Health Care Utilization After Nonfatal Firearm Injuries

We conducted a propensity score matched retrospective cohort study using the MarketScan Multistate Medicaid and Commercial Claims databases from January 1, 2015 to December 31, 2018. MarketScan databases are compiled and maintained by Truven Health Analytics and contain deidentified patient-level claims data for publicly and privately insured children representing more than 350 unique carriers. The MarketScan Commercial database contains data for dependents covered by employer-sponsored private health insurance throughout the United States, whereas MarketScan Medicaid encompasses Medicaid enrollees from multiple states. The databases link health care data from a variety of sources and include information on inpatient services, outpatient services, and outpatient prescriptions.

We included all individuals aged 0 to 17 years in the MarketScan Multi-State Commercial and Medicaid databases from January 1, 2015 to December 31, 2018. Youth who died during the initial health care encounter were excluded. The primary exposure was initial firearm-related injury defined by International Classification of Diseases, Tenth Revision (ICD-10) codes W32-W34, X72-X74, X93-X95, Y22, Y230-Y233, Y238, and Y239 corresponding to unintentional, self-inflicted, assault, and undetermined intents, respectively. This does not include youth with nonpowder gun injuries (ICD-10 codes W3401, W3411, X7401, X7402, X7409, X9501, X9502, X9509). We included all billing diagnostic codes because cause of injury is often listed as a patient’s nonprimary diagnosis. The index date for injured youth was defined as the date of their first claim related to a firearm injury between January 1, 2016 and December 31, 2017, whereas 2015 data were only used for preindex covariate generation. Comparison noninjured youth were assigned an index date of July 1. The term “index” is used to consistently describe time periods and covariates. Index, rather than “injury” is used as comparison youth do not have a clinical encounter, that is, injury, to identify. We examined all claims for inpatient, outpatient, and pharmacy services between 12 months before the index encounter (preindex) and 12 months after the index encounter (postindex). Figure 1 outlines cohort criteria and propensity methods.

The primary outcome was any health care encounter for inpatient hospitalizations, emergency department (ED) visits, and outpatient visits. The health care encounter outcomes were defined as any claims for hospitalizations, ED visits, and outpatient visits in the 12 months postindex encounter because of any diagnosis defined by visit codes. Encounter outcomes are dichotomous variables representing any utilization in the 12 months after the index encounter versus no utilization. To evaluate patterns of health care utilization after injury, we additionally examined utilization in the 12 months after the index encounter by primary diagnosis codes in each of the 3 settings. Encounters were grouped into 3 large categories of “Injury-Related,” “Mental-Health Related,” and “Other Diagnosis.” We also examined outpatient visits defined as count outcomes. Total health care cost was a secondary outcome defined as total per patient health care payments in the 12 months after and not including the index encounter. All outpatient, inpatient, or prescription payments made by either Medicaid or commercial insurance companies are included for cost outcomes that would include all out-of-pocket, coinsurance, and deductibles as applicable.

Medical condition and utilization covariates were assessed in the 12 months before the index encounter. Covariates included age, sex, race (as a social construct including non-Hispanic white, non-Hispanic Black, Hispanic, and other defined as Asian, Pacific Islander, American Indian, multiracial, and other unspecified race and ethnicity; available for Medicaid youth only), number of complex chronic conditions (CCCs),⁹  dichotomous CCC status, dichotomous transplantation status, number of days insured, presence or absence of any health care claim, number of pharmacy claims, number of outpatient claims, number of inpatient claims, any mental health condition,¹⁰  substance-related and addictive disorders,¹⁰  disruptive or impulse control and conduct disorders,¹⁰  and any nonfirearm injury. CCCs are defined as medical conditions that are expected to last at least 12 months, involve several organ systems or 1 system severely enough to require specialty pediatric care, and have a high probability of hospitalization.⁹  The Child and Adolescent Mental Health Disorders Classification System was used to identify mental health conditions.¹⁰  Preindex utilization variables were chosen to help account for gaps in insurance coverage and differences in health care seeking behavior. We did not include injury intent as a covariate given concerns about the accuracy of the coding of intent among patients with firearm injury.¹¹  All covariates for propensity score matching and for adjusted analyses were determined a priori based on a combination of clinical knowledge from the study team, previous literature, and examination of the imbalance prematch.¹²  Multiple of these covariates, including race, age, prior injury, and certain mental illnesses are known risk factors for firearm injury.^13–15  Race is an illustrative example of a potential confounding variable as it is associated with the exposure (firearm injury) and the outcome (health care utilization). Importantly, in this study, race is serving as a proxy for the complex systemic and structural inequities leading to higher rates of firearm injury and death for minority youth.¹⁶  The goal of covariate matching is to create an even distribution of each variable for both groups that precludes any further assessment of their impact on the outcome.

We applied propensity score methods to reduce group selection bias and controlled for potentially confounding variables. The propensity score was calculated by constructing a multivariable logistic regression model with firearm injury as the dependent variable. The calculated propensity score is the probability that the youth would sustain a firearm injury given the preindex characteristics listed above. The propensity score was created independent of the outcome measures. We matched firearm injured youth to youth who did not have a firearm injury (comparison noninjured youth) on a 1:1 basis using the greedy matching algorithm and caliper distance of 0.25 in SAS Proc PSMatch. Matching was completed separately based on insurance type (Medicaid or commercial) so that race could be retained and matched within the Medicaid group, and then groups were combined for outcome analysis. The balance of matching was examined by comparing the standard mean differences of each matching covariate between the 2 groups before and after matching. All covariates matched within 0.13 or less of a standardized mean difference, which is within a range considered well-balanced (0.10–0.25 depending on reference).¹⁷ 

Descriptive statistics were used to summarize demographics, preindex characteristics, and utilization using mean with SD for continuous variables and frequencies with percentages for categorical variables. Bivariate comparisons were completed using χ² tests for categorical variables and Wilcoxon-Mann-Whitney tests for continuous variables. After completing the propensity score matching, regression models were used to estimate relative risks of the dichotomous outcomes and mean number of visits and costs for continuous measures. Adjusted analyses of dichotomous outcomes used Poisson regression modeling with log link and robust error variance estimation. Visit count outcomes used negative binomial distribution and log link. Cost outcomes used γ distribution and log link. Utilizing doubly-robust estimation to account for any remaining imbalance, all outcome models controlled for age, sex, insurance type, 12 month preindex days insured, 12 month preindex dichotomous claims submitted, 12 month preindex number of inpatient claims, 12 month preindex number of outpatient claims, and 12 month preindex dichotomous mental health condition. All statistical analyses were performed using SAS v.9.4 (SAS Institute, Cary, NC), and 2-sided P values < .05 were considered statistically significant. Studies using these data have been designated as exempt from review by the institutional review board of the Medical University of South Carolina because they are nonhuman research.

There were 108 injured youth dropped because of inability to find an adequate match and 17 youth excluded because of death during the initial encounter (Fig 1). Postmatch, we identified a total of 2110 youth aged 0 to 17 with a nonfatal firearm injury with an additional 2110 matched comparison youth, resulting in a total sample of N = 4220. Table 1 describes the postmatch demographic and clinical characteristics of the cohort. The mean age was 13.5 years and 81% were male. Information for race and ethnicity was only available for the Medicaid insured youth, which represents ∼80% of the cohort. Of this group, 47.9% were Black, 21.9% were white, 2.8% were Hispanic, and 1.6% were in the other race category with 25.8% of Medicaid-insured youth missing race and ethnicity data. The overall cohort had 0.1 (SD 0.4) preindex inpatient hospitalization claims, 3.5 (SD 7.3) preindex outpatient claims, and 4.4 (SD 9.3) preindex pharmacy claims. Approximately one-third of the cohort had a preindex mental health condition and one-fifth had any preindex CCC.

Before the match, several covariates were significantly imbalanced between the groups (Supplemental Fig 3). The propensity score match achieved standardized differences for all covariates of less than 0.13 with overall reduction in standardized differences in propensity score of over 94% for both Medicaid and commercial cohorts; an excellent match (Supplemental Fig 3).¹⁷  Therefore, matched youth with and without firearm injury were similar in age, sex, race (only applicable to Medicaid youth), insurance status, CCC status, preindex mental health condition, preindex injury, and preindex utilization (Table 1). Although preindex outpatient, inpatient, and prescription utilization appear statistically different between groups, this is likely because of large sample size and is not clinically significant.

Table 2 demonstrates the unadjusted health care utilization outcomes following propensity score matching. In the 12 months postindex encounter, 11.6% of injured youth had at least 1 claim for inpatient hospitalization compared with 2.2% of noninjured youth (P < .0001). Regarding ED visits, 41.5% of youth with nonfatal firearm injury had at least 1 claim for ED visits compared with 27.8% of youth without nonfatal firearm injury (P < .0001). Additionally, 46.5% of youth with a nonfatal firearm injury had any claim for either ED visit or hospitalization compared with 28.9% of noninjured youth (P < .0001). Lastly, 79.1% of youth with nonfatal firearm injury had any claim for an outpatient visit compared with 74.7% of youth without a nonfatal firearm injury (P = .007).

Table 3 illustrates the results of the regression analyses of the primary dichotomous outcomes adjusted for age, sex, insurance status, preindex mental health condition, preindex dichotomous claims, preindex outpatient utilization, and preindex inpatient utilization. Compared with noninjured youth, youth with nonfatal firearm injury had a 5.31-fold increased risk (relative risk [RR] 5.31; 95% confidence interval [CI] 3.93–7.18) of an inpatient hospitalization in the 12 months after an index encounter. Additionally, youth with a nonfatal firearm injury had a 1.49-fold increased risk (RR 1.49; 95% CI 1.37–1.62) of ED visit and 1.06-fold increased risk (RR 1.06; 95% CI 1.03–1.10) of outpatient visit 12-months postindex encounter compared with noninjured youth. Youth with a nonfatal firearm injury had, on average, 4.03 (95% CI 3.84–4.22) outpatient visits versus 2.92 (95% CI 2.78–3.06) for comparison noninjured youth in the 12-months postindex encounter.

The unadjusted mean costs in the 12 months after an index encounter for youth with a nonfatal firearm injury were $8288 (SD $63 155) compared with $2180 (SD $7634) for comparison noninjured youth. Average adjusted 12-month postindex encounter costs were $7581 (CI $7101–$8092) for youth with a nonfatal firearm injury compared with $1990 (CI $1862–$2127) for comparison noninjured youth. This equates to an attributable 12-month increased cost of $5591 per injured youth (Fig 2). These cost values do not include the index encounter.

The descriptive analysis of utilization, grouped into 3 categories including injury, mental health and other, is provided in Table 4. In the 12 months after an index encounter, 1.1% of encounters were in the Injury-Related category, 11.5% were in the Mental Health-Related category, and 87.4% were in the Other Diagnosis category. Firearm injured youth had a higher proportion of Injury-Related encounters compared with noninjured youth (1.6% vs 0.4%) and a lower proportion of “Mental-Health Related” encounters (10.9% vs 12.3%).

In this propensity score matched cohort study utilizing a large claims database of commercial and Medicaid insured youth, we found that youth with a nonfatal firearm injury had significantly higher relative risks for health care utilization and costs in the 12 months after an index encounter compared with noninjured youth. Our findings uniquely expand upon the current literature highlighting the longitudinal impact of nonfatal firearm injuries in US youth by using a novel propensity score methods approach, including over 4000 youth with commercial and Medicaid insurance, and having a follow-up period of 12 months.

Our most notable finding is the over fivefold increased risk for hospitalization claims in the 12 months after an index encounter for youth with a nonfatal firearm injury compared with youth without a firearm injury. After hospitalization from a nonfatal firearm injury, 50% of youth are discharged with a disability, which is likely strongly contributing to our results.³  Our findings also shed light on the potential disruption to daily living for youth with nonfatal firearm injuries. Evidence has shown that exposure to firearm violence in childhood is associated with hypertension, posttraumatic stress, and future involvement in violent crime, but hospitalization is a unique stressor associated with sleep disturbances, separation from social support systems, as well as the large financial burden placed on families.^18,19  We also found that youth with a nonfatal firearm injury had 1.49 times higher risk for ED visit and 1.06 times higher risk for outpatient visits, suggesting prolonged adverse health consequences. Descriptively, in the 12 months after an index encounter firearm injured youth understandably have a higher proportion of encounters because of injury-related diagnoses. However, injury-related encounters were less than 2% of the total encounters, potentially because of many encounters being captured in the “Other Diagnosis” category either because they were more accurately described as complications of the injury or because of the inaccuracies with primary diagnoses codes, that is, misclassification bias.^20,21  It is difficult to draw more substantial conclusions regarding reasons for health care utilization after injury from these data given the limitations with primary diagnosis codes presenting an intriguing area for future study.²¹ 

In addition to health care encounters, we found a significant increase in health care expenditures for youth with a nonfatal firearm injury compared with comparison noninjured youth with an increased attributable cost of $5591 per injured youth. These costs may be a lower estimate than expected, especially given the risk of inpatient hospitalization. Note that cost is a mean estimate and when applied to the entire cohort, represents $11.8 million in total health care expenditures in the 12 months after injury for 2110 youth. Furthermore, this would equate to an estimated total population health care cost attributable to firearm injury in the 12 months after an injury alone of $62.9 million when applied to the ∼11 258 US youth aged 0 to 17 who suffered a nonfatal firearm injury in 2020.²  These costs are an underestimate of societal costs given our inability to capture intangible costs, such as wage losses for patients and caregivers, quality of life, effects on those indirectly impacted by the firearm injury, or criminal justice expenses. These figures are striking and represent the significant burden firearm injuries place on our society. Our estimate is also consistent with previous literature with a recent study by Pulcini et al estimating the prepost difference in health care expenditures for nonfatal firearm injuries in Medicaid insured youth aged 0 to 17 to be $16.6 million in the year after injury, with ∼72.7% of the health care expenditures from inpatient encounters.⁸  Additionally, our cohort is ∼80% Medicaid, which may contribute to this lower estimate given the lower reimbursement rate of Medicaid compared with commercial insurance plans.²²  Although we did not differentiate costs between Medicaid and commercially insured youth, this represents an interesting future area of study.

It is important to note that the increased health care utilization and costs after nonfatal firearm injuries we report here are not experienced equally by all communities. It has been well described that firearm injuries and deaths disproportionately impact Black, Medicaid-insured adolescent males.^23,24  A recent study by Andrews et al highlighted the widening racial disparities in firearm mortality for US youth.¹⁶  As such, the structural racism and discriminatory policies persisting in the United States for centuries will need to be considered to best apply potential solutions in an equitable manner. Comprehensive solutions to prevent youth firearm injury will need to take a public health approach, considering interventions at all levels from the individual, community, and society. At the individual level, access to unsecured firearms in the home is a major risk factor for firearm injury.^25,26  The American Academy of Pediatrics recommends all pediatric providers routinely screen and counsel on firearm safety.^4,27  At the community level, interventions should consider risk factors such as poverty, lack of economic opportunity, and social mobility that are associated with a higher likelihood of experiencing violence with their roots in structural inequities and racism.^28–30  Further, public policies at the local, state, and national level have been shown to reduce pediatric mortality from firearm injuries, but have not been widely implemented.^31–33  And finally, critical gaps remain in understanding the comprehensive burden at the individual and community level of this preventable cause of injury and death. Thus, further research funding is needed to elucidate answers to these questions.

This study should be interpreted in the context of a few limitations. First, MarketScan is a claims-based database; therefore, we were only able to capture encounters associated with insurance payment. Some health care services, specifically mental health support, are often not covered by insurance and therefore would be underestimated in our study. Additionally, a claims-based database only allows for longitudinal tracking if patients remain continually enrolled in their insurance plan. Second, the MarketScan Medicaid and Commercial database differ. Because race and ethnicity are only available for Medicaid-insured youth, it is difficult to examine race and ethnicity differences. Further, accuracy of race and ethnicity data are limited in claims sources, which limits the understanding of health care disparities. Next, we analyzed health care outcomes in a broad context of health care settings at the outpatient, ED, and hospitalization levels. Outcomes were only measured through 12 months; therefore, long-term utilization remains an area of future study. Additionally, although MarketScan has a wide geographic distribution, it may not be nationally representative. Lastly, we intentionally did not include analyses by intent as this has been shown to be unreliable.^11,34 

In the 12 months after an index encounter for a nonfatal firearm injury, injured youth have significantly increased relative risk for health care utilization including hospitalizations, emergency department visits, and outpatient visits. Additionally, youth with nonfatal firearm injuries had significantly higher health care costs in the 12 months after an index encounter compared with noninjured youth. It is important for health care providers, insurers, and lawmakers to understand the long-term consequences for youth with nonfatal firearm injury. Potentially even more critical, we must advocate for comprehensive solutions at levels of the patient, community, as well as local, state, and federal government to reduce the preventable health care utilization and costs among firearm injured US youth in an equitable manner.

CCC

complex chronic condition

ED

emergency department