Using a large administrative cohort of over six million pregnant individuals and their infants, we found that maternal HRSN increased the risk of several adverse infant outcomes, including preterm birth, LBW, SGA, NICU admission, and infant death.

When disaggregated into specific causes of death, observed estimates of mortality appeared attributable to SUID and ‘conditions originating in the perinatal period.’ These findings suggest that though a significant proportion of deaths among infants prenatally exposed to HRSN occur at or shortly after delivery, infants were also at risk for death after discharge and through the first year of life.

Maternal HRSN was protective against infant emergency department admission and infant rehospitalization. Previous literature has demonstrated a positive relation between increased number of SRFs and infant emergency department admission and infant hospitalization [27, 28] using patient-endorsed social needs collected by survey. We speculate that the protective association observed in our study may be explained by the effect of prolonged NICU hospitalizations among preterm births, early death, or loss of custody by the birthing parent with HRSN, potentially resulting in a less hazardous environment for the infant. Additionally, this protective association may be explained by various social determinants of health that make healthcare access difficult, including un- or underinsurance, economic hardship, poor health literacy and transportation insecurity. Particularly, un- and underinsurance and economic hardship are some of the strongest predictors of delayed or forgone necessary medical care [29, 30], which may have influenced our observed relationship between HRSN, emergency department admission, and rehospitalization.

Previous literature has demonstrated similar results to our study. Green et al. reported similar increases in homelessness in a national sample of pregnant patients who had comparable rates of preterm birth and perinatal characteristics [10]. Similarly, Pantell et al. reported similar rates of HRSN and perinatal complications among pregnant individuals with HRSNs in a California sample [12]. It is possible that various unmeasured factors, downstream to HRSN, play a role in the observed maternal and infant morbidity and mortality associated with HRSNs. These include decreased access to primary and prenatal care services [31,32,33], lack of social support [9], increased rates of medical comorbidities [33, 34], and systemic bias and discrimination from healthcare systems and providers [35]. Further, other literature has found that unstable housing and poor infant health may be cyclical and co-constitutive, suggesting that these positive feedback loops can irreparably cause both poor health and lack of social opportunity for low-income families [36]. Given the high rates of maternal morbidity in our sample, it is difficult to ascertain the temporality of HRSN and poor health outcomes in our population. Birthing people in this sample may have experienced poor health outcomes leading to economic strain and subsequent housing instability or unemployment, or vice versa. Previous research suggests that a bidirectional relationship is most likely [11, 37, 38].

Though HRSN documentation increased over the study period, this increase likely reflects increasing ICD social code use rather than increasing prevalence of HRSN, and current rates of documentation likely still severely underestimate the true prevalence of HRSNs. This is probably due to a lack of validated SRF screening tools, limited and vague guidelines for SRF screening, and varying methods for documentation [7, 39]. In the future, standardized application of validated screening tools in clinical settings would allow providers to identify HRSN in a timely fashion and provide the foundation for research necessary to inform widespread SRF screening guidelines. Our findings bolster arguments for increased presence of ancillary staff and social workers in healthcare settings, whose presence improves health and utilization outcomes [40]. Additionally, HRSN data have the potential to affect clinical decision-making and policy strategies, and influence the design of interventions to address them.

Currently, use of ICD social codes is voluntary, and Centers for Medicare and Medicaid Services (CMS) and commercial payers offer no independent financial incentive for their use. Though these codes are not billable on their own, as of January 2021, CMS guidelines allow HRSNs to warrant a higher level of complexity for an office visit. This allows providers to bill for a moderate level of complexity—rather than a minimal or low level—given a HRSN that complicates the diagnosis or management of their patient’s condition [41].

First, this study is limited by its reliance on ICD codes for HRSN documentation. Despite our robust sample size, the prevalence of HRSN documentation likely represents a tremendous underreporting of SRF. We recognize that this limits our ability to make conclusions about the risks associated with these exposures, but this bias is likely to be true in any large epidemiological sample of similar data. Regardless, characterizing the prevalence of HRSN documentation is one of the primary aims of this paper, as we believe this information is important to inform future screening and documentation practices. Given the scarcity of SDOH screening and documentation guidelines in and out of pregnancy, different providers, health systems, and electronic medical records likely employ varied processes to document HRSN. Two recent studies estimated sensitivity of ICD social codes to be as low at 10% compared to patient screening responses [39, 42]. Thus, given the high rates of maternal morbidity among individuals with HRSN, HRSN documentation may be biased toward individuals with comorbid diagnoses (e.g., substance use disorders, severe maternal morbidity) or other risk factors (e.g., public insurance, inadequate prenatal care), which likely have independent negative effects on infant outcomes. This discrepancy would overestimate the effects of the relationship between HRSN and infant outcomes. However, given the low prevalence of HRSN in this sample, it is likely that a significant number of individuals experiencing SRFs but without a documented HRSN are included in our referent population, which likely attenuates our observed effect estimates. Due to these competing biases, certainty regarding the causal relationship between HRSN and adverse infant outcomes is limited, and likely vary by provider screening and documentation protocols. Further, the misclassification as compared to self-reported surveys may have affected the protective association we observed among infant emergency department admission and infant rehospitalization. Of note, 99.8% of individuals in our sample delivered in medical centers that had previously used ICD social codes, indicating that HRSN documentation is likely not differential by hospital.

Second, given the low prevalence of ICD social code documentation in this study, we were unable to disaggregate the effects of specific SRFs on infant outcomes. Food insecurity, unemployment, and homelessness during pregnancy likely confer varying amounts of risk to infant outcomes, and experiencing two or three risk factors simultaneously may be riskier than experiencing one. Given the high proportion of homelessness among individuals in our HRSN sample, it is likely that much of the risk of adverse infant outcomes is attributable to homeless status during pregnancy. Further, only thirteen individuals in our sample had a diagnosis of food insecurity, which constitutes a tremendous underreporting compared to reliable estimates that point to nearly 30% of families with children experiencing food insecurity [4]. The “Hunger Vital Sign”, a two-item food insecurity screening tool, is widely used in clinical practice given its ease of administration and validity; an affirmative response to either question one or two has a sensitivity of 97% and a specificity of 83%. Thus, the underreporting of food insecurity in our sample is likely the result of inadequate documentation rather than inadequate screening. Given that food insecurity and homelessness frequently exist in tandem [43], it is likely that many individuals with documentation of homelessness were also experiencing food insecurity.

Third, we did not have information on postnatal exposure to SRFs. Especially with outcomes like SUID, inadequate housing may increase the prevalence of bedsharing, unsafe sleeping environments, and secondhand marijuana or tobacco smoke, which are all risk factors for SUID and cannot be accounted for in this study [44]. This limited our ability to make conclusions about the effect of maternal HRSN versus infant HRSN on infant outcomes. Conversely, given the extensive housing and economic resources available to pregnant people and families in California [45], it is possible that many infants with prenatal HRSN experienced more favorable postnatal environments, which would serve as a protective factor against poor health outcomes in infancy and beyond.

Despite these limitations, a strength of this study is the use of a population-based administrative cohort of California births. The study population was diverse with respect to race and ethnicity, socioeconomic status, and geographic region. The large size of the sample allowed for robust estimations of the temporal trends in ICD social code use in the state, an important first step in increasing SRF screening and documentation during pregnancy.