Neighborhood Deprivation and Association With Neonatal Intensive Care Unit Mortality and Morbidity for Extremely Premature Infants

Key Points Question Is socioeconomic deprivation at the neighborhood level, measured by an Area Deprivation Index, an independent risk factor for neonatal intensive care unit (NICU) mortality and morbidity among extremely premature infants? Findings In a cohort study of 2765 extremely premature infants (gestational age <29 weeks) admitted to 4 level IV NICUs in different US regions, national Area Deprivation Index percentile was associated with the risk of NICU mortality and morbidities after adjusting for multiple covariates. Meaning These findings have implications for public health measures, which should target areas of neighborhood deprivation to improve prenatal care with the aim of improving NICU outcomes.


Introduction
Socioeconomic disadvantage impacts maternal health and access to care, affecting the preterm delivery rate [1][2][3] and infant mortality. 4Once premature infants leave the neonatal intensive care unit (NICU), socioeconomic deprivation in the home environment adversely impacts neurodevelopmental outcomes. 5,6Evidence suggests that socioeconomic disparity increases adverse outcomes of extremely preterm infants during the NICU course, 7 but the interaction among deprivation, race, and other risk factors is not well established.3][14] In addition, due in part to a history of housing discrimination and structural racism in the US, members of racial minority groups more often live in socioeconomically disadvantaged communities. 15These complex interactions between race and socioeconomic status necessitate consideration when evaluating maternal and infant outcomes.
The Area Deprivation Index (ADI) is a metric that measures area disadvantage, determined by 17 census variables, including measures of poverty, educational level, housing, and employment status.Singh and Soahpush 16 first described this metric in relation to disparities in life expectancy that correlated with area deprivation.Kind et al 17 adapted, updated, and validated the ADI at the neighborhood level using block groups, the smallest geographical unit used by the US Census Bureau.
In the most recent iteration, ADI has been normalized at the national level to generate percentiles that can be easily compared, with higher values indicating greater disadvantage.National ADI percentiles and state ADI deciles using 2018 US Census Bureau data are published online in an interactive map. 18,19sociations between ADI and health outcomes have been investigated in many populations, including adult patients with Alzheimer disease, 20 cardiovascular disease, 21 and in-hospital COVID-19 mortality. 22The ADI has also been used to investigate disparity in neonatal health outcomes, such as the rates of exclusive breastfeeding, 23 neurobehavioral differences and brain structure in term infants, 24,25 respiratory morbidity after NICU discharge in infants with bronchopulmonary dysplasia, 26 and length of stay in infants with neonatal opioid withdrawal syndrome. 27 this project, we investigated the association of the ADI, a granular, comprehensive, geospatially linked measure of deprivation, and demographic characteristics with in-hospital mortality of extremely premature infants and 3 key morbidities (late-onset sepsis, necrotizing enterocolitis [NEC], and severe intraventricular hemorrhage [IVH]).We hypothesized that maternal residence in areas with greater deprivation is associated with increased NICU mortality and morbidity risk.Our study includes 8 years of data from 4 regional referral NICUs located in 4 US regions: Midwest, Northeast, Mid-Atlantic, and South.

Study Design and Patient Population
We conducted a multicenter, retrospective cohort study of non-Hispanic Black and White premature infants born at a gestational age of less than 29 weeks and admitted to 1 of 4 academic level IV NICUs-University of Virginia in Charlottesville; University of Alabama at Birmingham; Washington was used to determine the infants' race.Hispanic infants were excluded because this ethnicity category was not specified in the earlier years of the study, and they accounted for relatively few patients (5%-10%) at all but 1 NICU, at Columbia University.We also excluded Asian infants and other (eg, American Indian and Alaska Native) race categories with very few patients.We excluded infants with missing outcomes data from the analysis.Each institution's institutional review board reviewed and approved the study under a waiver of informed consent.

Address Geocoding and ADI Data Process
All geographic locations in the US are defined using a Federal Information Processing System code.
The degree of granularity of this code is determined by the number of digits, with a maximum of 15

Clinical and Outcome Data
Clinical data were extracted from unit databases or electronic health records, including demographic characteristics, perinatal variables, and outcomes.The primary outcome was mortality before NICU discharge, and secondary outcomes included late-onset sepsis or NEC and severe IVH.Late-onset sepsis was defined as positive blood culture findings after 3 days of age that was treated with antibiotics for at least 5 days (or less if the infant died during treatment); NEC, as radiographic evidence of pneumatosis, pneumoperitoneum, or portal venous gas plus clinical signs and symptoms of sepsis according to the modified Bell staging criteria for stage 2 or higher 29 ; and severe IVH, as grades III to IV IVH identified on cranial ultrasonography according to the Papile classification system. 30

Statistical Analysis
We performed univariate comparisons among ADI, demographic characteristics, and primary and secondary outcomes using Wilcoxon rank sum or χ 2 tests.We analyzed the multivariable association between ADI and NICU outcomes using bayesian logistic regression adjusted for race, birth weight, outborn status, and sex.For each variable included in the multivariable models, we also performed univariate bayesian logistic regression analysis with primary and secondary outcomes.For computational efficiency, quantitative variables were centered and scaled before analysis.Two-sided P < .05indicated statistical significance.
In contrast with frequentist models, bayesian models provide more information about parameters through posterior distributions, 31 more correctly quantify the uncertainty in model parameters, 32 and are often able to estimate models that may otherwise fail. 33We used weakly informative priors for all parameters.We considered a risk factor significant if the posterior 95% credible intervals (CrI) did not include zero and took the posterior median as the summary of the posterior in all regression analyses.We compared models with interaction terms to the baseline model and considered an interaction significant if the information criterion value was lower than that of the baseline model.

Patient Characteristics
We

Univariate Analysis
In pooled data across the 4 centers, the distribution of ADI was skewed, with 1770 infants (64.0%) born to families residing in areas of higher deprivation, or ADI above the fiftieth national percentile.
Mortality was similarly skewed, with more deaths among the infants with higher ADI (Figure 1   In addition to comparing variable distributions among binary subgroups, we performed univariate bayesian logistic regression to estimate the association between each variable used in multivariable modeling and the primary and secondary outcomes.Table 3 summarizes the univariate and multivariable modeling results for each outcome.

Results of Multivariable Analysis
We examined the association between ADI and NICU mortality while adjusting for the covariates identified in univariate analyses: birth weight, sex, race, and outborn status.Based on posterior median estimates, birth weight was the strongest risk factor for mortality (posterior estimate, −1.20 [95% CrI, −1.35 to −1.06]), but national ADI percentile (posterior estimate, 0.14 [95% CrI, 0.02-0.26]) and male sex (posterior estimate, 0.35 [95% CrI, 0.13-0.57])were also associated with mortality (Table 3).Neither race nor the interaction between race and ADI was significant, and the term did not improve the Watanabe-Akaike information criteria 34 of the model.
Birth weight and ADI were also associated with NICU morbidity in multivariable models (Table 3).Because we scaled and centered continuous variables, we were able to compare the size of the estimates among variables included in each model.By treating ADI and birth weight values as z scores, we report the effect of changing either variable by 1 SD, rather than by 1 percentile (ADI) or 1 g (birth weight).Therefore, in the model for mortality, we interpreted the effect of decreasing birth weight by 1 SD to be 9 times that of increasing ADI by 1 SD.In contrast, in the model for severe IVH, the effects of birth weight and ADI were similar and still less than the effect of outborn status.When modeling sepsis and NEC, the effect of decreasing birth weight was about 3 times that of increasing ADI, and male sex and outborn status had intermediate effects.Race did not significantly affect any of the three outcomes modeled.

Discussion
Socioeconomic factors play an important role in maternal health, access to prenatal care, rates of preterm birth, and infant morbidity and mortality outside of the NICU. 4,5,35While in the NICU, preterm infants are treated according to standardized care guidelines that might be expected to limit the association of socioeconomic deprivation with adverse outcomes.Few studies have evaluated the association of area deprivation with in-hospital outcomes. 36In this study, we found that residence in a socioeconomically disadvantaged neighborhood-as measured by the ADI-increases the risk of NICU mortality or morbidity for extremely preterm infants, even after accounting for other known risk factors.This indicates a pervasive and detrimental effect of area deprivation that may not be entirely eliminated by provision of standardized, high-quality NICU care.
The mechanism by which area deprivation affects neonatal outcomes may include the effect of maternal adversity on the developing fetus. 37,380][41] Lower maternal socioeconomic status increases the risk of preterm birth, 11,12,42 and among those born prematurely, it increases the risk of adverse neurodevelopmental outcomes. 43,44In contrast with other studies, we did not find an association between ADI and birth weight, likely due to the fact that we limited the analysis to extremely premature infants with a narrow range of birth weights.
We found that the association between higher deprivation and mortality remained in multivariable analysis while the association between Black race and mortality did not.These highly collinear variables make it difficult to estimate the effect of race or deprivation alone.Studies indicate that despite advances over time, racial disparities persist in neonatal outcomes and care practices, 10,45,46 but these analyses did not account for socioeconomic factors.As racial minority individuals proportionally endure more socioeconomic deprivation, it becomes essential to consider social disparities as a significant mediator of racial disparities.Public health efforts have been made to narrow disparities in antenatal corticosteroid use and cesarean deliveries, but gaps in these  46 Disparities in breastfeeding by race persist, 10 despite overall improvements in breastfeeding initiation over time, with the lowest rate of breastfeeding at NICU discharge among infants born to non-Hispanic Black mothers. 47In our cohort, infants born to Black mothers were exposed to higher levels of deprivation in utero and had lower birth weights.Still, they did not have higher associated morbidity or mortality when accounting for these and other confounding variables.
Our results suggest that the higher mortality observed among Black preterm infants compared with White infants may be partly confounded by factors captured in the ADI, including exposure to adverse social and economic conditions.

Strengths and Limitations
A strength of this study was the inclusion of 4 centers in geographically distinct areas of the US (Northeast, Mid-Atlantic, Midwest, and South) with different distributions of ADI, patient racial compositions, and population densities.Although all centers are level IV referral NICUs, there were differences in the percentage of inborn infants, ranging from 64.4% to 78.6%.Interestingly, the distribution of ADI was skewed in opposite directions for infants admitted to the centers in the Midwest and South compared with the Northeast, while the Mid-Atlantic center had a nearly symmetric ADI distribution (eFigure 1 in Supplement 1).The ADI distributions within each center appear to be representative of their respective regional distributions as displayed on the Neighborhood Atlas ADI map.We did not adjust for center in multivariable modeling, as national percentile ADI data has already been normalized across all geographic regions.Additionally, we did not use state ADI deciles in our analysis since these values do not translate across state lines and would not be appropriate for a pooled sample including data from multiple states.
Limitations of the study include the lack of granularity in the ADI and race data.Although block groups are the most geospatially compact census unit and the ADI can provide a localized overview of the degree of socioeconomic deprivation, its value is opaque.The relative contributions of each ADI component are not available for the most current values, making it infeasible to evaluate driving factors of the association between ADI and NICU morbidity and mortality.In addition, the exposures captured in the ADI may have differential impacts across the heterogeneous locations that we analyzed.For example, in the 2020 US Census data for just the 4 cities in which the study NICUs were located, the land area ranged from 10.24 to 302.64 square miles; population density ranged from 1365 to 29 302 per square mile; median household incomes ranged from $38 832 to $67 046; and the percentage of adults without health insurance ranged from 7.9% to 14.7%.Each of these differences likely plays a significant role in the factors that underlie the ADI calculation.
We classified race into broad groups of Black or White based on the self-reported maternal race and excluded American Indian and Alaska Native, Asian, and Hispanic patients, which may have left important interactions or outcomes undiscovered.Analyzing more refined racial and ethnic subgroups, including more than 1 race when the father's race differs from the mother's race, may have found differential associations between ADI and outcomes among more finely delineated racial and ethnic groups.For example, the "Hispanic paradox" refers to the paradoxical finding that Hispanic infant mortality rates lie at or below the rate of Non-Hispanic White infant mortality, despite higher socioeconomic deprivation. 48,49r results generate questions for future studies to evaluate how interventions to reduce area deprivation or improve equity might reduce disparities in NICU outcomes for premature infants.The study's findings also leave us wanting to know more about the mechanism of the association between ADI and NICU outcomes, but we could not answer such questions with the data available.

Conclusions
In this cohort study of preterm infants born at a gestational age of less than 29 weeks and admitted to 4 NICUs, ADI was a risk factor for NICU mortality and morbidity after adjusting for multiple covariates.These findings highlight the need for further research on the components of the ADI that

Figure 1 .
Figure 1.National Area Deprivation Index (ADI) and Neonatal Intensive Care Unit (NICU) Mortality

Figure 2 .
Figure 2. National Area Deprivation Index (ADI) Percentiles and Neonatal Intensive Care Unit (NICU) Morbidity Association of Neighborhood Deprivation With NICU Mortality and Morbidity University in St Louis, St Louis, Missouri; and Columbia University, New York City, New York-between January 1, 2012, and December 31, 2020.We followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.Maternal self-reported race JAMA Network Open.2023;6(5):e2311761.doi:10.1001/jamanetworkopen.2023.11761(Reprinted) May 11, 2023 2/12 Downloaded From: https://jamanetwork.com/ on 09/28/2023 analyzed ADI and clinical data from 2765 extremely premature infants admitted during the study period.The mean (SD) gestational age was 25.6 (1.7) weeks; birth weight, 805 (241) g.In terms of sex and race, 1391 infants (50.3%) were boys and 1374 (49.7%) were girls, and 1325 (47.9%) reported Black maternal race and 1440 (52.1%) reported White maternal race.A total of 498 infants (18.0%) died before NICU discharge, 692 (25.0%) were diagnosed with sepsis or NEC, and 353 (12.8%) developed severe IVH.Inborn status ranged from 261 of 405 infants (64.4%) at Washington University in St Louis to 396 of 504 (78.6%) at Columbia University.Table 1 summarizes cohort characteristics by center.Missing data occurred in 387 infants (14.0%), which required excluding them from multivariable modeling.Of these infants, 235 were missing data on birth hospital to determine inborn or outborn status, 83 were missing data on IVH grade, and 69 were missing data on whether or not sepsis or NEC was diagnosed during the NICU course.Only 37 infants (1.3%) were excluded due to their address being missing or not linked to a block group with an ADI percentile (PO boxes, rural routes).