In NICU outcomes that differed by sex (death before discharge, RDS, Apgar score < 7 at 5 minutes, intubation during initial resuscitation, hypothermia, severe IVH and surfactant therapy) this study found significantly higher estimated odds of adverse outcomes for males in unweighted analysis compared to IPW analysis. IPW analysis appropriately addresses confounding from BW and GA that occurs due to selection bias when artificially truncating the population of all births at birth weights less than 1500 grams to create the VLBW subpopulation. Thus, this finding reveals that studies using standard statistical methods to test for sex differences in VLBW infants are generally reporting estimates biased towards increased risk of adverse outcomes for males. The three outcomes (RDS, intubation during initial resuscitation and surfactant therapy) that became non-significant in IPW analysis results that remove the confounding effect of GA and BW due to selection bias are greatly influenced by underdevelopment of the lungs, with sex evidently playing less of a role.
The finding that VLBW males at the study institution were born at a significantly lower GA is consistent with previous research [9, 13, 17]. In this study, since sex was significantly associated with both death and GA, and GA was significantly associated with death after controlling for GA, GA would be considered a significant mediator or confounder of the relationship of sex with mortality as defined by causal steps analysis [18]. Although statistics cannot disentangle confounding from causally mediated effects [19], the results of this study provide evidence that GA is a confounder due to selection bias in the VLBW subpopulation. That fact that GA is also a causal mediator of sex differences in outcomes can be ascertained based on clinical knowledge. Among all births, many studies [11, 20–30] have found males to be born at a lower GA than females, with a hypothesized reason being that male fetuses are on average heavier than females of the same GA [21, 22]. That fact that lower GA is casually related to death and other adverse outcomes is evident beyond question. Therefore, GA is both a confounder and a casual mediator of sex differences in mortality and other adverse outcomes.
The results of the present study provide unbiased estimates of the casual effect of sex on mortality and other adverse outcomes in the NICU among VLBW infants. Although much remains unknown about the reason for these sex differences, hormones, genetics, immunology, physiology, microvascular function and growth factors have been proposed to play roles in the male disadvantage in outcomes among underdeveloped infants [12, 31, 32]. Research suggests male fetuses biologically prioritize growth pathways in order to facilitate an evolutionary advantage later in the reproductive stage of life, which renders males less adaptable in utero, thereby increasing the risk for perinatal morbidity and mortality compared to female fetuses that prioritize reserve capacity and adaptability [33]. Sex differences in placental development and function have been hypothesized to be a reason for increased risk of pregnancy complications for male fetuses [34].
Accurate estimation of sex differences in outcomes will facilitate investigation of the biological mechanisms involved so that targeted interventions can be developed to mitigate the male disadvantage. Some research has shown the sex gap in mortality among very preterm infants to be narrowing over time, resulting in speculation that males may preferentially benefit from new treatments such as antenatal steroids and surfactant [14, 35–37]. Studies in both animals [38–40] and humans [10, 41–43] have found differential treatment responses among fetal and neonatal males compared to females. There is even evidence that preterm males’ neonatal responses to maternal nutrition differs from that of females, and maternal breastmilk composition varies by infant sex [44]. Therefore, sex-specific strategies need to be investigated in order identify treatments that are best suited for an individual infant. This is not a new idea. In 2005 the Food and Drug Administration published recommendations [45] for research into sex differences in treatment responses and in 2014 the National Institute of Health created requirements [46] for biomedical research grant applications involving preclinical studies to include examination of sex as a variable. Despite these past recommendations, greater emphasis on research specifically investigating sex differences in treatment responses is still needed today.
This study was conducted at a single institution with a unique NICU patient population, so the findings might not generalize to other NICUs. This study is also limited by the retrospective observational study design since statistics cannot discriminate causal mediation from confounding with an observational study. However, since sex cannot be randomized as an intervention, the IPW analysis used in this study provides a valid method to estimate the causal effect of sex on NICU outcomes in the VLBW population.
This is the first published study to show that standard statistical methods that have been used in previous research yield biased estimates of sex differences in outcomes among VLBW infants. This study provides valid estimates of the casual effect of sex on mortality and other adverse NICU outcomes among VLBW infants. Accurate estimates are necessary to help us understand the causal pathways leading to sex differences in NICU outcomes. This knowledge will support development of targeted interventions that take infant sex into account.