INTRODUCTION

Incarceration and mental health are inextricably linked. Deinstitutionalization of psychiatric care placed the burden of mental health treatment on the criminal justice system.1,2 Roughly half of police encounters have a psychiatric basis,3 and more people with mental health disorders are incarcerated than in hospitals.4 Worsening matters, criminal justice involvement may aggravate mental health disorders.5 Prevalence of serious mental illness is 3 to 5 times higher in the US criminal justice system than in the general population.6 Moreover, given a history of justice involvement, mental illness is more prevalent among those living with versus without HIV.7

Incarceration and mental health are independently associated with HIV outcomes. Jail is considered one of the strongest predictors of HIV care disruption.8 People living with HIV (PLWH) and released from incarceration face barriers, including unemployment, homelessness, substance use disorder (SUD), and access to HIV care.9 Although retention in care (RIC) can improve during incarceration, it diminishes postrelease.10 Access to antiretroviral treatment (ART) is also negatively affected. Incarceration, particularly at intake and release, is associated with lengthy delays in prescription refills.11

Mental health diagnoses can precede, or result from, HIV infection.12 Several mental health diagnoses are linked to HIV progression through poor medication adherence13,14 and risky behaviors.13,15 After controlling for adherence and clinical and demographic characteristics, depression is linked to a decline in CD4 count16–18 and increased viral load and mortality.16,18–21 Mental illnesses are often underreported, but national estimates indicate that up to half of PLWH may be affected.22 Mental health diagnoses are often comorbid with SUD. Nearly half of PLWH in the United States are thought to misuse at least 1 substance,23,24 and both mental health and SUD play important roles in HIV treatment compliance.25–27 SUD is associated with increased transmission risk behavior, delayed HIV diagnosis, delayed linkage to care (LTC), and poor RIC.24,27

The overlap between justice involvement, mental health diagnosis, and SUD has been shown to contribute to excess HIV mortality among PLWH postrelease.28 Still, the interaction between these issues is unclear. We aim to assess whether the following are associated with RIC and undetectable viral load (UVL): (1) arrest history overall and by charge; (2) mental health diagnosis overall and by category; and (3) interaction effects between arrest history, mental health diagnosis, and SUD. Our findings may be relevant for discharge planning and community reentry for justice-involved PLWH with a mental health diagnosis.

METHODS Sample and Study Design

We performed a retrospective cohort study of PLWH ≥13 years of age residing in Marion County during 2018. We used archived HIV surveillance data from the Centers for Disease Control and Prevention's Enhanced HIV/AIDS Reporting System (eHARS). Demographics, HIV diagnosis date, and HIV laboratory dates and results are captured within eHARS. Of 5730 PLWH in eHARS at any point during 2018, 5283 (92%) were present during each of the 4 quarters (“continuously in eHARS”). The remaining 8% were not consistently present for 1 or more of the following reasons: (1) newly diagnosed, (2) in- or out-migration, or (3) death (see Results section).

Person-level data from the 2018 eHARS cohort were linked to arrest records from the Indianapolis Metropolitan Police Department and clinical data from the Indiana Network for Patient Care (INPC) for calendar years 2000–2018. INPC is one of the largest, longest-operating, and best-studied health information exchanges in the United States.29–31 Through a collaborative effort among all major hospital systems, INPC captures statewide encounter data to serve clinical, public health, and research needs. For validation purposes, we supplemented INPC data with raw electronic health record data from the 2 largest hospital systems in Marion County.

Data were stored and analyzed on a password-protected and encrypted server behind the university firewall. Only the core research team had access to identifiers, which were used solely to link records between the sources. The study was reviewed and approved by Indiana University's Institutional Review Board and Marion County Public Health Department's Research Review Committee.

Study Procedures

Record linkage was performed using individual identifiers, including first/middle/last name, gender, month/day/year of birth, and social security number. Twenty-four deterministic algorithms employed a conservative approach by matching exact combinations of identifier subsets. We then used RecMatch software and various combinations of blocking/matching schemes with 24 probabilistic algorithms to identify additional matches based on linkage score threshold.32,33 True matches from the remaining records were determined by manual review. Matched pairs were “daisy-chained” to identify an aggregation of records representing an individual, and each aggregation of records was assigned a unique identifier. In rare cases where demographic or clinical data elements differed between linked records for the same unique individual records, we used the most recent data elements.

Measures

Demographic characteristics, including age, gender (sex at birth), and race/ethnicity, originated from eHARS. Arrest history was defined as any 2010–2017 arrest, and categorized as violent, drug-related, prostitution, intoxication, and felony (any).

Three clinical measures (mental health diagnosis, SUD, and health care utilization) were assessed by ICD-9 or ICD-10 diagnosis codes during ≥1 encounters during 2010–2017. We chose a sufficient period to offset the possibility of infrequent care utilization and inconsistent coding practices. We defined having a mental health diagnosis as ≥1 encounters with a mental health ICD code. Mental health diagnosis was assessed overall and categorized as mood, anxiety, psychosis, attention deficit, and personality disorders. SUD was assessed overall and by substance, including alcohol, opioid, and nonopioid drug misuse. Health care utilization was defined as whether individuals had ≥1 clinical encounters and was categorized as inpatient, outpatient, routine, and emergency department (ED). Routine visits were defined using diagnosis codes consistent with annual well visits.

Outcomes of interest included LTC, RIC, and UVL. LTC was defined as receiving a HIV-related clinical encounter within 30 days of diagnosis.34 RIC was defined as ≥2 CD4 or viral load tests performed ≥90 days apart during 2018.34 UVL was defined as any viral load value of <50 copies per milliliter in 2018.34

Analysis

Descriptive analyses were performed overall, and by history of arrest and mental health diagnosis. We performed Bonferroni-corrected univariable analyses of descriptive and outcome measures using independent samples t tests. Multivariable regression analysis was used to evaluate main and interaction effects of 2010–2017 arrest and mental health diagnosis on odds of 2018 RIC and UVL. We performed marginal effects analysis to calculate predicted probabilities of RIC and UVL with and without an arrest and/or mental health diagnosis. Further analysis of LTC was omitted because the study period bridged a 2015 move in the goal from ≤90 to ≤30 days.22 Individuals with missing data were excluded from regression analyses, as were variables demonstrating collinearity. Our final model included arrest history and mental health diagnosis for the subset continuously in eHARS, while controlling for age, race, gender, and history of SUD (n = 5140). To prevent bias that might arise from missing data, descriptive and regression analyses were primarily restricted to PLWH who were continuously in eHARS during 2018.

We conducted post hoc analyses to explore counterintuitive findings of apparent protective associations with HIV care outcomes. Outcomes were stratified by outpatient care utilization, which was added to our final regression model.

All analyses were performed using Stata/MP 14.1 (College Station, TX). P values <0.05 were considered statistically significant.

RESULTS

There were 5730 PLWH ≥13 years old residing in Marion County in 2018. Of 447 not continuously in eHARS, 218 were newly diagnosed, 58 died, and 260 either in- or out-migrated during ≥1quarters (not mutually exclusive categories). Those continuously in eHARS, versus those who were not, were older (mean age 45 vs 38 years, P < 0.05) and more likely to have any clinical encounter (88% vs 60%, P < 0.05).

Those continuously in eHARS were predominately male (80%) and disproportionately black (50% versus 29% in Marion County)35 (Table 1). Mean age was 44.8 ± 12.7 years, with less than 6% <18 or >64. Nine of 10 (88%) had ≥1 clinical encounters, but only 34% had a routine visit. Two-thirds (68%) had ≥1 ED visits, and 43% had ≥1 inpatient stays.

TABLE 1. - Characteristics of PLWH in Marion County, IN, in 2018: Overall and by 2010–2017 Arrest History and Mental Health Diagnosis* Variable All PLWH Arrest History Mental Health Diagnosis (n = 5283) No (n = 4009) Yes (n = 1274) P No (n = 3434) Yes (n = 1849) P No. (%) No. (%) No. (%) No. (%) No. (%) Age (yrs) <0.05 0.6  13–17 22 (0.4) 20 (0.5) 2 (0.2) 16 (0.5) 6 (0.3)  18–24 230 (4.4) 140 (3.5) 90 (7.1) 131 (3.8) 99 (5.4)  25–34 1083 (20.5) 662 (16.5) 421 (33.0) 715 (20.8) 368 (19.9)  35–44 1126 (21.3) 829 (20.7) 297 (23.3) 740 (21.5) 386 (20.9)  45–54 1525 (28.9) 1235 (30.8) 290 (22.8) 967 (28.2) 558 (30.2)  55–64 1015 (19.2) 870 (21.7) 145 (11.4) 663 (19.3) 352 (19.0)  65+ 282 (5.3) 253 (6.3) 29 (2.3) 202 (5.9) 80 (4.3) Gender >0.9 <0.05  Male 4224 (80.0) 3185 (79.4) 1039 (81.6) 2835 (82.6) 1389 (75.1)  Female 1059 (20.0) 824 (20.6) 235 (18.4) 599 (17.4) 460 (24.9) Race/ethnicity <0.05 <0.05  Black 2566 (49.9) 1723 (44.5) 843 (66.7) 1690 (51.1) 876 (47.7)  Hispanic 455 (8.9) 375 (9.7) 80 (6.3) 343 (10.4) 112 (6.1)  White 1948 (37.9) 1651 (42.6) 297 (23.5) 1162 (35.2) 786 (42.8)  Other 171 (3.3) 127 (3.3) 44 (3.5) 110 (3.3) 61 (3.3) Mental health Dx 1849 (35.0) 1176 (29.3) 673 (52.8) <0.05 NA NA  Mood 1446 (27.4) 906 (22.6) 540 (42.4) <0.05 NA 1446 (78.2)  Anxiety 896 (17.0) 569 (14.2) 327 (25.7) <0.05 NA 896 (48.5)  Psychosis 415 (7.9) 194 (4.8) 221 (17.3) <0.05 NA 415 (22.4)  Attention deficit 113 (2.1) 59 (1.5) 54 (4.2) <0.05 NA 113 (6.1)  Personality 114 (2.2) 45 (1.1) 69 (5.4) <0.05 NA 114 (6.2) Substance use 983 (18.6) 447 (11.1) 536 (42.1) <0.05 261 (7.6) 722 (39.0) <0.05  Alcohol 489 (9.3) 208 (5.2) 281 (22.1) <0.05 123 (3.6) 366 (19.8) <0.05  Opioid 128 (2.4) 54 (1.3) 74 (5.8) <0.05 19 (0.6) 109 (5.9) <0.05  Other/unspec 710 (13.4) 306 (7.6) 404 (31.7) <0.05 157 (4.6) 553 (29.9) <0.05 Clinical encounter 4659 (88.2) 3396 (84.7) 1263 (99.1) <0.05 2810 (81.8) 1849 (100) <0.05  Outpatient 4560 (86.3) 3325 (82.9) 1235 (96.9) <0.05 2720 (79.2) 1840 (99.5) <0.05  Routine visit 1782 (33.7) 1326 (33.1) 456 (35.8) >0.9 875 (25.5) 907 (49.1) <0.05  Emergency department 3613 (68.4) 2419 (60.3) 1194 (93.7) <0.05 1948 (56.7) 1665 (90.0) <0.05  Inpatient 2243 (42.5) 1526 (38.1) 717 (56.3) <0.05 1053 (30.7) 1190 (64.4) <0.05 Arrest history 2082 (39.4) NA NA 1092 (31.8) 990 (53.5) <0.05  Violent 304 (5.8) NA 304 (23.9) 107 (3.1) 197 (10.7) <0.05  Prostitution 48 (0.9) NA 48 (3.8) 12 (0.3) 36 (1.9) <0.05  Drug related 307 (5.8) NA 307 (24.1) 134 (3.9) 173 (9.4) <0.05  Intoxication 274 (5.2) NA 274 (21.5) 111 (3.2) 163 (8.8) <0.05 Felony 650 (12.3) NA 650 (51.0) 279 (8.1) 371 (20.1) <0.05

Bonferroni-corrected P values. Bolded P-values are significant at P < 0.05 or less.

*Includes only PLWH with an eHARS record in Marion County during all 4 quarters of 2018.

Arrest History

One-fourth (24%) of PLWH were arrested during 2010–2017 (Table 1). Among these, half (51%) had ≥1 felony offenses, 24% had ≥1 violent offenses, 24% had ≥1 drug offenses, and 22% had ≥1 intoxication offenses. Four percent were charged with prostitution.

Mental Health Diagnosis

One-third (35%) of PLWH had a mental health diagnosis and two-thirds (64%) among those had ≥1 inpatient hospitalizations (Table 1). The most prevalent mental health diagnoses were mood (27%) and anxiety disorders (17%). More than half (53%) of those with an arrest had a mental health diagnosis (P < 0.05) and those with a mental health diagnosis were more likely to have been arrested for each arrest category, when compared with PLWH without a mental health diagnosis (all P < 0.05).

Substance Use Disorder

One in 5 PLWH (19%) had a SUD diagnosis. Nonopioid drug misuse (13%) was more prevalent than alcohol (9%) and opioid misuse (2%). PLWH and SUD were more likely to have an arrest (42% vs 11%, P < 0.05) or mental health diagnosis (39% vs 8%, P < 0.05; Table 1).

Health care Utilization

Among PLWH, 88% had ≥1 clinical encounters during 2010–2017. Outpatient visits accounted for most utilization, with 86% having ≥1 outpatient encounters (Table 1). Sixty-eight percent of PLWH had ≥1 ED visit; 43% had ≥1 inpatient hospitalizations, and 34% had ≥1 routine visits. Those with an arrest were more likely than those without to have any clinical encounter (99% vs 85%, P < 0.05). In fact, they were more likely than those without an arrest to have received inpatient, outpatient, or ED care (P < 0.05), although arrest history was unrelated to routine care. PLWH and a mental health diagnosis were more likely than those without a mental health diagnosis to have any type of clinical encounter (100% vs 82%, P < 0.05).

HIV Care Continuum

Individuals with an arrest were less likely than those without an arrest to have LTC regardless of whether they were diagnosed during (42% vs 63%, P = 0.02) or before (38% vs 43%, P = 0.012; Table 2) 2018. LTC did not differ by receipt of a mental health diagnosis regardless of when HIV diagnosis was received.

TABLE 2. - HIV Care Continuum Among PLWH in Marion County, IN, in 2018: Overall and by 2010–2017 Arrest History and Mental Health Diagnosis Independent Variable All PLWH Arrest History Mental Health Diagnosis No Yes P No Yes P No. (%) No. (%) No. (%) No. (%) No. (%) Linked to care within 30 d  2018 diagnosis (n = 215) 122 (56.7) 97 (62.6) 25 (41.7) 0.02 94 (55.0) 28 (63.6) >0.9  Pre-2018 diagnosis* (n = 5245) 2187 (41.7) 1705 (42.8) 482 (38.1) 0.012 1413 (41.5) 774 (42.1) >0.9 Retention in care* (n = 5283) 2179 (41.2) 1555 (38.8) 624 (49.0) 0.004 1216 (35.4) 963 (52.1) 0.004 Undetectable viral load* (n = 5283) 2296 (43.5) 1743 (43.5) 553 (43.4) >0.9 1343 (39.1) 953 (51.5) 0.004

Bonferroni-corrected P values. Bolded P-values are significant at P < 0.05 or less.

*Includes only PLWH with an eHARS record in Marion County during all 4 quarters of 2018.

RIC was improved among PLWH and an arrest history versus those with no arrest (49% vs 39%, P = 0.004), although an arrest had no significant association with UVL. Both RIC and UVL improved among those with versus without a mental health diagnosis (52% vs 35% RIC, P = 0.004; 52% vs 39% UVL, P = 0.004).

Multivariable Regression Analysis

Of 5283 PLWH in our cohort, 143 were excluded from regression analyses because of missing values. Imputation was not performed because this accounted for <3% of the cohort, minimizing the risk of bias. PLWH with an arrest history or mental health diagnosis, versus those without, experienced increased odds of RIC and UVL (Table 3). Specifically, those with an arrest had 1.54 times the odds of those without an arrest to have RIC (95% CI: 1.27 to 1.86) and 1.26 times the odds of an UVL (95% CI: 1.04 to 1.52). There was significant interaction between arrest and mental health diagnosis. When either predictor was present, the protective effect on RIC was enhanced, increasing the predicted probability (PP) of RIC to 54% (Fig. 1). This interaction reduced the protective effect on UVL such that a mental health diagnosis without an arrest history led to the highest PP of UVL (54%; Fig. 1).

TABLE 3. - Adjusted Odds of Retention in Care and Undetectable Viral Load by Arrest History, Mental Health Diagnosis, and Outpatient Care Utilization for People Living With HIV in Marion County, IN, in 2018 (n = 5140) Outcome: Odds of Retention in Care Model Controlling for Outpatient Care Utilization No Yes aOR 95% CI aOR 95% CI Arrest history 1.54 1.27 to 1.86 1.18 0.97 1.44 Mental health diagnosis 2.02 1.75 to 2.34 1.05 0.89 1.23 Arrest history*mental health diagnosis 0.75 0.57 to 0.97 1.05 0.80 1.38 Outpatient care visits = 1–9 — 5.14 3.78 6.98 Outpatient care visits = 10–24 — — 8.10 5.96 11.01 Outpatient care visits = 25–49 — — 13.58 10.01 18.43 Outpatient care visits = 50–99 — — 17.07 12.37 23.54 Outpatient care visits = 100+ — — 18.75 12.86 27.34 Outcome: Odds of Undetectable Viral Load Model Controlling for Outpatient Care Utilization No Yes aOR 95% CI aOR 95% CI Arrest history 1.26 1.04 to 1.52 0.91 0.74 1.11 Mental health diagnosis 1.95 1.68 to 2.25 0.90 0.77 1.07 Arrest history*mental health diagnosis 0.61 0.47 to 0.80 0.92 0.69 1.21 Outpatient care visits = 1–9 — — 5.55 4.12 7.47 Outpatient care visits = 10–24 — — 10.73 7.97 14.45 Outpatient care visits = 25–49 — — 16.74 12.43 22.55 Outpatient care visits = 50–99 — — 21.91 15.97 30.04 Outpatient care visits = 100+ — — 28.55 19.58 41.63

Exponentiated coefficients; multivariable logistic regression models adjusted for age, race/ethnicity, sex, and history of substance use; 143 observations were deleted because of missing values.

FIGURE 1.:

Adjusted predictions of retention in care and suppressed viral load among people living with HIV in Marion County, IN, in 2018: By arrest history, mental health diagnosis, and arrest*mental health diagnosis. Average marginal effects