et al., 2025 Global Narrative review Published literature Pregnant women Artificial intelligence frameworks Clinical, physiological Maternal health risk Khan et al., 2022 LMICs Narrative review Published studies Pregnant women and neonates AI and ML approaches Clinical, contextual Maternal and neonatal outcomes Westcott et al., 2020 United States Retrospective cohort Electronic health records Women delivering in hospital Machine learning classifiers Clinical, obstetric Maternal hemorrhage Shah et al., 2023 Kenya Retrospective cohort Hospital records Women giving birth Machine learning algorithms Clinical, obstetric Postpartum hemorrhage Togunwa et al., 2023 Not specified (LMIC focus) Secondary data analysis Public maternal health datasets Pregnant women Hybrid ANN–Random Forest model Clinical, demographic Maternal health risk classification Nirmala & Kambili, 2023 India Cross-sectional analysis Maternal health dataset Pregnant women Ensemble machine learning model Clinical, demographic Maternal risk level Rabbi et al., 2023 Not specified Comparative modeling study Maternal health datasets Pregnant women Multiple ML classifiers Clinical, demographic Maternal health risk Dawodi et al., 2020 Afghanistan Case study Health system records Pregnant women Data mining and ML techniques Clinical, system-level Maternal mortality and morbidity Lin et al., 2024 Global Systematic review Published literature Pregnant women AI-augmented decision support systems 3.3 Types of Maternal Risk Prediction Models Identified Across the included studies, a range of predictive approaches was identified, reflecting both traditional statistical methods and more recent artificial intelligence–driven techniques. Overall, three broad categories of models emerged: conventional statistical models, machine learning–based models, and hybrid or ensemble approaches. Conventional statistical models were less prominent among the included studies. Where present, they were typically embedded within broader analytical frameworks or used as comparators to evaluate the performance of machine learning models, rather than functioning as standalone predictive tools. This reflects a broader shift in the field toward more complex computational approaches for maternal risk prediction. Machine learning–based models constituted the dominant category, particularly among empirical studies. These included supervised learning algorithms such as decision trees, support vector machines, artificial neural networks, and random forest classifiers. Several studies developed and evaluated machine learning models to predict specific maternal complications, including maternal hemorrhage and postpartum hemorrhage, using routinely collected clinical and obstetric data (Westcott et al., 2020 ; Shah et al., 2023 ). Other studies applied machine learning techniques to classify overall maternal risk levels, rather than predicting a single outcome, with the aim of supporting early identification of high-risk pregnancies (Nirmala & Kambili, 2023 ; Rabbi et al., 2023 ). A subset of studies explored hybrid and ensemble approaches, combining multiple algorithms to enhance predictive performance. These models often integrated neural networks with tree-based methods or employed ensemble techniques that aggregated outputs from multiple classifiers. Such approaches were generally reported to achieve improved accuracy and robustness compared to single-model approaches, particularly when applied to heterogeneous maternal health datasets (Togunwa et al., 2023 ; Nirmala & Kambili, 2023 ). However, these studies largely emphasized technical performance metrics, with limited attention to implementation within routine maternal healthcare settings. Review-based studies further highlighted the expanding diversity of artificial intelligence applications in maternal health prediction. Systematic and narrative reviews documented increasing use of machine learning and artificial intelligence frameworks to predict maternal mortality, severe maternal morbidity, and pregnancy-related complications across different contexts (Silva Rocha et al., 2022 ; Khan et al., 2022 ; Lin et al., 2024 ). These reviews also noted considerable variability in data quality, outcome definitions, and validation approaches across studies. Despite methodological differences, most models were oriented toward facility-based prediction, with intended users typically assumed to be clinicians or health system decision-makers. Few studies explicitly designed models for use at the community level or integrated predictive outputs into referral systems or care escalation pathways. 3.4 Maternal Outcomes Predicted by Existing Models The included studies varied in the maternal outcomes they aimed to predict, with a clear concentration on acute, clinically observable complications rather than broader or longitudinal maternal risk trajectories. The most frequently predicted outcomes were maternal mortality, severe maternal morbidity, and specific obstetric complications known to contribute substantially to maternal deaths. Several studies focused explicitly on predicting maternal mortality or mortality-related risk during pregnancy and, in some cases, the immediate postpartum period. Review-based analyses synthesized evidence on artificial intelligence models designed to predict maternal death using clinical and demographic variables, highlighting growing interest in mortality prediction as a key application of machine learning in maternal health (Silva Rocha et al., 2022 ; Khan et al., 2022 ). However, these studies also reported considerable variability in outcome definitions and prediction timeframes. A prominent subset of empirical studies concentrated on hemorrhage-related outcomes, particularly postpartum hemorrhage, a leading cause of maternal mortality globally. Machine learning models were developed to identify women at increased risk using antenatal and intrapartum clinical data, with varying levels of predictive performance reported across settings (Westcott et al., 2020 ; Shah et al., 2023 ). In most cases, hemorrhage was modeled as a discrete event rather than as part of a broader continuum of maternal risk. Other studies adopted a more aggregated approach by predicting overall maternal risk or classifying pregnancies into high- and low-risk categories. These models used composite indicators to support early identification of women requiring enhanced monitoring or referral (Nirmala & Kambili, 2023 ; Rabbi et al., 2023 ; Togunwa et al., 2023 ). While this approach reflects a more holistic conceptualization of risk, outcome definitions were often inconsistent across studies, limiting comparability. Across the included literature, relatively few models explicitly addressed outcomes beyond the immediate postpartum period, despite evidence that a substantial proportion of maternal deaths occur days or weeks after childbirth. Similarly, intermediate outcomes—such as delayed referral, progression from moderate to severe morbidity, or failure to receive timely emergency care—were rarely modeled. Review-based studies further emphasized that most predictive efforts focused on outcomes readily captured in facility-based datasets, potentially overlooking socially mediated and system-level pathways to maternal harm (Silva Rocha et al., 2022 ; Lin et al., 2024 ). As a result, maternal risk was most commonly operationalized as a discrete endpoint rather than as a dynamic process influenced by interacting factors over time. 3.5 Risk Variables Incorporated in Maternal Risk Prediction Models Across the included studies, the selection of risk variables showed a strong emphasis on clinical and obstetric factors, with comparatively limited attention to social, contextual, and health system determinants. Most predictive models relied on variables routinely collected within facility-based settings, reflecting both data availability and the dominant biomedical orientation of maternal care. Clinical and obstetric variables formed the foundation of nearly all models. Commonly included factors were maternal age, parity, gestational age, blood pressure, hemoglobin levels, history of obstetric complications, mode of delivery, and indicators of current pregnancy status. These variables were used individually or in combination to train machine learning models or inform composite risk scores (Westcott et al., 2020 ; Shah et al., 2023 ; Nirmala & Kambili, 2023 ; Togunwa et al., 2023 ). In some cases, laboratory and physiological parameters were incorporated, particularly in models predicting acute complications such as hemorrhage. Demographic variables were included less consistently. While maternal age was almost universally incorporated, other indicators such as education level and marital status appeared less frequently and were typically treated as secondary predictors rather than central determinants of risk (Rabbi et al., 2023 ; Khan et al., 2022 ). Socioeconomic factors were rarely integrated in a systematic manner. Few studies included direct measures of income, employment, or household living conditions, despite strong evidence linking socioeconomic disadvantage to adverse maternal outcomes. Where such factors were acknowledged, they were more often discussed as contextual influences rather than operationalized within predictive models (Silva Rocha et al., 2022 ; Lin et al., 2024 ). Similarly, health system and access-related variables were largely absent. Factors such as distance to health facilities, referral delays, availability of emergency obstetric care, staffing capacity, and facility readiness were seldom incorporated into predictive frameworks. Even in studies conducted in low-resource settings, models frequently assumed timely access to care and consistent clinical monitoring—assumptions that may not reflect real-world maternal care pathways (Dawodi et al., 2020 ; Khan et al., 2022 ). Review-based studies highlighted this imbalance, noting that most maternal risk prediction models conceptualize risk primarily at the individual biological level. Broader determinants related to social vulnerability and health system performance were recognized as important but remain underrepresented in existing modeling approaches (Silva Rocha et al., 2022 ; Lin et al., 2024 ; Tzimourta et al., 2025 ). 3.6 Consideration of Context and Health System Factors in Existing Models The extent to which existing predictive models incorporated contextual and health system factors was limited across the included studies. Although many models were developed using facility-based or national datasets, the broader contextual dimensions influencing maternal outcomes were rarely translated into explicit predictive variables. A small number of studies acknowledged health system characteristics such as the availability of skilled birth attendants, facility delivery rates, and healthcare workforce density. However, these factors were more often examined descriptively or at aggregate levels rather than being directly integrated into predictive algorithms (Rosser et al., 2022; Dawodi et al., 2020 ). Even when available, such indicators were typically treated as background characteristics rather than as dynamic contributors to maternal risk. Geographic and spatial context received similarly limited attention. While some studies utilized national or regional datasets, variables such as rural residence, distance to health facilities, and regional disparities in service provision were inconsistently incorporated. When included, these were often simplified into categorical variables, limiting their ability to capture gradients in access to care and service availability (Mboya et al., 2020 ; Mfateneza et al., 2022 ). Social and cultural context was largely absent from model design. Variables reflecting gender norms, decision-making autonomy, health-seeking behavior, and community-level support systems were rarely operationalized. Review-based studies highlighted that, although these factors are well-established determinants of maternal outcomes, they remain difficult to incorporate due to limitations in routinely collected data and increased model complexity (Silva Rocha et al., 2022 ; Lin et al., 2024 ). Several studies conducted in low- and middle-income settings recognized the importance of delays in care—particularly delays in recognizing complications, reaching appropriate facilities, and receiving adequate treatment. However, these delay-related factors were seldom measured directly or incorporated into predictive frameworks, resulting in models that prioritize clinical presentation over care pathways (Dawodi et al., 2020 ; Khan et al., 2022 ). Overall, while contextual and health system factors were frequently acknowledged as critical determinants of maternal outcomes, their integration into predictive models remains limited. This gap was consistently identified in review-based studies as a key limitation of existing maternal risk prediction approaches (Silva Rocha et al., 2022 ; Tzimourta et al., 2025 ). 3.7 Identified Gaps in Existing Maternal Risk Prediction Models Across the included studies, several recurring gaps were evident in the design, scope, and application of maternal risk prediction models. A major limitation is the continued reliance on clinical and biomedical variables. Most models prioritize physiological measurements, laboratory parameters, obstetric history, and pregnancy-related complications. While these variables provide important predictive value, they offer a partial representation of maternal risk. Broader determinants, including socioeconomic conditions and structural inequities, are largely excluded due to data limitations and prevailing modelling priorities (Silva Rocha et al., 2022 ; Ramakrishnan et al., 2021 ; Mennickent et al., 2023 ). Closely related to this is the limited integration of social determinants of health. Variables such as education, income, occupation, marital status, and living conditions are inconsistently incorporated, even when datasets permit their inclusion. Studies using population-based survey data were more likely to include these variables, yet they were often secondary to clinical predictors and demonstrated variable influence on model performance (Mfateneza et al., 2022 ; Mboya et al., 2020 ; Raja et al., 2022). This limits the relevance of current models for public health planning and equity-focused interventions (Lin et al., 2024 ; Khan et al., 2022 ). Data-related constraints represent another important gap. Many studies rely on retrospective datasets characterized by missing data, inconsistent variable definitions, and underreporting of maternal outcomes. Facility-based datasets further exclude women who do not access formal healthcare services, thereby limiting the generalizability of findings to broader populations (Dawodi et al., 2020 ; Shah et al., 2023 ). Review studies have identified this as a structural limitation in maternal prediction research, particularly in low-resource settings (Silva Rocha et al., 2022 ; Tzimourta et al., 2025 ). Model validation and external applicability remain limited. While several studies report strong internal performance, fewer conduct external validation using independent datasets or across different geographic contexts. Models developed in single facilities or specific regions are rarely tested in other settings, raising concerns about their transferability across health systems with varying resources and care practices (Westcott et al., 2020 ; Machoron et al., 2022 ; Togunwa et al., 2023 ). The interpretability of machine learning models also presents a challenge. High-performing models, particularly those based on ensemble and deep learning methods, often function as black-box systems. Although these models may achieve improved predictive accuracy, they provide limited insight into how individual variables contribute to risk predictions. This lack of transparency can hinder clinical adoption and policy use, particularly in maternal health contexts where explainability is essential for decision-making and trust (Uddin et al., 2019 ; Mienye & Sun, 2022 ; Tilala et al., 2024 ). Finally, ethical considerations remain underexplored. Issues such as data privacy, algorithmic bias, and the potential reinforcement of existing health inequities are rarely addressed in primary modelling studies. While some recent reviews have highlighted these concerns, they are not yet systematically incorporated into the design, evaluation, or implementation of maternal risk prediction models (Rodríguez-Torres et al., 2022 ; Silva et al., 2023; Ganesan & Somasiri, 2024 ). 3.8 Methodological Trends in Maternal Risk Prediction Research The included studies demonstrated substantial methodological diversity in study design, data sources, modeling approaches, and outcome definitions. Most primary studies employed quantitative, retrospective designs using routinely collected clinical, demographic, or registry-based datasets. Facility-based hospital records and national demographic and health surveys were the most commonly used data sources, particularly in studies conducted in low- and middle-income settings (Mboya et al., 2020 ; Mfateneza et al., 2022 ; Dawodi et al., 2020 ). Analytically, there has been a clear transition from traditional statistical methods to more advanced machine learning approaches. Earlier studies relied primarily on logistic regression and rule-based classification, whereas more recent research has adopted supervised machine learning algorithms such as decision trees, support vector machines, random forests, gradient boosting, and artificial neural networks (Westcott et al., 2020 ; Uddin et al., 2019 ; Raja et al., 2022). Several studies further employed ensemble learning techniques, combining multiple classifiers to enhance predictive performance and model stability (Togunwa et al., 2023 ; Nirmala & Kambili, 2023 ; Mienye & Sun, 2022 ). Feature selection approaches varied considerably. Some studies relied on predefined clinical variables guided by obstetric knowledge, while others applied automated techniques such as recursive feature elimination, information gain ranking, and correlation-based filtering. Clinical and demographic variables—particularly maternal age, parity, and gestational age—were consistently retained, whereas socioeconomic and contextual variables were less frequently included in final models (Machoron et al., 2022 ; Mboya et al., 2020 ; Raja et al., 2022). Model evaluation practices were inconsistently reported. Most studies assessed performance using metrics such as accuracy, sensitivity, specificity, and area under the receiver operating characteristic curve. Cross-validation techniques, including k-fold validation, were commonly used; however, external validation using independent datasets or multi-site data was relatively rare (Westcott et al., 2020 ; Shah et al., 2023 ; Silva Rocha et al., 2022 ). Reporting of calibration measures and clinical utility indicators remained limited. Although several studies reported the use of programming environments such as Python and R, and occasionally cloud-based platforms or publicly available datasets, there was limited evidence of integration of predictive models into routine clinical workflows. Overall, methodological advances have been driven more by computational innovation than by implementation considerations. 3.9 Predicted Outcomes and Performance Measures Across Studies The studies included in the review targeted a broad range of maternal and pregnancy-related outcomes. Maternal mortality and severe maternal morbidity were central outcomes, particularly in studies conducted in high-burden or resource-constrained settings. Other commonly predicted outcomes included postpartum hemorrhage, hypertensive disorders of pregnancy such as preeclampsia, sepsis, preterm birth, and composite maternal risk classifications (Westcott et al., 2020 ; Kopanitsa et al., 2021 ; Shah et al., 2023 ). Some studies focused on maternal near-miss events, using them as proxies for mortality in contexts where death records were incomplete or unreliable. These models aimed to identify women at risk of life-threatening complications during pregnancy, delivery, or the immediate postpartum period (Silva Rocha et al., 2022 ; Ferreira et al., 2023 ). Other studies extended their focus to perinatal and neonatal outcomes, with maternal characteristics serving as key predictors, reflecting the close interrelationship between maternal and infant health (Mboya et al., 2020 ; Mfateneza et al., 2022 ). Performance evaluation metrics varied across studies. Accuracy and area under the receiver operating characteristic curve were the most frequently reported measures, followed by sensitivity, specificity, precision, recall, and F1 score. Several studies reported high predictive accuracy, particularly when using ensemble or deep learning approaches; however, metric selection often depended on outcome prevalence and study objectives (Uddin et al., 2019 ; Togunwa et al., 2023 ; Nirmala & Kambili, 2023 ). Sensitivity was frequently prioritized in models predicting acute obstetric complications, given the clinical implications of missed cases. In contrast, studies focusing on broader risk classification placed greater emphasis on balanced accuracy and specificity (Shah et al., 2023 ; Raja et al., 2022). Few studies reported calibration metrics or decision-curve analyses, limiting the assessment of clinical usefulness beyond discrimination performance. Comparative analyses indicated that ensemble methods, including random forest and gradient boosting, often outperformed single-model approaches, particularly in datasets characterized by nonlinear relationships or class imbalance. However, most studies relied on internal validation methods, and external validation across different populations or healthcare settings remained uncommon. As a result, reported performance metrics largely reflect controlled research environments rather than real-world implementation. 3.10 Summary of Evidence Gaps Emerging from Included Studies Across the included literature, several consistent evidence gaps were identified in the application of predictive modeling to maternal health. A key gap relates to the limited scope of risk factors incorporated into existing models. Most studies prioritize clinical and obstetric variables, while broader determinants such as socioeconomic status, education, occupation, geographic access to care, and structural inequities are infrequently included or insufficiently operationalized (Mboya et al., 2020 ; Mfateneza et al., 2022 ; Khan et al., 2022 ). This constrains the ability of models to capture the full complexity of maternal risk. Data-related limitations also remain significant. Many studies rely on retrospective, facility-based datasets characterized by missing data, inconsistent variable definitions, and underreporting of maternal outcomes. These datasets exclude women who do not access formal healthcare services, thereby limiting generalizability, particularly in low-resource settings (Dawodi et al., 2020 ; Shah et al., 2023 ; Silva Rocha et al., 2022 ). Another important gap concerns model validation and transferability. While internal validation methods are widely used, few studies conduct external validation across different populations or health system contexts. Models are typically developed within single institutions or regions, limiting evidence on their robustness and applicability in diverse settings (Westcott et al., 2020 ; Machoron et al., 2022 ; Togunwa et al., 2023 ). Model interpretability also remains a challenge. High-performing models, particularly those based on ensemble and deep learning techniques, often lack transparency regarding how predictions are generated. Although some studies report feature importance, comprehensive explainability analyses are uncommon, limiting their utility for clinical decision-making (Uddin et al., 2019 ; Mienye & Sun, 2022 ; Lin et al., 2024 ). Ethical and governance considerations are insufficiently addressed. Issues such as data privacy, algorithmic bias, and the potential reinforcement of existing health inequities are rarely incorporated into model design or evaluation. These concerns are more frequently discussed in review literature than in empirical studies (Rodríguez-Torres et al., 2022 ; Tilala et al., 2024 ; Ganesan & Somasiri, 2024 ). Finally, there is limited evidence on real-world implementation. Most studies focus on model development and performance evaluation, with minimal attention to feasibility, acceptability among healthcare providers, or integration into health systems. This gap persists across both high-income and low-resource settings and represents a critical barrier to the translation of predictive models into practice (Rabbi et al., 2023 ; Khan et al., 2022 ; Tzimourta et al., 2025 ). Table 2 summarizes the key evidence gaps identified across the included studies. Table 2 Summary of Evidence Gaps Identified in Maternal Risk Prediction Studies Evidence Domain Description of Identified Gap Illustrative References Scope of predictors Most models relied heavily on clinical and obstetric variables, with limited inclusion of social, economic, and contextual determinants of maternal health Mboya et al., 2020 ; Mfateneza et al., 2022 ; Khan et al., 2022 Integration of social determinants Socioeconomic status, education, occupation, and access to care were inconsistently incorporated, even when relevant data were available Raja et al., 2022; Lin et al., 2024 ; Ramakrishnan et al., 2021 Data representativeness Predominant use of facility-based and retrospective datasets limited population-level generalizability, particularly in low-resource settings Dawodi et al., 2020 ; Shah et al., 2023 ; Silva Rocha et al., 2022 Outcome reporting Definitions of maternal outcomes varied widely, including mortality, near-miss events, and composite risk scores, limiting cross-study comparability Westcott et al., 2020 ; Kopanitsa et al., 2021 ; Tzimourta et al., 2025 Model validation External validation using independent datasets or different geographic contexts was rarely conducted Machoron et al., 2022 ; Togunwa et al., 2023 ; Shah et al., 2023 Model interpretability High-performing models often lacked transparency and explainability, limiting clinical trust and adoption Uddin et al., 2019 ; Mienye & Sun, 2022 ; Tilala et al., 2024 Ethical considerations Ethical issues such as algorithmic bias, data privacy, and equity impacts were infrequently addressed in empirical modeling studies Rodríguez-Torres et al., 2022 ; Ganesan & Somasiri, 2024 Implementation evidence Limited reporting on real-world deployment, clinical integration, feasibility, or acceptability of predictive models Rabbi et al., 2023 ; Khan et al., 2022 ; Tzimourta et al., 2025 4.0 Discussion 4.1 Interpreting the Current Landscape of Maternal Risk Prediction This scoping review mapped existing evidence on maternal risk prediction and the application of artificial intelligence in maternal health. The findings indicate a growing body of research focused on predictive modeling, particularly in low- and middle-income countries where maternal mortality remains high. However, the evidence also reveals that current approaches are fragmented in scope, limited in contextual sensitivity, and uneven in methodological rigor. Most predictive models are structured around clinical and obstetric indicators, reflecting the continued dominance of biomedical paradigms in maternal health research. Variables such as maternal age, parity, gestational age, hypertensive disorders, and hemorrhage-related indicators are consistently prioritized across studies. While these factors are clinically important, their predominance reflects a narrow conceptualization of maternal risk, largely confined to physiological events occurring within health facility settings (Westcott et al., 2020 ; Shah et al., 2023 ; Togunwa et al., 2023 ). A key limitation across the literature is the insufficient integration of social and structural determinants of health. Maternal mortality is widely recognized as a multidimensional phenomenon shaped by poverty, education, gender inequality, geographic access to care, and health system capacity. Despite this, such determinants are rarely incorporated into predictive models in a systematic manner. Even in studies using demographic or population-based data, social variables are often treated as secondary predictors or excluded during feature selection due to methodological constraints (Mboya et al., 2020 ; Mfateneza et al., 2022 ; Khan et al., 2022 ). Data sources further constrain the representativeness of existing models. The heavy reliance on facility-based datasets limits the ability to capture risk among women who experience barriers to accessing care or who deliver outside formal health systems. This limitation is particularly significant in low-resource settings, where a substantial proportion of maternal deaths occur outside tertiary facilities and may be underreported in routine data systems (Dawodi et al., 2020 ; Silva Rocha et al., 2022 ). From a methodological perspective, advances in machine learning and ensemble modeling have improved predictive performance in several studies. However, these gains are often accompanied by reduced interpretability and limited contextual relevance. Many high-performing models function as black-box systems, providing little insight into how predictions are generated or how they can inform timely clinical or public health decision-making. This lack of transparency presents a barrier to adoption in maternal health settings, where trust, accountability, and explainability are essential (Uddin et al., 2019 ; Mienye & Sun, 2022 ; Tilala et al., 2024 ). Taken together, these findings suggest that while maternal risk prediction models demonstrate considerable technical promise, they remain insufficiently aligned with the complex and context-dependent realities of maternal mortality. A clear disconnect persists between predictive performance and practical applicability, particularly in settings where maternal risk is shaped by the interaction of clinical, social, and health system factors. 4.2 Need for a Context-Sensitive Predictive Framework in Maternal Health The findings of this scoping review underscore the need for a more comprehensive and context-sensitive predictive framework for maternal risk assessment. While existing models demonstrate technical feasibility and, in some cases, strong predictive performance, they are often constrained by narrow variable selection, limited validation, and weak alignment with real-world maternal health contexts. A central limitation across the reviewed studies is the fragmentation of risk domains. Clinical, demographic, and health system factors are frequently examined in isolation rather than as interacting determinants of maternal outcomes. However, maternal mortality is a cumulative and dynamic process, shaped by intersecting risks that evolve across the continuum of care—from pre-pregnancy through pregnancy, delivery, and the postpartum period. The absence of integrated frameworks limits the ability of current models to capture these complex and time-dependent risk trajectories (Khan et al., 2022 ; Ramakrishnan et al., 2021 ; Silva Rocha et al., 2022 ). The review also highlights the disconnect between model development and health system realities. Many predictive tools are derived from retrospective datasets without consideration of how predictions will be generated, interpreted, or acted upon in routine clinical or community settings. Consequently, model outputs are rarely linked to actionable care pathways such as referral systems, targeted monitoring, or early intervention strategies. This limits their practical utility, particularly in settings where timely decision-making is critical to preventing maternal morbidity and mortality (Dawodi et al., 2020 ; Shah et al., 2023 ). Equity considerations further reinforce the need for a revised framework. Predictive models that rely predominantly on facility-based data risk systematically excluding vulnerable populations, including women with limited access to care or those residing in marginalized communities. Without deliberate integration of social and structural determinants, such models may inadvertently reinforce existing disparities in maternal health outcomes rather than mitigate them (Rodríguez-Torres et al., 2022 ; Tilala et al., 2024 ). Another important limitation is the restricted generalizability of existing models. Many are developed and validated within single institutions or specific geographic contexts, with limited testing across diverse health system environments. This raises concerns regarding their transferability to settings with different resource constraints, care pathways, and population characteristics. A context-sensitive framework must therefore prioritize adaptability, enabling models to function across varied data environments and healthcare systems (Machoron et al., 2022 ; Togunwa et al., 2023 ). Finally, the emphasis on predictive performance as the primary indicator of model success is insufficient for maternal health applications. While accuracy remains important, it must be balanced with interpretability, transparency, and usability. Predictive tools must generate outputs that can be understood and acted upon by clinicians, midwives, and public health practitioners. Without this, even highly accurate models may fail to influence practice or improve outcomes (Uddin et al., 2019 ; Mienye & Sun, 2022 ). Taken together, these findings highlight the need for a predictive framework that is integrative, context-aware, and implementation-oriented. Such a framework should move beyond isolated clinical indicators to incorporate social determinants and health system factors, while also ensuring that predictive outputs are actionable, equitable, and adaptable across diverse maternal health settings. 4.3 Conceptual Framework for Maternal Risk Prediction Building on the gaps identified in the reviewed literature, this study proposes a conceptual framework for maternal risk prediction that reflects the multidimensional and dynamic nature of maternal health outcomes. The framework is grounded in evidence demonstrating that maternal risk is not determined by isolated clinical events, but by interactions between individual, social, and health system factors across the continuum of care. At the individual level, the framework incorporates clinical and biological determinants that are consistently identified across existing predictive models. These include maternal age, parity, gravidity, gestational age, pre-existing medical conditions, and pregnancy-related complications such as hypertensive disorders and hemorrhage. These variables remain central to risk prediction and are often the strongest immediate predictors of adverse maternal outcomes (Westcott et al., 2020 ; Shah et al., 2023 ; Togunwa et al., 2023 ). Beyond clinical factors, the framework explicitly integrates social and demographic determinants of health, which are underrepresented in current predictive approaches. These include education, socioeconomic status, occupation, marital status, geographic location, and access to transportation and care. Evidence from population-based studies indicates that these factors influence maternal outcomes both directly and indirectly by shaping health-seeking behavior, timing of care, and continuity of service utilization (Mboya et al., 2020 ; Mfateneza et al., 2022 ; Khan et al., 2022 ). The framework further incorporates health system and care pathway factors, recognizing that maternal risk evolves within specific service delivery contexts. These include the availability of skilled birth attendants, facility readiness, referral systems, quality of antenatal and intrapartum care, and the timeliness of emergency obstetric services. Prior research highlights that failures in these domains—such as delays in recognition, referral, or treatment—contribute significantly to the progression from complication to mortality (Dawodi et al., 2020 ; Silva Rocha et al., 2022 ). A key feature of the proposed framework is its dynamic orientation. Maternal risk is conceptualized as evolving over time, shaped by interactions between clinical status, social context, and health system responsiveness at different stages of pregnancy and the postpartum period. This temporal perspective addresses a major limitation of existing models, which often rely on single time-point assessments and retrospective data snapshots (Ramakrishnan et al., 2021 ; Mennickent et al., 2023 ). Within this framework, predictive modeling is positioned as a decision-support tool embedded within care processes rather than as a standalone technical output. Risk predictions are intended to inform early identification, targeted surveillance, and timely intervention, while remaining interpretable and actionable for healthcare providers. This emphasis on usability and transparency responds directly to concerns in the literature regarding the adoption and ethical use of artificial intelligence in maternal health (Uddin et al., 2019 ; Rodríguez-Torres et al., 2022 ; Tilala et al., 2024 ). Overall, the proposed framework conceptualizes maternal risk as the product of interacting determinants operating across multiple levels and time points. By integrating clinical, social, and health system factors within a single structure, it addresses key limitations of existing predictive models and provides a foundation for the development of more context-sensitive, equitable, and implementable maternal risk prediction tools. As illustrated in Fig. 1 , the framework highlights the interaction between these domains across pregnancy and the postpartum period, emphasizing the need for predictive approaches that extend beyond facility-based, single-factor models toward more integrated and adaptive systems. The figure illustrates a multidimensional framework in which maternal risk is shaped by the interaction of clinical and biological factors, social and demographic determinants, and health system and care pathway factors. These domains operate dynamically across pregnancy and the postpartum period, collectively influencing maternal risk trajectories. The framework positions risk prediction as an integrated and evolving process intended to support early identification, timely intervention, and equitable maternal care. 4.4 Implications for Future Research and Practice The findings of this scoping review have important implications for both future research and maternal health practice. First, there is a clear need to move beyond narrowly defined clinical predictors toward more comprehensive models that systematically incorporate social determinants and health system factors. Future research should prioritize the integration of socioeconomic, geographic, and access-related variables with clinical data, particularly in high-burden settings where maternal risk is shaped by structural inequities. From a methodological perspective, there is a need for greater use of prospective and longitudinal study designs. Such approaches would allow maternal risk to be assessed dynamically across pregnancy and the postpartum period, rather than relying on single time-point predictions derived from retrospective datasets. Longitudinal modeling would enable better understanding of how risk evolves and interacts over time, thereby improving the predictive and practical value of these models. Strengthening model validation is also essential. Future studies should prioritize external validation using independent datasets and diverse geographic contexts to enhance generalizability. Multi-site and cross-country validation efforts are particularly important for ensuring that predictive models are applicable across varying health system environments, especially in low- and middle-income countries where resource constraints and care pathways differ significantly. For clinical and public health practice, the findings highlight the importance of developing predictive tools that are not only accurate but also interpretable and actionable. Models should be designed to support decision-making by clinicians, midwives, and community health workers, with clear pathways linking risk classification to intervention, referral, and monitoring strategies. Without integration into existing care workflows, predictive outputs are unlikely to translate into meaningful improvements in maternal outcomes. The review also underscores the importance of embedding ethical considerations into predictive model development and implementation. Issues related to data privacy, algorithmic bias, and equitable representation must be explicitly addressed to avoid reinforcing existing disparities in maternal health. Engaging stakeholders—including frontline healthcare providers and communities—in the design and deployment of predictive systems can further support trust, acceptability, and responsible use. Finally, advancing maternal risk prediction will require interdisciplinary collaboration. Effective model development and implementation depend on the combined expertise of public health researchers, clinicians, data scientists, and policymakers. Aligning predictive innovation with health system priorities and contextual realities will be critical for ensuring that these tools contribute to improved quality of care and reductions in preventable maternal mortality. Limitations of the Review This scoping review had several limitations that should be acknowledged. First, the review relied on published literature and publicly available datasets, which may have excluded relevant unpublished studies or ongoing projects related to maternal risk prediction. As a result, some emerging or locally implemented predictive approaches may not have been captured. Second, variability in study designs, outcome definitions, and modeling approaches limited direct comparability across included studies. Maternal outcomes ranged from mortality and near-miss events to composite risk classifications, and predictive performance metrics were inconsistently reported. These differences constrained the ability to synthesize findings quantitatively. Third, most included studies were based on retrospective and facility-derived data, reflecting limitations in the underlying evidence base rather than the review process itself. As such, the review may underrepresent maternal risk among women who did not access formal health services or who experienced barriers to care. Finally, although the review aimed to capture global evidence, the distribution of studies was uneven across regions. Some high-burden settings were underrepresented, potentially limiting the generalizability of the mapped evidence. Conclusion This scoping review examined existing evidence on maternal risk prediction and the application of artificial intelligence and machine learning in maternal health. The findings indicated increasing use of predictive models to identify maternal risk, particularly for acute obstetric complications and mortality-related outcomes. However, existing approaches were largely constrained by narrow variable selection, limited incorporation of social and health system determinants, and insufficient external validation. By synthesizing these gaps, the review highlighted the need for more integrated and context-sensitive predictive frameworks that reflect the multidimensional nature of maternal risk. The proposed conceptual framework provided a structured basis for future model development by linking clinical, social, and system-level factors across the continuum of care. Advancing maternal risk prediction will require methodological rigor, ethical sensitivity, and alignment with real-world care pathways. When grounded in context and equity, predictive models hold potential to support earlier identification of high-risk pregnancies, inform targeted interventions, and contribute to reducing preventable maternal deaths, particularly in high-burden settings. References Ajegbile ML (2023) Closing the gap in maternal health access and quality through targeted investments in low-resource settings. J Global Health Rep 7. https://doi.org/10.29392/001c.88917 Camargo BV, Justo AM (2013) IRaMuTeQ: Um software gratuito para análise de dados textuais. Temas em Psicologia 21(2):513–518 Dawodi M, Wada T, Baktash JA (2020) Applicability of ICT, data mining and machine learning to reduce maternal mortality and morbidity: Case study Afghanistan. Int Inform Inst (Tokyo) 23:33–45 Ferreira MES, Coutinho RZ, Queiroz BL (2023) Morbimortalidade materna no Brasil e a urgência de um sistema nacional de vigilância do near miss materno. Cadernos de saúde pública, 39(8), e00013923 Ganesan S, Somasiri N (2024) Navigating the integration of machine learning in health care: Challenges, strategies and ethical considerations. J Comput Cogn Eng 1–16. https://doi.org/10.47852/bonviewJCCE42023600 Gomez-Jemes L, Oprescu AM, Chimenea-Toscano A, Lutgardo G, Romero-Tornero M (2022) Machine learning to predict pre-eclampsia and uterine growth restriction in pregnant women. Electronics, 11 . https://doi.org/10.3390/electronics11193240 James G, Umoren I, Ekong A, Inyang S, Aloysius O (2024) Analysis of support vector machine and random forest models for classification of the impact of technostress in COVID and post-COVID era. J Nigerian Soc Phys Sci 6:2102. https://doi.org/10.46481/jnsps.2024.2102 Khan M, Khurshid M, Vatsa M, Singh R, Duggal M, Singh K (2022) On AI approaches for promoting maternal and neonatal health in low-resource settings: A review. Front Public Health 10:880034 Kopanitsa G, Metsker O, Paskochev D, Greschischeva S (2021) Identification of risk factors and prediction of sepsis in pregnancy using machine learning. Procedia Comput Sci 193:393–401. https://doi.org/10.1016/j.procs.2021.10.040 Lin X, Liang C, Liu J, Lyu T, Ghumman N, Campbell B (2024) Artificial intelligence–augmented clinical decision support systems for pregnancy care: A systematic review. J Med Internet Res 26:e54737 Machoron JJE, Inbaraj CN, Jeng XA (2022) A semi-supervised machine learning approach in predicting high-risk pregnancies in the Philippines. Diagnostics 12. https://doi.org/10.3390/diagnostics2112782 Mboya IB, Mahande MJ, Mohanad M, Obure J, Mwanbi HG (2020) Prediction of perinatal death using machine learning models: A birth registry-based cohort study in Northern Tanzania. BMJ Open. https://doi.org/10.1136/bmjopen-2020-040132 Mennickent D, Rodríguez A, Opazo MC et al (2023) Machine learning applied in maternal and fetal health: A narrative review focused on pregnancy diseases and complications. Front Endocrinol 14:1130139 Mfateneza E, Rutayisire PC, Biracyaza E, Musafiri S, Mpabuka WG (2022) Application of machine learning methods for predicting infant mortality in Rwanda. BMC Pregnancy and Childbirth, 22 . https://doi.org/10.1186/s12884-022-04699-8 Mienye DI, Sun Y (2022) A survey of ensemble learning: Concepts, algorithms, applications and prospects. IEEE Access 99129–99137. https://doi.org/10.1109/ACCESS.2022.3207287 Nirmala, Kambili RS (2023) Maternal risk level prediction using ensemble model. Int J Creative Res Thoughts 11:916–920 Rabbi F, Hossian NU, Sokolov A, Chidester J (2023) Predicting maternal health risk by utilizing different machine learning classifiers. Proceedings of the American Society of Engineering Management Annual Conference Ramakrishnan R, Rao S, He JR (2021) Perinatal health predictors using artificial intelligence: A review. Women’s Health 17. https://doi.org/10.1177/17455065211046132 Rodríguez-Torres H, González-González M, Sánchez-López J (2022) Artificial intelligence in medicine: Ethical and methodological challenges. Gac Med Mex 158(11):527–534 Shah SY, Saxena S, Rani SP et al (2023) Prediction of postpartum hemorrhage using machine learning algorithms in a Kenyan population. Front Global Women’s Health. https://doi.org/10.3389/fgwh.2023.1161157 Silva Rocha ED, de Morais Melo FL, de Mello MEF et al (2022) On usage of artificial intelligence for predicting mortality during and post-pregnancy: A systematic review. BMC Med Inf Decis Mak 22(1):334 Tilala MH, Chenchala PK, Choppadandi A et al (2024) Ethical considerations in the use of artificial intelligence and machine learning in health care. Cureus 16(6). https://doi.org/10.7759/cureus.62443 Togunwa TO, Babatunde AO, Abdullah KK (2023) Deep hybrid model for maternal health risk classification in pregnancy. Front Artif Intell 6:1213436. https://doi.org/10.3389/frai.2023.1213436 Tzimourta KD, Tsipouras MG, Angelidis P et al (2025) Maternal health risk detection: Advancing midwifery with artificial intelligence. Healthcare 13(7):833 Uddin S, Khan A, Hossain M, Moni M (2019) Comparing different supervised machine learning algorithms for disease prediction. BMC Medical Informatics and Decision Making, 19 . https://doi.org/10.1186/s12911-019-1004-8 Westcott JM, Hughes F, Liu W et al (2020) Prediction of maternal haemorrhage using machine learning. American Journal of Obstetrics and Gynecology, 222 . https://doi.org/10.1016/j.ajog.2019.11.653 World Health Organization (2017) Trends in maternal mortality 2000–2017. WHO World Health Organization (2023) Maternal health statistics . https://www.who.int Additional Declarations The authors declare no competing interests. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-9408936","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Systematic Review","associatedPublications":[],"authors":[{"id":622611284,"identity":"92471c34-c459-466f-ace7-abdfc6e5a3f3","order_by":0,"name":"Eric Kwasi Elliason","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA4klEQVRIie3RvQrCMBDA8QsH1yXU1aDWV4gUHOurKAUnB8FFcbBScPVtMlc6uBRnwcWPJ+ggKDiYKjo4tHUTzH84bsgPEgJgMv1iUTYkcEI23+uV22WJY1sYy4xQOQLgOhXqV7OtkNi7iJ3GQ6+3QN6enAdenQAPx20OEZsuthLpP8iuoXx9MXLdQQ6RCZAIZPQkQqEmnGoFxLq+yEioWSlCTBOXkPosVXExEQkL9cV8hxDjGlNr/UEFb7ETXKXBzePN5WqeXtW0U7HCwymPALDgvSJ/zNzjn/ryzWmTyWT6m+6uxD5PcrBDRAAAAABJRU5ErkJggg==","orcid":"https://orcid.org/0009-0006-3424-972X","institution":"Desh Bhagat University","correspondingAuthor":true,"prefix":"","firstName":"Eric","middleName":"Kwasi","lastName":"Elliason","suffix":""}],"badges":[],"createdAt":"2026-04-14 00:34:43","currentVersionCode":1,"declarations":{"humanSubjects":false,"vertebrateSubjects":false,"conflictsOfInterestStatement":false,"humanSubjectEthicalGuidelines":false,"humanSubjectConsent":false,"humanSubjectClinicalTrial":false,"humanSubjectCaseReport":false,"vertebrateSubjectEthicalGuidelines":false},"doi":"10.21203/rs.3.rs-9408936/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9408936/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":106935266,"identity":"cc34d9b0-28ad-4cd0-9ffb-23a45923506e","added_by":"auto","created_at":"2026-04-15 03:17:47","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":93172,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003e\u003cstrong\u003ePRISMA flow diagram of study selection.\u003c/strong\u003e\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eFlow diagram illustrating the process of study identification, screening, eligibility assessment, and inclusion for the scoping review, conducted in accordance with PRISMA-ScR guidelines. A total of 186 records were identified, 150 were screened after removal of duplicates, 32 full-text articles were assessed for eligibility, and 10 studies were included in the final review.\u003c/em\u003e\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-9408936/v1/4ab64b3ae8d596dac334c4d1.png"},{"id":106935265,"identity":"05277af4-1a31-4bbb-8305-b5c73e902702","added_by":"auto","created_at":"2026-04-15 03:17:47","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":641919,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFigure 1. Conceptual framework for maternal risk prediction.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eThe figure illustrates a multidimensional framework in which maternal risk is shaped by the interaction of clinical and biological factors, social and demographic determinants, and health system and care pathway factors. These domains operate dynamically across pregnancy and the postpartum period, collectively influencing maternal risk trajectories. The framework positions risk prediction as an integrated and evolving process intended to support early identification, timely intervention, and equitable maternal care.\u003c/em\u003e\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-9408936/v1/52335c6d12b71e6ec09ec18e.png"},{"id":106961388,"identity":"7fcfe25c-21b2-486d-a436-1f959eb3d795","added_by":"auto","created_at":"2026-04-15 09:25:24","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1768826,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9408936/v1/5dd25d03-1ad0-434a-8690-e6fcf152a2cb.pdf"}],"financialInterests":"The authors declare no competing interests.","formattedTitle":"\u003cp\u003e\u003cstrong\u003eMaternal Risk Prediction Models in Africa: A Scoping Review of Approaches, Variables, and Contextual Gaps\u003c/strong\u003e\u003c/p\u003e","fulltext":[{"header":"1.0 Introduction","content":"\u003cp\u003eMaternal mortality remains one of the most persistent and troubling public health challenges in sub-Saharan Africa. Despite decades of global commitment, including the Safe Motherhood Initiative and the Sustainable Development Goals, the region continues to account for nearly two thirds of global maternal deaths (World Health Organization [WHO], 2023). In several countries, progress has stalled or reversed, raising concerns about the effectiveness of current strategies for identifying and managing maternal risk.\u003c/p\u003e \u003cp\u003eTraditionally, maternal mortality reduction efforts have focused on improving access to antenatal care, skilled birth attendance, and emergency obstetric services. While these interventions are essential, they often rely on late recognition of complications, when preventive opportunities have already been lost (Knight et al., 2019). Many maternal deaths occur not because risks were entirely absent, but because warning signs were either underestimated, poorly monitored over time, or disconnected from timely action within the health system (Thaddeus \u0026amp; Maine, 1994).\u003c/p\u003e \u003cp\u003eRisk identification in maternal health has largely been guided by static clinical checklists and categorical classifications. Factors such as maternal age, parity, anemia, hypertension, and obstetric history are commonly used to label women as low or high risk during pregnancy. However, evidence increasingly suggests that these approaches have limited predictive value, particularly in low resource settings where social vulnerability and health system constraints play a decisive role in outcomes (Souza et al., 2014). Many women classified as low risk at antenatal visits still experience severe maternal morbidity or death, while others labeled high risk may never develop complications.\u003c/p\u003e \u003cp\u003eIn response to these limitations, predictive models and risk scoring tools have been developed to support earlier detection of maternal danger. These range from simple rule based algorithms to more recent machine learning approaches designed to estimate the probability of adverse maternal outcomes. However, most existing models have been developed using hospital based data from high income settings, focus narrowly on biomedical variables, and treat risk as a single time point event rather than a dynamic process that evolves across pregnancy, childbirth, and the postpartum period (Graham et al., 2016; Tun\u0026ccedil;alp et al., 2015). Their applicability to African contexts, where delays in care, social determinants, and facility readiness strongly shape outcomes, remains uncertain.\u003c/p\u003e \u003cp\u003eFurthermore, maternal risk in African settings is deeply embedded within broader structural and contextual factors. Poverty, nutritional insecurity, limited autonomy in health decision making, geographic barriers to care, and fragmented referral systems interact with clinical conditions to amplify vulnerability (Kruk et al., 2018). Predictive approaches that do not account for these layered realities risk oversimplifying maternal danger and may fail to support meaningful prevention.\u003c/p\u003e \u003cp\u003eDespite growing interest in predictive analytics for maternal health, there is limited clarity on what types of maternal risk prediction models currently exist, what variables they incorporate, what outcomes they aim to predict, and how well they reflect the African context. A comprehensive mapping of this evidence is needed to understand the conceptual assumptions, methodological approaches, and practical limitations of existing models.\u003c/p\u003e \u003cp\u003eThis scoping review therefore aims to systematically map maternal risk prediction models applied in African and comparable low- and middle-income settings. Specifically, it seeks to examine the types of predictive approaches used, the domains of risk variables included, the maternal outcomes predicted, and the extent to which social and health system factors are incorporated. By identifying gaps in current models, this review intends to inform the development of more context sensitive, dynamic, and actionable frameworks for predicting maternal risk and preventing avoidable maternal deaths.\u003c/p\u003e"},{"header":"2.0 Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1 Study Design\u003c/h2\u003e \u003cp\u003eThis scoping review was conducted to map existing evidence on maternal risk prediction models, with particular attention to their conceptual foundations, methodological approaches, and relevance to African and other low- and middle-income settings. The review followed the methodological framework proposed by Arksey and O\u0026rsquo;Malley (2005), further refined by Levac et al. (2010). Reporting was guided by the Preferred Reporting Items for Systematic Reviews and Meta-Analyses Extension for Scoping Reviews (PRISMA-ScR) to ensure transparency and methodological rigor.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e2.2 Review Questions\u003c/h2\u003e \u003cp\u003eThe review was guided by the following questions:\u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003eWhat types of maternal risk prediction models have been developed or applied in African and comparable low- and middle-income contexts?\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eWhat maternal outcomes are these models designed to predict?\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eWhat categories of risk variables are incorporated into existing models?\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eTo what extent do these models account for social, contextual, and health system factors?\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e2.3 Eligibility Criteria\u003c/h2\u003e \u003cp\u003eEligibility criteria were defined using the Population\u0026ndash;Concept\u0026ndash;Context (PCC) framework.\u003c/p\u003e \u003cp\u003eThe \u003cb\u003epopulation\u003c/b\u003e included pregnant women, women in labor, and postpartum women. Studies focusing exclusively on neonatal outcomes without maternal risk prediction were excluded.\u003c/p\u003e \u003cp\u003eThe \u003cb\u003econcept\u003c/b\u003e of interest was maternal risk prediction, including statistical models, risk scoring tools, algorithms, and machine learning approaches developed to predict maternal mortality, severe maternal morbidity, or pregnancy-related complications.\u003c/p\u003e \u003cp\u003eThe \u003cb\u003econtext\u003c/b\u003e included African countries and other low- and middle-income settings. Studies conducted exclusively in high-income settings were excluded unless they explicitly addressed applicability to low-resource contexts.\u003c/p\u003e \u003cp\u003eOnly peer-reviewed articles were included. Commentaries, editorials, conference abstracts, and opinion pieces without empirical or methodological content were excluded.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e2.4 Information Sources and Search Strategy\u003c/h2\u003e \u003cp\u003eA systematic search was conducted across multiple electronic databases, including PubMed, Scopus, Web of Science, and CINAHL. Additional records were identified through Google Scholar and manual screening of reference lists of included studies.\u003c/p\u003e \u003cp\u003eThe search strategy combined controlled vocabulary terms and free-text keywords related to maternal health, risk prediction, and modeling approaches. Key terms included \u0026ldquo;maternal mortality,\u0026rdquo; \u0026ldquo;maternal morbidity,\u0026rdquo; \u0026ldquo;risk prediction,\u0026rdquo; \u0026ldquo;prediction model,\u0026rdquo; \u0026ldquo;algorithm,\u0026rdquo; \u0026ldquo;machine learning,\u0026rdquo; and \u0026ldquo;artificial intelligence,\u0026rdquo; combined with geographic terms such as \u0026ldquo;Africa\u0026rdquo; and \u0026ldquo;low- and middle-income countries.\u0026rdquo;\u003c/p\u003e \u003cp\u003eThe final search was completed in [insert month and year], and all retrieved records were exported into reference management software for screening and deduplication.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003e2.5 Study Selection\u003c/h2\u003e \u003cp\u003eThe study selection process followed PRISMA-ScR guidelines. A total of \u003cb\u003e186 records\u003c/b\u003e were identified through database searching and additional sources. After removal of \u003cb\u003e36 duplicates\u003c/b\u003e, \u003cb\u003e150 records\u003c/b\u003e remained for title and abstract screening.\u003c/p\u003e \u003cp\u003eDuring the screening stage, \u003cb\u003e118 records were excluded\u003c/b\u003e based on lack of relevance to maternal risk prediction, absence of predictive modeling components, exclusive focus on neonatal outcomes, or non-empirical study design.\u003c/p\u003e \u003cp\u003eA total of \u003cb\u003e32 full-text articles\u003c/b\u003e were assessed for eligibility. Of these, \u003cb\u003e22 studies were excluded\u003c/b\u003e for the following reasons: lack of a predictive modeling component (\u003cb\u003en\u0026thinsp;=\u0026thinsp;9\u003c/b\u003e), exclusive focus on neonatal outcomes (\u003cb\u003en\u0026thinsp;=\u0026thinsp;6\u003c/b\u003e), and limited relevance to maternal health decision-making in low- and middle-income contexts (\u003cb\u003en\u0026thinsp;=\u0026thinsp;7\u003c/b\u003e).\u003c/p\u003e \u003cp\u003eA final total of \u003cb\u003e10 studies\u003c/b\u003e met the inclusion criteria and were included in the review. The study selection process is presented in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e2.6 Data Charting\u003c/h2\u003e \u003cp\u003eA data charting form was developed and iteratively refined to extract key information from the included studies. Extracted data included author and year of publication, country or region of study, study design, type of predictive model, data sources, maternal outcomes predicted, categories of risk variables included, and reported limitations.\u003c/p\u003e \u003cp\u003eParticular attention was given to whether studies incorporated non-clinical variables such as socioeconomic status, geographic access, and health system characteristics.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003e2.7 Data Synthesis\u003c/h2\u003e \u003cp\u003eExtracted data were synthesized descriptively and thematically. Studies were grouped according to the type of predictive approach used, including statistical models, machine learning methods, and hybrid or ensemble approaches.\u003c/p\u003e \u003cp\u003ePatterns in outcome focus, variable selection, and contextual sensitivity were examined to identify key trends and gaps in the literature. Consistent with scoping review methodology, no formal assessment of study quality or risk of bias was conducted.\u003c/p\u003e \u003c/div\u003e"},{"header":"3.0 Results","content":"\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Study Selection\u003c/h2\u003e \u003cp\u003eThe database search and supplementary screening identified a total of 186 records, including studies retrieved from PubMed, Scopus, Web of Science, and CINAHL, as well as additional sources such as Google Scholar and reference list screening. After removal of 36 duplicates, 150 records remained for title and abstract screening.\u003c/p\u003e \u003cp\u003eDuring the screening phase, 118 records were excluded due to lack of relevance to maternal risk prediction, absence of predictive modeling components, exclusive focus on neonatal outcomes, or non-empirical study designs. A total of 32 full-text articles were assessed for eligibility.\u003c/p\u003e \u003cp\u003eFollowing full-text review, 22 articles were excluded for the following reasons: lack of a predictive modeling component (n\u0026thinsp;=\u0026thinsp;9), exclusive focus on neonatal outcomes (n\u0026thinsp;=\u0026thinsp;6), and limited relevance to maternal health decision-making in low- and middle-income contexts (n\u0026thinsp;=\u0026thinsp;7).\u003c/p\u003e \u003cp\u003eA final total of 10 studies met the inclusion criteria and were included in the scoping review. The study selection process followed PRISMA-ScR guidance and is illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cem\u003eFlow diagram illustrating the process of study identification, screening, eligibility assessment, and inclusion for the scoping review, conducted in accordance with PRISMA-ScR guidelines. A total of 186 records were identified, 150 were screened after removal of duplicates, 32 full-text articles were assessed for eligibility, and 10 studies were included in the final review.\u003c/em\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003e3.2 Characteristics of Included Studies\u003c/h2\u003e \u003cp\u003eThe characteristics of the included studies are summarized in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. The studies were published between 2020 and 2025, reflecting increasing interest in predictive and artificial intelligence\u0026ndash;based approaches to maternal health. The evidence base comprised both primary empirical studies and review-based studies, indicating that the field remains in a phase of methodological development and consolidation.\u003c/p\u003e \u003cp\u003eGeographically, the included studies represented a mix of global analyses, low- and middle-income country\u0026ndash;focused research, and context-specific applications. Some studies were conducted in African settings, including Kenya and Afghanistan (Dawodi et al., \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Shah et al., \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), while others adopted global or multi-context perspectives through systematic or narrative reviews (Silva Rocha et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Khan et al., \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Lin et al., \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2024\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eIn terms of study design, review-based studies were prominent, particularly those synthesizing evidence on artificial intelligence and machine learning applications in maternal health (Silva Rocha et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Tzimourta et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Lin et al., \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Empirical studies most commonly employed retrospective cohort designs using hospital records or electronic health data. Other studies conducted secondary analyses of existing maternal health datasets to develop and evaluate predictive models (Westcott et al., \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Togunwa et al., \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Nirmala \u0026amp; Kambili, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eAcross studies, the primary populations of interest were pregnant women receiving antenatal or intrapartum care, with limited explicit attention to the postpartum period. Data sources were predominantly facility-based, including hospital records and electronic health systems, with fewer studies incorporating community-level data or longitudinal follow-up.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eCharacteristics of Included Studies on Maternal Risk Prediction and Artificial Intelligence\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"8\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAuthor(s), Year\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCountry / Region\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eStudy Design\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eData Source\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eStudy Population\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eType of Predictive Model\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eKey Risk Variable Domains\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003eMaternal Outcome Predicted\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSilva Rocha et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2022\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGlobal\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSystematic review\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003ePublished literature\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePregnant and postpartum women\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eMachine learning models (various)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eClinical, demographic\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eMaternal mortality during and post-pregnancy\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTzimourta et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2025\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGlobal\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNarrative review\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003ePublished literature\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePregnant women\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eArtificial intelligence frameworks\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eClinical, physiological\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eMaternal health risk\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eKhan et al., \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2022\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLMICs\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNarrative review\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003ePublished studies\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePregnant women and neonates\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eAI and ML approaches\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eClinical, contextual\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eMaternal and neonatal outcomes\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWestcott et al., \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2020\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eUnited States\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eRetrospective cohort\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eElectronic health records\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eWomen delivering in hospital\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eMachine learning classifiers\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eClinical, obstetric\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eMaternal hemorrhage\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eShah et al., \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2023\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eKenya\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eRetrospective cohort\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eHospital records\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eWomen giving birth\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eMachine learning algorithms\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eClinical, obstetric\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003ePostpartum hemorrhage\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTogunwa et al., \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2023\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNot specified (LMIC focus)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSecondary data analysis\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003ePublic maternal health datasets\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePregnant women\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eHybrid ANN\u0026ndash;Random Forest model\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eClinical, demographic\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eMaternal health risk classification\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNirmala \u0026amp; Kambili, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2023\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eIndia\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eCross-sectional analysis\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eMaternal health dataset\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePregnant women\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eEnsemble machine learning model\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eClinical, demographic\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eMaternal risk level\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRabbi et al., \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2023\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNot specified\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eComparative modeling study\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eMaternal health datasets\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePregnant women\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eMultiple ML classifiers\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eClinical, demographic\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eMaternal health risk\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDawodi et al., \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2020\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAfghanistan\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eCase study\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eHealth system records\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePregnant women\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eData mining and ML techniques\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eClinical, system-level\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eMaternal mortality and morbidity\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLin et al., \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2024\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGlobal\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSystematic review\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003ePublished literature\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePregnant women\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eAI-augmented decision support systems\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003e3.3 Types of Maternal Risk Prediction Models Identified\u003c/h2\u003e \u003cp\u003eAcross the included studies, a range of predictive approaches was identified, reflecting both traditional statistical methods and more recent artificial intelligence\u0026ndash;driven techniques. Overall, three broad categories of models emerged: conventional statistical models, machine learning\u0026ndash;based models, and hybrid or ensemble approaches.\u003c/p\u003e \u003cp\u003eConventional statistical models were less prominent among the included studies. Where present, they were typically embedded within broader analytical frameworks or used as comparators to evaluate the performance of machine learning models, rather than functioning as standalone predictive tools. This reflects a broader shift in the field toward more complex computational approaches for maternal risk prediction.\u003c/p\u003e \u003cp\u003eMachine learning\u0026ndash;based models constituted the dominant category, particularly among empirical studies. These included supervised learning algorithms such as decision trees, support vector machines, artificial neural networks, and random forest classifiers. Several studies developed and evaluated machine learning models to predict specific maternal complications, including maternal hemorrhage and postpartum hemorrhage, using routinely collected clinical and obstetric data (Westcott et al., \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Shah et al., \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Other studies applied machine learning techniques to classify overall maternal risk levels, rather than predicting a single outcome, with the aim of supporting early identification of high-risk pregnancies (Nirmala \u0026amp; Kambili, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Rabbi et al., \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eA subset of studies explored hybrid and ensemble approaches, combining multiple algorithms to enhance predictive performance. These models often integrated neural networks with tree-based methods or employed ensemble techniques that aggregated outputs from multiple classifiers. Such approaches were generally reported to achieve improved accuracy and robustness compared to single-model approaches, particularly when applied to heterogeneous maternal health datasets (Togunwa et al., \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Nirmala \u0026amp; Kambili, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). However, these studies largely emphasized technical performance metrics, with limited attention to implementation within routine maternal healthcare settings.\u003c/p\u003e \u003cp\u003eReview-based studies further highlighted the expanding diversity of artificial intelligence applications in maternal health prediction. Systematic and narrative reviews documented increasing use of machine learning and artificial intelligence frameworks to predict maternal mortality, severe maternal morbidity, and pregnancy-related complications across different contexts (Silva Rocha et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Khan et al., \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Lin et al., \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). These reviews also noted considerable variability in data quality, outcome definitions, and validation approaches across studies.\u003c/p\u003e \u003cp\u003eDespite methodological differences, most models were oriented toward facility-based prediction, with intended users typically assumed to be clinicians or health system decision-makers. Few studies explicitly designed models for use at the community level or integrated predictive outputs into referral systems or care escalation pathways.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003e3.4 Maternal Outcomes Predicted by Existing Models\u003c/h2\u003e \u003cp\u003eThe included studies varied in the maternal outcomes they aimed to predict, with a clear concentration on acute, clinically observable complications rather than broader or longitudinal maternal risk trajectories. The most frequently predicted outcomes were maternal mortality, severe maternal morbidity, and specific obstetric complications known to contribute substantially to maternal deaths.\u003c/p\u003e \u003cp\u003eSeveral studies focused explicitly on predicting maternal mortality or mortality-related risk during pregnancy and, in some cases, the immediate postpartum period. Review-based analyses synthesized evidence on artificial intelligence models designed to predict maternal death using clinical and demographic variables, highlighting growing interest in mortality prediction as a key application of machine learning in maternal health (Silva Rocha et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Khan et al., \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). However, these studies also reported considerable variability in outcome definitions and prediction timeframes.\u003c/p\u003e \u003cp\u003eA prominent subset of empirical studies concentrated on hemorrhage-related outcomes, particularly postpartum hemorrhage, a leading cause of maternal mortality globally. Machine learning models were developed to identify women at increased risk using antenatal and intrapartum clinical data, with varying levels of predictive performance reported across settings (Westcott et al., \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Shah et al., \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). In most cases, hemorrhage was modeled as a discrete event rather than as part of a broader continuum of maternal risk.\u003c/p\u003e \u003cp\u003eOther studies adopted a more aggregated approach by predicting overall maternal risk or classifying pregnancies into high- and low-risk categories. These models used composite indicators to support early identification of women requiring enhanced monitoring or referral (Nirmala \u0026amp; Kambili, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Rabbi et al., \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Togunwa et al., \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). While this approach reflects a more holistic conceptualization of risk, outcome definitions were often inconsistent across studies, limiting comparability.\u003c/p\u003e \u003cp\u003eAcross the included literature, relatively few models explicitly addressed outcomes beyond the immediate postpartum period, despite evidence that a substantial proportion of maternal deaths occur days or weeks after childbirth. Similarly, intermediate outcomes\u0026mdash;such as delayed referral, progression from moderate to severe morbidity, or failure to receive timely emergency care\u0026mdash;were rarely modeled.\u003c/p\u003e \u003cp\u003eReview-based studies further emphasized that most predictive efforts focused on outcomes readily captured in facility-based datasets, potentially overlooking socially mediated and system-level pathways to maternal harm (Silva Rocha et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Lin et al., \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). As a result, maternal risk was most commonly operationalized as a discrete endpoint rather than as a dynamic process influenced by interacting factors over time.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003e3.5 Risk Variables Incorporated in Maternal Risk Prediction Models\u003c/h2\u003e \u003cp\u003eAcross the included studies, the selection of risk variables showed a strong emphasis on clinical and obstetric factors, with comparatively limited attention to social, contextual, and health system determinants. Most predictive models relied on variables routinely collected within facility-based settings, reflecting both data availability and the dominant biomedical orientation of maternal care.\u003c/p\u003e \u003cp\u003eClinical and obstetric variables formed the foundation of nearly all models. Commonly included factors were maternal age, parity, gestational age, blood pressure, hemoglobin levels, history of obstetric complications, mode of delivery, and indicators of current pregnancy status. These variables were used individually or in combination to train machine learning models or inform composite risk scores (Westcott et al., \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Shah et al., \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Nirmala \u0026amp; Kambili, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Togunwa et al., \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). In some cases, laboratory and physiological parameters were incorporated, particularly in models predicting acute complications such as hemorrhage.\u003c/p\u003e \u003cp\u003eDemographic variables were included less consistently. While maternal age was almost universally incorporated, other indicators such as education level and marital status appeared less frequently and were typically treated as secondary predictors rather than central determinants of risk (Rabbi et al., \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Khan et al., \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2022\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eSocioeconomic factors were rarely integrated in a systematic manner. Few studies included direct measures of income, employment, or household living conditions, despite strong evidence linking socioeconomic disadvantage to adverse maternal outcomes. Where such factors were acknowledged, they were more often discussed as contextual influences rather than operationalized within predictive models (Silva Rocha et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Lin et al., \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2024\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eSimilarly, health system and access-related variables were largely absent. Factors such as distance to health facilities, referral delays, availability of emergency obstetric care, staffing capacity, and facility readiness were seldom incorporated into predictive frameworks. Even in studies conducted in low-resource settings, models frequently assumed timely access to care and consistent clinical monitoring\u0026mdash;assumptions that may not reflect real-world maternal care pathways (Dawodi et al., \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Khan et al., \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2022\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eReview-based studies highlighted this imbalance, noting that most maternal risk prediction models conceptualize risk primarily at the individual biological level. Broader determinants related to social vulnerability and health system performance were recognized as important but remain underrepresented in existing modeling approaches (Silva Rocha et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Lin et al., \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Tzimourta et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2025\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003e3.6 Consideration of Context and Health System Factors in Existing Models\u003c/h2\u003e \u003cp\u003eThe extent to which existing predictive models incorporated contextual and health system factors was limited across the included studies. Although many models were developed using facility-based or national datasets, the broader contextual dimensions influencing maternal outcomes were rarely translated into explicit predictive variables.\u003c/p\u003e \u003cp\u003eA small number of studies acknowledged health system characteristics such as the availability of skilled birth attendants, facility delivery rates, and healthcare workforce density. However, these factors were more often examined descriptively or at aggregate levels rather than being directly integrated into predictive algorithms (Rosser et al., 2022; Dawodi et al., \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). Even when available, such indicators were typically treated as background characteristics rather than as dynamic contributors to maternal risk.\u003c/p\u003e \u003cp\u003eGeographic and spatial context received similarly limited attention. While some studies utilized national or regional datasets, variables such as rural residence, distance to health facilities, and regional disparities in service provision were inconsistently incorporated. When included, these were often simplified into categorical variables, limiting their ability to capture gradients in access to care and service availability (Mboya et al., \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Mfateneza et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2022\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eSocial and cultural context was largely absent from model design. Variables reflecting gender norms, decision-making autonomy, health-seeking behavior, and community-level support systems were rarely operationalized. Review-based studies highlighted that, although these factors are well-established determinants of maternal outcomes, they remain difficult to incorporate due to limitations in routinely collected data and increased model complexity (Silva Rocha et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Lin et al., \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2024\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eSeveral studies conducted in low- and middle-income settings recognized the importance of delays in care\u0026mdash;particularly delays in recognizing complications, reaching appropriate facilities, and receiving adequate treatment. However, these delay-related factors were seldom measured directly or incorporated into predictive frameworks, resulting in models that prioritize clinical presentation over care pathways (Dawodi et al., \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Khan et al., \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2022\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eOverall, while contextual and health system factors were frequently acknowledged as critical determinants of maternal outcomes, their integration into predictive models remains limited. This gap was consistently identified in review-based studies as a key limitation of existing maternal risk prediction approaches (Silva Rocha et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Tzimourta et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2025\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003e3.7 Identified Gaps in Existing Maternal Risk Prediction Models\u003c/h2\u003e \u003cp\u003eAcross the included studies, several recurring gaps were evident in the design, scope, and application of maternal risk prediction models.\u003c/p\u003e \u003cp\u003eA major limitation is the continued reliance on clinical and biomedical variables. Most models prioritize physiological measurements, laboratory parameters, obstetric history, and pregnancy-related complications. While these variables provide important predictive value, they offer a partial representation of maternal risk. Broader determinants, including socioeconomic conditions and structural inequities, are largely excluded due to data limitations and prevailing modelling priorities (Silva Rocha et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Ramakrishnan et al., \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Mennickent et al., \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eClosely related to this is the limited integration of social determinants of health. Variables such as education, income, occupation, marital status, and living conditions are inconsistently incorporated, even when datasets permit their inclusion. Studies using population-based survey data were more likely to include these variables, yet they were often secondary to clinical predictors and demonstrated variable influence on model performance (Mfateneza et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Mboya et al., \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Raja et al., 2022). This limits the relevance of current models for public health planning and equity-focused interventions (Lin et al., \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Khan et al., \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2022\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eData-related constraints represent another important gap. Many studies rely on retrospective datasets characterized by missing data, inconsistent variable definitions, and underreporting of maternal outcomes. Facility-based datasets further exclude women who do not access formal healthcare services, thereby limiting the generalizability of findings to broader populations (Dawodi et al., \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Shah et al., \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Review studies have identified this as a structural limitation in maternal prediction research, particularly in low-resource settings (Silva Rocha et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Tzimourta et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2025\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eModel validation and external applicability remain limited. While several studies report strong internal performance, fewer conduct external validation using independent datasets or across different geographic contexts. Models developed in single facilities or specific regions are rarely tested in other settings, raising concerns about their transferability across health systems with varying resources and care practices (Westcott et al., \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Machoron et al., \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Togunwa et al., \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThe interpretability of machine learning models also presents a challenge. High-performing models, particularly those based on ensemble and deep learning methods, often function as black-box systems. Although these models may achieve improved predictive accuracy, they provide limited insight into how individual variables contribute to risk predictions. This lack of transparency can hinder clinical adoption and policy use, particularly in maternal health contexts where explainability is essential for decision-making and trust (Uddin et al., \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Mienye \u0026amp; Sun, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Tilala et al., \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2024\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eFinally, ethical considerations remain underexplored. Issues such as data privacy, algorithmic bias, and the potential reinforcement of existing health inequities are rarely addressed in primary modelling studies. While some recent reviews have highlighted these concerns, they are not yet systematically incorporated into the design, evaluation, or implementation of maternal risk prediction models (Rodr\u0026iacute;guez-Torres et al., \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Silva et al., 2023; Ganesan \u0026amp; Somasiri, \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2024\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec18\" class=\"Section2\"\u003e \u003ch2\u003e3.8 Methodological Trends in Maternal Risk Prediction Research\u003c/h2\u003e \u003cp\u003eThe included studies demonstrated substantial methodological diversity in study design, data sources, modeling approaches, and outcome definitions. Most primary studies employed quantitative, retrospective designs using routinely collected clinical, demographic, or registry-based datasets. Facility-based hospital records and national demographic and health surveys were the most commonly used data sources, particularly in studies conducted in low- and middle-income settings (Mboya et al., \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Mfateneza et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Dawodi et al., \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2020\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eAnalytically, there has been a clear transition from traditional statistical methods to more advanced machine learning approaches. Earlier studies relied primarily on logistic regression and rule-based classification, whereas more recent research has adopted supervised machine learning algorithms such as decision trees, support vector machines, random forests, gradient boosting, and artificial neural networks (Westcott et al., \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Uddin et al., \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Raja et al., 2022). Several studies further employed ensemble learning techniques, combining multiple classifiers to enhance predictive performance and model stability (Togunwa et al., \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Nirmala \u0026amp; Kambili, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Mienye \u0026amp; Sun, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2022\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eFeature selection approaches varied considerably. Some studies relied on predefined clinical variables guided by obstetric knowledge, while others applied automated techniques such as recursive feature elimination, information gain ranking, and correlation-based filtering. Clinical and demographic variables\u0026mdash;particularly maternal age, parity, and gestational age\u0026mdash;were consistently retained, whereas socioeconomic and contextual variables were less frequently included in final models (Machoron et al., \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Mboya et al., \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Raja et al., 2022).\u003c/p\u003e \u003cp\u003eModel evaluation practices were inconsistently reported. Most studies assessed performance using metrics such as accuracy, sensitivity, specificity, and area under the receiver operating characteristic curve. Cross-validation techniques, including k-fold validation, were commonly used; however, external validation using independent datasets or multi-site data was relatively rare (Westcott et al., \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Shah et al., \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Silva Rocha et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Reporting of calibration measures and clinical utility indicators remained limited.\u003c/p\u003e \u003cp\u003eAlthough several studies reported the use of programming environments such as Python and R, and occasionally cloud-based platforms or publicly available datasets, there was limited evidence of integration of predictive models into routine clinical workflows. Overall, methodological advances have been driven more by computational innovation than by implementation considerations.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec19\" class=\"Section2\"\u003e \u003ch2\u003e3.9 Predicted Outcomes and Performance Measures Across Studies\u003c/h2\u003e \u003cp\u003eThe studies included in the review targeted a broad range of maternal and pregnancy-related outcomes. Maternal mortality and severe maternal morbidity were central outcomes, particularly in studies conducted in high-burden or resource-constrained settings. Other commonly predicted outcomes included postpartum hemorrhage, hypertensive disorders of pregnancy such as preeclampsia, sepsis, preterm birth, and composite maternal risk classifications (Westcott et al., \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Kopanitsa et al., \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Shah et al., \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eSome studies focused on maternal near-miss events, using them as proxies for mortality in contexts where death records were incomplete or unreliable. These models aimed to identify women at risk of life-threatening complications during pregnancy, delivery, or the immediate postpartum period (Silva Rocha et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Ferreira et al., \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Other studies extended their focus to perinatal and neonatal outcomes, with maternal characteristics serving as key predictors, reflecting the close interrelationship between maternal and infant health (Mboya et al., \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Mfateneza et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2022\u003c/span\u003e).\u003c/p\u003e \u003cp\u003ePerformance evaluation metrics varied across studies. Accuracy and area under the receiver operating characteristic curve were the most frequently reported measures, followed by sensitivity, specificity, precision, recall, and F1 score. Several studies reported high predictive accuracy, particularly when using ensemble or deep learning approaches; however, metric selection often depended on outcome prevalence and study objectives (Uddin et al., \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Togunwa et al., \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Nirmala \u0026amp; Kambili, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eSensitivity was frequently prioritized in models predicting acute obstetric complications, given the clinical implications of missed cases. In contrast, studies focusing on broader risk classification placed greater emphasis on balanced accuracy and specificity (Shah et al., \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Raja et al., 2022). Few studies reported calibration metrics or decision-curve analyses, limiting the assessment of clinical usefulness beyond discrimination performance.\u003c/p\u003e \u003cp\u003eComparative analyses indicated that ensemble methods, including random forest and gradient boosting, often outperformed single-model approaches, particularly in datasets characterized by nonlinear relationships or class imbalance. However, most studies relied on internal validation methods, and external validation across different populations or healthcare settings remained uncommon. As a result, reported performance metrics largely reflect controlled research environments rather than real-world implementation.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec20\" class=\"Section2\"\u003e \u003ch2\u003e3.10 Summary of Evidence Gaps Emerging from Included Studies\u003c/h2\u003e \u003cp\u003eAcross the included literature, several consistent evidence gaps were identified in the application of predictive modeling to maternal health.\u003c/p\u003e \u003cp\u003eA key gap relates to the limited scope of risk factors incorporated into existing models. Most studies prioritize clinical and obstetric variables, while broader determinants such as socioeconomic status, education, occupation, geographic access to care, and structural inequities are infrequently included or insufficiently operationalized (Mboya et al., \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Mfateneza et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Khan et al., \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). This constrains the ability of models to capture the full complexity of maternal risk.\u003c/p\u003e \u003cp\u003eData-related limitations also remain significant. Many studies rely on retrospective, facility-based datasets characterized by missing data, inconsistent variable definitions, and underreporting of maternal outcomes. These datasets exclude women who do not access formal healthcare services, thereby limiting generalizability, particularly in low-resource settings (Dawodi et al., \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Shah et al., \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Silva Rocha et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2022\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eAnother important gap concerns model validation and transferability. While internal validation methods are widely used, few studies conduct external validation across different populations or health system contexts. Models are typically developed within single institutions or regions, limiting evidence on their robustness and applicability in diverse settings (Westcott et al., \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Machoron et al., \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Togunwa et al., \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eModel interpretability also remains a challenge. High-performing models, particularly those based on ensemble and deep learning techniques, often lack transparency regarding how predictions are generated. Although some studies report feature importance, comprehensive explainability analyses are uncommon, limiting their utility for clinical decision-making (Uddin et al., \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Mienye \u0026amp; Sun, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Lin et al., \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2024\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eEthical and governance considerations are insufficiently addressed. Issues such as data privacy, algorithmic bias, and the potential reinforcement of existing health inequities are rarely incorporated into model design or evaluation. These concerns are more frequently discussed in review literature than in empirical studies (Rodr\u0026iacute;guez-Torres et al., \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Tilala et al., \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Ganesan \u0026amp; Somasiri, \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2024\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eFinally, there is limited evidence on real-world implementation. Most studies focus on model development and performance evaluation, with minimal attention to feasibility, acceptability among healthcare providers, or integration into health systems. This gap persists across both high-income and low-resource settings and represents a critical barrier to the translation of predictive models into practice (Rabbi et al., \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Khan et al., \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Tzimourta et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e summarizes the key evidence gaps identified across the included studies.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eSummary of Evidence Gaps Identified in Maternal Risk Prediction Studies\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eEvidence Domain\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDescription of Identified Gap\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eIllustrative References\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eScope of predictors\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMost models relied heavily on clinical and obstetric variables, with limited inclusion of social, economic, and contextual determinants of maternal health\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMboya et al., \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Mfateneza et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Khan et al., \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2022\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eIntegration of social determinants\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSocioeconomic status, education, occupation, and access to care were inconsistently incorporated, even when relevant data were available\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eRaja et al., 2022; Lin et al., \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Ramakrishnan et al., \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2021\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eData representativeness\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePredominant use of facility-based and retrospective datasets limited population-level generalizability, particularly in low-resource settings\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eDawodi et al., \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Shah et al., \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Silva Rocha et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2022\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOutcome reporting\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDefinitions of maternal outcomes varied widely, including mortality, near-miss events, and composite risk scores, limiting cross-study comparability\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eWestcott et al., \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Kopanitsa et al., \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Tzimourta et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2025\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eModel validation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eExternal validation using independent datasets or different geographic contexts was rarely conducted\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMachoron et al., \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Togunwa et al., \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Shah et al., \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2023\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eModel interpretability\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHigh-performing models often lacked transparency and explainability, limiting clinical trust and adoption\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eUddin et al., \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Mienye \u0026amp; Sun, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Tilala et al., \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2024\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eEthical considerations\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eEthical issues such as algorithmic bias, data privacy, and equity impacts were infrequently addressed in empirical modeling studies\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eRodr\u0026iacute;guez-Torres et al., \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Ganesan \u0026amp; Somasiri, \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2024\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eImplementation evidence\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLimited reporting on real-world deployment, clinical integration, feasibility, or acceptability of predictive models\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eRabbi et al., \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Khan et al., \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Tzimourta et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2025\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"4.0 Discussion","content":"\u003cdiv id=\"Sec22\" class=\"Section2\"\u003e \u003ch2\u003e4.1 Interpreting the Current Landscape of Maternal Risk Prediction\u003c/h2\u003e \u003cp\u003eThis scoping review mapped existing evidence on maternal risk prediction and the application of artificial intelligence in maternal health. The findings indicate a growing body of research focused on predictive modeling, particularly in low- and middle-income countries where maternal mortality remains high. However, the evidence also reveals that current approaches are fragmented in scope, limited in contextual sensitivity, and uneven in methodological rigor.\u003c/p\u003e \u003cp\u003eMost predictive models are structured around clinical and obstetric indicators, reflecting the continued dominance of biomedical paradigms in maternal health research. Variables such as maternal age, parity, gestational age, hypertensive disorders, and hemorrhage-related indicators are consistently prioritized across studies. While these factors are clinically important, their predominance reflects a narrow conceptualization of maternal risk, largely confined to physiological events occurring within health facility settings (Westcott et al., \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Shah et al., \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Togunwa et al., \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eA key limitation across the literature is the insufficient integration of social and structural determinants of health. Maternal mortality is widely recognized as a multidimensional phenomenon shaped by poverty, education, gender inequality, geographic access to care, and health system capacity. Despite this, such determinants are rarely incorporated into predictive models in a systematic manner. Even in studies using demographic or population-based data, social variables are often treated as secondary predictors or excluded during feature selection due to methodological constraints (Mboya et al., \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Mfateneza et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Khan et al., \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2022\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eData sources further constrain the representativeness of existing models. The heavy reliance on facility-based datasets limits the ability to capture risk among women who experience barriers to accessing care or who deliver outside formal health systems. This limitation is particularly significant in low-resource settings, where a substantial proportion of maternal deaths occur outside tertiary facilities and may be underreported in routine data systems (Dawodi et al., \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Silva Rocha et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2022\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eFrom a methodological perspective, advances in machine learning and ensemble modeling have improved predictive performance in several studies. However, these gains are often accompanied by reduced interpretability and limited contextual relevance. Many high-performing models function as black-box systems, providing little insight into how predictions are generated or how they can inform timely clinical or public health decision-making. This lack of transparency presents a barrier to adoption in maternal health settings, where trust, accountability, and explainability are essential (Uddin et al., \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Mienye \u0026amp; Sun, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Tilala et al., \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2024\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eTaken together, these findings suggest that while maternal risk prediction models demonstrate considerable technical promise, they remain insufficiently aligned with the complex and context-dependent realities of maternal mortality. A clear disconnect persists between predictive performance and practical applicability, particularly in settings where maternal risk is shaped by the interaction of clinical, social, and health system factors.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec23\" class=\"Section2\"\u003e \u003ch2\u003e4.2 Need for a Context-Sensitive Predictive Framework in Maternal Health\u003c/h2\u003e \u003cp\u003eThe findings of this scoping review underscore the need for a more comprehensive and context-sensitive predictive framework for maternal risk assessment. While existing models demonstrate technical feasibility and, in some cases, strong predictive performance, they are often constrained by narrow variable selection, limited validation, and weak alignment with real-world maternal health contexts.\u003c/p\u003e \u003cp\u003eA central limitation across the reviewed studies is the fragmentation of risk domains. Clinical, demographic, and health system factors are frequently examined in isolation rather than as interacting determinants of maternal outcomes. However, maternal mortality is a cumulative and dynamic process, shaped by intersecting risks that evolve across the continuum of care\u0026mdash;from pre-pregnancy through pregnancy, delivery, and the postpartum period. The absence of integrated frameworks limits the ability of current models to capture these complex and time-dependent risk trajectories (Khan et al., \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Ramakrishnan et al., \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Silva Rocha et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2022\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThe review also highlights the disconnect between model development and health system realities. Many predictive tools are derived from retrospective datasets without consideration of how predictions will be generated, interpreted, or acted upon in routine clinical or community settings. Consequently, model outputs are rarely linked to actionable care pathways such as referral systems, targeted monitoring, or early intervention strategies. This limits their practical utility, particularly in settings where timely decision-making is critical to preventing maternal morbidity and mortality (Dawodi et al., \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Shah et al., \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eEquity considerations further reinforce the need for a revised framework. Predictive models that rely predominantly on facility-based data risk systematically excluding vulnerable populations, including women with limited access to care or those residing in marginalized communities. Without deliberate integration of social and structural determinants, such models may inadvertently reinforce existing disparities in maternal health outcomes rather than mitigate them (Rodr\u0026iacute;guez-Torres et al., \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Tilala et al., \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2024\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eAnother important limitation is the restricted generalizability of existing models. Many are developed and validated within single institutions or specific geographic contexts, with limited testing across diverse health system environments. This raises concerns regarding their transferability to settings with different resource constraints, care pathways, and population characteristics. A context-sensitive framework must therefore prioritize adaptability, enabling models to function across varied data environments and healthcare systems (Machoron et al., \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Togunwa et al., \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eFinally, the emphasis on predictive performance as the primary indicator of model success is insufficient for maternal health applications. While accuracy remains important, it must be balanced with interpretability, transparency, and usability. Predictive tools must generate outputs that can be understood and acted upon by clinicians, midwives, and public health practitioners. Without this, even highly accurate models may fail to influence practice or improve outcomes (Uddin et al., \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Mienye \u0026amp; Sun, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2022\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eTaken together, these findings highlight the need for a predictive framework that is integrative, context-aware, and implementation-oriented. Such a framework should move beyond isolated clinical indicators to incorporate social determinants and health system factors, while also ensuring that predictive outputs are actionable, equitable, and adaptable across diverse maternal health settings.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec24\" class=\"Section2\"\u003e \u003ch2\u003e4.3 Conceptual Framework for Maternal Risk Prediction\u003c/h2\u003e \u003cp\u003eBuilding on the gaps identified in the reviewed literature, this study proposes a conceptual framework for maternal risk prediction that reflects the multidimensional and dynamic nature of maternal health outcomes. The framework is grounded in evidence demonstrating that maternal risk is not determined by isolated clinical events, but by interactions between individual, social, and health system factors across the continuum of care.\u003c/p\u003e \u003cp\u003eAt the individual level, the framework incorporates clinical and biological determinants that are consistently identified across existing predictive models. These include maternal age, parity, gravidity, gestational age, pre-existing medical conditions, and pregnancy-related complications such as hypertensive disorders and hemorrhage. These variables remain central to risk prediction and are often the strongest immediate predictors of adverse maternal outcomes (Westcott et al., \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Shah et al., \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Togunwa et al., \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eBeyond clinical factors, the framework explicitly integrates social and demographic determinants of health, which are underrepresented in current predictive approaches. These include education, socioeconomic status, occupation, marital status, geographic location, and access to transportation and care. Evidence from population-based studies indicates that these factors influence maternal outcomes both directly and indirectly by shaping health-seeking behavior, timing of care, and continuity of service utilization (Mboya et al., \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Mfateneza et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Khan et al., \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2022\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThe framework further incorporates health system and care pathway factors, recognizing that maternal risk evolves within specific service delivery contexts. These include the availability of skilled birth attendants, facility readiness, referral systems, quality of antenatal and intrapartum care, and the timeliness of emergency obstetric services. Prior research highlights that failures in these domains\u0026mdash;such as delays in recognition, referral, or treatment\u0026mdash;contribute significantly to the progression from complication to mortality (Dawodi et al., \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Silva Rocha et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2022\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eA key feature of the proposed framework is its dynamic orientation. Maternal risk is conceptualized as evolving over time, shaped by interactions between clinical status, social context, and health system responsiveness at different stages of pregnancy and the postpartum period. This temporal perspective addresses a major limitation of existing models, which often rely on single time-point assessments and retrospective data snapshots (Ramakrishnan et al., \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Mennickent et al., \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eWithin this framework, predictive modeling is positioned as a decision-support tool embedded within care processes rather than as a standalone technical output. Risk predictions are intended to inform early identification, targeted surveillance, and timely intervention, while remaining interpretable and actionable for healthcare providers. This emphasis on usability and transparency responds directly to concerns in the literature regarding the adoption and ethical use of artificial intelligence in maternal health (Uddin et al., \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Rodr\u0026iacute;guez-Torres et al., \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Tilala et al., \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2024\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eOverall, the proposed framework conceptualizes maternal risk as the product of interacting determinants operating across multiple levels and time points. By integrating clinical, social, and health system factors within a single structure, it addresses key limitations of existing predictive models and provides a foundation for the development of more context-sensitive, equitable, and implementable maternal risk prediction tools.\u003c/p\u003e \u003cp\u003eAs illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e1\u003c/span\u003e, the framework highlights the interaction between these domains across pregnancy and the postpartum period, emphasizing the need for predictive approaches that extend beyond facility-based, single-factor models toward more integrated and adaptive systems.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cem\u003eThe figure illustrates a multidimensional framework in which maternal risk is shaped by the interaction of clinical and biological factors, social and demographic determinants, and health system and care pathway factors. These domains operate dynamically across pregnancy and the postpartum period, collectively influencing maternal risk trajectories. The framework positions risk prediction as an integrated and evolving process intended to support early identification, timely intervention, and equitable maternal care.\u003c/em\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec25\" class=\"Section2\"\u003e \u003ch2\u003e4.4 Implications for Future Research and Practice\u003c/h2\u003e \u003cp\u003eThe findings of this scoping review have important implications for both future research and maternal health practice. First, there is a clear need to move beyond narrowly defined clinical predictors toward more comprehensive models that systematically incorporate social determinants and health system factors. Future research should prioritize the integration of socioeconomic, geographic, and access-related variables with clinical data, particularly in high-burden settings where maternal risk is shaped by structural inequities.\u003c/p\u003e \u003cp\u003eFrom a methodological perspective, there is a need for greater use of prospective and longitudinal study designs. Such approaches would allow maternal risk to be assessed dynamically across pregnancy and the postpartum period, rather than relying on single time-point predictions derived from retrospective datasets. Longitudinal modeling would enable better understanding of how risk evolves and interacts over time, thereby improving the predictive and practical value of these models.\u003c/p\u003e \u003cp\u003eStrengthening model validation is also essential. Future studies should prioritize external validation using independent datasets and diverse geographic contexts to enhance generalizability. Multi-site and cross-country validation efforts are particularly important for ensuring that predictive models are applicable across varying health system environments, especially in low- and middle-income countries where resource constraints and care pathways differ significantly.\u003c/p\u003e \u003cp\u003eFor clinical and public health practice, the findings highlight the importance of developing predictive tools that are not only accurate but also interpretable and actionable. Models should be designed to support decision-making by clinicians, midwives, and community health workers, with clear pathways linking risk classification to intervention, referral, and monitoring strategies. Without integration into existing care workflows, predictive outputs are unlikely to translate into meaningful improvements in maternal outcomes.\u003c/p\u003e \u003cp\u003eThe review also underscores the importance of embedding ethical considerations into predictive model development and implementation. Issues related to data privacy, algorithmic bias, and equitable representation must be explicitly addressed to avoid reinforcing existing disparities in maternal health. Engaging stakeholders\u0026mdash;including frontline healthcare providers and communities\u0026mdash;in the design and deployment of predictive systems can further support trust, acceptability, and responsible use.\u003c/p\u003e \u003cp\u003eFinally, advancing maternal risk prediction will require interdisciplinary collaboration. Effective model development and implementation depend on the combined expertise of public health researchers, clinicians, data scientists, and policymakers. Aligning predictive innovation with health system priorities and contextual realities will be critical for ensuring that these tools contribute to improved quality of care and reductions in preventable maternal mortality.\u003c/p\u003e \u003cp\u003e \u003cb\u003eLimitations of the Review\u003c/b\u003e \u003c/p\u003e \u003cp\u003eThis scoping review had several limitations that should be acknowledged. First, the review relied on published literature and publicly available datasets, which may have excluded relevant unpublished studies or ongoing projects related to maternal risk prediction. As a result, some emerging or locally implemented predictive approaches may not have been captured.\u003c/p\u003e \u003cp\u003eSecond, variability in study designs, outcome definitions, and modeling approaches limited direct comparability across included studies. Maternal outcomes ranged from mortality and near-miss events to composite risk classifications, and predictive performance metrics were inconsistently reported. These differences constrained the ability to synthesize findings quantitatively.\u003c/p\u003e \u003cp\u003eThird, most included studies were based on retrospective and facility-derived data, reflecting limitations in the underlying evidence base rather than the review process itself. As such, the review may underrepresent maternal risk among women who did not access formal health services or who experienced barriers to care.\u003c/p\u003e \u003cp\u003eFinally, although the review aimed to capture global evidence, the distribution of studies was uneven across regions. Some high-burden settings were underrepresented, potentially limiting the generalizability of the mapped evidence.\u003c/p\u003e \u003c/div\u003e"},{"header":"Conclusion","content":"\u003cp\u003eThis scoping review examined existing evidence on maternal risk prediction and the application of artificial intelligence and machine learning in maternal health. The findings indicated increasing use of predictive models to identify maternal risk, particularly for acute obstetric complications and mortality-related outcomes. However, existing approaches were largely constrained by narrow variable selection, limited incorporation of social and health system determinants, and insufficient external validation.\u003c/p\u003e \u003cp\u003eBy synthesizing these gaps, the review highlighted the need for more integrated and context-sensitive predictive frameworks that reflect the multidimensional nature of maternal risk. The proposed conceptual framework provided a structured basis for future model development by linking clinical, social, and system-level factors across the continuum of care.\u003c/p\u003e \u003cp\u003eAdvancing maternal risk prediction will require methodological rigor, ethical sensitivity, and alignment with real-world care pathways. When grounded in context and equity, predictive models hold potential to support earlier identification of high-risk pregnancies, inform targeted interventions, and contribute to reducing preventable maternal deaths, particularly in high-burden settings.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eAjegbile ML (2023) Closing the gap in maternal health access and quality through targeted investments in low-resource settings. J Global Health Rep 7. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.29392/001c.88917\u003c/span\u003e\u003cspan address=\"10.29392/001c.88917\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCamargo BV, Justo AM (2013) IRaMuTeQ: Um software gratuito para an\u0026aacute;lise de dados textuais. Temas em Psicologia 21(2):513\u0026ndash;518\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDawodi M, Wada T, Baktash JA (2020) Applicability of ICT, data mining and machine learning to reduce maternal mortality and morbidity: Case study Afghanistan. Int Inform Inst (Tokyo) 23:33\u0026ndash;45\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFerreira MES, Coutinho RZ, Queiroz BL (2023) Morbimortalidade materna no Brasil e a urg\u0026ecirc;ncia de um sistema nacional de vigil\u0026acirc;ncia do near miss materno. Cadernos de sa\u0026uacute;de p\u0026uacute;blica, 39(8), e00013923\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGanesan S, Somasiri N (2024) Navigating the integration of machine learning in health care: Challenges, strategies and ethical considerations. J Comput Cogn Eng 1\u0026ndash;16. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.47852/bonviewJCCE42023600\u003c/span\u003e\u003cspan address=\"10.47852/bonviewJCCE42023600\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGomez-Jemes L, Oprescu AM, Chimenea-Toscano A, Lutgardo G, Romero-Tornero M (2022) Machine learning to predict pre-eclampsia and uterine growth restriction in pregnant women. \u003cem\u003eElectronics, 11\u003c/em\u003e. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/electronics11193240\u003c/span\u003e\u003cspan address=\"10.3390/electronics11193240\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJames G, Umoren I, Ekong A, Inyang S, Aloysius O (2024) Analysis of support vector machine and random forest models for classification of the impact of technostress in COVID and post-COVID era. J Nigerian Soc Phys Sci 6:2102. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.46481/jnsps.2024.2102\u003c/span\u003e\u003cspan address=\"10.46481/jnsps.2024.2102\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKhan M, Khurshid M, Vatsa M, Singh R, Duggal M, Singh K (2022) On AI approaches for promoting maternal and neonatal health in low-resource settings: A review. Front Public Health 10:880034\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKopanitsa G, Metsker O, Paskochev D, Greschischeva S (2021) Identification of risk factors and prediction of sepsis in pregnancy using machine learning. Procedia Comput Sci 193:393\u0026ndash;401. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.procs.2021.10.040\u003c/span\u003e\u003cspan address=\"10.1016/j.procs.2021.10.040\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLin X, Liang C, Liu J, Lyu T, Ghumman N, Campbell B (2024) Artificial intelligence\u0026ndash;augmented clinical decision support systems for pregnancy care: A systematic review. J Med Internet Res 26:e54737\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMachoron JJE, Inbaraj CN, Jeng XA (2022) A semi-supervised machine learning approach in predicting high-risk pregnancies in the Philippines. Diagnostics 12. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/diagnostics2112782\u003c/span\u003e\u003cspan address=\"10.3390/diagnostics2112782\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMboya IB, Mahande MJ, Mohanad M, Obure J, Mwanbi HG (2020) Prediction of perinatal death using machine learning models: A birth registry-based cohort study in Northern Tanzania. BMJ Open. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1136/bmjopen-2020-040132\u003c/span\u003e\u003cspan address=\"10.1136/bmjopen-2020-040132\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMennickent D, Rodr\u0026iacute;guez A, Opazo MC et al (2023) Machine learning applied in maternal and fetal health: A narrative review focused on pregnancy diseases and complications. Front Endocrinol 14:1130139\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMfateneza E, Rutayisire PC, Biracyaza E, Musafiri S, Mpabuka WG (2022) Application of machine learning methods for predicting infant mortality in Rwanda. \u003cem\u003eBMC Pregnancy and Childbirth, 22\u003c/em\u003e. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s12884-022-04699-8\u003c/span\u003e\u003cspan address=\"10.1186/s12884-022-04699-8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMienye DI, Sun Y (2022) A survey of ensemble learning: Concepts, algorithms, applications and prospects. IEEE Access 99129\u0026ndash;99137. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/ACCESS.2022.3207287\u003c/span\u003e\u003cspan address=\"10.1109/ACCESS.2022.3207287\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNirmala, Kambili RS (2023) Maternal risk level prediction using ensemble model. Int J Creative Res Thoughts 11:916\u0026ndash;920\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRabbi F, Hossian NU, Sokolov A, Chidester J (2023) Predicting maternal health risk by utilizing different machine learning classifiers. \u003cem\u003eProceedings of the American Society of Engineering Management Annual Conference\u003c/em\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRamakrishnan R, Rao S, He JR (2021) Perinatal health predictors using artificial intelligence: A review. Women\u0026rsquo;s Health 17. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1177/17455065211046132\u003c/span\u003e\u003cspan address=\"10.1177/17455065211046132\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRodr\u0026iacute;guez-Torres H, Gonz\u0026aacute;lez-Gonz\u0026aacute;lez M, S\u0026aacute;nchez-L\u0026oacute;pez J (2022) Artificial intelligence in medicine: Ethical and methodological challenges. Gac Med Mex 158(11):527\u0026ndash;534\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShah SY, Saxena S, Rani SP et al (2023) Prediction of postpartum hemorrhage using machine learning algorithms in a Kenyan population. Front Global Women\u0026rsquo;s Health. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3389/fgwh.2023.1161157\u003c/span\u003e\u003cspan address=\"10.3389/fgwh.2023.1161157\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSilva Rocha ED, de Morais Melo FL, de Mello MEF et al (2022) On usage of artificial intelligence for predicting mortality during and post-pregnancy: A systematic review. BMC Med Inf Decis Mak 22(1):334\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTilala MH, Chenchala PK, Choppadandi A et al (2024) Ethical considerations in the use of artificial intelligence and machine learning in health care. Cureus 16(6). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.7759/cureus.62443\u003c/span\u003e\u003cspan address=\"10.7759/cureus.62443\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTogunwa TO, Babatunde AO, Abdullah KK (2023) Deep hybrid model for maternal health risk classification in pregnancy. Front Artif Intell 6:1213436. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3389/frai.2023.1213436\u003c/span\u003e\u003cspan address=\"10.3389/frai.2023.1213436\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTzimourta KD, Tsipouras MG, Angelidis P et al (2025) Maternal health risk detection: Advancing midwifery with artificial intelligence. Healthcare 13(7):833\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eUddin S, Khan A, Hossain M, Moni M (2019) Comparing different supervised machine learning algorithms for disease prediction. \u003cem\u003eBMC Medical Informatics and Decision Making, 19\u003c/em\u003e. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s12911-019-1004-8\u003c/span\u003e\u003cspan address=\"10.1186/s12911-019-1004-8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWestcott JM, Hughes F, Liu W et al (2020) Prediction of maternal haemorrhage using machine learning. \u003cem\u003eAmerican Journal of Obstetrics and Gynecology, 222\u003c/em\u003e. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.ajog.2019.11.653\u003c/span\u003e\u003cspan address=\"10.1016/j.ajog.2019.11.653\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWorld Health Organization (2017) Trends in maternal mortality 2000\u0026ndash;2017. WHO\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWorld Health Organization (2023) \u003cem\u003eMaternal health statistics\u003c/em\u003e. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.who.int\u003c/span\u003e\u003cspan address=\"https://www.who.int\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Maternal mortality, Maternal risk prediction, Artificial intelligence, Machine learning, Scoping review, Health systems, Social determinants of health","lastPublishedDoi":"10.21203/rs.3.rs-9408936/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9408936/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e \u003cp\u003eMaternal mortality remains unacceptably high in many low- and middle-income countries, particularly in sub-Saharan Africa, despite global commitments to its reduction. In recent years, artificial intelligence and machine learning\u0026ndash;based predictive models have been increasingly applied to maternal health, with the aim of identifying women at high risk of adverse outcomes. However, the scope, design, and contextual relevance of these models remain unclear.\u003c/p\u003e\u003ch2\u003eObjective\u003c/h2\u003e \u003cp\u003eThis scoping review aimed to map existing evidence on maternal risk prediction models, examine the types of outcomes and predictors used, identify methodological trends and gaps, and propose a conceptual framework to guide the development of more context-sensitive predictive approaches.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eA scoping review was conducted following established methodological guidance. Peer-reviewed studies and relevant reports focusing on maternal risk prediction, maternal mortality, severe maternal morbidity, and the application of artificial intelligence or machine learning in maternal health were included. Data were charted on study characteristics, data sources, predictors, modeling approaches, predicted outcomes, and performance measures. Findings were synthesized narratively.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eThe review identified a growing body of literature applying machine learning techniques to predict maternal mortality, near-miss events, and pregnancy-related complications such as postpartum hemorrhage, preeclampsia, and sepsis. Most models relied predominantly on clinical and obstetric variables and were developed using retrospective, facility-based datasets. Social determinants of health and health system factors were inconsistently incorporated, and external validation across diverse contexts was limited. High-performing models often lacked interpretability, and evidence on real-world implementation was scarce. These gaps highlighted a disconnect between predictive accuracy and practical applicability in maternal health settings.\u003c/p\u003e\u003ch2\u003eConclusion\u003c/h2\u003e \u003cp\u003eExisting maternal risk prediction models demonstrated technical promise but remained limited in scope, contextual sensitivity, and equity orientation. This review highlighted the need for integrated predictive frameworks that incorporate clinical, social, and health system determinants of maternal risk. The proposed conceptual framework provides a foundation for developing more context-aware and actionable predictive models to support timely intervention and reduce preventable maternal deaths.\u003c/p\u003e","manuscriptTitle":"Maternal Risk Prediction Models in Africa: A Scoping Review of Approaches, Variables, and Contextual Gaps","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-04-15 03:17:43","doi":"10.21203/rs.3.rs-9408936/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"5c36d315-e90c-44af-a2a6-2c7aeadca987","owner":[],"postedDate":"April 15th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":66254293,"name":"Maternal \u0026 Fetal Medicine"},{"id":66254294,"name":"Health Policy"},{"id":66254295,"name":"Health Economics and Outcomes Research"},{"id":66254296,"name":"Health Law"}],"tags":[],"updatedAt":"2026-04-15T03:17:43+00:00","versionOfRecord":[],"versionCreatedAt":"2026-04-15 03:17:43","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9408936","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9408936","identity":"rs-9408936","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}