The Effect of Health Insurance on Health Care Utilization in the Justice-Involved Population: United States, 2014–2016

Abstract

Objectives. To examine the impact of health insurance coverage on utilization of outpatient, hospital, and emergency department care among justice-involved individuals in the United States.

Methods. We performed repeated cross-sectional analyses with data from the National Survey of Drug Use and Health. The study population included 6086 adults with justice involvement within the past year. We used logistic regression to examine the odds of health care utilization based on either a dichotomous or categorical measure of health insurance coverage. We used negative binomial regression to examine the number of times a specific type of care was utilized with both a dichotomous measure of health insurance coverage and a categorical measure of type of health insurance.

Results. Health insurance was associated with increased utilization of outpatient, inpatient, and emergency department care.

Conclusions. Health insurance coverage was associated with increased utilization of outpatient, inpatient, and emergency department health care among justice-involved individuals. Therefore, expanding access to health insurance in this population has the potential to increase care utilization of all types and decrease barriers to medical services.

In 2016, 1 in 38 adults in the United States was under correctional supervision.¹ Of those, roughly 2.2 million adults were incarcerated, with 1.5 million in prison and more than 740 000 in jails.^1,2 Although these numbers represent a small decline from recent years, the US prison population remains the highest in the world.³ Returning citizens represent a medically vulnerable population, as they suffer from increased rates of many chronic medical conditions including hypertension, diabetes, asthma, cancer, and mental health conditions, as well as infectious diseases such as HIV and hepatitis.^4–6 Release from prison itself carries an increased risk of hospitalization and death upon release.^7–9

Returning citizens also face multiple challenges upon re-entry to the community including poverty, homelessness, and employment issues.¹⁰ Typically, access to medical care among this population has posed challenges, as many returning citizens lack health insurance.^11–14 While the Affordable Care Act (ACA) has increased insurance rates, the overall rate of uninsurance among returning citizens remains double that of the general public.¹¹ This obstacle in access leads to increased rates of emergency department (ED) and inpatient hospital utilization for medical care upon release, often for preventable conditions that can be managed in an outpatient setting.^15,16

Previous studies have sought to examine the impact of increased insurance coverage following the implementation of the ACA on utilization of mental health and substance use treatment of justice-involved young men,¹¹ as well as utilization of the ED in the justice-involved population.¹² Although other studies have demonstrated that health insurance is associated with increased care use for low-income populations,¹⁷ to our knowledge, no study has yet examined whether health insurance coverage affects the total amount of medical care that justice-involved individuals receive, or the impact of insurance on the source of care utilized. It is critical to examine the utilization of inpatient, emergency, and outpatient care in this population, as these patients are medically complex and hold an increased risk of hospitalization and death upon release.^7,8

Current literature suggests that decreased rates of insurance among justice-involved individuals create barriers to accessing primary care, which in turn causes increased utilization of acute medical care.^11,12 What is yet to be addressed is how insurance coverage for justice-involved individuals affects utilization across inpatient, outpatient, and ED settings. To address this gap in knowledge, we analyzed multiple years (2014, 2015, and 2016 surveys) of National Survey of Drug Use and Health (NSDUH) data to examine how health insurance coverage for justice-involved individuals affects utilization of outpatient, hospital, and ED services and whether the impact of health insurance on medical care utilization varies by type of insurance coverage (i.e., public, private, public and private, other) among justice-involved individuals.

METHODS

We took the data from the cross-sectional NSDUH in 3 recent waves (2014, 2015, 2016) after the ACA was fully implemented. This survey is the primary source of nationally representative annual estimates of drug use and mental illness among the noninstitutionalized US population aged 12 years and older.^11,18 Following previous studies,¹¹ we defined justice involvement broadly to include individuals arrested, on probation, or on parole during the past year. To ensure that was appropriate, we conducted exploratory descriptive analyses (not shown) that demonstrated that each of these justice-involved groups has similar health insurance rates to one another and health insurance rates that are much lower than for the adult population as a whole.¹⁹ Therefore, we restricted our sample to the 6086 adults (aged 18 years and older) who fit these criteria. In robustness checks (which we discuss in the Sensitivity Analyses section), we ran analyses separately by type of criminal justice involvement to ensure that our results were not driven by our decision to collapse justice involvement into 1 category.

For each analysis, we further restricted our sample to individuals who were not missing data on the outcome of interest and employed multiple imputation to fill in missing data for covariates.²⁰ Our final analytic sample varies slightly by outcome, ranging from 5924 to 6058 individuals.

Outcome Measures

In this study, we had 6 outcome measures capturing 3 different types of care utilization. To capture outpatient care usage, we included a measure indicating whether an individual visited a doctor, nurse, physician assistant, or nurse practitioner during the past 12 months and a count measure indicating the number of times an individual received this type of care. To capture ED usage, we included a measure indicating whether an individual had been treated in an ED during the past 12 months and a count measure capturing the number of times an individual received this type of care. To capture overnight stays in the hospital, we included a measure indicating whether an individual stayed overnight or longer as an inpatient in a hospital during the past year and a count measure capturing the number of nights an individual received this type of care.

Key Independent Variables

We used 2 different self-report measures to capture health insurance status. First, we included a binary measure that indicated whether an individual had any type of health insurance coverage. Second, we created a categorical variable indicating what type of health insurance an individual had: no health insurance coverage, public health insurance coverage only, private health insurance coverage only, both public and private health insurance coverage, or other health insurance coverage. The type of health insurance that individuals in the “other” category had is not available in the NSDUH data, but consists of respondents who did not report having coverage through Medicare, Medicaid or Children’s Health Insurance Program, private health insurance, Champus, Tricare, Veterans Administration health care, or military.¹⁸ It is important to be cautious when interpreting coefficients associated with “other” health insurance because it is unclear what type of insurance is being captured.

Control Variables

On the basis of previous literature and theoretical expectations, we controlled for a number of factors that may be associated with both health insurance coverage and health care usage. To capture severity of criminal justice involvement, we created a 3-category variable indicating whether individuals were arrested, on probation, or on parole during the past 12 months. To capture substance use over the past year, we controlled for alcohol dependence or use, nicotine dependence or use, illicit drug dependence or use, and marijuana dependence or use. To capture respondent mental and physical health, we controlled for serious psychological distress during the past year and self-reported health (coded excellent, very good, good, or fair or poor).

We also controlled for demographic characteristics, which included age (coded 18–25, 26–34, 35–49, and 50 years or older), gender, race/ethnicity (coded White, Black, Latino/a, or other), educational attainment (coded less than high school, high school completed, some college, and college completed or more), and marital status (coded married, single, or divorced, separated, or widowed). We captured economic circumstances by controlling for the income-to-needs ratio (coded below the poverty threshold, up to 2 times the poverty threshold, or more than 2 times the poverty threshold [according to the US Census Bureau in the year the person was surveyed]) and employment status (coded full-time, part-time, unemployed, or other). We also controlled for the number of times respondents moved over the past year (coded 0, 1, 2, or 3 or more times). Lastly, following previous research, we included a scale capturing respondent’s propensity for taking risks, coded from 0 to 2, with 0 representing the lowest levels of risk-taking propensity and 2 representing the highest.^21,22

Analytic Strategy

For each outcome measure, we used repeated cross-sectional data and conducted 4 sets of analyses, each including the full set of control variables described previously, as well as year fixed effects. In the first 2 sets of analyses, we used logistic regression to examine the odds of having used a particular type of care. The only difference between the first 2 sets of analyses is that the first explored the association between our dichotomous measure of health insurance coverage and care usage, while our second examined the association for our categorical measure capturing different types of health insurance coverage. In our third and fourth sets of analyses, we used negative binomial regression to examine the number of times an individual used a specific type of care. The only difference between the third and fourth sets of analyses is that the third explored the association between our dichotomous measure of health insurance coverage and care usage, while the fourth examined the association for our categorical measure capturing different types of health insurance coverage. For each analysis, we used a statistical significance threshold of P less than .05, but also indicated whether a coefficient was significant at P less than .01. We also tested whether there were statistical differences in the coefficients of different types of health insurance by changing the reference group (e.g., making public-only the reference group, private-only the reference group).

RESULTS

In Table 1, we provide descriptive statistics for the sample and separately for individuals with and without health insurance. The descriptive statistics suggest that people with health insurance were significantly more likely to use care and used care more frequently, regardless of the type of care. Nearly 75% of those who had health insurance used outpatient care compared with approximately half of those without health insurance. Moreover, the median number of times a respondent used outpatient care during the year among people with health insurance was 2, while for those without insurance it was 1.

TABLE 1—

Descriptive Statistics by Health Insurance Status in the National Survey of Drug Use and Health 2014, 2015, and 2016 Survey Waves: United States

	Full Sample, % (No.) or Median (Range)	Health Insurance, % (No.) or Median (Range)	No Health Insurance, % (No.) or Median (Range)
Outpatient (no, yes)*	67.81 (3987)	74.31 (3225)	50.11 (762)
Outpatient no.*	2 (0–19)	2 (0–19)	1 (0–19)
ED (no, yes)*	43.96 (2626)	46.01 (2018)	38.35 (608)
ED no.*	0 (0–31)	0 (0–31)	0 (0–30)
Hospital overnight (no, yes)*	13.28 (741)	15.04 (619)	8.49 (122)
Hospital overnight no.*	0 (0–31)	0 (0–31)	0 (0–30)
Health insurance type*
None	26.88 (1546)	0 (0)	100.00 (1546)
Public	33.42 (2015)	45.70 (2015)	0.00 (0)
Private	31.38 (2015)	42.91 (2015)	0.00 (0)
Public and private	4.37 (214)	5.98 (214)	0.00 (0)
Other	3.96 (296)	5.41 (296)	0.00 (0)
Criminal justice involvement
Arrest	41.84 (2595)	41.29 (1934)	43.36 (661)
Probation	40.11 (2432)	40.82 (1845)	38.18 (587)
Parole	18.05 (1059)	17.89 (761)	18.46 (298)
Substance use
Alcohol dependence or abuse	24.14 (1538)	24.17 (1147)	24.06 (391)
Nicotine dependence or abuse*	26.83 (1643)	25.69 (1183)	29.95 (460)
Illicit drug use	12.43 (756)	12.21 (564)	13.04 (192)
Marijuana dependence or abuse	7.85 (624)	8.21 (480)	6.90 (144)
Psychological distress	24.87 (1642)	25.72 (1270)	22.54 (372)
Health*
Excellent	16.23 (1086)	16.45 (822)	15.64 (264)
Very good	30.52 (2028)	31.50 (1520)	27.86 (508)
Good	33.19 (1968)	31.31 (1433)	38.29 (535)
Fair or poor	20.05 (1002)	20.73 (764)	18.21 (238)
Employment*
Full time	45.29 (2691)	44.20 (1956)	48.25 (735)
Part time	12.69 (877)	12.56 (666)	13.03 (211)
Unemployed	14.49 (984)	12.52 (672)	19.86 (312)
Other	27.53 (1534)	30.72 (1246)	18.86 (288)
Marital status*
Married	23.33 (1089)	23.91 (812)	21.74 (277)
Never married	52.70 (3953)	50.99 (2929)	57.35 (1024)
Divorced, separated, or widowed	23.97 (1044)	25.10 (799)	20.90 (245)
Female*	29.98 (2066)	32.20 (1672)	23.94 (394)
Age, y*
18–25	28.09 (2896)	29.04 (2220)	25.50 (676)
26–34	26.75 (1550)	23.83 (1059)	34.69 (491)
35–49	25.68 (1206)	25.07 (887)	27.34 (319)
≥ 50	19.48 (434)	22.06 (374)	12.47 (60)
Race/ethnicity*
White	53.03 (3011)	55.14 (2309)	47.28 (702)
Black	21.80 (1267)	22.03 (935)	21.16 (332)
Latino	19.18 (1126)	16.13 (732)	27.48 (394)
Other	5.99 (682)	6.70 (564)	4.08 (118)
Education*
< high school	26.54 (1640)	23.04 (1090)	36.07 (550)
High school	35.24 (2233)	34.76 (1647)	36.54 (586)
Some college	30.76 (1839)	33.45 (1481)	23.45 (358)
≥ college	7.45 (374)	8.74 (322)	3.93 (52)
Povertya,*
In poverty	33.68 (2192)	30.86 (1564)	41.32 (628)
Up to 2 times poverty threshold	26.82 (1660)	25.88 (1164)	29.37 (496)
More than 2 times poverty threshold	39.5 (2202)	43.26 (1781)	29.31 (421)
No. of residential moves*
0	53.73 (2921)	55.76 (2259)	48.15 (662)
1	24.65 (1579)	23.90 (1153)	26.70 (426)
2	11.86 (764)	10.71 (528)	15.01 (236)
≥ 3	9.77 (688)	9.63 (507)	10.14 (181)
Risk-taking scale
0	61.61 (3501)	61.38 (2600)	62.24 (901)
1	16.75 (1027)	16.75 (774)	16.77 (253)
2	21.64 (1543)	21.87 (1157)	20.99 (386)

Note. ED = emergency department. Sample size n = 6058. Descriptive statistics were weighted accounting for complex survey design. We used the χ² test to determine statistical significance for categorical variables and the t test for continuous variables. For ED and hospital night number, 31 is used by the National Survey of Drug Use and Health (NSDUH) to capture 31 or more visits. For outpatient number, NSDUH uses 16 to capture 16 to 20 times, 17 to capture 21 to 25 times, 18 to capture 26 to 30 times, and 19 to capture 31 or more times. The “other” race/ethnicity category captures all individuals who are not in the first 3 categories. “Single” includes people who are partnered if they have never been married, divorced, separated, or widowed. Unweighted frequencies are in parentheses for categorical variables. Minimum and maximum are in parentheses for continuous variables.

^aAccording to the US Census Bureau in the year the person was surveyed.

^*Significant difference between people with health insurance and people without health insurance at P < .05.

In terms of ED usage, 46% of those with health insurance used care compared with 38% without insurance. Although the median amount of times a respondent used the ED was 0 regardless of health insurance status, the mean for those who had health insurance was 1.14 times during the year, while for those without insurance it was 0.91 times. Those with health insurance were also more likely to stay overnight at the hospital (15.04%) compared with those without insurance (8.49%). Although the median number of times a respondent stayed overnight at the hospital was 0 regardless of health insurance status, the mean number of nights spent was higher among those with health insurance (0.81 for those with health insurance and 0.40 for those without insurance). In addition to care usage, there were significant differences by health insurance status across other characteristics, such as age, nicotine dependence or use, self-reported health, race/ethnicity, income level, number of residential moves, marital status, gender, employment status, and educational status. Therefore, multivariate analyses are needed to isolate the association between health insurance status and use of care.

Multivariate Results

In Table 2, we explored whether health insurance status was associated with outpatient care usage among justice-involved individuals. In this table, we only present health insurance coefficients to preserve space, but the coefficients associated with the control variables can be found in Table A (available as a supplement to the online version of this article at http://www.ajph.org). In the first set of analyses, we examined whether health insurance coverage and type of health insurance were associated with having used outpatient care in the past year. The results suggest that individuals with health insurance had 2.39 times higher odds of having used outpatient care than individuals who lacked insurance (P < .01). The results from model 2 suggest that this association held for all types of insurance (P < .01), but was significantly larger for people who had both public and private insurance than for all other insurance types.

TABLE 2—

The Effect of Health Insurance Coverage on the Use of Outpatient Care in the National Survey of Drug Use and Health 2014, 2015, and 2016 Survey Waves: United States

	Any Outpatient Care Use, OR (95% CI)	No. of Times Outpatient Care Used, OR (95% CI)
Any health insurance	2.39 (1.98, 2.88)	1.67 (1.49, 1.88)
McFadden’s R2	0.12	0.05
Type of health insurance (Ref = none)
Public	2.43 (1.93, 3.05)	1.76 (1.57, 1.99)
Private	2.27 (1.82, 2.82)	1.55 (1.33, 1.81)
Public and private	5.20 (2.86, 9.46)	2.21 (1.80, 2.73)
Other	1.95 (1.24, 3.05)	1.17 (0.93, 1.48)
McFadden’s R2	0.12	0.05

Note. CI = confidence interval; OR = odds ratio. Sample size n = 5924. Regression analyses were weighted accounting for complex survey design. Logistic regression was used for any outpatient care use analyses, negative binomial regression was used for analyses of number of times outpatient care was used, and coefficients for number of times outpatient care was used are reported as incident rate ratios. Analyses examining impact of any health insurance and type of health insurance were conducted separately. All models included the full set of controls (level of criminal justice involvement, substance use, serious psychological distress, self-reported health, age, gender, race/ethnicity, educational attainment, marital status, income-to-needs ratio, employment status, number of times respondents moved over the past year, and risk-taking behavior) as well as year fixed effects. Results with control variable coefficients can be found in Table A, available as a supplement to the online version of this article at http://www.ajph.org.

In our next set of analyses, we examined whether health insurance coverage and type of health insurance was associated with the number of times an individual used outpatient care in the past year. The results indicate that having health insurance was associated with a 67% increase in the number of times an individual used outpatient care (P < .01). The results also show that this association held for individuals who had public health insurance only (P < .01), private health insurance only (P < .01), and both public and private health insurance (P < .01), but not for individuals who indicated having other health insurance. Individuals with all types of health insurance used outpatient care significantly more than those with other insurance, while individuals with public and private insurance also used significantly more outpatient care than those who only had public or private insurance.

In Table B (available as a supplement to the online version of this article at http://www.ajph.org), we examined whether health insurance status was associated with ED usage among justice-involved individuals. In models 1 and 2, we explored whether health insurance coverage and type of health insurance was associated with having used the ED in the past year. Our results from model 1 show that individuals with health insurance had 42% higher odds of having used the ED than individuals without insurance (P < .01). Model 2 demonstrates that this association was strongest for people who had public health insurance (P < .01 for those with public health insurance only and P < .05 for those reporting both public and private health insurance coverage). We did not find significant associations for those who only had private health insurance and for those who indicated having other health insurance. In addition, we found that those with public health insurance only were significantly more likely to have used the ED than individuals with private insurance only.

In models 3 and 4, we explored whether health insurance coverage and type of health insurance was associated with the number of times an individual used the ED in the past year. The results from model 3 indicate that having health insurance was associated with a 17% increase in the number of times an individual used the ED (P < .05). The findings from model 4 suggest that this association only holds for individuals who had public health insurance only (P < .05) and that individuals who had public health insurance only used the ED significantly more times than individuals with private insurance only.

In Table C (available as a supplement to the online version of this article at http://www.ajph.org), we examined whether health insurance status was associated with hospital overnight stays among justice-involved individuals. In models 1 and 2, we explored whether health insurance coverage and type of health insurance was associated with having spent the night in the hospital during the past year. The findings from model 1 suggest that the individuals with health insurance had 63% higher odds of having spent a night in the hospital than individuals without insurance (P < .01). The results from model 2 suggest that these associations only held for individuals with public health insurance (P < .01 for those with public health insurance only and P < .05 for those reporting both public and private health insurance). Moreover, the results suggest that individuals who had public insurance only and public and private insurance were significantly more likely to use the ED than individuals who had private insurance only.

In models 3 and 4, we explored whether health insurance coverage and type of health insurance was associated with the number of times an individual stayed in the hospital overnight in the past year. The results from model 3 indicate that having health insurance was associated with a 74% increase in the number of times an individual stayed overnight in the hospital (P < .01). The results from model 4 show that this association held for individuals who had public health insurance coverage only (P < .01) and public and private health insurance (P < .05). Moreover, individuals who had public health insurance only spent significantly more days overnight at the hospital than individuals with private insurance only.

Sensitivity Analyses

For the results we report in this article, we collapsed individuals who experienced criminal justice involvement in the past year into 1 group. In supplemental analyses, we examined whether the effects of health insurance on health care service usage differed across types of criminal justice involvement. To do so, we ran separate multivariate logistic regression and negative binomial regression analyses (not shown) for each type of criminal justice involvement and tested for equality of the health insurance coefficients.²³ Using this approach, we found very little evidence of statistically significant differences in the impact of health insurance on health care service usage across types of criminal justice involvement. Specifically, out of 198 comparisons, there were statistical differences across criminal justice involvement types in only 13 instances (6.57%).

DISCUSSION

In this study, we examined whether health insurance coverage was associated with an increased use of medical care among justice-involved individuals by using nationally representative data from the NSDUH survey waves (2014, 2015, and 2016) administered after implementation of the ACA. Our results from multivariate analyses suggest that justice-involved individuals were more likely to use all types of care, including outpatient care, the ED, and overnight hospitalization, when they had health insurance. The impact of health insurance on outpatient care was similar for both public and private insurance, while for ED use and overnight hospitalization the impact tended to be particularly large for individuals with public insurance.

On a broad level, the findings from this study are consistent with previous scholarship suggesting that strengthening social welfare programs can mitigate the negative implications of mass incarceration.²⁴ More narrowly, our results align with research suggesting that health insurance coverage is associated with an increase in substance use treatment and mental health treatment usage among justice-involved individuals who self-report these behavioral health issues.¹¹ Previous studies have demonstrated that justice involvement is associated with higher rates of inpatient and ED utilization, and that formerly incarcerated individuals are less likely to report primary care utilization and more likely to report ED use.^15,16 However, to our knowledge, no study to date has examined utilization rates of acute versus outpatient primary care, and our study is the first to examine how health insurance coverage affects total care usage among justice-involved individuals, as well as the first to explore how health insurance coverage affects frequency of care usage by source of care.

This study may have important implications for policymakers. Previous research has found that justice-involved individuals are far less likely to have health insurance coverage than are individuals who are not justice-involved.^11–14 Our results suggest that expanding health insurance access for this population has the potential to increase care utilization. This is critical because research has consistently shown that justice-involved populations are medically vulnerable, with particularly high risks of HIV, hepatitis, hypertension, diabetes, asthma, cancer, mental health challenges, and substance use issues.^4–9 While our study does demonstrate that insurance is associated with increased utilization of all levels of care, it is important to note that this increase also includes primary care. Previous studies have focused on how justice-involved individuals are more likely to utilize acute levels of care; however, our results suggest that health insurance coverage can lead to an increase in the use of outpatient care among this population.^11,12 Presumably, increasing utilization of outpatient care may allow for chronic and preventive issues. Because justice-involved individuals are disproportionately of low socioeconomic status and people of color, expanding access to health insurance has the potential to reduce health disparities.²⁵

In addition to discussing the significance of our findings, it is important to acknowledge this study’s limitations. Perhaps most important, our study is based on nonexperimental data. Thus, despite the fact that we accounted for a wide range of potentially confounding factors, we cannot be certain that health insurance coverage has a causal impact on care usage among justice-involved individuals and that our findings were not driven by omitted variables. The cross-sectional nature of the survey also makes it impossible for us to determine the exact timing of health insurance coverage enrollment and care usage, so it is possible that some respondents were enrolling in health insurance in response to health service usage. We also did not have data indicating the state of residence of each respondent and therefore could not account for state-level differences in ACA implementation and criminal justice policies. Lastly, we found some evidence suggesting that public insurance is associated with higher rates of service use than private insurance, but our data did not allow us to determine why. Future research should examine this issue further.

Although the implications of our findings for health outcomes are promising, the results are nuanced as to how expanding access to health insurance might influence the health care system. Multiple studies have demonstrated that justice-involved populations have an increased rate of inpatient and ED utilization.^{7,11,12,15,16} These are expensive sources of care, and our results suggest that expanding health insurance access might increase utilization of this type of care, and therefore increase health care costs. However, our results suggest that justice-involved individuals with health insurance are also more likely to use outpatient primary care, which is less expensive. Essentially, taken together, our results suggest that expanding health insurance access is likely to lead to increased utilization of care, regardless of type. From a cost standpoint, the key then is for policymakers to ensure that it is easier and more desirable for justice-involved populations to prioritize primary care over more expensive types of care.

It is also important to note that, although our data suggest that increasing health insurance will increase utilization, this does not necessarily improve health outcomes. Multiple socioeconomic factors contribute to health outcomes. Increasing health care access is a piece of improving population health, but not the only solution. Future work should examine how policies can be developed to expand insurance coverage while also investigating the multiple factors that contribute to health inequality.