Association Between Acute Exposure to Crime and Individual Systolic Blood Pressure

Abstract

Introduction:

Hypertension is associated with adverse cardiovascular outcomes and is geographically concentrated in urban underserved neighborhoods. This study examines the temporal–spatial association between individual exposure to violent crime and blood pressure.

Methods:

A retrospective observational cohort study analyzed 39,211 patients with 227,595 blood pressure measurements from 2014 to 2016 at 3 outpatient clinics at an academic medical center in Chicago. Patients were included in the study if they had documentation of blood pressure in the medical record and resided in census tracts with >1,000 observations. Geocoded violent crime events were obtained from the Chicago Police Department. Individual-level exposure was defined on the basis of spatial and temporal buffers around each patient’s home. Spatial buffers included 100-, 250-, 500-, and 1,000-meter disc radii, and temporal buffers included 7, 30, and 60 days preceding each outpatient appointment. Systolic blood pressure measurements (mmHg) were abstracted from the electronic health record. Analysis was performed in 2019–2020.

Results:

For each violent crime event within 100 meters from home, systolic blood pressure increased by 0.14 mmHg within 7 days of exposure compared with 0.08 mmHg at 30 days of exposure. In analyses stratified by neighborhood cluster, systolic blood pressure increased by 0.37 mmHg among patients in the suburban affluent cluster relative to that among those in an extreme poverty cluster for the same spatial and temporal buffer.

Conclusions:

Exposure to a violent crime event was associated with increased blood pressure, with gradient effects by both distance and time from exposure.

INTRODUCTION

Hypertension is associated with lower socio-economic status (SES) and is often concentrated in underserved communities.¹ Areas at increased risk for developing hypertension have higher concentrations of poverty, fewer opportunities for physical activity, less social cohesion, and less neighborhood safety than areas with better environments.^2-4 In a 2016 study, lower perceptions of neighborhood safety were associated with higher blood pressure (BP).³ Similarly, a 2012 study found that people who reported that their neighborhood was unsafe had higher odds of delaying prescription refills.⁵ However, longitudinal studies have shown mixed associations between violent crime and BP. A multicenter cohort study of nearly 3,500 adults found no association between perceived neighborhood safety and hypertension after 10 years of follow-up,⁶ and a 2018 analysis of the same cohort found that increased police-recorded crime was conversely associated with reductions in BP among women, with high variability based on neighborhood size.⁷ Some of these inconsistencies can be attributed to survey data, which rely on subjective measures of violent crime, such as perceived safety. However, because geospatial analysis has become more readily accessible, objective measures of violent crime have emerged to supplement self-reported data.⁸

In 2018, a cross-sectional study of 14,799 adults found that patients in census tracts with the highest rates of violent crime had 25% higher odds of having elevated BP than those in tracts with the lowest rates.⁹ Another recent study examined violent crime rates in 295 unique census tracts in Chicago during the 2014–2016 crime surge. At the population level, a 20-unit increase in the violent crime rate during this surge was associated with 3% higher odds of elevated BP and 8% higher odds of missed outpatient appointments.¹⁰ Moreover, when comparing census tracts with low baseline crime rates relative to those with high baseline crime rates, the same 20-unit increase in violent crime was associated with 5% higher odds of elevated BP in tracts with low baseline crime rates. These findings suggest that communities experiencing lower baseline crime may be more sensitive to changes in exogenous crime incidents, whereas those with higher baseline crime rates may have already adapted and are therefore less sensitive to dynamic change. Previous work has shown neuroendocrine responses to stressful events may affect BP control in the short and long term. Because patients’ chronic and acute exposure to nearby violent crime events varies, acute and chronic neuroendocrine responses may vary as well and may relate to patients’ level of crime exposure. However, no studies, to the authors’ knowledge, have examined whether individuals who have been exposed to a violent incident near their home experience dynamic changes in BP soon after the incident.

The purpose of this study is to examine the relationship between spatiotemporal proximity to violent crime and dynamic changes in BP at the patient level. The hypothesis is that a recent violent incident occurring near a patient’s home would cause a greater BP increase than a remote violent incident occurring farther away. A secondary objective includes an examination of how the relationship between crime incidents and BP varies by neighborhood type.

METHODS

Study Sample

A retrospective longitudinal cohort study was performed analyzing 39,211 adult patients with 227,595 BP measurements from 2014 to 2016. All BP measurements were extracted from the electronic health record (EHR) of 3 outpatient clinics at an academic hospital in Chicago. Longitudinal EHR data were paired with longitudinal crime data from the City of Chicago Police Data Portal, enabling a comparison of multiple BP measurements across varying levels of violent crime exposure. For example, an individual patient may have had 8 BP measurements over the 3-year study period, with 8 different levels of violent crime exposure preceding each measurement (e.g., 3 violent crimes within 100 meters from home and 7 days before BP measurement). These data were obtained through routine clinical care agnostic to the study design.

Patients were included if they had documentation of BP in the EHR and resided in a census tract with >1,000 observations. Because this longitudinal study measured changes in BP over time, at least 1 repeated BP measurement had to occur within 120 days after the first BP measurement. Each patient’s residential address was obtained from administrative data in the EHR; patients with no listed address, patients with >1 address, and patients who changed addresses during the study period were excluded. This study followed the STROBE reporting guidelines for observational studies and was conducted with approval from the University of Chicago IRB.

Measures

The primary exposure variable was a violent crime incident, obtained from the City of Chicago Police Data Portal, which contains a geocoded record of all unique crime incidents in Chicago.¹¹ Violent crime was defined by the Chicago Police Department’s Crime Type Categories and included assault, battery, domestic sexual assault, robbery, and homicide. Crimes were included from January 1, 2014 to December 30, 2016. First, total violent crime incidents were geocoded, mapped, and aggregated within a series of spatial buffers around patient residence locations and indexed by the time of crime incident in daily intervals. Each spatial buffer was calculated as a disc of increasing radius from each patient’s residential address and included 100-, 250-, 500-, or 1,000-meter intervals. Second, violent crime incidents were counted within a series of time lag buffers. Each time lag was measured as the number of days before each BP measurement and included 0–7, 0–30, and 0–60 days.

The primary outcome variable was a change in individual systolic BP (SBP). All BP measurements were taken at the time of outpatient clinic appointments. Although SBP, diastolic BP, and heart rate have all been shown to change in response to acute psychological stress, previous studies have shown SBP to be the best of the 3 factors at predicting future cardiac events.¹² As such, SBP was used as the primary outcome variable.

Covariates extracted from patient health records included sex, age, BMI, primary insurance status, and BP medication status. To account for potential social and economic confounders, spatial neighborhood clusters designed by Kolak et al.¹³ were used. By aggregating the 5-year mean of 15 different socioeconomic variables at the census tract level in a national-scale analysis, Kolak and colleagues¹³ developed 7 neighborhood typologies that reflect differing dimensions of social determinants of health (SDH), including residential patterns of socioeconomic advantage and aspects of mobility, urban core opportunity, social cohesion, and accessibility (summarized in Appendix Table 1, available online). Of the 7 typologies found for the continental U.S., 5 were found in the study area (although all typologies were present in Chicago). Within the study area, the extreme poverty cluster is characterized by a high proportion of residents with minority status, low income, high poverty, and high unemployment. By contrast, the suburban affluent cluster is characterized by high income, a high proportion of children and vehicles per adult, and low poverty.

Each SDH index was independently associated with age-adjusted neighborhood mortality after controlling for spatial size and violent crime in Chicago (R² =0.63, p<0.001). During modeling, there was significant multicollinearity with neighborhood cluster type after adding race as a covariate. Therefore, because neighborhood clusters have a validated association with SDH and premature mortality, omitting race as a single covariate decreased collinear regression without sacrificing measurable effects of hidden SDH.¹³ Time effects for month and year were also included to account for seasonal variation and crime trends, respectively.

Statistical Analysis

For primary analyses, linear random effects mixed models were used to assess the relative change in SBP from the initial visit as a function of the number of crime incidents within each spatial–temporal buffer, adjusting for patient characteristics and neighborhood cluster. All models implemented robust SEs. Continuous variables were expressed as means (SDs) or median (IQRs) depending on normality, and categorical variables were expressed as frequencies and percentages. A multilevel mixed-model regression model (at the measurement level and clustered by patient) assessed change in SBP as a function of crime count per spatial–temporal buffer, using random effects for patient differences, fixed effects for months to account for seasonal variation, and robust SEs for spatiotemporal buffers, adjusting for patient characteristics and neighborhood characteristics as described earlier. Buffers were tested separately in sequential models to test for sensitivity of parameter specification.

To account for statistical interaction between neighborhood clusters, a second linear model that included an interaction term for both neighborhood cluster type and spatial–temporal buffer was used. Cluster 4, extreme poverty, was used as the reference category because this cluster was the most common in this data set. Cases with missing covariates were deleted casewise. Statistical significance was defined as a p<0.05 for 2-tailed tests. Data were analyzed using Stata/SE, version 13.1, and R, version 3.5.1.

RESULTS

The original sample included 79,283 patients. From the original sample, 12,318 (15.5%) did not have documentation of BP during their clinic appointment and were excluded. An additional 25,938 (32.7%) patients were excluded for residing in census tracts with <1,000 outpatient observations, largely reflecting those outside the immediate medical center catchment zone. Several patients had missing covariates: 1 patient was excluded for no documented insurance, 2 more patients were excluded for no reported sex, and 1,803 patients (2.2%) did not have a documented BMI.

The final sample included 39,221 patients with 227,595 BP measurements (Table 1). The median SBP was 130 mmHg (IQR=117–143 mmHg), and the median diastolic BP was 74 mmHg (IQR=67–82 mmHg). The mean age was 60 years (IQR=40–75 years). Most patients were aged >65 years (42.5%), were African American (71.6%), had a BMI <30 (61.5%), and were insured by Medicare (47.1%). Most patients (50.8%) resided in neighborhood Cluster 4, extreme poverty. Generally, the average number of crimes increased as the spatial radius from residence increased and as the time to an outpatient appointment increased (Appendix Table 2, available online). Patients who were excluded for the reasons mentioned earlier were more likely to be White and to have private insurance (Appendix Table 3, available online) than those included.

Table 1.

Baseline Characteristics of Study Population

Characteristics	N=227,595, n (%)	Median (IQR)
Age, years		60.00 (40.00, 74.00)
18–24	47,371 (20.8)
35–49	31,404 (13.8)
50–64	51,985 (22.8)
≥65	96,835 (42.5)
Sex
Male	73,694 (32.4)
Female	15.3901 (67.6)
Race
White non-Hispanic	43,705 (19.2)
Black non-Hispanic	16.3023 (71.6)
Hispanic or Latino	5,840 (2.6)
Other	12,719 (5.6)
Unknown	2,308 (1.0)
Insurance status
Private	37,896 (16.7)
Medicaid	27,130 (11.9)
Medicare	107,264 (47.1)
Managed care	2,411 (1.1)
Other	52,894 (23.2)
BP, mmHg
Systolic BP <140 and diastolic BP <90	151,631 (66.6)
Systolic BP ≥140 or diastolic BP ≥91	75,964 (33.4)
Median systolic BP		130.00 (117.00, 143.00)
Median diastolic BP		74.00 (67.00, 82.00)
HR, bpm
Normal (HR ≤100)	210,866 (92.5)
Tachycardia (HR >100)	16,729 (7.5)
Median HR		78.00 (69.00, 88.00)
BMI, kg/m2
Not obese (BMI <30)	139,938 (61.5)
Obese (BMI ≥30)	87,657 (38.5)
Median BMI		27.74 (23.63, 33.21)
Neighborhood cluster
Rural affordable	24,335 (10.7)
Vibrant urban core	6,507 (2.9)
Extreme poverty	115,554 (50.8)
Suburban affluent	78,639 (34.6)
Sparse areas	2,560 (1.1)
Medications prescribed
Beta-receptor antagonist	54,313 (23.9)
ACE/ARB receptor antagonist	57,548 (25.3)
Aldosterone antagonist	8,753 (3.8)
HMG-CoA reductase inhibitor	67,805 (29.8)
Antiplatelet	29,425 (12.9)
Anticoagulation	19,613 (8.6)
Insulin	20,058 (8.8)
Calcium channel antagonist	51,353 (22.6)
Baseline laboratory values
Creatinine		1.00 (0.80, 1.20)
Brain natriuretic peptide		727.00 (164.00, 3,325.00)
Platelets		237.00 (194.00, 287.00)
Sodium		140.00 (138.00, 142.00)
Hemoglobin		12.40 (11.10, 13.50)
Potassium		4.10 (3.80, 4.40)

ACE/ARB, angiotensin-converting enzyme inhibitor or angiotensin receptor blocker; BP, blood pressure; bpm, beat per minute; HMG-CoA, hydroxymethylglutaryl-CoA; HR, heart rate.

For violent crimes occurring within 7 days before an appointment, most patients (62.7%) were exposed to 0 violent crimes at the smallest spatial buffer of 100 meters; the remaining 37.3% were exposed to a median of 1 violent crime (range=1–11 violent crimes). For each violent crime occurring within this smallest buffer, SBP increased by 0.14 mmHg (95% CI=0.0015, 0.2845, p=0.048) per violent crime. At a larger buffer radius of 500 meters, SBP increased by 0.05 mmHg (95% CI=0.0251, 0.0769, p<0.001) per violent crime (Table 2). For patients living in neighborhood Cluster 6, suburban affluent, each violent crime occurring within the smallest buffer (i.e., 100 meters and 7 days) was associated with an increase in SBP of 0.37 mmHg (95% CI=0.4910, 0.6913, p=0.024) relative to that for patients living in neighborhood Cluster 4, extreme poverty, which was the reference group for the analysis. At 250 meters and 7 days, each crime incident was associated with an SBP increase of 0.21 mmHg (95% CI=0.0796, 0.3335, p=0.001) relative to that for the reference group. Finally, at 500 meters and 7 days, each exposure was associated with an SBP increase in 0.16 mmHg (95% CI=0.1031, 0.2213, p<0.001) relative to that for the reference group (Table 3).

Table 2.

SBP Change Per Buffer

Radius from home, meters	Time to next appointment, days	SBP increase, mmHg	p-value	95% CI
100	7	0.14	0.048	0.0015, 0.2846
250	7	0.19	0.69	−0.0356, 0.0741
500	7	0.05	<0.001	0.0251, 0.0769
1,000	7	0.03	<0.001	(0.0138, 0.0376
100	30	0.08	0.019	0.0136, 0.1491
250	30	0.02	0.053	−0.0003, 0.4784
500	30	0.02	<0.001	0.0127, 0.0329
1,000	30	0.08	0.019	0.0136, 0.1491
100	60	0.04	0.091	−0.0063, 0.0847
250	60	0.02	0.006	0.0059, 0.0359
500	60	0.01	<0.001	0.0081, 0.0198
1,000	60	0.01	<0.001	0.0041, 0.0085

Note: Boldface indicates statistical significance (p<0.05).

SBP, systolic blood pressure.

Table 3.

SBP Increase by Spatial–Temporal Buffer in Suburban Affluent Neighborhood Versus That in the Extreme Poverty Reference Group

Radius from home, meters	Time to next appointment, days	SBP increase, mmHg	p-value	(95% CI)
100	7	0.37	0.024	(0.0491, 0.6913)
250	7	0.21	0.001	(0.0796, 0.3334)
500	7	0.16	<0.001	(0.1030, 0.2213)
100	30	0.28	<0.001	(0.1311, 0.4368)
250	30	0.11	<0.001	(0.0499, 0.1602)
500	30	0.06	<0.001	(0.0426, 0.0871)

Note: Boldface indicates statistical significance (p<0.05).

SBP, systolic blood pressure.

For violent crimes occurring within 30 days before an appointment, patients remained exposed to a median of 1 violent crime, but the range increased (range=1–20 violent crimes). For each violent crime occurring within 100 meters from home, SBP increased by 0.081 mmHg (95% CI=0.0136, 0.1490, p=0.019). At a larger buffer radius of 500 meters, SBP increased by 0.023 mmHg (95% CI=0.0127, 0.0329, p<0.01) for each violent crime (Table 2). For patients living in neighborhood Cluster 6, suburban affluent, each violent crime within 100 meters and 30 days was associated with an increase of 0.28 mmHg (95% CI=0.1311, 0.4368, p<0.01) in SBP relative to that for the reference group. At 250 meters, each individual crime event was associated with an increase in SBP of 0.10 mmHg (95% CI=0.0498, 0.1602, p<0.01) relative to that for the reference group. Finally, at 500 meters, individual crime events were associated with an increase in SBP of 0.06 mmHg (95% CI=0.0426, 0.0871, p<0.01) relative to that for the reference group (Table 3).

The association between violent crime and change in BP is shown graphically in Figure 1A for various combinations of spatial and temporal buffers. In general, crimes occurring close to a patient’s home and shortly before a clinic appointment were associated with higher SBP (shown as the large spike in the graphic). This effect diminished as spatial and temporal buffers increased. As shown in Figure 1B, patients living in suburban affluent clusters demonstrated an exaggerated association between crime and BP compared with the population overall. The graphs for patients living in other clusters are shown in Appendix Figures 1-3 (available online). Generally, patients living in neighborhood clusters that have better resources experienced a greater BP increase per crime incident experienced for the same spatial–temporal buffer than those living in all neighborhood clusters.

Figure 1. SBP change (mmHg) per spatiotemporal buffer (A) for all patients and (B) for the suburban affluent cluster.

Note: The association between violent crime and change in blood pressure is shown graphically in Figure 1A for various combinations of spatial and temporal buffers. In general, crimes occurring close to a patient’s home and shortly before a clinic appointment were associated with higher SBP (shown as the large spike in the graphic). This effect diminished as spatial and temporal buffers increased. As shown in Figure 1B, patients living in suburban affluent clusters showed an exaggerated association between crime and blood pressure compared with the population overall.

SBP, systolic blood pressure.

DISCUSSION

In this sample of urban-dwelling adults, this study found that nearby and recent exposure to violent crime was associated with an increase in individual SBP. This study adds to the current literature by quantifying the longitudinal associations between nearby violent crime events and acute changes in individual BP and by showing a gradient effect of violent crime: patients who experienced crime more recently and closer to their residence have a larger change in SBP.

This study adds to the current understanding in several ways. First, observed changes in individual BP support evidence of an acute physiologic stress response to violent crime at the patient level. These findings are consistent with allostatic load theory, which describes the accumulation of wear and tear after repeated or chronic activation of acute stress responses through the hypothalamo–pituitary–adrenal axis.¹⁴ In support of this theory, this study shows a graded impact of closer and more recent crime exposure. By contrast, violent crimes in the distant past demonstrated a waning effect on SBP over time, suggesting a degree of physiologic resiliency if patients are given enough time and space to buffer these events. Psychological stress can trigger excessive concentrations of cortisol and catecholamines, which impair glucose metabolism and exacerbate vascular reactivity through oxidative mechanisms.¹⁵ These findings are consistent with those of previous work that showed that various types of crime exposure may be associated with clinical and laboratory risk factors for cardiovascular disease, including higher BP and heart rate, elevated C-reactive protein, metabolic syndrome, and obesity.^16-20

Finally, this study shows that individual crime events occurring in affluent neighborhoods exert a greater increase in individual SBP than events experienced in poorer neighborhoods. Similarly, patients living in higher-SES neighborhood clusters experienced a greater increase in SBP for each violent crime within the same spatiotemporal buffer than those living in lower-SES neighborhood clusters.¹⁰ It is possible that having a baseline experience of and thus expectation for violent crime may confer a blunted response to acute stress.

Previous research showed that surges in violent crime rates for high-crime census tracts are associated with a 2.5 mmHg increase in BP, which is similar to the effect in observational cohorts evaluating the effects of dietary salt intake on BP.^10,21 Although the effect shown in this study at the individual level is predictably smaller, these findings show for the first time that even a single violent crime has a measurable impact on nearby patients’ BP. Because these effects are cumulative, a patient experiencing ≥3 crimes together would likely have a hypertensive excursion by an order of magnitude on par with dietary indiscretion depending on their neighborhood of residence. Importantly, although BP overall tends to be less well controlled in high-crime neighborhoods, the additional impacts of a single violent crime in these neighborhoods had a far smaller impact on BP than did a single crime in more affluent neighborhoods that less frequently experience violent crime.

Limitations

This study had several limitations. Some degree of ecologic fallacy is possible by assigning neighborhood-level characteristics to individual outcomes. Initial analysis included individual covariates such as patient race, but these models showed unacceptably high multicollinearity with neighborhood clusters. Individual-level race may have a stronger association with individual BP readings than the neighborhood-level race/ethnicity variable embedded in the neighborhood typologies. Although each patient’s activity space (routine locations a person visits throughout daily life) may be a more realistic measure of exposure to neighborhood violence, measurement of activity space is challenging and proved unreliable using the data available for this study. Furthermore, 15.5% of the initial sample were excluded for missing SBP, and 32.7% of patients were excluded owing to living in census tracts with <1,000 observations. Although the demographics of the excluded patients differed somewhat from those included (Appendix Table 3, available online), the stability of the modeling depended on the large number of observations in each census tract. Future studies could focus a similar analysis on regions farther from this medical center. This center serves a largely urban population, and the neighborhood typologies in this study reflect that reality. The distribution of typologies in this study differs from those of Chicago and those of the U.S. (Appendix Table 4, available online); the authors view this fact as a strength of this study because this population was enriched with a high percentage of underserved patients who were often exposed to violent crime. This may however limit the generalizability of the study. Exposure to crime was dependent entirely on police reporting, which could precipitate differential misclassification of exposure. Although this data set contained too few instances of International Classification of Diseases codes corresponding to assault, battery, or sexual assaults for thorough analysis, future study could examine the impacts of personal victimization or witness of violent crime on BP. BP measurements obtained in the clinic may not mirror BP levels in the ambulatory setting—future work could involve a similar analysis using ambulatory BP monitoring to account for white-coat hypertension or masked hypertension. Smoking and other lifestyle behaviors were also not considered in this study.

CONCLUSIONS

In Chicago, nearby and recent violent crimes are associated with an acute rise in BP. Although BP control overall is worse in high-crime neighborhoods, the acute impact of a single violent crime on BP is greater for patients in more affluent neighborhoods who are less often exposed to crime.