Parental Determinants and Adaptive Clinical Thresholds for Low Birth Weight

Low birth weight (LBW) is a major public health concern that is associated with increased neonatal morbidity, mortality, and long-term negative outcomes. Despite extensive research identifying several maternal, socioeconomic, and demographic factors associated with LBW, most studies have examined these variables in isolation using small datasets, which complicates our ability to understand the determinants of LBW for the following reasons. First, few studies have examined how maternal and paternal characteristics together influence the risk of low birth weight across diverse populations. Second, current clinical practices often rely on fixed birth weight thresholds that do not account for neonatal health status or biological variation across maternal profiles. Finally, the role of paternal anthropometric factors remains poorly understood, with inconsistent findings and limited exploration of whether parental influences vary by infant sex. Together, these gaps underscore the necessity for an integrated analytical framework that combines comprehensive population-level analysis, clinically adaptive risk stratification, and sex-specific modeling of parental contributions to fetal growth.This thesis addresses these existing gaps through a multi-stage analytical framework across three retrospective studies. First, a population-level predictive modeling approach was applied to a large-scale natality dataset using supervised machine learning models to identify maternal, paternal, and socioeconomic factors consistently associated with LBW. To identify consistent predictors across these analytical methods, model interpretation techniques such as odds-ratio analysis, SHAP values etc were used. Second, a two-staged clinical modeling strategy was developed to derive adaptive birth weight thresholds. A conditional inference tree and a fuzzy inference model were first used to characterize the relationship between maternal height, birth weight, and Apgar score based on neonatal health status. Subsequently, ensemble regression models were applied to estimate maternal height-specific birth weight cutoffs associated with non-risk neonatal outcomes. Finally, the fourth chapter focused on interpretable deep learning regression architectures were designed to quantify the contributions of maternal and paternal anthropometry to birth weight. Feature attribution was assessed using mother-to-father importance ratios, SHAP analysis, and ablation experiments to evaluate sex-specific and joint parental effects.Our first study identified that maternal height, pre-pregnancy weight, gestational weight gain, parental ethnicity, and key socioeconomic and behavioral factors were the most consistent determinants of birth weight, highlighting the joint influence of biological and social conditions. Notably, maternal height emerged as a central factor underlying variation in birth weight risk across population groups.This variation in risk profiles shows that a single clinical cutoff cannot adequately reflect biological differences among mothers. Addressing this limitation, the second study derived clinically adaptive thresholds by integrating maternal height with Apgar score–based health assessment. The result demonstrated that the conventional 2,500 g cutoff systematically misclassifies risk across maternal profiles, with taller mothers requiring higher birth weights to achieve healthy Apgar outcomes and shorter mothers frequently delivering healthy newborns below this standard. While maternal biology explained substantial variation in neonatal risk, these findings also raised the need to examine whether fetal growth is shaped jointly by both parents and whether these influences differ by infant sex. Therefore, in the third study the interpretable deep learning models revealed that paternal anthropometric characteristics, particularly body mass index and weight, contribute meaningfully to birth weight prediction, especially for male infants. In contrast, maternal features remained dominant for female infants, indicating clear sex-specific parental influences on fetal growth.Overall, this work establishes an integrated, data-driven framework for birth weight risk assessment that accounts for biological variation, socioeconomic context, and sex-specific parental influences beyond traditional fixed-threshold approaches.