Negative illness feedbacks: High‐frisk policing reduces civilian reliance on ED services

Abstract Objective This paper demonstrates that localized and chronic stop‐question‐and‐frisk (SQF) practices are associated with community members’ utilization of emergency department (ED) resources. To explain this relationship, we explore the empirical applicability of a legal epidemiological framework, or the study of legal institutional influences on the distribution of disease and injury. Data and Study Design Analyses are derived from merging data from the Philadelphia Vehicle and Pedestrians Investigation, the National Historical Geographic Information System, and the Southeastern Philadelphia Community Health database to zip code identifiers common to all datasets. Weighted multilevel negative binomial regressions measure the influence that local SQF practices have on ED use for this population. Analytic methods incorporate patient demographic covariates including household size, health insurance status, and having a doctor as a usual source of care. Principal Findings Findings reveal that both tract‐level frisking and poor health are linked to more frequent use of hospital EDs, per respondent report. Despite their health care needs, however, reporting poor/fair health status is associated with a substantial decrease in the rate of emergency department visits as neighborhood frisk concentration increases (IRR = 0.923; 95% CI: 0.891, 0.957). Moreover, more sickly people in high‐frisk neighborhoods live in tracts that have greater racial disparities in frisking—a pattern that accounts for the moderating role of neighborhood frisking in sick people's usage of the emergency room. Conclusions Findings indicating the robust association reported above interrogate the chronic incompatibility of local health and human service system aims. The study also provides an interdisciplinary theoretical lens through which stakeholders can make sense of these challenges and their implications.


| INTRODUC TI ON
Police violence is an unyielding public health crisis in a number of marginalized and vulnerable American communities. [1][2][3][4][5][6] Research on spillovers of the criminal legal system indicates that direct contact with police officers is associated with poor health outcomes including psychological distress, [7][8][9][10][11] compounded illness conditions, 12,13 infectious disease transmission, 14,15 and death. 16,17 Even indirect police contact poses adverse health risks for those embedded in networks that endure aggressive policing practices [18][19][20][21][22][23][24][25][26] especially, among men living in aggressively surveilled areas and women living in lethally surveilled areas. 27 Police officers have the latitude to initiate referrals to health care, suggest the type of institution best equipped to provide care, and determine whether transfers are executed voluntarily or coercively. 28,29 Less, however, is known about the association between embeddedness in intrusive police prevalence and surveilled community members' willingness to rely on hospital emergency room care.
Thus, we consider connectivity to intrusive policing as it correlates with visits to hospital emergency departments (EDs).

Emergency room admission protocols open opportunities for legal
surveillance. 30,31 For example, Leibschutz et al 26

shared that when
Black patients in an ED learned that police could and were questioning individuals arriving in municipal ambulances, their suspicion of collusion between law enforcement and health care providers peaked.
Furthermore, Lara-Millán 25 illustrates that even when the urban poor covertly seek help for health problems, law enforcement has developed strategies to identify the location of the facilities they frequent.
Officers stationed at Lara-Millán's field site, routinely worked toward "systematically thinning the front room wait list" (2014:877) of patients whose activity did not justify criminal legal system intervention. When police conduct unsanctioned background checks of waiting room attendees, ED utilization within these surveilled communities poses an imminent threat. These sorts of institutional practices, which are sustained by police officers and accepted by hospital staff, can deter civilians from seeking much-needed help through ED care. This study measures how localized police activity shapes such hospital utilization.
Legal epidemiology is a framework that could enhance our theoretical understanding of how collusion between law enforcement and health services influences patients' decisions to rely on ED care.
Legal epidemiology posits that law and legal practices shape health outcomes. [32][33][34] Simultaneously, research has emerged focusing on how racialized policing in racialized contexts exacerbates health problems. 1,4,22,23,[35][36][37] While the current study was not designed to test the theoretical constructs of legal epidemiology, we examine its applicability for interpreting a patient-centered and spatially defined relationship between state-operated local policing practices and hospital utilization for an ethnoracially diverse urban sample.
Brayne's 38 conceptualization of "system avoidance" holds that justice-impacted individuals avoid surveilling institutions due to their maintenance of formal, systematized, and identifiable records.
Similarly, we expect to observe a relationship between systemic place-based police intrusion and depressed reliance on hospital EDs, including those that facilitate access to trauma services.
The current study will test whether police officers' stop-question-and-frisk (SQF) practices dampen ED visitation. This association is an indicator of negative illness feedbacks where police tasked with fostering a safe environment engage practices that curtail the procurement of services for those facing health challenges. Such countervailing relationships produce new forms of systemic violence that reveal the systematic marginalization that governs the lives of the inequitably surveilled. In such communities, police behaviors weaken, and even disentangle, the relationship between self-rated health and hospitalization. We account for within and between neighborhood variation in known sociodemographic factors associated with our measures of interest.

| Study design
This study is a cross-sectional observational study, describing the rate of ED visits due to residential exposure to frisk (ie, a bodily pat down). The study focuses on the 18-65 population of the Philadelphia

What is Known on This Topic
• Police violence is an unyielding public health crisis in ethnoracially marginalized and economically vulnerable American communities.
• Direct contact with police officers is associated with a number of poor health outcomes, including psychological distress, substance misuse and co-morbidity, infectious disease transmission, and death.
• Indirect contact with police officers, via residence in a neighborhood with higher concentrations of proactive and lethal policing, is associated with poor health outcomes as well (negative evaluations of health, diabetes, hypertension, acute asthma, obesity, and psychological distress)-especially, among men living in aggressively surveilled areas and women living in lethally surveilled areas.

What This Study Adds
• An empirical application of legal epidemiology, which examines the chronic incompatibility of local health, safety, and human service system aims.
• Evidence that sicker people who live in neighborhoods with high stop-and-frisk rates report less frequent use of the hospital ED, while ED utilization among healthier people is unrelated to their residential exposure to frisking. Agencies tasked with fostering a safe environment appear to engage practices that curtail the use of services by those facing serious health challenges.

| Data sources
The individual data source for this study comes from the Southeastern Pennsylvania Household Health Survey (SEPA HHS), a repeated cross-sectional telephone survey of more than 10 000 households. through the National Health Geographic Information Survey. 39 Forty-four zip code tabulation areas with shared zip code identifiers were retained for analysis.

| Measures
The outcome of interest is the number of ED visits in the past year.
Respondents were asked: "How many visits, if any, did you have to an emergency room during the past twelve months, that is, since (date one year ago) 2013/2014?" This is reported to SEPA HHS by the respondent as a count variable characterizing the number of times in the past year that the respondent has visited the emergency room (ED) for any reason. Higher values indicate more ED utilization.
The individual-level independent variable is a five-category self-rated health, an ordinal variable with Likert response options (1 = Excellent; 2 = Very Good; 3 = Good; 4 = Fair; and 5 = Poor).
Self-rated health is also measured as a dichotomous indicator, where 1 = Poor/Fair and 0 = Excellent/Very Good/Good. Higher values of both outcomes indicate worse health. Self-rated health captures lay constructions of illness and is a well-known strong predictor of mortality. 40 The contextual independent variable is neighborhood frisk concentration and is measured at the zip code-level. This continuous indicator characterizes the percentage of stops in a neighborhood where police officers engage in frisking-that is, patting down a pedestrian or a driver. Unlike prior research on the illness risks of policing pedestrians, 20,22,23 we evaluate data that include stops of both pedestrian and vehicular stops, instead of focusing only on pedestrians. As such, our data capture a more robust measure of frisking than prior research. To our knowledge, this is the first study that evaluates the health care costs of frisking drivers, even while much of the research on inequalities in police surveillance is built for research on drivers. [41][42][43] Higher values indicate more exposure to frisking by police for people living in an area.
Zip code frisk concentration is also treated as an effect modifier in the analysis. Household income includes all sources of income from anyone living at the address who is related to the respondent by blood, marriage, or adoption and is transformed so that a one-unit change is equivalent to $10 000. Respondents are provided 26 categories from which to choose, and these categories are midpoint recoded (see Appendix S1 for values). Since there is a substantial amount of missing data on income (n i = 654. 21.3 percent; Table 1), mean substitution replacement is used to retain respondents with missing information on household income in the analytical sample, where all respondents with missing data are assigned the mean value of the nonmissing data. We also include an indicator of missingness on household income (0 = Missing; 1 = Data Present). Furthermore, we include an imputed measure of poverty status, which is provided by PHMC to include in analysis to reduce bias due to missingness. The Gender, health insurance, regular source of medical care, and chronic health conditions are dichotomous variables. Gender is characterized as male (1) or female (0). A person is indicated as having health care insurance (1) if they indicated they received health insurance through work, school, or a union; that they bought health insurance directly and paid for it in total by themselves or their family regardless of whether they did so with government assistance; that they had access to health insurance through a government program; or that they obtained health insurance through some other insurance, group, or place not including those already mentioned.
All others are coded as uninsured (0) Table 1 indicates the reference category (0) for dichotomous and nominal covariates.
Neighborhood-level confounders derived from the 2015 ACS dataset account for covariance related to education and poverty status. Neighborhood educational attainment is measured by the percentage of the population over the age of 25 who has at least a college degree. Neighborhood poverty status is measured as the percentage of families whose income to poverty level ratio is less than one (1). A moderate negative correlation exists between neighborhood education and neighborhood poverty (r = −.59), and a moderate correlation exists between neighborhood frisk concentration and neighborhood socioeconomic status (r education.j = −.49; r poverty.j = .53).

| Sample
The study sample focused on people between 18

| Analytic strategy
The statistical modeling framework employed in this study an-  We conduct three sets of regression analyses to evaluate the role of illness and neighborhood frisk exposure on ED utilization.
The first regression analysis (Model 1) evaluates the past-year rate of ED visits to self-rated health. We expect that respondents with worse self-rated health will utilize the ED more in the 12 months prior to participating in the SEPA HHS survey (Hypothesis 2).
The second regression analysis (Model 2) evaluates whether the rate of ED visits associated with worse self-rated health is independent of neighborhood frisk concentration. We expect that, among those respondents with the same reported health profile, living in zip codes where police are more likely to frisk during stops will be associated with an increased rate of visiting the ED (Hypothesis 3).
Essentially, self-rated health and neighborhood frisking likelihoods are separate sources of variation in ED utilization behavior.

The third regression analysis (Model 3) evaluates an effect mod-
ifier-that is, whether the number of visits to the ED in the past year is associated with exposure to police frisking behavior in the neighborhood is different according to lay reports of health. We expect that respondents of worse health who live in zip codes where police engage more frequently in frisking more often will be less likely to visit the ED (Hypothesis 4). Confirmation of Hypothesis 4 would evidence that neighborhood frisk concentration represses, or buffers against, ED utilization.
Incidence rate ratios are shown, and 95% confidence intervals are provided in brackets. Estimates shown above one (1) indicate higher rates of ED visits, while estimates below one indicate lower rates of ED visits. Estimates of control covariates are omitted from tabulated results but are available upon request.  Panel B of Table 2 employs a nonlinear functional form for selfrated health that distinguishes the rate of ED visits for two status groups-those reporting "poor" or "fair" health and those reporting "excellent," "very good," or "good" health.   Abbreviations: E, Excellent; F, Fair; G, Good; P, Poor; SRH, Self-Rated Health; VG, Very Good.

| RE SULTS
*P < .05, **P < .01, ***P < .001 (two-tailed test for statistical significance).  We conclude by considering the ethical and future research implications of the results for the behaviors of health care professionals and providers. Our findings highlight the health care consequences of two interspersed sources-embeddedness in a legally sanctioned highly surveilled community and reliance on the medical community for access to the sick role. However, most health care administrators and professionals do not draft intake protocols in collaboration with community members who understandably fear medical intervention and engage in self-isolating practices in the presence of and as a result of localized police presence. Our results, then, provide support for the utility of empirical inquiry into social medicine, which calls medical professions to go into the community instead of waiting for its residents to pass through their doors. 55,56 Process and impact evaluations could assess the merits of these processual remedies, including a dedicated publicly funded investment in regulating and supporting the viable operation of these centers, the patients who would rely on their services, and the law enforcement agents responsible for the safety of all parties moving through those sites. A legal epidemiological framework for imagining the law's potential to positively shape these health-promoting possibilities merits scholarly and political attention.

| D ISCUSS I ON
Namely, explicit policy and protocols that limit police presence inside the emergency and trauma care setting could reduce hospital utilization disparities. We also recommend a committed consideration of alternatives to police presence that not only reduce the routine disruptions that are linked to criminal legal system intrusion, but allow for the cultivation of community-based, community-built, and community-directed health and wellness promoting institutions. [57][58][59] Such alternatives must engage health care professionals to remove their role in the criminalization process. Finally, we advance the belief that people know what they need. Yet, the sickest among them who know they are in need must also weigh that urgency against a reasonable desire to keep themselves and those dearest to them, alive. This calculation is not a choice that anyone should consider, let alone routinely make.

ACK N OWLED G M ENTS
Joint Acknowledgment/Disclosure Statement: The National Historical Geographic Information System (NHGIS) dataset is available free of charge, but there are access conditions that limit its usage. In order to access NHGIS data, Sewell registered to establish an account with the data host, IPUMS, and agreed to abide by their usage license conditions. As a registrant, Sewell is permitted to publish a subset of the data to meet journal requirements for accessing data related to a particular publication.