Association of Residence in High–Police Contact Neighborhoods With Preterm Birth Among Black and White Individuals in Minneapolis

Key Points Question Is living in a neighborhood with high police presence associated with increased risk of preterm birth? Findings In this cross-sectional study of 1059 Minneapolis residents who gave birth to a live singleton in 2016, the odds of preterm birth for pregnant people living in a neighborhood with high police presence was significantly higher compared with the odds of their racial counterparts in a low-presence neighborhood (90% increase for White individuals, 100% increase for US-born Black individuals, and 10% for Black individuals born outside of the US). The higher the proportion of Black residents in the neighborhood, the greater the number of police incident reports. Meaning These findings suggest that greater police presence in Black vs White neighborhoods may contribute to the persistent Black-White preterm birth disparity in Minneapolis.


Introduction
Black pregnant people in the United States experience preterm birth (PTB; birth at <37 weeks' gestation) at rates approximately 2 times that of White pregnant people. 1 Black pregnant people are also twice as likely to experience the death of an infant younger than 1 year, a disparity primarily associated with preterm-related causes of death. 2 These entrenched disparities remain unmoved by decades of public health research and persist despite increased access to prenatal care. They are not explained by differences in health behaviors, and they affect Black individuals of all socioeconomic statuses (SESs). 3 The persistent and racialized nature of these inequities points to racism as a root cause.
Racism shows up in many ways in our society. One manifestation is residential racial segregation (ie, geographic separation by race). Black pregnant people who live in areas with high levels of racial segregation are more likely to give birth prematurely. 4 Residential segregation relegates Black people to neighborhoods disproportionately affected by poverty, violence, and crime. 5,6 In lieu of policy solutions to address these issues, greater police presence has been the answer in many communities-a practice known as proactive policing. 7 Proactive policing has led to a policing system that attempts to aggressively maintain order by preventing crime before it occurs through what is commonly referred to as broken windows policing. 8 It also encourages more frequent involuntary contact between law enforcement and the public. Due to socially and often racially biased notions of criminality and deviance, these policies have resulted in the disproportionate police presence in low-income and majority Black neighborhoods. 7 In this study, we define overpolicing as the disproportionate targeting of Black communities by law enforcement that does not correlate to the incidence of crime within neighborhoods but rather reflects racist beliefs about Black deviance. 9,10 Research has shown that men report increased psychological distress in neighborhood-level increases in aggressive policing (ie, frisking and use of force by police). 11 However, to our knowledge, there is no literature that seeks to understand how aggressive policing and increased police surveillance affect maternal and infant health, specifically PTB. Recent evidence suggests that states with more killings of unarmed Black people by police have larger Black-White PTB disparities. 12 Given that PTB is believed to be influenced by a wide range of environmental and psychosocial factors and their interactions, [13][14][15][16] it is important to understand whether living in a community with a disproportionate amount of police presence affects the risk of PTB.
This study examined the association between police contact and PTB in US-born Black, Black born outside the US, and White pregnant people in Minneapolis, Minnesota. Minneapolis is a wellpositioned geographic location for our study for several reasons. First, despite its reputation as a politically progressive city with forward-thinking urban planning, Minneapolis has some of the worst scores for residential racial segregation measures in the nation. 6 Second, Minneapolis is home to an ethnically diverse Black population-notably one of the largest Somali communities outside of Somalia 17 -providing a unique opportunity to assess the intersectionality between race and nativity and its joint association with police contact patterns and racial inequity in PTB. Third, the stark disparity in PTB rates for Black and White pregnant people and infant mortality in Minneapolis has been well documented. 18 As the city continues to reimagine public safety, particularly in response to the killing of George Floyd, Jr, results from this study will add to the body of evidence that will inform the redesign of public policies in Minneapolis and elsewhere in the United States. 19 Leveraging patient medical records from a large health care system in the Minneapolis-St Paul area, detailed spatial data, and city-level policing data, we aimed to answer 2 questions: (1) is   excessive police contact associated with increased risk of PTB for White pregnant people, US-born   Black pregnant people, and Black pregnant people born outside the US in Minneapolis, and (2) is police contact racialized (ie, more police contact in Black than White neighborhoods) in a way that could suggest increased exposure and subsequent risk of PTB?

Study Cohort
Our study cohort included pregnant people who gave birth to live singletons from January 1 to December 31, 2016, at Fairview Health System facilities (8545 individuals). We further restricted our analytical sample to Black and White individuals (based on their self-reported race) whose residential address on the medical record fell within one of the 116 census tracts that make up the city of

Police Contact
We operationalized and measured community-level police contact at the census tract level, using the police incident data from the City of Minneapolis Police Incident Report. 21 This database contained information on the date and time the incidents were reported by police officers, the type of offense (Figure 1), and the geographic coordinates of the incidents. These incidents do not represent all contacts between police and members of the public, only those for which police officers filed official  reports and to which uniform crime reporting coding was assigned. To ensure that we captured the extent of police contact from all Minneapolis census tracts, we pooled police incident reports from 2012 to 2016 and assigned a census tract identifier to each incident. The police incidents reported in this database were not precisely located on a street boundary. When we intersected the small buffers with the census tracts, buffers that appeared to be on the border of Minneapolis were assigned census tract identification for 15 tracts located outside Minneapolis.
The 5-year cumulative count of police incidents of all offense types was calculated for each census tract. For this period, the value ranged from 10 to more than 9000 incidents. To account for increased incidents due to a larger census tract population size, we calculated per-capita police contact by dividing the cumulative count of police incidents in each census tract by the census tract population estimate from the 2012-2016 American Community Survey (ACS) 5-year estimates. 22 Based on the per-capita police contact from 116 Minneapolis and 15 bordering census tracts, we dichotomized census tracts into those with high (fourth quartile) vs low (first to third quartile) police contact. For sensitivity purposes, we also assigned census tracts' police contact status based on tertiles (ie, third tertile vs first and second tertiles). This tertile-based measure put individuals who lived in a census tract with moderately high police contact (ie, from the 67th to 74th percentile) in the high exposure group. Hence, census tracts with police contact per capita in the third tertile and those in the fourth quartile may be treated as high and very high exposure neighborhoods, respectively.

Neighborhood Racial Composition
We

Other Covariates
Based on previous studies 4, 23 and available data in the medical records, we measured age at the time of delivery (<25 years, 25-29 years, 30-34 years; Ն35 years), marital status (married vs not married), and insurance status (ever received public insurance vs never). Measures of SES in medical records were sparse, as patients provide this information voluntarily. Given a high degree of missingness for SES in the medical records, we opted not to impute the missing data because the algorithm would likely generate biased predictions. Instead, we supplemented the medical record data with the census tract median household income from the 2012-2016 ACS 5-year estimates, which has been shown to be highly associated with the residents' actual household income. 24 Despite the availability of health diagnosis (eg, hypertension and diabetes) and health behavior (eg, smoking) information in the medical record, we did not include such covariates in our analysis. These factors are associated with a higher risk of PTB [25][26][27] and are likely mediators in the association between police contact and PTB; including them in the regression would have biased the association of police contact with PTB toward the null. 28

Statistical Analysis
First, we conducted a descriptive analysis to characterize sociodemographic characteristics, police contact levels, and PTB incidence among pregnant people in our sample. The distribution of these characteristics by racial group was compared using Pearson χ 2 test and t test for categorical and continuous measures, respectively.
Second, we examined the association between police contact and PTB using logistic regression, stratified by racial group. In our unadjusted model, we regressed PTB on police contact. In the fully adjusted model, we controlled for age at the time of delivery, marital status, insurance status, and census tract median household income. For both models, we calculated odds ratios (ORs) and 95% CIs. Standard errors were calculated using the Huber-White method to account for the heteroscedasticity at the census tract level. 29 To test the sensitivity of our regression findings, we executed similar models using a tertile-based measure of police contact. Lastly, we tested for spatial autocorrelation between the proportion of Black residents in all Minneapolis census tracts and the cumulative police incident counts. Spatial autocorrelations were measured globally to test for clustering and locally using the Moran I statistics. 30 The global measure was assessed using a Moran scatterplot and significance tested through a permutation test. 30 For visual examination, we also created local indicator of spatial association (LISA) maps, which indicate the level of correlation between the proportion of Black residents and the police incident count in neighboring locations. 30 All statistical tests were 2-sided, with α = .05 as the level of statistical significance. Data management and analyses of the medical record data were conducted in Stata version 13.1

JAMA Network Open | Public Health
(StataCorp) and R version 3.6.2 (R Project for Statistical Computing). Spatial analysis was conducted in GeoDa version 1.14.0 (Center for Spatial Data Science, University of Chicago).

Results
Of 1059   When we reclassified the police contact variable so that the census tracts in the third tertile (rather than the fourth quartile) were considered high exposure ( Our geospatial analysis revealed a positive autocorrelation between the proportion of Black residents and the number of police incidents in the neighboring census tract with a Moran I of 0.237. This measure was tested with a permutation test, and clustering was significantly different than a random distribution at a pseudo-P value of .001. Figure 2 shows local clustering of census tracts that demonstrated a higher correlation between the proportion of Black residents and the police incident counts (the LISA statistic with P < .05). We observed clusters with a high proportion of Black residents and a high incident count (ie, high-high; 13 census tracts), a collection of clusters with a low proportion Black residents and a low incident count (low-low; 18 census tracts), clusters with a low proportion Black residents and high incident count (low-high; 6 census tracts), and clusters with a high proportion Black residents and a low incident count (high-low; 1 census tracts). Other clusters showed no significant clustering.

Discussion
This study provides additional insight into the association between police contact, neighborhood context, and risk of PTB. To our knowledge, we are the first to assess the association of police contact with PTB among US-born Black pregnant people and Black pregnant people born outside the US, an important intragroup distinction. Our analyses suggest that high police contact may affect not only Black but also White pregnant people. This finding is consistent with similar studies conducted in urban populations associating racial disparities in birth outcomes with neighborhood context. 23,31,32 Our analyses also suggest that the degree to which police contact was associated with increased risk of PTB varied for different racial groups. We found the OR of PTB was highest for US-born Black born outside the US in our sample lived in neighborhoods with high police contact, but the incidence of PTB was low, even less than that of White pregnant people. We found that Black pregnant people who were born outside the US, even when living in neighborhoods with high levels of police contact, were less likely to experience PTB, even when compared with their White counterparts. These results reinforce the notion that Black identity is a social, not a biological, fact. Although both foreign-and US-born individuals occupy the same racial category (Black), the lived experience of anti-Black racism   Several studies point directly to the aggressive, degrading, and racist nature of police stops and indicate that these factors make police encounters psychologically difficult for Black people to endure. [33][34][35][36] Police contact-specifically routine stops conducted by police departments as a crime deterrent measure in certain neighborhoods-signal to Black people that in the eyes of the law, they are seen as inherently criminal and dangerous. 37 This negative socialization coupled with the aggressive behavior that officers exhibit during these stops increases stress, anxiety, and adverse physical health for people living in these communities. 33,34,38,39 Now more than ever, communities across the country are grappling with how to ensure public safety in light of racialized incidents of police brutality brought to their awareness in the aftermath of the murder of George Floyd, Jr, in Minneapolis, where this study took place.

Limitations
Our study has several limitations. First, our study measured levels of police contact based on the number of police incidents in the census tract in which the pregnant people lived. However, given our ecological study design, we did not (and would not have been able to) ascertain whether the pregnant people in our study had personal contact with the police before or during pregnancy. Furthermore, it is important to note that community-level police contact is likely associated with other forms of structural racism that cannot be assessed in this study. Second, because the police incident data we used to derive the police contact measure were from before the year of birth, we made an implicit assumption that the individuals in our study had stable housing and stayed at the same address throughout the period of exposure. This was done mainly because of the cross-sectional nature of our data. The degree of housing stability likely varied for White pregnant people, US-born Black pregnant people, and Black pregnant people who were born outside the US in our sample. As a result, misclassification of the exposure variable in our analysis is possible, but it is difficult to assess given the nature of our data. Third, our sample was from 1 health system in the Minneapolis-St Paul area and may not be representative of all pregnant people in Minneapolis. Furthermore, given our small sample size, estimates from our multivariate regression may be unstable. Future research should explore the use of longitudinal data that track participants' residence and levels of police contact dynamically to address these limitations. We also recommend the use of data from other representative sources (eg, restricted vital statistic data with geocoding variable) with a larger sample size to assess whether the association reported in our study is true in other locales, especially in urban areas with high proportions of Black immigrants (eg, New York City; Washington, DC; Seattle). Additionally, not all types of police contact may carry the same stressinducing effects and may be associated with the odds of PTB to at a different degree. Future studies should examine the potential heterogeneous associations of police contact with health outcomes by offense type.

Conclusions
In conclusion, we found that police contact was associated with increased risk of PTB among White pregnant people, US-born Black pregnant people, and Black pregnant people who were born outside the US. However, because neighborhoods with a greater proportion of Black residents were more likely to be policed, the higher incidence of PTB among Black pregnant people than White pregnant people may be attributed to racialized exposure rather than a differential effect of police contact between racial groups. These findings suggest that racialized police patterns borne from a history of structural racism in the United States may contribute to racial disparity in PTB.