Reproductive performance and predictive models for litter size in sows: a narrative review

1 Introduction

Global pig production has increased steadily over recent decades due to rising demand for pork and animal protein (Zhang et al., 2023; Mateos et al., 2024). The sector supports the livelihoods of many households by meeting their nutritional, economic, and social needs. Alongside increasing input costs and market volatility, these pressures have intensified the need for improved efficiency in pig production. Poor reproductive performance, including reduction in total number born (TNB), lower live birth rates, or higher stillbirths, leads to increased replacement costs (Knox, 2025), additional labor requirements, and reduced economic returns. At the same time, pig production is undergoing rapid accelerated genetic improvement, particularly in highly prolific maternal lines such as Large White, Landrace, and Duroc (Collingbourne, 2019; Dos Santos et al., 2023; Du et al., 2025).

These highly prolific breeds have been selectively improved for enhanced fertility and maternal performance, enabling higher litter sizes and improved piglet survival (Knol et al., 2022). Selection for increased productivity has also introduced new biological trade-offs, including greater susceptibility to environmental stressors and health challenges (Tucker et al., 2021; Gombos et al., 2025). Climate stress, particularly heat stress, further compounds these issues by negatively affecting sow fertility, embryo survival, and lactation performance, thereby reducing reproductive efficiency (Lucy and Safranski, 2017; Ghulam Mohyuddin et al., 2022; Nkosi et al., 2026). Advances in precision livestock farming offer new opportunities to address these challenges. The integration of sensor technologies, automated monitoring systems, and real-time data analytics enables more precise management of sow health, behavior, and reproductive status. These technologies facilitate early detection of reproductive issues and support data-driven decision-making, improving both individual sow performance and overall herd productivity (Verhoeven et al., 2023; Mia et al., 2025). In parallel, the increasing availability of large-scale phenotypic and genomic data has accelerated the development of predictive models aimed at forecasting reproductive outcomes, particularly litter size (Gauthier et al., 2022; Su et al., 2025). Reproductive performance is influenced by complex interactions among genetic, physiological, environmental, and management factors (Kraeling and Webel, 2015; Tummaruk et al., 2023). Key performance indicators such as litter size (total born, born alive, and weaned), along with fertility and survival traits, remain central to evaluating productivity (Yu et al., 2022; Sanz-Fernández et al., 2024). Despite their importance, the genetic architecture underlying these traits is not yet fully understood, and predictive accuracy remains variable across models and production systems (Zhang et al., 2023). Furthermore, reproductive traits exhibit substantial variation across parities, reflecting underlying physiological changes associated with sexual maturity, ovulation rate, embryo survival, and uterine function. Studies have shown that litter size generally increases from early parities to peak reproductive age before declining in older sows, while piglet loss tends to increase at higher parities. The non-linear nature and low repeatability of reproductive traits make accurate prediction challenging and highlight the need for robust predictive modeling approaches that account for biological variation throughout a sow’s productive lifetime (Thiengpimol et al., 2024).

Despite the growing body of research on sow reproductive performance, important gaps remain in how key reproductive traits are evaluated and predicted under commercial production conditions. Many studies examine genetic, physiological, environmental, nutritional, and management factors in isolation, with limited integration of these determinants into comprehensive predictive frameworks. Moreover, existing predictive studies frequently focus on specific breeds, parity structures, or individual modeling approaches (Alsahaf et al., 2018; Thiengpimol et al., 2024; Chen et al., 2025a). Parity is known as a fundamental biological predictor rather than a descriptive production variable, as reproductive performance follows non-linear trajectories throughout a sow’s productive life. Litter size increases from early parities to peak production before declining in later parities, and polynomial and longitudinal models have demonstrated superior ability to characterize these parity-dependent reproductive patterns compared with conventional approaches (Thiengpimol et al., 2024). Likewise, reproductive performance in the first parity has been shown to predict subsequent farrowing performance and lifetime productivity (Iida and Koketsu, 2015). Beyond litter size, parity also improves the prediction of preweaning mortality using machine learning models, identifying parity among the most influential predictors alongside litter size, birth weight, gestation length, and sow health status (Rahman et al., 2023). Furthermore, deterministic modeling studies have demonstrated that incorporating parity distribution into herd-level prediction models enhances the estimation of replacement strategies, herd structure, productivity, and resource requirements under commercial production systems (Muniz et al., 2024). Although predictive models are increasingly applied to traits such as litter size and piglet survival, their practical applicability and consistency across production systems remain constrained because most models incorporate only selected biological variables or production scenarios, limiting their transferability across breeds, management systems, and environmental conditions.

Although several reviews have examined individual aspects of sow reproduction, including reproductive physiology, prolificacy, genetic improvement, genomic selection, reproductive management, and the application of artificial intelligence in livestock production, these topics are generally addressed independently. To date, there has been limited synthesis linking the biological determinants of reproductive performance with the statistical, genomic, and machine learning approaches used to predict litter size under commercial production conditions. As predictive analytics become increasingly integrated into precision livestock farming, a comprehensive evaluation that connects biological mechanisms with predictive methodologies is needed to improve model interpretability, identify biologically relevant predictors, and highlight knowledge gaps limiting their application across diverse production systems. This review addresses this gap by integrating current understanding of reproductive biology with contemporary predictive modeling approaches within a single framework.

This review aims to (i) synthesize the key reproductive performance metrics used to evaluate sow productivity; (ii) examine the major biological, environmental, and management-related factors influencing litter size and piglet survival; and (iii) evaluate commonly used statistical, genomic, and predictive modeling approaches for estimating reproductive performance, while identifying current limitations and future research priorities. Collectively, this review provides an integrated framework that links the biological basis of sow reproductive performance with contemporary predictive modeling approaches, with the aim of informing future research, precision breeding strategies, and evidence-based reproductive management. This manuscript aligns with the United Nations Sustainable Development Goals (SDGs) by promoting sustainable food production systems (SDG 2: Zero Hunger) and improving resource efficiency (SDG 12: Responsible Consumption and Production) in pig production systems. By optimizing sow reproductive performance, farmers can enhance food security while minimizing the environmental footprint (SDG 13: Climate Action) of intensive animal agriculture.

2 Materials and methods

This review was conducted using a narrative literature review approach to provide a comprehensive and critical synthesis of current evidence on sow reproductive performance and predictive modeling for litter size. A narrative review design was considered appropriate as the topic encompasses diverse research domains, including reproductive physiology, genetics, nutrition, environmental and management factors, precision livestock farming, statistical modeling, genomic prediction, and machine learning applications. The heterogeneity of study designs, analytical approaches, production systems, and outcome measures limits the suitability of quantitative evidence synthesis, while a narrative approach allows the integration of multidisciplinary evidence and facilitates the identification of emerging concepts, knowledge gaps, and future research directions. A structured literature search was conducted using Scopus, Web of Science, PubMed, ScienceDirect, and Google Scholar to identify relevant publications published between January 2018 and July 2026. Search strategies combined keywords and Boolean operators related to sow reproductive performance, including sow reproductive performance, litter size, born alive piglets, piglet survival, parity, fertility, genomic prediction, machine learning, statistical models, predictive modeling, and precision livestock farming. Additional relevant publications were identified through manual screening of reference lists from key articles to ensure comprehensive coverage of the available literature. The literature selection was guided by the objectives of this review. Studies were considered eligible if they examined (i) reproductive performance indicators used to evaluate sow productivity; (ii) biological, physiological, genetic, nutritional, environmental, or management factors influencing litter size and piglet survival; or (iii) statistical, genomic, artificial intelligence, or machine learning approaches for predicting reproductive performance. Priority was given to peer-reviewed articles published in English that presented original research, methodological developments, meta-analyses, or authoritative review papers with direct relevance to commercial pig production and reproductive prediction. Rather than applying a formal systematic screening protocol, publications were selected based on their scientific relevance, methodological rigor, and contribution to advancing current understanding of sow reproductive performance and predictive modeling. Editorials, conference abstracts, opinion articles, non-English publications, and studies that did not directly address the review objectives were excluded. The selected literature was critically appraised and synthesized to identify areas of consensus, conflicting evidence, methodological advances, and persistent knowledge gaps. Emphasis was placed on integrating biological determinants of reproductive performance with contemporary predictive modeling approaches, including conventional statistical models, genomic prediction methods, and machine learning techniques, to provide a comprehensive framework for understanding and improving reproductive efficiency in modern swine production systems.

3 Reproductive performance metrics

Reproductive performance in sows is evaluated using a combination of litter-level, efficiency, and lifetime productivity metrics, each capturing distinct aspects of reproductive biology and production efficiency (Sanz-Fernández et al., 2024; Thiengpimol et al., 2024). Due to the biological complexity of reproductive traits based on multiple interacting factors, no single metric can adequately capture overall sow productivity. Therefore, a combination of reproductive indicators is required to provide a comprehensive assessment of herd performance and economic efficiency (Tinh et al., 2021; Gustina et al., 2023). The importance of evaluating multiple reproductive indicators is further highlighted by the substantial variation observed across parities. Studies have shown that litter size generally increases from early parities to peak reproductive age before declining in older sows, whereas piglet loss and stillbirth rates tend to increase at later parities (Warda et al., 2021; Ju et al., 2022; Knap et al., 2023; Buthelezi et al., 2024a). These non-linear patterns indicate that reproductive performance should be interpreted within the context of sow age and parity rather than through single measurements alone (Thiengpimol et al., 2024).

Table 1 demonstrates that sow reproductive performance cannot be adequately evaluated using a single metric because each indicator reflects a different biological and reproductive component of the reproductive cycle (Sanz-Fernández et al., 2024; Thiengpimol et al., 2024). Litter-based traits such as total number born and number born alive (NBA) primarily describe reproductive potential at farrowing, whereas number weaned (NOW) and piglets per sow per year (PSY) more accurately reflect realized productivity and economic return through combined effects of fertility, maternal ability, and piglet survival (Yu et al., 2022; Sanz-Fernández et al., 2024). Fertility-related measures, including conception rate, farrowing rate, weaning-to-estrus interval (WEI), and farrowing interval, provide insights into reproductive efficiency and the extent of non-productive days within the herd (Gianluppi et al., 2020; Mellagi et al., 2023). Most traits are strongly influenced by interacting gene, physiological, nutritional, environmental, and management factors, which limit their interpretation when considered individually (Kraeling and Webel, 2015; Tummaruk et al., 2023). Furthermore, several reproductive traits are genetically correlated and may respond differently to selection. For example, selection for larger litter sizes can increase the number of piglets born but may also increase within-litter variation, reduce average birth weight, and increase pre-weaning mortality (Kemp et al., 2018; Harper and Bunter, 2024). Therefore, evaluating reproductive performance requires balancing productivity and piglet viability rather than maximizing a single trait. Consequently, the main message of this table is that reproductive performance should be assessed using an integrated and multitrait approach that combines fertility, litter size, piglet survival, and lifetime productivity indicators to obtain a more accurate understanding of sow efficiency and herd profitability (Koketsu et al., 2017; Sanz-Fernández et al., 2024). The multidimensional nature of reproductive performance presents a major challenge for prediction and genetic improvement. As individual traits are influenced by numerous interacting biological and environmental factors, modern breeding programs increasingly rely on integrated datasets and predictive modeling approaches to identify animals with superior reproductive potential and lifetime productivity.

Table 1. Key reproductive performance metrics in sows: biological meaning, economic relevance, and limitations.

Although the importance of reproductive performance metrics is widely recognized, the strength and direction of their associations with litter size are not always consistent across studies. These inconsistencies are largely attributable to differences in breed composition, parity distribution, herd health status, nutritional management, housing systems, climatic conditions, and production objectives among commercial and experimental populations. For example, the reproductive advantages associated with highly prolific genetic lines may be partially offset under conditions of heat stress, suboptimal nutrition, or increased disease pressure, where physiological demands exceed the sow’s capacity to maintain pregnancy and support fetal development (Lucy and Safranski, 2017; Tucker et al., 2021; Gombos et al., 2025). Parity-related responses may differ depending on gilt development, body condition, and management practices that influence endocrine function, ovulation rate, uterine capacity, and embryo survival (Kraeling and Webel, 2015; Tummaruk et al., 2023; Thiengpimol et al., 2024). These interacting biological mechanisms highlight that reproductive traits are not independent but are regulated through complex relationships among genetics, metabolism, hormonal regulation, and the uterine environment (Kraeling and Webel, 2015; Lucy and Safranski, 2017; Tummaruk et al., 2023). Variation among studies should be interpreted within the context of the production system rather than viewed as contradictory evidence, emphasizing the need for integrated evaluation of reproductive performance when developing management strategies and predictive models.

3.1 Determinants of reproductive performance in sows

While reproductive performance metrics provide essential tools for quantifying sow productivity, they do not explain the underlying causes of variation observed within and between herds (Koketsu et al., 2017). Reproductive outcomes such as litter size, fertility, and piglet survival are shaped by a complex interaction of genetic, physiological, environmental, and management-related factors (Kraeling and Webel, 2015; Tummaruk et al., 2023). Understanding these determinants is essential for identifying performance constraints and improving reproductive efficiency in modern pig production systems. Among these determinants, breed is one of the most influential factors. Breed, parity, nutrition, environmental conditions, and management practices are consistently identified as major determinants of reproductive performance, and their reported effects vary considerably among studies (Kraeling and Webel, 2015; Koketsu et al., 2017; Tummaruk et al., 2023). Such variability reflects differences in genetic background, production systems, herd health status, climatic conditions, nutritional strategies, and management intensity, as well as interactions among these factors rather than the influence of any single determinant. For example, genetic selection for increased prolificacy may improve ovulation rate and litter size, but its benefits can be constrained by limited uterine capacity, increased embryonic competition, or environmental stressors that compromise embryo survival and fetal development (Kemp et al., 2018; Lucy and Safranski, 2017; Harper and Bunter, 2024). Nutritional interventions and management practices may produce different reproductive responses depending on the physiological stage, body condition, and parity of the sow (Kraeling and Webel, 2015; Gianluppi et al., 2020; Tummaruk et al., 2023). These observations demonstrate that reproductive performance is regulated through interconnected genetic, endocrine, metabolic, and uterine mechanisms, highlighting the importance of interpreting individual determinants within the broader biological and production context rather than in isolation (Kraeling and Webel, 2015; Lucy and Safranski, 2017; Tummaruk et al., 2023; Harper and Bunter, 2024). It establishes the genetic potential upon which physiological, environmental, and management effects subsequently act.

3.1.1 Breed effects on reproductive performance

Genetic differences between pig breeds play a central role in shaping reproductive performance, with studies consistently demonstrating that breeds differ in reproductive traits due to their underlying genetic variation (Flowers, 2008; Rothschild and Ruvinsky, 2011; Harper and Bunter, 2024). Maternal breeds such as Landrace and Large White generally outperform terminal sire breeds such as Duroc, Hampshire, and Berkshire in litter traits, including TNB, NBA, and piglets weaned (Lee and Haley, 1995; Bell et al., 2015). Similarly, Chinese indigenous breeds such as Meishan exhibit markedly higher prolificacy and maternal ability compared with Western commercial breeds such as Large White, Landrace, and Duroc (Hall et al., 2002; Rothschild and Ruvinsky, 2011). These breed differences are associated with higher ovulation rates, improved uterine capacity, enhanced embryo survival, and favorable endocrine and maternal characteristics in highly prolific breeds (Bidanel, 2011; Bell et al., 2015). However, breed differences are not solely determined by prolificacy genes. Selection history and breeding objectives also contribute substantially to variation in reproductive performance. Maternal lines have traditionally been selected for fertility, litter size, and mothering ability, whereas terminal sire breeds have primarily been selected for growth rate, carcass composition, and feed efficiency. Consequently, differences in reproductive performance often reflect long-term breeding priorities rather than inherent biological superiority. Comparative performance data across breeds further highlight these differences, with highly prolific breeds showing larger litter sizes and improved piglet survival under comparable management conditions (Harper and Bunter, 2024).

While these differences are partly linked to genetic potential for prolificacy, realized reproductive performance is also strongly influenced by breed-specific functional and adaptive characteristics. Environmental robustness influences the ability of a sow to maintain fertility and litter performance under heat stress, disease exposure, and nutritional challenges (Rauw and Gomez-Raya, 2015). Landrace and Large White sows generally maintain more stable reproductive performance across production environments, whereas less adapted breeds may exhibit greater variability in fertility and piglet survival (Nowak et al., 2020). Highly prolific maternal breeds, particularly German and Danish Landrace, have also been intensively selected for reproductive efficiency, resulting in increased litter size and improved maternal traits (Tinh et al., 2021; Chen et al., 2025a). Breed differences in feed utilization during lactation may also influence piglet growth and survival, particularly in highly prolific sows (Arnaud et al., 2023). Calmer and more docile breeds are generally associated with improved nursing behavior, reduced piglet crushing, and better maternal care, all of which contribute to improved pre-weaning survival and effective litter size at weaning (Osotsi et al., 2025). Despite advances in genetic selection, increased prolificacy is often associated with biological trade-offs, including reduced piglet uniformity and increased pre-weaning mortality due to greater prenatal competition and elevated metabolic demands (Tucker et al., 2021). Therefore, reproductive performance reflects not only the biological capacity for prolificacy but also the interaction between breed adaptability, maternal functionality, environmental resilience, and management conditions.

Lactation feed efficiency is another important functional trait, as breeds that efficiently convert nutrients into milk production support improved piglet growth and survival during the suckling period, thereby increasing the number weaned (Arnaud et al., 2023). This becomes particularly important in highly prolific sows where nutrient demand during lactation is substantially elevated. Temperament and handling ease further influence maternal performance and piglet survival (Tokach et al., 2019).

3.1.2 Genetic architecture of reproductive traits

Genetic architecture plays a fundamental role in determining reproductive performance in sows (Yang et al., 2023; Zhang et al., 2023; Chen et al., 2025a). Reproductive traits are generally low to moderately heritable and polygenic, being controlled by many genes of small effect associated with follicular development, implantation, placental efficiency, and maternal behavior (Sun et al., 2023; Hong et al., 2024). Heritability estimates for litter-size traits are typically low, indicating a substantial influence of environmental and management factors. For example, heritability estimates range from 0.01 to 0.06 for litter-size traits in Large White pigs (Hong et al., 2025), while other researchers (Yang et al., 2023) reported estimates ranging from 0.005 to 0.159 for reproductive traits across Duroc, Landrace, and Yorkshire populations. Similarly, heritability of 0.066 for the total number born and 0.049 for the ratio of healthy births in Large White pigs, confirming the low genetic determinism of reproductive traits, was reported (Dong et al., 2020). Studies in commercial Yorkshire populations have also reported low heritability for number born alive (0.076), reinforcing the strong influence of environmental and management factors on reproductive performance (Yang et al., 2024).

The low heritability of reproductive traits partly explains the relatively slow rate of genetic improvement achieved through conventional phenotypic selection. Nevertheless, sufficient additive genetic variation exists to enable long-term genetic progress. Strong positive genetic correlations have been reported among TNB, NBA, number born healthy (NBH), and litter birth weight (LBW), suggesting opportunities for multitrait selection strategies that simultaneously improve litter size and piglet viability (Yang et al., 2023, Yang et al., 2024). However, selection solely for increased litter size may result in undesirable consequences, including greater litter variability, increased stillbirth rates, lower birth weights, and reduced piglet uniformity (Sell-Kubiak, 2021; Hong et al., 2025). These findings highlight the importance of balanced breeding objectives that optimize both productivity and animal welfare.

Consequently, modern breeding programs increasingly utilize genomic selection, genomic estimated breeding values (GEBVs), and best linear unbiased prediction (BLUP)-based evaluation systems to improve selection accuracy for reproductive traits. These approaches are particularly valuable for low-heritability and sex-limited traits such as litter size because they exploit information from thousands of genetic markers across the genome. Recent evidence demonstrated that single-step genomic BLUP (ssGBLUP) improved prediction accuracy for litter-size traits by 6.38%–13.33% compared with conventional BLUP methods in Large White pigs (Hong et al., 2025). Similarly, multitrait genomic evaluation models incorporating growth traits have improved prediction accuracy for number born alive, further demonstrating the value of integrating correlated traits into genetic evaluation programs (Yang et al., 2024).

Recent genomic studies have further identified fertility-related loci and candidate genes influencing litter size, prenatal survival, and reproductive efficiency in modern pig populations (Chen et al., 2025b). Genomic studies in Landrace populations have identified genetic markers associated with fertility, growth, and gestation characteristics (Nenova et al., 2023). Consequently, genetic improvement of reproductive traits is typically slower than improvement of growth or carcass traits, necessitating the use of advanced breeding approaches such as genomic selection, GEBVs, and BLUP-based evaluation systems (Bidanel, 2011; Sell-Kubiak et al., 2015; Song et al., 2017). In addition, genotype × environment interactions play a critical role in shaping reproductive outcomes. Genetic potential may not be fully expressed under suboptimal environmental or management conditions, highlighting the importance of aligning breeding strategies with production environments (Gombos et al., 2025). From a predictive modeling perspective, these interactions further demonstrate that reproductive performance cannot be accurately explained using genetic information alone. Integrating breed characteristics, genomic markers, environmental conditions, and management variables into predictive models is therefore essential for improving the accuracy and biological relevance of reproductive performance prediction in modern pig production systems (Gauthier et al., 2022; Su et al., 2025).

Despite these advances, several limitations continue to constrain the application of genomic prediction for reproductive traits in commercial pig breeding. Prediction accuracy is highly dependent on the size and genetic composition of the reference population, with models generally performing best in populations that are closely related to those used for model development. Consequently, genomic prediction equations developed for one breed or breeding population may not be directly transferable to genetically distinct populations because of differences in allele frequencies, linkage disequilibrium patterns, and selection histories (Meuwissen et al., 2001; Hayes et al., 2009; Daetwyler et al., 2013; Yang et al., 2024; Hong et al., 2025). Furthermore, genotype × environment interactions may alter the expression of reproductive traits across different production systems, climates, nutritional regimes, and management conditions, thereby reducing prediction accuracy when models are applied outside the environments in which they were developed (Falconer, 1952; Mulder and Bijma, 2005; Hayes et al., 2009; Tummaruk et al., 2023; Su et al., 2025). Although genomic information substantially improves selection accuracy over conventional pedigree-based approaches by capturing Mendelian sampling and identifying favorable chromosome segments at an early age, the greatest predictive performance is achieved when genomic data are integrated with phenotypic, environmental, and management information. This integrated approach improves the biological interpretation of reproductive traits and enhances the robustness and practical applicability of predictive models across diverse commercial production systems (Bidanel, 2011; Sell-Kubiak et al., 2015; Gauthier et al., 2022; Yang et al., 2024; Hong et al., 2025; Su et al., 2025).

Table 2 demonstrates that reproductive performance in pigs is regulated by a complex polygenic network involving genes associated with endocrine regulation, follicular development, embryo implantation, placental function, maternal behavior, and lactation. Genes such as ESR1, FSHR, LHCGR, and GNRHR are involved in the regulation of the hypothalamic–pituitary–gonadal axis and ovarian function and have been associated with variation in ovulation rate, conception success, and litter size across pig populations (Lidaka et al., 2021; Hristov and Stojanov, 2025). Genes including RBP4, TROAP, and IGF1 are more closely associated with embryonic development, implantation, placental efficiency, and prenatal survival, which directly influence the number born alive and piglet viability (Knap et al., 2023; Hernández-Montiel et al., 2026). Although these genes have been repeatedly associated with reproductive traits, their individual effects generally explain only a small proportion of the observed phenotypic variation. This reflects the highly polygenic nature of reproductive traits and highlights the limitations of relying solely on candidate gene approaches for genetic improvement. Maternal-function genes such as PRLR contribute to lactation performance, maternal behavior, and piglet survival, highlighting that reproductive efficiency extends beyond fertility alone. The table further illustrates that no single gene independently determines reproductive success; rather, reproductive traits result from the cumulative interaction of multiple genes of small effect together with environmental and management influences (Bidanel, 2011; Chen et al., 2022; Knap et al., 2023). Consequently, modern breeding programs increasingly rely on genomic selection, GWAS, and predictive modeling approaches to integrate these complex genetic interactions and improve reproductive performance more accurately in commercial pig populations (Zhang et al., 2023; Chen et al., 2025a).

Table 2. Genes and/or QTLs associated with reproductive performance in sows, their biological pathway, roles, and regulation.

Table 3 demonstrates that reproductive performance varies considerably among pig breed types due to differences in genetic potential, maternal ability, environmental adaptability, and physiological efficiency. Maternal breeds such as Large White and Landrace generally show superior fertility, litter size, and piglet survival because of intensive selection for reproductive efficiency and maternal traits (Lee and Haley, 1995; Tinh et al., 2021). Prolific breeds such as Meishan exhibit exceptionally high TNB and NBA, although their performance is more dependent on optimal management and nutrition (Hall et al., 2002; Bidanel, 2011). Terminal sire breeds are primarily selected for growth and carcass quality rather than reproductive performance, resulting in lower litter productivity and more variable maternal behavior (Rothschild and Ruvinsky, 2011). Indigenous breeds display lower reproductive output but greater environmental resilience, while crossbred sows benefit from heterosis, combining improved fertility, adaptability, and piglet survival across production systems (Harper and Bunter, 2024). It has been established that breed influences ovulation rate, litter size, piglet survival, maternal ability, and adaptation to environmental conditions. Therefore, breed information should be routinely incorporated as a predictor variable in statistical, genomic, and machine learning models of reproductive performance. Accounting for breed-specific differences improves the accuracy of litter size prediction and enhances the applicability of predictive models across diverse production systems. Conversely, failure to consider breed effects may reduce prediction accuracy and limit the transferability of predictive models between populations and management environments.

Table 3. General impact of breed characteristics on key reproductive performance metrics.

3.1.3 Physiological and parity-related dynamics

Reproductive performance varies significantly across the sow’s productive lifespan due to physiological and endocrine changes (Klimas et al., 2020; Warda et al., 2021; Gustina et al., 2023; Mabunda et al., 2025). Parity dynamics are important, as reproductive performance typically improves from first parity to mid-parity and declines in older sows (Gustina et al., 2023; Buthelezi et al., 2024a). These changes are largely mediated through the hypothalamic–pituitary–gonadal (HPG) axis, which regulates key reproductive processes via coordinated hormonal signaling. The hypothalamus secretes gonadotropin-releasing hormone (GnRH) in a pulsatile manner, stimulating the anterior pituitary to release follicle-stimulating hormone (FSH) and luteinizing hormone (LH) (Al-Suhaimi and Khan, 2022). FSH promotes follicular recruitment and growth, thereby determining the pool of ovulatory follicles, while LH triggers final follicular maturation and the pre-ovulatory LH surge required for ovulation (Duncan, 2019). Consequently, ovulation rate and ultimately potential litter size are directly dependent on the efficiency of GnRH–FSH–LH signaling, which varies with the physiological state of the sows.

Developing ovarian follicles produce estrogen, which plays a dual role by stimulating reproductive tract development and exerting positive feedback on the hypothalamus and pituitary to induce the LH surge (Ziecik et al., 2016; Li et al., 2024). Therefore, progesterone concentrations are closely associated with embryo survival and litter uniformity, while inadequate luteal function can result in early embryonic loss (Bazer and Johnson, 2014; Ziecik et al., 2016).

Parity-related differences in reproductive performance can thus be explained by variation in endocrine function. First-parity sows often exhibit reduced GnRH pulse frequency and lower LH secretion, resulting in smaller follicular pools, reduced ovulation rates, and consequently smaller litter sizes (Buthelezi et al., 2024a; Buthelezi et al., 2024b). In contrast, mid-parity sows exhibit optimal endocrine function, characterized by efficient GnRH pulsatility, pronounced LH surges, and stable progesterone secretion, resulting in higher ovulation rates, improved embryo survival, and increased litter size (Knox, 2015; Ziecik et al., 2016; White et al., 2023). In older sows, reduced ovarian responsiveness and altered progesterone dynamics compromise oocyte quality and embryo viability, leading to greater variability in litter outcomes (Knox, 2015; Koketsu et al., 2017).

Reproductive processes are also influenced by metabolic status, as body condition affects endocrine signaling. Both under-conditioning and over-conditioning can disrupt GnRH secretion and gonadotropin release, leading to impaired follicular development and reduced fertility (Muro et al., 2023). A critical biological trade-off exists between ovulation rate and embryo survival. Although increased ovulation rates can increase the number of fertilized embryos, the uterus can support only a limited number of conceptuses throughout gestation. Thus, competition for uterine space and placental resources may increase embryonic mortality and reduce piglet birth weight in highly prolific sows (Foxcroft et al., 2006; Chaiyapatmaetee et al., 2025). While an increased ovulation rate driven by enhanced gonadotropin activity can elevate potential litter size, it may exceed uterine capacity, resulting in increased embryonic mortality and reduced litter uniformity (Da Silva et al., 2016; Vernunft et al., 2018). Thus, reproductive efficiency reflects a balance between endocrine-driven ovulation and the physiological capacity to sustain developing embryos.

Table 4 highlights clear parity-related physiological patterns in sow reproductive performance, but several important nuances emerge. Primiparous sows (P1) generally produce smaller litters than multiparous sows but may achieve acceptable reproductive performance when managed appropriately (Rutherford et al., 2013; Nowland et al., 2019). Multiparous sows (P2–P5) represent the most biologically efficient stage, where improved coordination of the HPG axis supports higher ovulation rates, better embryo survival, and more consistent reproductive outcomes (Foxcroft et al., 2006; Soede et al., 2011b; Da Silva et al., 2016). Advanced parity sows (P6+) tend to show declining reproductive efficiency, driven by reduced oocyte quality, altered hormonal responsiveness, and cumulative physiological stress associated with aging (Knox, 2015; Ziecik et al., 2016). Parameters such as prolonged farrowing duration, extended birth intervals, and delayed placental expulsion in early and late parities reflect underlying endocrine and metabolic constraints that can negatively affect piglet survival (Nowak et al., 2020; Mota-Rojas et al., 2022). The table underscores that optimal reproductive performance is not solely determined by litter size but by the balance between endocrine function, physiological capacity, and maternal efficiency across parity stages, mainly in high-producing modern genotypes. Parity strongly influences ovulation rate, embryo survival, litter size, maternal performance, and piglet survival; therefore, it is routinely incorporated as a predictor variable in statistical, genomic, and machine learning models of reproductive performance. Accounting for parity-specific physiological differences improves prediction accuracy and enables more reliable estimation of litter size across a sow’s productive lifespan.

Table 4. Physiological and parity-related dynamics in sows.

3.1.4 Environmental stressors and climatic influences

Environmental conditions exert a strong influence on reproductive performance, with thermal stress being one of the most significant limiting factors in pig production systems (Ross et al., 2015). Environmental stressors affecting sow reproduction include heat stress, cold stress, humidity, photoperiod, ventilation quality, and stocking density, all of which can alter physiological and endocrine function and ultimately influence fertility, litter size, and piglet survival (Renaudeau et al., 2012; Tummaruk et al., 2023; Nkosi et al., 2026). Among these factors, heat stress has received the greatest attention because pigs possess limited thermoregulatory capacity due to their low sweat gland activity and high metabolic heat production (Ross et al., 2015; Liu et al., 2022). The importance of environmental stressors is expected to increase under climate change scenarios, as rising global temperatures and more frequent heat-wave events are projected to intensify thermal stress in pig production systems. Consequently, environmental adaptation has become an increasingly important component of reproductive management and breeding strategies (Ross et al., 2015; Gourdine et al., 2021).

Elevated ambient temperatures reduce feed intake, disrupt endocrine function, and impair oocyte quality, leading to reduced conception rates and smaller litter sizes (Lucy and Safranski, 2017; Ghulam Mohyuddin et al., 2022). Cold stress increases maintenance energy requirements, which can reduce body condition and impair reproductive recovery, especially during lactation and the post-weaning period (Prunier and Quesnel, 2000; Renaudeau et al., 2012; Pearce et al., 2013). Seasonal variation and photoperiod also affect reproductive efficiency, with declines often observed during warmer periods and changes in daylight length influencing melatonin secretion and reproductive hormone regulation. Changes in photoperiod can alter melatonin secretion and subsequently affect GnRH release, ovarian activity, and seasonal fertility, particularly in outdoor and semi-intensive production systems (Auvigne et al., 2010; Padet Tummaruk et al., 2023).

Humidity and ventilation further modify the impact of thermal stress. High relative humidity reduces the sow’s ability to dissipate heat through evaporative cooling, thereby intensifying heat stress and its negative effects on fertility and feed intake (Huynh et al., 2005; David Renaudeau et al., 2012). Poor ventilation and high stocking density can additionally increase respiratory stress, aggression, and physiological discomfort, contributing to reduced reproductive efficiency and piglet performance (Spoolder et al., 2009; Maes et al., 2011). Environmental stress not only impacts fertility but also influences piglet survival and growth, particularly in systems with inadequate climate control. These effects are further amplified in high-producing sows, which have greater metabolic demands and are more sensitive to stress. Primiparous and older sows are generally more susceptible to heat stress due to physiological immaturity or age-related declines in thermoregulatory capacity, whereas mid-parity sows tend to exhibit greater resilience and reproductive stability under environmental challenges.

Table 5 demonstrates that environmental stressors influence reproductive performance through both direct physiological effects and indirect endocrine disruption. Among these, heat stress emerges as the most critical factor, as it simultaneously reduces feed intake, alters GnRH and LH secretion, and compromises oocyte quality, thereby affecting both ovulation rate and conception success (Renaudeau et al., 2012; Lucy and Safranski, 2017; Gourdine et al., 2021). While other factors, such as photoperiod and humidity, exert more subtle effects, they interact synergistically with temperature to amplify reproductive inefficiencies (Gourdine et al., 2021; Liu et al., 2022; Knox, 2024). Management-related stressors, including poor ventilation and high stocking density, further exacerbate these effects by increasing physiological stress and disrupting hormonal balance (Brown-Brandl et al., 2013; Cho et al., 2023; Sossidou et al., 2025). Importantly, these stressors do not act in isolation; rather, their combined impact determines overall reproductive outcomes. This highlights that reproductive inefficiency in sows is not solely biologically determined but is strongly modulated by environmental conditions, particularly in high-producing animals with elevated metabolic demands (Gourdine et al., 2021; Chavez-Flores et al., 2026). Environmental thresholds represent commonly reported ranges in pig production systems. Effects are based on physiological and endocrine responses, influencing fertility, litter size, and piglet survival.

Table 5. Environmental stressors and climatic influences affecting sow reproductive performance.

↑, increase/improve; ↓, decrease/reduce/impair; >, greater than; <, less than.

Environmental conditions directly influence endocrine function, feed intake, fertility, embryo survival, and litter size, and environmental variables are routinely incorporated into predictive models of sow reproductive performance. Factors such as ambient temperature, humidity, season, ventilation quality, and stocking density have been shown to improve the accuracy of statistical, genomic, and machine learning models when predicting litter size and reproductive efficiency. Incorporating environmental information is therefore essential for developing robust prediction models that remain reliable across diverse production systems and climatic conditions. Environmental stressors rarely occur independently. In commercial production systems, heat stress is frequently accompanied by elevated humidity, poor air quality, and altered feeding behavior, resulting in cumulative physiological effects. Consequently, reproductive performance should be viewed as the outcome of multiple interacting environmental pressures rather than isolated climatic factors.

3.1.5 Nutritional and metabolic regulation

Nutrition is a key determinant of reproductive success, influencing both fertility and lactation performance (Robinson et al., 2006; Huber, 2025; Monteiro et al., 2025; Mpofu et al., 2026). Adequate feed intake during lactation is essential for maintaining body condition and supporting milk production, which in turn affects piglet survival and growth (Chisoro et al., 2023; Ampode et al., 2024; Johannsen et al., 2024; Mpofu et al., 2026). Insufficient feed intake can result in excessive body condition loss, delaying the return to estrus and prolonging the weaning-to-estrus interval (WEI) (Prunier and Quesnel, 2000; Ye et al., 2024).

Energy balance plays a central role in reproductive function. Negative energy balance during lactation is associated with impaired ovarian activity, delayed ovulation, and reduced conception rates (Monteiro et al., 2025). Excessive energy intake may lead to metabolic disorders that negatively affect reproductive performance (Muro et al., 2023; Gu et al., 2024; Hu et al., 2025; Mpofu et al., 2026). Therefore, optimizing nutritional strategies across production stages is critical for maintaining reproductive efficiency. Modern hyperprolific sows are susceptible to negative energy balance as nutrient requirements for milk production often exceed feed intake capacity during lactation. Consequently, excessive mobilization of body reserves may impair follicular development, delay post-weaning reproductive recovery, and reduce subsequent litter performance. This highlights an important biological influence between current lactation demands and future reproductive efficiency. In addition to energy and protein supply, micronutrients such as selenium, zinc, calcium, phosphorus, and antioxidant vitamins contribute to reproductive function through their roles in immune competence, oxidative stress regulation, embryo development, and uterine health. Deficiencies or imbalances may compromise fertility, embryo survival, and farrowing outcomes, particularly under intensive production conditions (Surai et al., 2016; Monteiro et al., 2025). Nutritional status directly influences endocrine function, ovulation rate, embryo survival, litter size, and lactational performance, and nutritional and metabolic indicators are increasingly incorporated into predictive models of reproductive performance. Variables such as feed intake, body weight change, backfat thickness, energy balance, and metabolic biomarkers have been shown to improve the accuracy of statistical, genomic, and machine learning models used to predict litter size and sow productivity. Integrating nutritional information into predictive frameworks is therefore essential for accurately forecasting reproductive outcomes under commercial production conditions.

Table 6 demonstrates that nutritional regulation of reproduction is fundamentally mediated through metabolic–endocrine integration rather than isolated nutrient effects. Energy and protein intake act as primary drivers by influencing insulin and IGF-1 signaling, which directly regulate ovarian activity and embryo development (Gormley et al., 2024; Monteiro et al., 2025). Body condition indicators such as weight change and backfat thickness serve as integrative markers of metabolic status, with both under- and over-conditioning disrupting hormonal balance and reproductive efficiency (Mun et al., 2024; Ye et al., 2024; Osotsi et al., 2025; Muller et al., 2025). Overall, the table highlights that optimal reproductive performance depends on maintaining a balanced metabolic state that supports both endocrine function and physiological capacity.

Table 6. Nutritional and metabolic determinants influencing reproductive performance in sows.

↑, increase/improve/higher; ↓, decrease/reduce/lower; →, leads to/results in.

Table 7 demonstrates that feedstuffs influence reproductive performance primarily through their effects on metabolic and endocrine pathways rather than through direct reproductive mechanisms alone. Energy-rich feeds enhance insulin signaling and stimulate the hypothalamic–pituitary–gonadal axis, thereby improving ovulation and conception rates, while adequate protein and amino acid supply support follicular development and embryo survival through IGF-1 and mTOR pathways (Hu et al., 2025; Monteiro et al., 2025). Micronutrients and functional additives contribute by reducing oxidative stress and improving uterine and placental function, highlighting the importance of diet quality in addition to quantity (Theil et al., 2014; Surai et al., 2016). Anti-nutritional factors such as mycotoxins disrupt endocrine balance and can severely impair fertility (Gajęcka et al., 2016). Overall, the table underscores that optimal reproductive performance depends on a balanced diet that supports both metabolic status and hormonal regulation, reinforcing the concept that nutrition is a central driver of reproductive efficiency in sows (Prunier and Quesnel, 2000; Gormley et al., 2024).

Table 7. Effects of feedstuffs on reproductive performance in sows.

↑, increase/improved/higher; ↓, decrease/reduced/lower; →, leads to/results in.

3.1.6 Reproductive management and insemination strategy

Management practices are crucial for optimizing reproductive outcomes at the herd level (da Costa Freitas and Barbosa, 2025). Accurate estrus detection and proper timing of insemination are essential for maximizing conception rates and farrowing success (Kraeling and Webel, 2015; Mellagi et al., 2023). Errors in insemination timing or poor semen quality can significantly reduce reproductive efficiency in sows by decreasing fertilization success, farrowing rate, embryo survival, and litter size (Koketsu et al., 2017; Soede et al., 1995). In addition, prolonged semen storage and inappropriate timing of artificial insemination relative to ovulation may compromise sperm viability and reduce reproductive performance in commercial pig production systems (Lucca et al., 2020; Singh et al., 2022; Bortolozzo et al., 2024). Advances in artificial insemination techniques and reproductive monitoring have improved breeding outcomes; however, their effectiveness depends on proper implementation and skilled management (Mellagi et al., 2023). In addition, herd management practices such as grouping strategies, weaning protocols, and farrowing supervision play a key role in minimizing reproductive losses and improving piglet survival (Koketsu and Iida, 2017).

Table 8 demonstrates that reproductive management influences sow performance through insemination accuracy, semen quality, stress control, farrowing supervision, and herd health. Correct estrus detection and insemination timing improve fertilization success and conception rate, whereas prolonged semen storage, poor technique, and inadequate estrus detection reduce fertility and litter size (Knox, 2016; Flowers, 2022; Mellagi et al., 2023). Grouping, weaning, and biosecurity practices also affect reproductive outcomes by modifying stress exposure, body condition, and disease risk (Spoolder et al., 2009; Maes et al., 2011; Koketsu et al., 2017; Xu et al., 2024). Therefore, effective management complements genetic and physiological potential by reducing avoidable reproductive losses.

Table 8. Reproductive management and insemination strategies affecting sow performance.

↑, increase/improve/higher; ↓, decrease/reduce/lower; →, leads to/results in.

3.1.7 Peripartum and early-life determinants

The peripartum period, which includes the final stage of gestation, farrowing, and early lactation, is a critical phase influencing both sow and piglet outcomes (van den Bosch et al., 2023). Factors such as farrowing duration, maternal behavior, uterine health, colostrum intake, and birth weight directly affect piglet survival and stillbirth rates (Tucker et al., 2021; Langendijk and Soede, 2023; Vodolazska et al., 2023). Poor farrowing management can increase perinatal mortality and reduce litter quality (Baxter et al., 2011; Mota-Rojas et al., 2022; Langendijk and Soede, 2023). Effective management during this stage is therefore essential for maximizing the number of piglets weaned and improving overall herd productivity (Koketsu et al., 2017; Nowak et al., 2020).

Early-life conditions, including colostrum intake and birth weight, are also major determinants of piglet survival and long-term performance (Salgado-López et al., 2026). Piglets with low birth weight or inadequate colostrum intake are more susceptible to mortality and poor growth (Tucker et al., 2021; Vodolazska et al., 2023). Effective management during this stage is therefore essential for maximizing the number of piglets weaned and improving overall herd productivity.

Table 9 highlights that reproductive performance is shaped by both early-life developmental factors and peripartum physiological conditions. Early-life traits such as birth weight, colostrum intake, and pre-weaning growth establish the foundation for future reproductive capacity by influencing growth, immune competence, and age at puberty (Devillers et al., 2007; Fix et al., 2010; Calderón Díaz et al. (2017)). Peripartum factors act more immediately by affecting metabolic and endocrine balance during critical reproductive stages (Grahofer and Plush, 2023; Theil et al., 2023). Among these, lactation feed intake and thermal stress are particularly influential, as they directly alter energy balance and hormonal regulation, thereby affecting ovulation and conception (Lucy and Safranski, 2017; Gormley et al., 2024; Muller et al., 2025). Parity further integrates these effects across the sow’s lifespan, reflecting cumulative physiological changes (Koketsu et al., 2017; Knox, 2019). Overall, the table demonstrates that optimal reproductive performance requires both strong early-life development and effective peripartum management to maintain metabolic stability and reproductive efficiency (Corrales-Hernández et al., 2025; Monteiro et al., 2025).

Table 9. Early-life and peripartum determinants of reproductive performance in sows.

↑, increase/improve/higher; ↓, decrease/reduce/lower; →, leads to/results in.

3.2 Predictive modeling approaches for litter size

A wide range of modeling approaches have been applied to predict reproductive performance in pigs, each differing in complexity, data requirements, and predictive capability (Gauthier et al., 2022; Mia et al., 2025). Classical statistical models remain widely used due to their interpretability (Dobson and Barnett, 2018; Hoarau and Dumont, 2023), while genomic and machine learning approaches offer improved predictive accuracy by incorporating high-dimensional data (Verardo et al., 2016; Kousenidis et al., 2022; Su et al., 2025). Biological and simulation models provide complementary insights by integrating physiological mechanisms, nutritional regulation, environmental stressors, and herd-level production dynamics to predict reproductive performance over time (Knap and Doeschl-Wilson, 2020; Burns and Spajić, 2024; Hassanpour, 2025). Collectively, these approaches demonstrate that litter size and reproductive performance are influenced by complex interactions among genetic, physiological, environmental, and management-related variables, necessitating increasingly integrative predictive frameworks (Wang et al., 2022; Sun et al., 2025).

Classical statistical approaches formed the foundation of reproductive prediction studies in pigs because they allowed researchers to identify relationships between reproductive traits and explanatory variables. Early regression and mixed-model studies demonstrated that sow parity, ovulation rate, insemination timing, lactation feed intake, and environmental conditions significantly influence litter size and piglet survival (Rodriguez-Zas et al., 2003; Koketsu et al., 2017). Statistical modeling studies further showed that reproductive performance is strongly associated with herd management and physiological status, particularly in highly prolific sows where endocrine and metabolic factors interact with environmental stressors (Kraeling and Webel, 2015; Tummaruk et al., 2023). These models remain valuable because they provide biologically interpretable relationships and enable estimation of fixed and random effects influencing fertility and litter outcomes (Dobson and Barnett, 2018; Hoarau and Dumont, 2023).

Genomic prediction models have increasingly been utilized to improve prediction accuracy for low-heritability reproductive traits (Shi et al., 2025). Genome-wide association studies (GWAS) identified numerous loci and candidate genes associated with litter size, prenatal survival, and reproductive efficiency, confirming that reproductive traits are polygenic and controlled by many genes with small effects (Verardo et al., 2016; Zhang et al., 2023). Studies in Landrace and Yorkshire populations identified fertility-related genomic regions associated with TNB, NBA, prenatal survival, and overall reproductive efficiency, confirming the polygenic nature of sow fertility traits (Yang et al., 2023; Chen et al., 2025a; Kim et al., 2026). Because reproductive traits generally exhibit low heritability, genomic prediction models based solely on genomic marker information often show limited predictive accuracy (Sun et al., 2025).

Recent studies have therefore focused on integrating genomic selection with machine learning approaches. Wang et al. (2022) demonstrated that machine learning algorithms, including support vector regression (SVR), random forest (RF), kernel ridge regression (KRR), and ensemble learning methods, outperformed conventional genomic prediction methods such as GBLUP and BayesHE in predicting reproductive traits in Chinese Yorkshire pigs (Shi et al., 2025; Sun et al., 2025; Wang et al., 2025). Their study reported that machine learning models improved prediction accuracy by approximately 15%–21% compared with traditional genomic selection approaches, particularly for total number born and number born alive traits. Importantly, the study also showed that prediction performance improved substantially after optimization of model hyperparameters, highlighting the importance of algorithm tuning in livestock prediction systems.

Similarly, Sun et al. (2025) and Wu et al. (2025) investigated genomic prediction of reproductive traits in Large White pigs using genotype imputation and preselected marker strategies. Their findings demonstrated that incorporating imputed whole-genome sequence data and GWAS-selected markers improved genomic prediction accuracy for reproductive traits, particularly when high-quality imputation panels were utilized. The study further confirmed that litter size traits generally exhibit low heritability, reinforcing the need for advanced genomic and machine learning approaches to improve selection efficiency in commercial pig breeding systems.

Machine learning applications have also expanded beyond genomic prediction to precision livestock farming systems. Machine learning models have been applied to predict litter performance in lactating sows and demonstrated that integrating sow characteristics, lactation variables, and production history improved the prediction of litter outcomes in precision farming environments (Gauthier et al., 2022). Likewise, Su et al. (2025) utilized ensemble learning models, including random forest and gradient boosting regression, to predict weaning performance in lactating sows using farm management and nutritional data collected from more than 10,000 observations across commercial farms. Their study identified lactation duration, parity, litter birth weight, and sow backfat thickness as major predictors of weaning litter size and weaning litter weight. Ensemble models achieved relatively high predictive performance, with coefficients of determination (R²) ranging from 0.40 to 0.80, demonstrating the value of integrating nutritional, physiological, and management-related variables into predictive systems.

Other studies further demonstrated the value of machine learning in modeling complex biological interactions affecting reproductive outcomes. Kousenidis et al. (2022) applied machine learning to ultrasound imaging for the prediction of litter size in pigs and reported that automated image analysis systems could accurately estimate litter size during early gestation. Rahman et al. (2023) used machine learning approaches to identify predictors of piglet mortality and demonstrated that parity, birth weight, litter size, and stillbirth incidence were among the most influential predictors of pre-weaning survival. These findings highlight that machine learning approaches are particularly effective for modeling non-linear biological interactions and multidimensional production datasets that are difficult to capture using traditional statistical methods.

Although statistical, genomic, machine learning, and simulation models all contribute to reproductive prediction, they differ substantially in their strengths and limitations. Statistical models are highly interpretable and require relatively small datasets but may not be able to capture complex non-linear relationships. Genomic models improve genetic evaluation for low-heritability traits but often explain only a portion of the observed variation. Machine learning approaches generally provide higher predictive accuracy when large datasets are available, although model interpretability can be limited. In contrast, biological and simulation models provide mechanistic insight into reproductive processes but require extensive biological assumptions and parameterization.

No single modeling approach consistently provides the highest predictive performance across all reproductive traits and production settings, because model utility depends on the prediction objective, data structure, sample size, trait architecture, and validation strategy (Daetwyler et al., 2013; Wientjes et al., 2013; Hickey et al., 2017). Classical statistical and mixed models remain valuable for routine herd analysis because they are transparent, computationally accessible, and capable of accounting for known fixed and random effects, although they may inadequately represent complex non-linear relationships (Dobson and Barnett, 2018; Hoarau and Dumont, 2023). Genomic approaches such as GBLUP and ssGBLUP provide practical advantages for breeding value estimation by incorporating genome-wide relationships and increasing selection accuracy for low-heritability traits, but their performance depends strongly on reference population size, genetic relatedness, marker quality, and population representativeness (Daetwyler et al., 2013; Misztal et al., 2020). Machine learning algorithms can outperform conventional approaches when large, high-quality datasets contain complex interactions among genomic, phenotypic, environmental, and management variables; however, such improvements are not universal and may diminish during independent or across-population validation (Daetwyler et al., 2013; Whalen et al., 2022; Wang et al., 2022; Su et al., 2025). Their commercial use is also constrained by limited interpretability, sensitivity to hyperparameter selection, risk of overfitting, and greater computational and data management requirements (Whalen et al., 2022; Novakovsky et al., 2023; Mia et al., 2025). Mechanistic models offer stronger biological interpretation and scenario-testing capacity but depend heavily on accurate assumptions and parameter estimates (Knap and Doeschl-Wilson, 2020; Bernabò et al., 2021). Therefore, the greatest practical value is likely to arise from models selected according to the intended application and validated under representative commercial conditions, with integrated or ensemble frameworks offering particular promise when they combine predictive accuracy with interpretability, robustness, and operational feasibility (Daetwyler et al., 2013; Gauthier et al., 2022; Su et al., 2025).

Biological and simulation models provide additional insight because they incorporate mechanistic understanding of reproductive physiology and production system dynamics. Mechanistic reproductive models have been used to simulate endocrine regulation, follicular development, embryo survival, and metabolic interactions influencing fertility (Bernabò et al., 2021; Muro et al., 2023). Simulation models such as ZPLAN have also been applied to evaluate long-term breeding strategies, genetic gain, and economic efficiency in pig production systems (Jembere, 2016; Hassanpour, 2025). These approaches are valuable because they provide biologically interpretable frameworks capable of integrating environmental, genetic, and management-related processes over time. Future predictive systems are likely to combine genomic information, sensor-derived behavioral data, environmental measurements, nutritional indicators, and reproductive records within integrated artificial intelligence frameworks. Such multisource models have the potential to provide real-time decision support for breeding, reproductive management, and herd-level optimization. However, challenges related to data quality, model transferability, computational requirements, and biological interpretability must be addressed before these approaches can be widely implemented in commercial pig production systems.

Table 10 demonstrates that predictive modeling approaches for sow reproductive performance differ in complexity, interpretability, and predictive capability. Classical statistical models remain widely used because they are simple, interpretable, and effective for analyzing structured reproductive data, although they are limited in capturing complex non-linear interactions (Dobson and Barnett, 2018; Hoarau and Dumont, 2023). Genomic models such as BLUP, GBLUP, and GWAS improve the identification of fertility-related genes and breeding values, but reproductive traits remain highly polygenic and strongly influenced by environmental factors (Chen et al., 2025a; Hong et al., 2025; Wei et al., 2025). Machine learning models have gained importance due to their ability to analyze large and complex datasets with improved predictive accuracy, particularly when integrating genomic, phenotypic, and environmental data, although their interpretability remains limited (Rahman et al., 2023; Mia et al., 2025). Mechanistic and systems-based models further contribute to biological understanding by simulating physiological and management interactions over time (Bernabò et al., 2021; Gauthier et al., 2022). Overall, the table highlights that integrating statistical, genomic, machine learning, and mechanistic approaches is likely to provide the most effective framework for predicting litter size and improving reproductive efficiency in modern pig production systems.

Table 10. Comparison of predictive modeling approaches for sow reproductive performance.

4 Conceptual integration of biological determinants and predictive modeling

The complex interactions among genetic, physiological, nutritional, environmental, and management factors affect key biological processes such as ovulation rate, fertilization success, embryo survival, fetal development, and maternal performance (Kraeling and Webel, 2015; Tummaruk et al., 2023), which ultimately influence reproductive outcomes including litter size, number born alive, piglet survival, and lifetime productivity. Importantly, many of these determinants also serve as the predictor variables underpinning contemporary reproductive prediction models. Variables such as breed, genomic information, parity, body condition, feed intake, environmental conditions, reproductive history, and management-related data are routinely incorporated into statistical, genomic, and machine learning models to predict reproductive performance (Panigrahi et al., 2025; Wang et al., 2025). Thus, predictive modeling and reproductive biology should be regarded as complementary and interconnected components of modern precision pig production rather than as independent fields of study. Predictive modeling should be viewed as a tool for quantifying and forecasting the effects of biological processes on reproductive outcomes in modern precision livestock farming (Curti et al., 2023).

Figure 1 illustrates the conceptual framework underpinning this review. Biological determinants influence reproductive processes and performance traits, which generate the phenotypic, genomic, environmental, and management data used as inputs for predictive models. Classical statistical models generally utilize biologically interpretable predictors such as parity, breed, and environmental factors, whereas genomic models incorporate molecular information associated with fertility and prolificacy. Machine learning approaches further integrate multiple data sources to capture complex relationships among biological and production variables.

Therefore, the accuracy and usefulness of predictive models depend largely on their ability to incorporate biologically meaningful information. Improved understanding of the biological mechanisms underlying reproductive performance can enhance model development and interpretation, while model outputs can support breeding, nutritional, environmental, and reproductive management decisions. This highlights the complementary relationship between reproductive biology and predictive modeling and underscores the need for integrated approaches to improve sow productivity and litter size prediction. Model outputs support precision breeding, reproductive management, and decision-making, creating a feedback cycle that can further influence reproductive performance and productivity.

5 Future directions in research

The evidence synthesized in this review demonstrates that improving sow reproductive performance will depend on integrating reproductive biology with advanced predictive modeling rather than relying on increasingly sophisticated algorithms alone. Although considerable progress has been made in identifying genetic, physiological, nutritional, environmental, and management determinants of reproductive performance, current predictive models remain constrained by limited biological integration, inadequate external validation, and inconsistent performance across breeds and production systems (Zhang et al., 2023; Thiengpimol et al., 2024a; Wang et al., 2025). Future research should therefore prioritize biologically informed, interpretable, and externally validated prediction systems capable of integrating multiple sources of biological and production data within precision livestock farming.

As discussed throughout this review, reproductive performance is determined by interactions among genetics, endocrine function, metabolism, nutrition, parity, environmental stress, and management rather than isolated production variables (Kraeling and Webel, 2015; Lucy and Safranski, 2017; Tummaruk et al., 2023). Thus, future prediction systems should integrate biologically important indicators such as body condition, backfat thickness, metabolic and endocrine biomarkers, reproductive history, and parity-dependent reproductive trajectories, together with genomic and production records. Such biologically informed models are expected to improve both predictive accuracy and mechanistic interpretation while enhancing their applicability across commercial production systems (Willis et al., 2003; Quesnel et al., 2007; Björkman et al., 2023).

Furthermore, while genome-wide association studies have identified candidate genes associated with fertility and litter size, these loci explain only a proportion of the observed phenotypic variation because reproductive traits are highly polygenic and influenced by environmental interactions (Verardo et al., 2016; Zhang et al., 2023; Hong et al., 2025; Cieleń et al., 2026). Future studies should integrate genomic, transcriptomic, epigenetic, physiological, environmental, and longitudinal production data to better characterize the biological pathways regulating ovulation, embryo survival, placental development, and maternal performance. Emphasis should be placed on evaluating genotype × environment interactions to improve prediction robustness under diverse climatic and management conditions.

Machine learning algorithms have demonstrated considerable potential for modeling the complex non-linear relationships underlying litter size and piglet survival. However, their practical implementation remains limited by poor biological interpretability and insufficient external validation (Rahman et al., 2023; Mia et al., 2025). Future research should move beyond maximizing predictive accuracy and instead evaluate artificial intelligence models according to their biological relevance, interpretability, robustness, and generalizability across independent populations and commercial production systems. The incorporation of explainable artificial intelligence frameworks will facilitate identification of the biological variables driving model predictions and improve confidence among breeders, veterinarians, and producers responsible for implementing precision breeding strategies (Verhoeven et al., 2023; Su et al., 2025).

Also, continuous monitoring of behavioral, physiological, nutritional, and environmental indicators using sensors and automated recording systems offers opportunities to detect reproductive disorders earlier and optimize breeding decisions (Neethirajan, 2020; Verhoeven et al., 2023; Tedeschi et al., 2025). Given the increasing impact of climate change on sow fertility, future prediction models should explicitly incorporate environmental variables such as temperature, humidity, and seasonal variation, together with parity-dependent reproductive trajectories and management practices, to improve model robustness under commercial conditions (Lucy and Safranski, 2017; Ghulam Mohyuddin et al., 2022; Mateos et al., 2024). Integrating these data streams within user-friendly decision support systems will facilitate precision breeding, nutritional management, replacement decisions, and reproductive interventions tailored to individual animals and herd-level objectives.

Finally, future advances in sow reproductive prediction will therefore depend on developing integrated frameworks that combine biological understanding with robust predictive analytics. Rather than treating reproductive biology and predictive modeling as separate disciplines, future research should establish biologically interpretable, externally validated, and climate-resilient prediction systems capable of integrating genomic, physiological, environmental, and management information. Such systems will provide the greatest opportunity to improve litter size, piglet survival, lifetime sow productivity, and the long-term sustainability of commercial pig production.

6 Conclusion

Reproductive efficiency remains a cornerstone of productivity, profitability, and sustainability in modern pig production systems, particularly in highly prolific maternal breeds. The reproductive performance cannot be adequately captured by a single indicator but instead requires a comprehensive evaluation using interconnected metrics, including litter size, fertility rates, reproductive intervals, and lifetime productivity. While these metrics provide essential tools for quantifying performance, their interpretation is inherently limited without a clear understanding of the biological and management factors driving their variations. The reproductive outcomes are shaped by a complex and dynamic interplay of genetics, biological status (and modifications of the biological status), physiological status, environmental conditions, and herd management practices. Genetic selection has successfully improved prolificacy; however, these gains are often accompanied by biological trade-offs, including increased variability in piglet survival and greater sensitivity to stress. Physiological factors such as parity dynamics, endocrine regulation, and energy balance further influence reproductive efficiency, while external pressures, particularly heat stress and management quality, remain critical constraints in modern production systems. Collectively, the evidence synthesized in this review demonstrates that meaningful improvements in sow reproductive performance depend on integrating biological understanding with statistical, genomic, and machine learning approaches rather than considering these components independently. For producers and breeding organizations, these findings emphasize the importance of combining genetic improvement with precision management strategies. Enhancing reproductive efficiency will require not only selection for prolificacy but also improved environmental management, nutritional support, health control, and the application of data-driven decision-making tools. Challenges remain in model generalizability, data integration, and the practical application of predictive outputs, particularly in commercial farm settings. Despite substantial advances in predictive analytics, challenges related to data quality, biological variability, model interpretability, and external validation continue to limit their widespread application across diverse production systems. Addressing these challenges will require the integration of biologically relevant phenotypic, physiological, genomic, environmental, and precision livestock farming data within robust and externally validated prediction frameworks. Future progress should therefore prioritize biologically informed, interpretable, and practically applicable prediction systems capable of supporting evidence-based breeding and reproductive management across diverse commercial production systems. Unlike previous reviews that have considered reproductive determinants or predictive modeling approaches separately, this review provides an integrated framework linking reproductive biology with contemporary predictive methodologies, thereby offering a foundation for precision breeding and evidence-based reproductive management. Future research should prioritize biologically informed, interpretable, and practically applicable prediction systems capable of supporting sustainable, resilient, and data-driven pig production across diverse commercial environments.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Francesca Conte, University of Messina, Italy

Reviewed by: Fernando Sánchez-Esquiliche, Sánchez Romero Carvajal Jabugo SAU, Spain

Gerardo Ordaz Ochoa, National Institute of Forestry and Agricultural Research (INIFAP), Mexico