Racial Segregation and Intraventricular Hemorrhage in Preterm Infants

We constructed a deidentified cohort of deliveries using maternal and infant hospital data linked to birth and death certificate data from California, Missouri, and Pennsylvania from 1995 to 2009. According to the policies of the Children’s Hospital of Philadelphia Institutional Review Board, the abstracted data did not meet standards for human subjects research. The institutional review boards of the departments of health in California, Missouri, and Pennsylvania approved this study. Maternal and infant hospital records were linked to infant birth certificates and death certificates by using previously described methods that allow for >98% matching of birth certificates to hospital records.¹⁷ 

Live singleton births delivered between 24 + 0/7 and 31 + 6/7 weeks’ gestational age and weighing at least 500 g were eligible for inclusion. Neonates were excluded if they had congenital anomalies, died in the delivery room, or were missing birth certificate or hospital data as per the method previously described.¹⁸  Infants were also excluded if their recorded BW was above 5 SDs from the mean BW of their gestational age group because these values were likely a recording error.¹⁹  To preserve comparability to existing segregation literature, infants were only included if maternal self-reported race was either non-Hispanic black or non-Hispanic white. Because data on nation of origin were not available, no distinction can be made among African Americans, Caribbean Islanders, or natives of the African continent; thus, the designation of black race was used for all groups. In total, 105 245 infants (76 128 white and 29 117 black) were included.

The main outcome of interest was the presence of IVH, as defined by International Classification of Diseases, Ninth Revision, codes 772.10 to 772.14. Because information on IVH severity was not available for any observations before 2000, all grades of IVH were included.

The primary predictor of interest was segregation, defined as the dissimilarity index calculated by the US Census Bureau using the 2000 decennial census data and presented at the metropolitan statistical area (MSA) level. An MSA consists of an urban core and surrounding areas with strong economic ties. Dissimilarity is the percent of a racial group that would have to move so that the proportion for each group in a subarea matches that of the whole. Values range from 0 to 1, with higher values signifying greater levels of minority segregation. Of the 19 segregation indices calculated, dissimilarity was chosen as the primary predictor of interest because it is the most commonly published measure in segregation studies.²⁰ 

RRS levels were linked to maternal zip code (ZC) via a published MSA-ZC crosswalk table from the US Census Bureau for a total match of 67.2% (70 775) of individuals. A total of 32.8% were unable to be matched either because the ZC represented a rural area outside of the urban MSA or because the ZC, which is subject to reallocation, deletion, and divisions as populations fluctuate, was not in use in the year (2000) when the crosswalk table was published. Dissimilarity scores were used to divide individuals into quartiles by degree of segregation.

To account for potential confounding variables, maternal, pregnancy, and infant characteristics were included in the final model. In previous research, it has been suggested that maternal history of hypertension (either chronic or secondary to pregnancy) is protective against the development of IVH, whereas chorioamnionitis is a risk factor for IVH.³  The method of delivery (vaginal versus cesarean delivery) and preterm labor were included as potential pregnancy risk factors for IVH.^3,7  BW was included in the adjustment model because this measure is strongly correlated with IVH independent of gestational age.³  Lastly, to disentangle the impact of community socioeconomic factors that may track with RRS levels, we included the following ZC data obtained from the US Census Bureau and the various American Community Surveys: median income, percent below poverty, percent completing high school, percent completing college, mean home value, and percent renting a home. These data were linked by ZC and delivery year.

All statistical analyses were performed by using Stata 14 (Stat Corp, College Station, TX). Subjects were stratified into quartiles of segregation. Descriptive statistics were calculated by using χ² tests for dichotomous variables and analysis of variance for continuous variables.

There were large, statistically significant differences in the baseline demographic, medical, and social characteristics across the segregation quartile groups that could not be adequately adjusted for in multivariate regression. We used propensity score matching, which is a method to control for such imbalance in confounding factors that might affect treatment assignment status in observational studies.²¹  The propensity score was calculated for each patient with complete data by using a logistic regression model to estimate the probability of receiving the “treatment” of living in a highly segregated community (defined dissimilarity index above the group median). Patients were considered “unexposed” if they resided in a community with a dissimilarity index below the group median. By using previously published literature and a priori assumptions, the following characteristics were included as covariates in the propensity score regression model: chorioamnionitis, pregnancy-induced hypertension, chronic hypertension, preterm labor, cesarean delivery, community-level median income, and college attendance. To preserve comparability between groups, exact matching was forced on race and BW category (<1250, 1251–2500, >2500 g).

“Exposed” and unexposed patients were matched 1:1 without replacement by using a 0.05 caliper, for a total of 17 918 pairs (35 836 individuals). To ensure that covariates were balanced, or controlled for, after matching, we assessed both the standardized differences in each covariate between the high- and low-segregation groups and the percent reduction in bias (per previously described methods).²²  A standardized difference threshold of 0.1 was used for this test, as in previous studies.^23,24  Conditional Poisson modeling determined the risk ratio (RR) of IVH in the group exposed to high segregation compared with the unexposed group within each matched pair. The analysis was performed separately for each racial group to determine if the association varied by race.

Of the 70 775 non-Hispanic white and black premature infants in the cohort, 36.10% (n = 25 458) were born to mothers who identified as non-Hispanic black. As shown in Table 1, the proportion of black infants increased significantly between the least and most segregated quartile (18.3% vs 64.3%; P < .001), as is consistent with the definition of segregation. Although the least segregated quartiles were composed mostly of infants born in California (89.8%), states were evenly represented in the most segregated quartile: 36.2% Californians, 29.4% Missourians, and 34.5% Pennsylvanians. Compared with mothers living in the least segregated areas, those individuals living in highly segregated areas were younger and more likely to have chorioamnionitis, pregnancy-induced hypertension, chronic hypertension, and preterm labor. Infants born to mothers in the highest quartile of segregation had BWs that were, on average, 215 g lower than those of infants in the lowest quartile. There were also significant community differences: mothers living in the most segregated quartile experienced significantly higher rates of poverty, lower median income, lower home values, and lower rates of college attendance compared with mothers living in the least segregated quartile.

A total of 8058 (11.4%) infants developed IVH. As shown in Fig 1, infants born to mothers living in the highest segregation quartile had significantly higher rates of IVH compared with those from the lowest quartile (12.9% vs 10.4%; P < .001; Table 1). For both races, the rate of IVH tended to increase with segregation (for white infants, 0.51% increased rate of IVH per quartile, P = .003; for black infants, 1.12% increase per quartile, P = .001).

Propensity scores were calculated for 70 746 infants (35 440 in the untreated group; 35 306 in the treated group) as shown in Table 2. Propensity scores ranged from 0.000087 to 0.99873 (mean = 0.4989598) with an SD of 0.2459. Of the infants with propensity scores, 35 836 could be matched for 17 918 pairs. The matched pairs were well balanced on all infant, maternal, and community factors included in the model, with standardized differences all less than the 0.1 threshold. The bias in the propensity to reside in a highly segregated community within matched pairs was reduced 99.5% from the baseline, with similar reductions in bias for other individual covariates within pairs (Table 2).

For all matched pairs, the risk of developing IVH was significantly greater in the group exposed to a highly segregated environment (RR = 1.08; 95% confidence interval [CI]: 1.01–1.15) (Table 3). When stratified by race, this effect did not persist in white infant pairs (RR = 1.05; 95% CI: 0.97–1.13). In black infant pairs, the group exposed to high RRS had 1.16 times the risk of developing IVH compared with the group living in less segregated communities (95% CI: 1.03–1.30).

At baseline, there were large variations in medical, maternal, and social covariates between the lowest and highest segregation quartiles and a significant difference in IVH rates. Propensity score matching, which matched pairs on their likelihood of receiving the treatment of living in a segregated community, balanced these baseline differences. The increased risk of developing IVH in infants born to mothers living in a segregated community persisted after adequate balancing. The effect of segregation on IVH was much greater for black infants than for white infants, thus suggesting a differential effect of segregation by maternal race.

The existing literature on IVH has focused on identifying patient- and hospital-level risk factors.^5–7,25  Whereas increased understanding of medical risk may guide clinical response to IVH, characterization of the upstream social determinants of IVH can be useful in shaping prevention strategies. Only one previous study has revealed an association between maternal social characteristics, such as lower education and public insurance, and risk of IVH in a subset of black patients.³  Although this is consistent with our finding that social risk factors vary by race, the aforementioned study did not control for community-level factors such as neighborhood poverty or median home values.

RRS, which has been called the “fundamental cause of racial disparities in health,”⁸  was investigated to better illuminate the community forces that may increase an infant’s susceptibility to IVH. RRS has been previously identified as a risk factor for other adverse events in the neonatal period, such as preterm delivery, low BW, and infant mortality.^16,26–29  These works have been criticized for failing to adequately adjust for other social variables, such as poverty and education, which colocalize with higher levels of segregation.³⁰  Propensity score matching is a method to circumvent these issues, but it has been underused as a tool in the segregation literature.^30,31 

Unlike previously studied outcomes that are measurable at the time of delivery, this study linked segregation to an event that occurs in the early neonatal period, thereby suggesting intergenerational transmission of the impact of segregation. The propensity score analysis allowed for adjustment of differences in BW, a variable that is both an established risk factor for IVH and a measure negatively associated with segregation. Because the association between IVH and segregation persisted after matching for BW, segregation likely induces susceptibility to IVH independent of its effect on BW.

Current theories regarding how RRS impacts health are focused on concurrent measures of socioeconomic opportunity experienced by the residents in these segregated areas, such as educational and economic opportunity, rates of crime, and housing quality.^8,10,11  These models are useful to identify areas that are modifiable through policy and public health interventions; however, they fail to account for the intergenerational transmission of the effect of segregation that is suggested by this analysis. The theory of chronic stress ties community characteristics to concrete physiologic alterations, postulating that severe, consistent stress causes maladaptive neuroendocrine, inflammatory, and vascular alterations that increase susceptibility to disease.^32–35  The maternal experience of chronic stress induces epigenetic alterations in her unborn fetus, indicating intergenerational propagation.³⁶  Chronic stress has been linked to other adverse neonatal outcomes that have been associated with segregation, such as low BW and preterm delivery.^37–39  The pathophysiology underlying IVH is in part attributed to fragile vascularity and immature homeostasis at the germinal matrix, which overlap with the known effects of chronic stress.⁴⁰  Further work is needed to understand the changes that chronic stress induces at the molecular level and how this can lead to injury at the developing germinal matrix.

White women living in segregated environments had infants who were at a lower risk of IVH, even when adjusting for social factors, suggesting that white race is inherently protective regardless of environment. The enclave effect postulates that underrepresented groups living in segregation are more closely connected, with increased social cohesion and political and social willpower.^41,42  The paradoxical association between Mexican American immigrants with low socioeconomic status and increased BW of their infants has been attributed to this enclave effect.^43,44  White families living in mostly black neighborhoods might provide each other with better support and control a greater proportion of political power, which could insulate against the harmful neighborhood conditions. Alternatively, black mothers may be uniquely susceptible to the impact of segregation because it represents institutionalized racism and personal discrimination, which is an established predictor of poor health in adults and children and through mothers to their unborn children.^45–49  Given US history of discriminatory housing practices, RRS is sometimes measured as a proxy for structural racism.⁵⁰  Both the enclave effect and racial discrimination warrant further investigation because they could uncover strategies to protect black infants in vulnerable neighborhoods that do not require physical relocation of individuals.

This was an observational study and is therefore limited in its ability to establish a causal link between segregation and IVH. We employed propensity score matching to better balance the covariates between segregated and nonsegregated groups and reduce ecologic fallacy. It is possible that the matched pairs are still unbalanced with regard to some unmeasured confounder, but we included any available risk factors for IVH in our propensity score calculations. Because the least segregated quartiles consisted mostly of California-born infants, whereas the most segregated quartiles were more balanced geographically, many pairs were discordant on state residence at birth, which could introduce confounders that reflect differences in clinical practice or access to care that vary by region. However, it is clear that there is variation in segregation level between states, which may offer opportunities to evaluate state-specific factors that promote less segregated communities.

The diagnosis of IVH is predicated on ultrasonography screening in high-risk infants; institutional practices and individual skill may have introduced inconsistencies in diagnosis and grading. Our IVH rate was similar to previously published results,¹  and by aggregating all grades of IVH, we eliminated potential observer biases in grading. Given limitations in International Classification of Diseases, Ninth Revision, codes for IVH severity, we were unable to distinguish severe bleeds from mild bleeds, prohibiting comment on the clinical significance of these findings. However, even small bleeds are associated with long-term, moderate-to-severe neurodevelopmental impairment.⁵¹  Data for segregation were only available by MSA, prohibiting an individual- or neighborhood-level assessment. By assigning the same segregation exposure to women living both in the urban core and the suburbs within a single MSA, our results were likely biased toward the null.

In this study, we found that the maternal experience of segregation is a novel risk factor for the development of IVH in premature black infants and may explain observed racial disparities in IVH incidence and mortality. Continued research is needed to illuminate the pathway that connects segregation to IVH, with an emphasis on understanding the impact of toxic stress at the molecular level. Black mothers were more likely to have infants with IVH than white mothers, even when living in similarly segregated environments; this observation warrants further investigation to determine if the difference stems from a protective interaction in white mothers or is due to injustices suffered by black mothers. Propensity score analysis offers a valid method for building causal inference and may be useful in disentangling the complex relationships between race and health. Eliminating segregation will require significant, sustained political and financial willpower. In the interim, pediatricians can effect change by employing an understanding of environmental barriers, including RRS, to target evidence-based clinical interventions, advocate for policy changes, and provide community-specific education.

BW

birth weight

CI

confidence interval

IVH

intraventricular hemorrhage

MSA

metropolitan statistical area

RR

risk ratio

RRS

residential racial segregation

ZC

zip code