Relationships Between Opioid-Related Hospitalizations and Intimate Partner Violence and Child Maltreatment Hospitalizations in Pennsylvania Across Space and Time

Abstract

Intimate partner violence (IPV) and child maltreatment outcomes are markedly associated with substance abuse disorders. However, few studies have explored these serious family violence outcomes in connection to the opioid epidemic or population-level geographic connections between these variables. This study assesses associations of ZIP code-level IPV and child maltreatment hospitalization outcomes with opioid- and alcohol-related diagnoses as well as economic and demographic neighborhood characteristics. We used 11 years (2004–2014) of ZIP code-level Pennsylvania hospital discharge data and U.S. Census neighborhood characteristics data. As nearby ZIP codes are more likely to be similar than those that are distant, we incorporated spatial autocorrelation using conditionally autoregressive Bayesian hierarchical space–time models. There was a positive relationship between ZIP code-level opioid-related diagnoses and both IPV (relative risk 1.061; 95% credible interval [1.015, 1.106]) and child maltreatment (relative risk 1.055; 95% credible interval [1.035, 1.070]) hospitalizations. There was a positive relationship between alcohol-related diagnoses and IPV but not child maltreatment. Higher median household incomes were associated with lower counts of both IPV and child maltreatment hospitalizations. To illustrate geographic heterogeneity of model estimates, posterior distributions were used to compare variability of effects across ZIP codes. Our findings emphasize the secondary implications of the opioid epidemic in the form of family violence within communities.

Introduction

Intimate partner violence (IPV) and child maltreatment are serious and highly prevalent public health issues (Kim et al., 2017; S. G. Smith et al., 2018; U.S. Department of Health & Human Services, 2020). It is well-known that substance use and substance use disorders are connected to both IPV (Kraanen et al., 2010; Stith et al., 2004) and child maltreatment (Chaffin et al., 1996; Dubowitz et al., 2011; Ondersma, 2002). However, differential associations are apparent across drug types, with alcohol being the most commonly used substance associated with both perpetration and victimization (Kraanen et al., 2014; P. H. Smith et al., 2012). In response to steeply rising rates of opioid use disorders (OUDs), opioid poisoning and overdose, and associated mortality (Kochanek et al., 2017), the connections between opioid misuse and IPV and child maltreatment are increasingly important to understand.

There is evidence that both IPV and child maltreatment are associated with opioids, though these relationships vary across violence type, population, and measure of opioid use. IPV victimization among women has been consistently shown to be connected to OUDs (P. H. Smith et al., 2012; Stone & Rothman, 2019; Pallatino et al., 2019). Jessell et al. (2015) found a connection between nonmedical use of prescription opioids and heroin and increased sexual violence victimization among young adults; this association was stronger for women but was present for men as well. Conversely, Williams et al. (2020) found a stronger association between opioid use and IPV victimization among males than females after accounting for sexual assault and childhood trauma. Another study found that opioid-dependent fathers exhibited more physical, psychological, and sexual aggression toward the mothers of their youngest children than a control group (Moore et al., 2011). Opioid-related hospitalization rates are related to higher report rates of child maltreatment and substantiated reports (Ghertner et al., 2018). Higher opioid prescribing rates are associated with greater risk for substantiated child abuse (Morris, Marco, Bailey, et al., 2019) and child removals (Ghertner et al., 2018; Quast et al., 2018, 2019). Prescription opioid overdoses were positively associated with child maltreatment and injury hospitalizations in California communities (Wolf et al., 2016).

IPV, child maltreatment, and opioid misuse trends all exhibit considerable geographic heterogeneity. Specific characteristics of neighborhood environments, such as economic disadvantage, alcohol outlet densities, and social disorganization, are associated with neighborhood IPV rates (Cunradi et al., 2013; Fox & Benson, 2006; Miles-Doan & Kelly, 1997; Miles-Doan, 1998). Even after accounting for these factors, spatial patterning of IPV often remains (Gracia et al., 2015). Likewise, neighborhood characteristics, such as poverty and unemployment, are associated with child maltreatment risk (Coulton et al., 2007; Freisthler et al., 2006; Maguire-Jack et al., 2015; Molnar et al., 2016). Spatial patterns of substantiated cases of child abuse and neglect vary across ZIP codes (Morris, Marco, Maguire-Jack, Kouros, Bailey, et al., 2019) and counties (Morris, Marco, Maguire-Jack, Kouros, Im, et al., 2019; B. D. Smith et al., 2017). Spatial clustering (Brownstein et al., 2010), geographic heterogeneity, and associations with specific area characteristics (Cerdá et al., 2013; Green et al., 2011) are also apparent for opioid misuse.

As previously mentioned, opioid misuse is connected to both IPV and child maltreatment; furthermore, geographic heterogeneity and spatial patterning of each of these conditions exist. However, little is known about the relationship between population-level geographic distributions of opioid misuse and these two types of violence. To capture the nuances of geographic heterogeneity in distributions of IPV and child maltreatment, analyses must be carried out at an appropriate spatial resolution. For example, data available at the state or county level may not be sufficiently resolved to capture local conditions. While representative of more severe cases of family violence, hospital discharge data (HDD) are a source of consistently collected records at the ZIP code level, which can be used to obtain counts of IPV and child maltreatment, as well as diagnoses related to opioid and alcohol misuse. The present study assesses the ZIP code-level relationships between hospitalizations for two forms of violence, IPV and child maltreatment, and opioid-related hospitalizations in Pennsylvania. We conducted spatiotemporal analyses using 11 years of HDD, adjusting for demographic and environmental covariates and substance use disorder patterns.

Method

Data Sources and Variables

We used Pennsylvania Health Care Cost Containment Council (Pennsylvania Health Care Cost Containment Council, 2014) hospital discharge data from 2004 through 2014 to obtain patient-level diagnoses for hospitalizations that required an overnight stay. We excluded 951,478 (4.6%) of 20,840,763 observations as they had ZIP codes that were either missing our outside of Pennsylvania. Diagnoses that appeared as primary, secondary, and/or E-codes (external causes of injuries and poisonings codes) were included. Counts were aggregated to the patient ZIP code level. There were 16,275 space–time polygons across the 11 years.

Diagnoses relevant to IPV included the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes for adult maltreatment, abuse, and neglect (995.80–995.85) and the E-code for the perpetrator of child and adult abuse being a spouse or partner (E967.3). For counts of child maltreatment, relevant codes included: child maltreatment syndrome (995.5x), E-codes for which the perpetrator of child and adult abuse was the father, stepfather, or boyfriend (E967.0), the mother, step-mother, or girlfriend (E967.2), or a grandparent (E967.6).

Opioid-related diagnosis codes were as follows: opioid abuse (305.5x), dependence (304.0x), dependence with combinations of opioids with other drugs (304.7x); poisoning by heroin (965.01), accidental heroin poisoning (E850.0); and nonheroin opioid poisoning (965.00, 965.02, 965.09, E850.1, E850.2). Counts of unique diagnoses for each space–time unit were then divided by each unit’s hospitalization count to obtain rates per 100 hospitalizations. We also calculated temporally lagged opioid-related hospitalization rates. As the shape and number of ZIP codes were inconsistent across time, we matched ZIP codes to their respective previous-year ZIP codes based on greatest area overlap.

Codes related to alcohol use disorders (AUDs) included acute alcoholic intoxication (303.0x), nondependent alcohol abuse (305.0x), other and unspecified alcohol dependence (303.9x), and E-codes accidental poisoning by alcoholic beverages (E860.0), accidental poisoning by other and unspecified ethyl alcohol and its products (E860.1), accidental poisoning by other specified alcohols (E860.8), and accidental poisoning by unspecified alcohol (E860.9). These codes were denominated by the total hospitalization count by each ZIP code and year.

Additional diagnostic ICD-9-CM codes were aggregated for outcomes used in sensitivity analyses. ICD-9-CM E-codes for motor vehicle traffic accidents included E81x.x. For influenza, we used ICD-9-CM codes 487.xx and 488.xx. Codes included for falls were E88x.x. We also included the overall hospitalization rate to control for each ZIP code’s inpatient care access.

We used GeoLytics projections (GeoLytics, 2014) to create annual estimates of demographic variables. As these variables were available at the Census block group rather than the ZIP code level, we aggregated them to year-specific ZIP code boundaries using a method successfully employed in prior analyses (Mair et al., 2013). Variables included three economic conditions: median household income (per US$10,000), the percent of the population below 150% of the poverty line, and the percent of the population that was unemployed. Other demographic conditions included racial distributions (percent Black, Hispanic, and non-Hispanic White), age distributions (percent ages 0–19, 20–24, 25–44, and 65 or older), and percent male. To reflect qualitatively differing levels of urbanicity, population density was categorized into quintiles (0–52.5, 52.6–124.1, 124.2–333.3, 333.4–1414.2, and 1414.3+ people/square mile).

To represent each ZIP code’s overall retail environment, we aggregated counts of establishments with North American Industry Classification System (NAICS; NAICS, U.S. Census Bureau, 2014) codes for Retail Trade (Sectors 44–45) and Accommodation and Food Services (Sector 72). These counts were then converted to densities per square mile. Finally, we included a misalignment measure of the geographic instability of each ZIP code’s population between consecutive years (Mair et al., 2013). This measure was calculated as the percent of the 2000 Census block-level populations within each year-specific ZIP code that would have fallen outside of the boundaries of the previous year’s best-matched ZIP code.

As 83 of the 16,275 year-specific ZIP codes had population counts of zero, they were assigned a population of five to allow for nonzero population risk in all areas. These ZIP codes were then assigned mean values by year for all variables originating from the Census.

Data Analysis

We used Bayesian hierarchical space–time misalignment models for all analyses. In contrast to standard frequentist approaches, Bayesian analyses estimate posterior distributions of effects. As such, for each effect, we estimate a 95% credible interval, which reflects the points between which 95% of each effect’s posterior distribution falls. Although sharing an acronym with confidence intervals, credible intervals are conceptually different. We also present the median value of the posterior distribution, which roughly corresponds to a frequentist point estimate.

As the shape and number of ZIP code polygons commonly vary with time (e.g., in Pennsylvania, there were 1,469 ZIP codes in 2004 and 1,490 in 2014), we used an approach that addresses this spatial misalignment (Zhu et al., 2013). Observation-level conditionally autoregressive (CAR) spatial effects as well as nonspatial random effects were included. All fixed and random effects were assigned noninformative priors.

As all outcomes were counts, Poisson regression was used:

$$Y_{i,t}\mid {\mu }_{i,t}~\mathit{\text{Poisson}}\left(E_{i,t}\text{exp}\left({\mu }_{i,t}\right)\right).$$

For each spatial unit i at time t, Y_i,t refers to the count of each outcome, while E_i,t is the expected count, calculated using the assumption that hospitalizations across ZIP codes are distributed in direct proportion to population size.

The log relative rate (RR), μ_i,t, for each space–time unit is as follows:

$${\mu }_{i,t}=\beta X_{i,t}+{\varphi }_{i,t}+{\theta }_{i,t}.$$

β is a vector of coefficient estimates for the corresponding matrix of observed fixed effects, X_i,t. ϕ_i,t is a vector of observation-level CAR spatial random effects, while θ_i,t is a vector of nonspatial random effects.

We used the R package R-INLA to estimate all results (Rue et al., 2009). We estimated separate models for IPV and child maltreatment hospitalization counts. To ensure that our findings were not correlates of overall hospitalization trends, in addition to including rates of overall hospitalizations, we conducted sensitivity analyses for the following outcomes, which we hypothesized would not be connected to opioid misuse: motor vehicle traffic accidents, influenza (in the overall population and among children only), and falls (in the overall population and among the elderly only). Each model adjusted for key demographic, economic, environmental, and substance use variables.

Results

There was a mean of 0.17 (SD = 0.57) IPV-related hospitalizations and 0.18 (SD = 0.61) child-maltreatment-related hospitalizations per ZIP code in Pennsylvania from 2004 through 2014 (Table 1). The mean number of hospitalizations related to IPV and child maltreatment per ZIP code increased from 0.14 to 0.17 and from 0.15 to 0.18, respectively, from 2004 to 2014. Spatial autocorrelation was noted for both IPV (2014 Moran’s I = 0.238) and child maltreatment (2014 Moran’s I = 0.212) at the ZIP code level.

Table 1.

Descriptive Statistics, ZIP Codes in Pennsylvania, 2004–2014 (n = 16,275 ZIP Codes).

Covariate	M	SD	Minimum	Maximum	Mean Change, 2014 vs. 2004
Number of hospitalizations with IPV diagnoses	0.17	0.57	0	10	0.05
Number of hospitalizations with child maltreatment diagnoses	0.18	0.61	0	9	0.03
Opioid-related diagnoses, per 100 hospitalizations	1.21	1.40	0	50	0.93
Alcohol use disorder-related diagnoses, per 100 hospitalizations	3.64	2.98	0	100	1.48
Overall hospitalization rate, per 100 people	18.83	54.48	0.00	1,780.00	−2.30
Economic conditions
Below 150% of the poverty level, %	21.33	10.03	0.00	86.65	3.50
Median household income (US$10,000)	4.48	1.67	0.00	18.46	0.859
Unemployment, %	7.42	5.24	0.00	100.00	2.30
Race/ethnicity, %
Black	4.28	11.77	0.00	98.01	−0.07
Hispanic	2.24	5.50	0.00	71.45	0.37
White	89.12	15.27	0.40	100.00	−6.27
Age, %
0–19	24.02	3.71	0.00	50.0	−1.69
20–24	6.60	1.34	0.00	35.5	0.19
25–44	24.81	3.56	0.00	77.4	−1.83
45–64	27.41	3.15	0.43	68.8	0.80
>65	17.16	4.35	0.00	89.17	2.52
Male, %	49.42	2.73	5.26	100.00	0.01
Population size	8,490.78	11,258.67	5.00	73,131.91	221.08
Population density, per square mile	1,278.49	3,111.11	0.10	36,466.28	21.50
Retail clutter, number of establishments per square mile (×10)	104.20	584.60	0.00	18,565.47	−6.49

Note. IPV = intimate partner violence.

There was substantial geographic heterogeneity in economic conditions across ZIP codes in Pennsylvania. The mean percentage of residents living below 150% of the poverty level ranged from 0 to 86.65, with a mean of 21.33 (SD = 10.03), while the median household income ranged from US$0 to US$184,588, with a mean of US$44,812 (Table 1). The mean percentage of unemployed residents per ZIP code was 7.42 (SD = 5.24). Pennsylvania has a small minority population; the racial/ethnic composition per ZIP code was, on average, 89.12% White during the study period (with a decrease of 6.27% from 2004 to 2014). The mean rate of opioid-related hospitalizations per 100 hospitalizations was 1.21 (SD = 1.40) per ZIP code, increasing by 0.93 per 100 from 2004 to 2014. There were, on average, 3.64 (SD = 2.98) hospitalizations related to AUD per 100 hospitalizations, with an increase of 1.48 per 100 during the study period.

Results from two models are displayed in Table 2. The first model has IPV hospitalizations as the outcome, while the second is an analogous model with an outcome of child maltreatment hospitalizations. As our analyses are Bayesian, the reported median RRs correspond to the 50th percentile of each effect’s posterior distribution; the lower and upper bounds of each 95% credible interval correspond to the values at which 2.5% and 97.5% of the posterior distribution fall, respectively. Effects are considered well-supported if at least 95% of their posterior distributions fall entirely above or entirely below 1.000. Concurrent opioid-related hospitalization rates were well-supported and positive in both models. One additional opioid-related diagnosis per 100 hospitalizations was associated with a median 6.1% increase in IPV-related hospitalizations (95% credible interval [CI] [1.015, 1.106]) and a 5.9% increase in child maltreatment-related hospitalizations (95% CI [1.012, 1.107]). Based on results from previous work using Pennsylvania Health Care Cost Containment Council data (Sumetsky et al., 2019), we ran additional models that also adjusted for the past-year opioid-related hospitalization rate (results not shown). When included, the past-year opioid-related hospitalization rate was not well-supported for either outcome. One additional AUD-related diagnosis per 100 hospitalizations was associated with a 5.5% increase in IPV hospitalizations per ZIP code, but was not well-supported in the child maltreatment model (median RR 1.016; 95% CI [0.985, 1.044]).

Table 2.

RRs (95% Credible Intervals) and ln(RR), IPV, and Child Maltreatment Hospitalizations, Bayesian Spatial Misalignment (n = 16,275 ZIP Codes) Models.

	Model 1: IPV	Model 2: Child Maltreatment
Covariate	RR [95% CI]	RR [95% CI]
Opioid-related diagnoses, per 100 hospitalizations	1.061 [1.015, 1.106]a	1.059 [1.012, 1.107]a
Alcohol use disorder-related diagnoses, per 100 hospitalizations	1.055 [1.035, 1.070]a	1.016 [0.985, 1.044]
Year	1.005 [0.985, 1.026]	1.034 [1.014, 1.054]a
Demographic characteristics
Economic conditions
Below poverty level, %	0.999 [0.991, 1.007]	1.007 [0.999, 1.015]
Median HH income (US$10,000)	0.821 [0.778, 0.866]a	0.759 [0.717, 0.803]a
Unemployment, %	0.996 [0.985, 1.008]	0.987 [0.975, 0.998]a
Race/ethnicity, %
Black	1.000 [0.990, 1.011]	1.003 [0.992, 1.014]
Hispanic	0.993 [0.983, 1.004]	0.995 [0.985, 1.005]
White	0.994 [0.983, 1.004]	0.997 [0.986, 1.008]
Age, %
0–19	1.015 [0.987, 1.044]	1.039 [1.011, 1.067]a
20–24	0.981 [0.938, 1.023]	0.996 [0.954, 1.037]
25–44	0.990 [0.965, 1.014]	0.982 [0.957, 1.007]
>65	1.016 [0.986, 1.047]	1.008 [0.978, 1.039]
Male, %	1.022 [0.997, 1.046]	0.972 [0.944, 1.001]
Population densityb
Quintile 2	1.180 [0.843, 1.688]	1.219 [0.925, 1.629]
Quintile 3	1.532 [1.116, 2.160]a	1.319 [1.011, 1.751]a
Quintile 4	1.843 [1.348, 2.590]a	1.463 [1.122, 1.940]a
Quintile 5	2.049 [1.478, 2.918]a	1.482 [1.114, 2.002]a
Overall hospitalization rate, per 100 people	1.005 [1.003, 1.006]a	1.004 [1.002, 1.005]a
Retail clutter, number of establishments per square mile (× 10)	1.000 [0.999, 1.000]	1.001 [1.001, 1.002]a
Misalignment effect: ZIP code instability	1.007 [0.986, 1.025]	0.986 [0.961, 1.007]
Random effects
Spatial random effects (s.d. CAR process)	0.415 [0.410, 0.420]	0.229 [0.226, 0.231]
ZIP code-level random effects (s.d.)	0.038 [0.009, 0.139]	0.371 [0.278, 0.513]
Spatial-to-total random variability ratioc	0.992 [0.900, 1.000]	0.276 [0.166, 0.404]
DIC	10,893.07	11,374.39

Note. RR = relative rate; HH = household; IPV = intimate partner violence; CI = credible interval; CAR = conditional autoregressive; DIC = deviance information criterion.

^aIndicates findings that are well-supported by the data as evidenced by credible intervals that exclude one for relative risks.

^bPopulation density was divided into approximately equal quintiles as follows: 1 (referent; <52.5 people/square mile), 2 (52.6–124.1 people/square mile), 3 (124.2–333.3 people/square mile), 4 (333.4–1414.2 people/square mile), and 5 (1414.3+ people/square mile).

^cCalculated as the approximate variance ratio of spatial to both spatial and nonspatial random effects.

Effects of opioid-related diagnoses were not positively well-supported in any sensitivity analyses, which assessed outcomes of motor vehicle traffic accidents, influenza, and falls. The AUD hospitalization rate was well-supported and positive in the motor vehicle accidents model, well-supported and negative in falls models, and not well-supported in the influenza models.

In both models, a higher median household income was associated with fewer counts of respective outcomes. Higher proportions of unemployment were connected to fewer child maltreatment hospitalizations; the relationship with area unemployment was not well-supported in the IPV model. None of the racial/ethnic distribution effects were well-supported. None of the age distributions were well-supported in the IPV model; however, a greater proportion of individuals between the ages of 0 and 19 years was associated with more child maltreatment hospitalizations (median RR 1.039; 95% CI [1.011, 1.067]). ZIP codes with greater population densities were associated with a greater number of IPV- and child maltreatment-related hospitalizations.

The spatial-to-total random variability ratio is a crude measure of the proportion of error that is spatially dependent as the spatial and spatially unstructured effects in CAR models are not directly comparable. However, the high ratio (median 0.992; 95% CI [0.900, 1.000]) noted in the IPV model suggests considerable spatial dependence, with less substantial spatial dependence noted in the child maltreatment model (median 0.276; 95% CI [0.166, 0.404]).

We used posterior distributions from the two models to visualize the contributions of fixed effects for specific covariates that were well-supported in either or both models across a range of ZIP codes (Figure 1). For each outcome, we selected the five ZIP codes with the lowest estimated rates and 10 ZIP codes with the 10 highest estimated rates (excluding ZIP codes with population counts below 100) in 2014. Each ZIP code-specific bar depicts the median RR contributions of a number of fixed effects. Effects connected to elevated risk are above the relative risk of 1, while those connected to reduced risk are below 1. In both the IPV and child maltreatment models, median household income tended to have a greater protective effect in the lowest risk compared with the highest risk ZIP codes. The opioid-related and AUD hospitalization rates contributed more to risk for IPV and child maltreatment in the highest risk ZIP codes compared with those with the lowest risk.

Figure 1.

RRs broken down by contributions of effects that are well-supported in either or both the IPV and child maltreatment models.

Note. Covariate-specific posterior effects for 2014 Pennsylvania ZIP codes with the five lowest RRs and the 10 highest RRs are presented for (A) IPV and (B) child maltreatment outcomes. RRs = relative rates; IPV = intimate partner violence.

Discussion

Our analyses indicate a positive association between hospitalizations related to opioid misuse and outcomes of IPV and child maltreatment. Hospitalizations with codes for both IPV and child maltreatment outcomes are linked to the opioid epidemic (Table 2). Other ecological similarities between the two outcomes (e.g., no well-supported racial or ethnic density associations) are present as well. Yet, our spatial models of IPV and child maltreatment are also notably distinct. For example, while both outcomes were positively associated with lower median household incomes, only child maltreatment was negatively connected to unemployment. Unsurprisingly, child maltreatment (but not IPV) was associated with a greater percentage of youth (ages 0–19 years) in a ZIP code. Finally, though 2014 Moran’s I values were fairly similar for both outcomes, a crude measure of spatial-to-total random variability suggested that the adjusted IPV model exhibited considerably more spatial dependence than the adjusted child maltreatment model. This indicates the importance of accounting for spatial dependence in studying these family violence outcomes, particularly IPV. Results from studies ignoring spatial autocorrelation may be erroneous.

To compare effects across ZIP codes, we evaluated posterior distributions for effects that were well-supported in either the IPV or child maltreatment model. To do this, we plotted median RRs of these effects for high- and low-risk ZIP codes based on overall model estimates (Figure 1). This allows us to visualize the relative magnitude of these effects across areas. For example, opioid- and AUD-related hospitalizations consistently contributed to the elevated IPV risk in ZIP codes with the highest RRs; conversely, contributions of these substance abuse conditions were slight for low-RR ZIP codes. Compared with ZIP codes with the highest RRs for IPV, those with the lowest RRs had far higher median household incomes. The same income dichotomy was evident in the child maltreatment model. Likewise, regions with the highest child maltreatment RRs were more heavily affected by opioid-related hospitalization effects. Yet, the effect of AUD-related hospitalizations (not well-supported in the child maltreatment model) was negligible in high risk ZIP codes, especially compared with the larger effect seen in high-RR ZIP codes for IPV.

The opioid epidemic has obvious direct impacts, such as an increase in overdose-related fatalities (Kochanek et al., 2017). Yet, there are also key secondary consequences that have not been studied extensively. IPV and child maltreatment hospitalizations, which have discernable spatial patterns across Pennsylvania, are connected not only to economic conditions such as median household income but also to trends of substance abuse disorders, such as opioid-related hospitalizations. In this study, we show that patterns of both family violence outcomes are associated with concurrent (but not past-year) opioid-related hospitalizations. This differs from temporal (past-year) associations connecting the opioid epidemic to other outcomes, such as HIV, hepatitis C virus (HCV), and mental disorder hospitalizations (Sumetsky et al., 2019).

To ensure that our findings were not merely a reflection of varying hospitalization trends, we conducted analyses using hospitalization outcomes that we would not expect to be associated with opioid misuse, such as motor vehicle traffic accidents, influenza, and falls. None of these outcomes had positive well-supported associations with opioid-related hospitalization rates. Both concurrent and past-year opioid-related diagnosis rates were, however, connected to other related hospitalization outcomes, such as HIV, HCV, and mental disorders (Sumetsky et al., 2019). The differential connections between concurrent and past-year opioid misuse issues and a variety of health outcomes highlight important secondary implications of opioid misuse trends.

HDD are a favorable data source for spatial analysis as they are consistently collected over space and time at a fairly fine spatial resolution (ZIP codes). However, due to underreporting and tendency of HDD to represent more severe cases, rates of IPV and child maltreatment are likely far higher in the overall population. Furthermore, hospitalization is subject to racial and ethnic disparities. For example, Hispanic individuals experiencing IPV tend to be hospitalized less often (Lipsky et al., 2009). While our findings highlight the geographic heterogeneity of patterns of family violence, our results do not necessarily extend to other geographic regions outside Pennsylvania. Our results also cannot be extended to the individual level as they are based on aggregate data. Finally, we cannot extrapolate findings to estimate the magnitude of the impact of the opioid epidemic on all family violence.

Family violence, such as IPV and child maltreatment, and opioid misuse are serious and connected public health concerns. Although efforts to combat the direct impacts of the opioid epidemic have been sizable, secondary consequences have received relatively little attention. Our findings point to the interconnectedness of opioid misuse and family violence within communities, and the role of area-level economic conditions on these public health problems. Consideration of this information is crucial for public health efforts to prevent family violence.

Author Biographies

Natalie Sumetsky is a project analyst, Department of Behavioral and Community Health Sciences, and PhD student, Department of Epidemiology, University of Pittsburgh Graduate School of Public Health.

Jessica G. Burke is a professor and associate chair, Department of Behavioral and Community Health Sciences, Deputy Director, Center for Social Dynamics and Community Health, and Associate Dean for Academic Affairs, University of Pittsburgh Graduate School of Public Health.

Christina Mair is an associate professor, Department of Behavioral and Community Health Sciences, and Director, Center for Social Dynamics and Community Health, University of Pittsburgh Graduate School of Public Health.