Retention in HIV care during the 3 years following release from incarceration: A cohort study

Background Sustained retention in HIV care (RIC) and viral suppression (VS) are central to US national HIV prevention strategies, but have not been comprehensively assessed in criminal justice (CJ) populations with known health disparities. The purpose of this study is to identify predictors of RIC and VS following release from prison or jail. Methods and findings This is a retrospective cohort study of all adult people living with HIV (PLWH) incarcerated in Connecticut, US, during the period January 1, 2007, to December 31, 2011, and observed through December 31, 2014 (n = 1,094). Most cohort participants were unmarried (83.7%) men (77.0%) who were black or Hispanic (78.1%) and acquired HIV from injection drug use (72.6%). Prison-based pharmacy and custody databases were linked with community HIV surveillance monitoring and case management databases. Post-release RIC declined steadily over 3 years of follow-up (67.2% retained for year 1, 51.3% retained for years 1–2, and 42.5% retained for years 1–3). Compared with individuals who were not re-incarcerated, individuals who were re-incarcerated were more likely to meet RIC criteria (48% versus 34%; p < 0.001) but less likely to have VS (72% versus 81%; p = 0.048). Using multivariable logistic regression models (individual-level analysis for 1,001 individuals after excluding 93 deaths), both sustained RIC and VS at 3 years post-release were independently associated with older age (RIC: adjusted odds ratio [AOR] = 1.61, 95% CI = 1.22–2.12; VS: AOR = 1.37, 95% CI = 1.06–1.78), having health insurance (RIC: AOR = 2.15, 95% CI = 1.60–2.89; VS: AOR = 2.01, 95% CI = 1.53–2.64), and receiving an increased number of transitional case management visits. The same factors were significant when we assessed RIC and VS outcomes in each 6-month period using generalized estimating equations (for 1,094 individuals contributing 6,227 6-month periods prior to death or censoring). Additionally, receipt of antiretroviral therapy during incarceration (RIC: AOR = 1.33, 95% CI 1.07–1.65; VS: AOR = 1.91, 95% CI = 1.56–2.34), early linkage to care post-release (RIC: AOR = 2.64, 95% CI = 2.03–3.43; VS: AOR = 1.79; 95% CI = 1.45–2.21), and absolute time and proportion of follow-up time spent re-incarcerated were highly correlated with better treatment outcomes. Limited data were available on changes over time in injection drug use or other substance use disorders, psychiatric disorders, or housing status. Conclusions In a large cohort of CJ-involved PLWH with a 3-year post-release evaluation, RIC diminished significantly over time, but was associated with HIV care during incarceration, health insurance, case management services, and early linkage to care post-release. While re-incarceration and conditional release provide opportunities to engage in care, reducing recidivism and supporting community-based RIC efforts are key to improving longitudinal treatment outcomes among CJ-involved PLWH.

highly structured setting in which to diagnose and initiate treatment for HIV, 22,23 CJS involvement can be an opportunity for facilitating retention in HIV care. 17,22 Studies from our group and others within the last decade have shown that PLH generally have well-managed disease during incarceration and that it is the period of time immediately after their release during which they are prone to experiencing poor HIV treatment outcomes; virologic failure is common post-release but often only identified and addressed upon reincarceration. 16,24 A 2004 retrospective cohort study showed that, of 292 PLH who were treated for HIV while in prison, released, and subsequently reincarcerated, 59% achieved viral suppression prior to release. 24 However, upon re-entering the CT prison system after spending ≥3 months in the community, any benefit from ART that was observed during incarceration had been lost. 24 This suggests that it may be release into the community, rather than incarceration itself, that makes individuals vulnerable to poor HIV treatment outcomes. These DOC-based studies, however, were limited in that they could only follow PLH while they were incarcerated; assumptions had to be made, especially about non-recidivists, regarding the success of their HIV treatment during their time in the community.
A4. The longitudinal impact of incarceration and community re-entry on inmates' continuity of HIV care remains poorly understood. Successful treatment of HIV requires immediate linkage to HIV care after release and regular (every 3-6 months) monitoring by HIV-1 RNA viral load (VL). 25 Correctional facilities generally provide each inmate with a 10-to 30-day ART supply at their time of release, under the assumption that inmates will receive an ART refill upon linking themselves to a healthcare provider post-release. 22 In a national study, former jail detainees who met with an HIV provider within 30 days of release had a higher probability of achieving viral suppression within 6 months. 26 In one Texas county, however, only 20% of recently released inmates enrolled in an HIV clinic, 27 and only 17.7% filled ART prescriptions 28 within this 30-day window.
Despite the need to better understand HIV management after release, no observational study to date has effectively collected statewide, longitudinal data on HIV outcomes for PLH during and after their transition from a correctional facility into the community. Research on this topic has been primarily limited to cross-sectional and community-based cohort studies, many of which have recall bias, limited follow-up time, and suboptimal study participant retention rates (typically 50%, with missing data either ignored or imputed). The previously described Yale University DOC-based studies have the benefit of using documented custody and biological data, thus minimizing recall bias, but they have lacked access to post-release disease monitoring data until now.
A5. The effectiveness of transitional linkage-to-care services aimed to improve HIV treatment outcomes post-release is also unclear. Qualitative and mixed-methods studies suggest that poor prison discharge planning can lead to poor linkage to care upon community reentry. 29 Individuals often have difficulty accessing services in order to meet their basic needs, such as housing and employment, and may relapse to substance abuse or engage in transactional sex work soon after their release. 30 These behaviors increase their risk of transmitting HIV, being reincarcerated, 30 and, likely, defaulting from ART. In some US states, transitional linkage-to-care (TLC) and/or case management programs have been implemented to help connect inmates to the healthcare, substance abuse treatment, mental health, housing, and vocational services they require upon release. 31 Studies on the effectiveness of such programs have primarily followed former jail detainees in the community for ≤6 months and have found mixed results. Most of these studies have been based in the EnhanceLink Initiative, a national project that evaluated TLC interventions at ten US jail sites. 31 Despite using TLC services, only 58% of jail detainees visited an HIV clinic within the first 3 months post-release and only 47% had a visit in the second 3-month period. 32 In contrast, the only known study that has been conducted in prison settings showed high care retention rates irrespective of TLC services. 33 Thus, TLC services appear to be beneficial, at least in jail settings, 29 but more research is needed order to appropriately target and manage the needs of inmates most at risk for poor HIV treatment outcomes. A6. Connecticut is an ideal setting for exploring the longitudinal impact of incarceration on HIV treatment outcomes. Consistent with the majority of the country, substance abuse in CT is a major contributor to both HIV infection 34 and incarceration. 35 In 2012-2013, CT's incarceration rate for adult US residents averaged 431 per 100,000. 36 In addition, CT has one of the highest recidivism rates nationally; of inmates released in 2004, 43.7% were reincarcerated within 3 years. 37 The general population in CT has an HIV prevalence of 0.36%, which ranks 11 th among all US states and 2 nd among the six US states with integrated correctional health systems. 11,38 While at least 1.6% of male inmates are known to the DOC as having HIV, the prevalence rate is more than twice as high (3.9%) among female inmates. 11 In addition, blacks and Hispanics disproportionately comprise 24% of the state population but approximately 65% of all PLH in CT. 34 In the previously mentioned prisoner cohort studies conducted at Yale University, which used CT DOC data similar to data we are proposing to use in this study, it was discovered that only half of PLH who were virally suppressed prior to release were still suppressed upon reincarceration. 19,24 While only one-third of prison entrants were virally suppressed at time of incarceration, ART regimens, after being optimized, became effective for 70% of inmates during their incarceration period. 23 In addition, disparities were noted in that HIV-infected CT inmates were more likely than members of the general CT prison population to be female, of ethnic or racial minorities, and substance abusers. 24 Yet blacks and Hispanics were less likely than whites to be virally suppressed upon entry to the DOC, and women who had achieved viral suppression during a prior incarceration were less likely than men to be observed upon reincarceration to have maintained suppression. 39 The timeline of HIV-related events and the combination of factors that ultimately lead to these poor outcomes being observed upon reincarceration remain unclear. Linking correctional data to community-based clinical data has recently been recognized as an innovative but infrequently used strategy to address unanswered questions regarding longitudinal HIV outcomes in CJS-involved populations. 40 In addition to having access to and experience with linking correctional databases based on inmate ID number, 19,23,39 Drs. Altice and Meyer at Yale University recently acquired access to longitudinal, community-based HIV surveillance data from the CT Department of Public Health (DPH). Thus, the goal of the proposed study is to use these data to follow CJS-involved PLH longitudinally after release into the community in order to gain a better understanding of the timing of linkage to and retention in HIV care and rates of viral suppression in this vulnerable population.

A7. Innovation
The majority of inmates experience virologic failure after release into the community. While this is presumably due to insufficient engagement in HIV care, the timing of care engagement and treatment failure remains poorly understood. Previous studies have been limited to examining HIV treatment outcomes in specific sub-populations of PLH (e.g., drug users) either during incarceration or after release. In contrast, the proposed study will allow all CJS-involved PLH to be followed longitudinally in the community, regardless of whether they re-enter the CJS at a later date. This study will be the first to link data from an integrated DOC and a DPH HIV surveillance system at a statewide level, and is highly innovative in its ability to: 1) assess the likelihood and timing of post-release linkage to HIV care, retention in care, and virologic failure for all CJS-involved PLH, and 2) evaluate the effects of factors such as TLC services utilization, viral suppression at time of release, and the full complexity of one's interaction with the CJS (e.g., length and frequency of incarceration, year of release, and correctional supervision in the community by means of probation or parole) on post-release HIV treatment outcomes. Additional strengths include the potential ability to explore sex and racial/ethnic disparities, which is typically limited by small sample size in observational cohort studies.

B. Aims
The innovative combined use of CJS-based DOC data and community-based eHARs data will allow me to objectively assess HIV treatment outcomes for all PLH released from a CT DOC facility during 2007-2015, according to the following specific aims: Aim 1. Assess post-release linkage to HIV care: (a.) Determine the proportion of PLH with a VL monitored within three critical time windows following release: 1, 3, and 6 months postrelease. 25 (b.) Among the PLH who are linked to care (as defined above based on VL monitoring), determine the proportion that demonstrate virologic failure (VL>400 copies/mL 25 ) at any point during these time windows. (c.) Estimate the median time to first VL post-release.
Aim 2. Assess post-release retention in care, using two different accepted definitions of retention: (a.) Examine the proportion of PLH with ≥2 VLs drawn at least 3 months apart during the first 12 months post-release and, among the PLH who fulfill this definition of retention in care, the proportion that demonstrate virologic failure during these 12 months. (b.) For PLH who are linked to care within 3-6 months after release (per Aim 1a. criteria), examine the proportion of PLH with a VL monitored within 3-6 months after the initial linkage-to-care VL. Additionally, among the PLH who fulfill this definition of retention in care, examine the proportion that demonstrate virologic failure during these time frames. (c.) Estimate the median length of time to the next VL that is drawn after the first post-release VL.

C.1. Study population:
The subjects of interest are PLH in CT with a history of involvement in the CT DOC, an integrated correctional system composed of 17 facilities (16 for men, 1 for women) for both pre-trial jail detainees and sentenced prisoners. Of the 16,594 individuals (both HIV-infected and uninfected) incarcerated in 2014, 94% were men, and 41% self-identified as non-Hispanic black, 31% as non-Hispanic white, and 26% as Hispanic. 41 By 2010 year-end estimates, at least 1.7% of persons in CT DOC custody had diagnosed HIV (1.6% of men, 3.9% of women). 11 This is likely an underestimate of the true HIV prevalence due to under-reporting, under-diagnosis, and non-disclosure of HIV status. In addition, CT DOC data spanning the years 2005-2012 demonstrate that the number of women and racial/ethnic minorities in the CJS with diagnosed HIV is disproportionately higher that the number of minorities in the general CJS population (19.1% of incarcerated PLH are female, 47.5% non-Hispanic black, and 32.3% Hispanic). 23 C.2. Data sources, database linkage, and selection of study participants from existing data: C.2.1. Description of the CT DOC database: Upon intake into the CT DOC, data are collected on individuals' demographic characteristics and medical and psychiatric health, including substance use disorders. The DOC also maintains a pharmacy database of all medications, including ART, prescribed during incarceration and has an extensive custody database, including dates of incarceration and release, transfer within the DOC or to a hospital, criminal charges, whether the person was in pre-trial detention or received a prison sentence, and whether the person was released on parole or probation (i.e., community supervision). For prior studies, Drs. Meyer, Altice, and their colleagues successfully merged these data sources to create a comprehensive database containing demographic, biological, pharmacy, and custody data corresponding to all incarceration periods in CT during 2005-2012. 19,23,39 However, this DOC cohort does not allow us to follow PLH who are released to the community and never re-incarcerated. Thus, we propose to link these data to the extensive statewide database described next.

C.2.2. Description of CT DPH-managed data (eHARS and TLC):
In the 1980s, the Centers for Disease Control and Prevention (CDC) established an HIV surveillance program to collect, analyze, and distribute data on HIV based on mandatory disease reporting. By 2008, all states had implemented confidential HIV infection reporting as part of a name-based electronic surveillance system called the enhanced HIV/AIDS Reporting System (eHARS). 42 This DPH-managed HIV surveillance database includes demographic information (i.e., sex, race/ethnicity, and age) for all PLH in CT, and, for individuals who have been incarcerated, their unique inmate ID number. eHARS also contains HIV/AIDS diagnosis data, including the mode of HIV exposure and initial VL and CD4 count at time of diagnosis. After diagnosis, all results from HIV monitoring tests (VL and AIDS-defining CD4) conducted at any CT laboratory facility are reported to the DPH and entered into eHARS, which allows for continuous monitoring of HIV disease progression over time. Mortality data are also collected and confirmed by cross-referencing against National Death Index and Social Security records. The DPH also documents inmates' utilization of the DPH-funded Transitional Linkage into the Community statewide program (Project TLC). TLC assigns a case manager to follow PLH recently released from the CT DOC for up to 30 days post-release, assisting them with linkages and referrals to community-based medical and social services. TLC data also include dates of clinic visits with HIV providers working at clinics funded by the Ryan White HIV/AIDS program, which is recognized as the most common source of HIV care for CJSinvolved PLH. 31 C.2.3. Database linkage and selection of study participants from existing data: Drs. Meyer and Altice at Yale have already obtained data for all PLH incarcerated in the CT DOC during the period 6/1/1999 -12/31/2012. Four primary sources of CT DOC data were used to compile this database: 1) statewide database of all demographic data (e.g., birth date, sex, race, marital status, education status, medical insurance status upon DOC entry/exit, and presence of medical and psychiatric health issues, including substance abuse); 2) laboratory database with all VL/CD4 counts collected during incarceration; 3) pharmacy database, including all prescription information (e.g., ART); and 4) custody and criminal offense database, including dates and types of offense charges, intra-system transfers, and community supervision data. These databases were merged for analysis based on inmate number, after which all unique identifiers were removed. Through ongoing collaboration with the CT DOC, we will update create an updated database for individuals released from the CT DOC between 01/01/2007-12/31/2015 and link these data to the eHARS data available from CT DPH.
As shown in Figure 1, we will work with an experienced data manager at the CT DPH, using SAS 9.4 and the Link Plus probabilistic record linkage program 43 to link CT DOC data with eHARS HIV monitoring, mortality, TLC, and clinic data based on inmate ID number (in accordance with the Disclosure of Identifiable Health Data Section 19a-25-3(3)). If necessary, name and date of birth will be used to confirm that the data have been accurately merged. Subsequently, all identifying information including inmate ID number will be replaced with anonymous participant ID numbers prior to analysis. Specifically, the following variables will be requested from eHARS: sex, race/ethnicity, date of first positive HIV test, age at HIV diagnosis, source of HIV report (e.g., private practice, hospital, state facility, federal laboratory, blood bank, etc.), county of HIV diagnosis, disease stage at diagnosis (HIV vs. AIDS), and mode of HIV exposure (e.g., heterosexual, intravenous drug use [IDU], no identified risk factor, or no reported risk). All reported CD4 T-cell and VL values following HIV diagnosis will also be extracted. From the Project TLC database, the types of TLC services provided and the dates of pre-and post-release encounters with TLC case managers will be extracted. Dates of post-release encounters with Ryan White HIV providers will be requested to provide insight into the extent of TLC and HIV care services utilized.
Persons included in the final analysis will meet the following criteria: 1) confirmed HIV seropositive; 2) incarcerated and released at least once from a CT DOC facility during 2007-2015; 3) have at least one set of lab data available during at least one incarceration period; 4) have pharmacy data available during at least one incarceration period; and 5) spend ≥6 months postrelease in the community prior to the end of the study period. Based on CT DOC data from 2005-2012, we estimate there will be at least 1,125 eligible subjects released during 2007-2015. After accounting for recidivism rates, we anticipate that these subjects will experience a total of 2,025 incarceration periods with corresponding post-release periods that will be eligible for analysis. We will also conduct a subanalysis limited to non-recidivists (i.e., individuals who are released, spend ≥6 months in the community, and are never reincarcerated during the study period), since data on this subpopulation of CJS-involved PLH are particularly limited in the literature.

C.3. Statistical analysis:
C.3.1. Construction of outcome variables: Throughout the study follow-up period (1/1/2007 -12/31/2015), each inmate will likely have a series of VL measurements with corresponding test dates. These reported VLs represent times at which individuals interfaced with the healthcare system (both within prison or jail and in the community); because a patient cannot have a VL measured without seeing a provider, VL can be viewed as a proxy for clinic visits and receipt of an ART prescription.
Prisoners receive 30 days of ART post-release and should be evaluated by a provider within this timeframe. In addition, the US Department of Health and Human Services (DHHS) HIV monitoring guidelines dictate that a VL should be repeatedly drawn every 3-4 months for HIV monitoring. 25 For persons on a stable, suppressive ART regimen, the timing between VL monitoring can sometimes be extended to every 6 months. 25, 44 Optimal viral suppression of HIV is defined as a VL persistently below the level of detection (which can be as high as 400 copies/mL depending on the assay used). 25 Our primary study outcomes of interest are: (1) linkage to care, with different degrees of linkage defined as having a VL drawn within 1, 3, or 6 months of release (Aim 1a); and (2) retention in care, which will be explored using two different definitions: (Aim 2a) having two or more VLs drawn at least 3 months apart within the first year after release, and (Aim 2b) having a VL drawn within 3 or 6 months after the first post-release VL. These definitions of linkage to and retention in care imply that the maximum allowable period of time between VLs should generally be 6 months in order to be sufficiently engaged in care in accordance with DHHS HIV monitoring guidelines. 25 A secondary study outcome will be "consistent viral suppression" (<400 copies/mL) during the previously described time frames of interest (Aims 1b, 2a, and 2b). We will also estimate the median length of time between (1) the release date and first post-release VL (Aim 1c), and (2) the first and second post-release VLs (Aim 2c). These study outcomes will provide insight into how many individuals are sufficiently engaging in HIV care after release as well as the time points at which individuals most frequently drop out of care and/or develop virologic failure, and the times at which interventions may be most effective at improving these HIV treatment outcomes.
As illustrated in Figure 3, each individual's date of release from jail or prison will be used as a start date for the linkage to care and retention in care outcomes. For example, a PLH released on January 1 st , 2009, will be evaluated to determine whether a VL was drawn by February 1 st , April 1 st , and July 1 st , 2009 (per Aim 1a's definition of successful linkage to care within 1, 3, and 6 months post-release, respectively). Retention in care will be determined based on whether the individual has at least two reported VLs drawn at least 3 months apart prior to January 1 st , 2010 (Aim 2a). For Aim 2b (not shown in Figure 3), the first post-release VL (e.g., April 1 st , 2009) will be used as a start date to determine whether the second post-release VL was drawn within 3-6 months after the first post-release VL (e.g., by July 1 st , 2009, or October 1 st , 2009). In order to evaluate whether individuals are successfully virally suppressed during these time frames (Aims  1b, 2a, and 2b), all reported continuous VL measurements will be categorized as "viral suppression" or "virologic failure" using an HIV-1 RNA level threshold of 400 copies/mL; if any VL drawn within the time frame under analysis is ≥400 copies/mL, the entire time frame will be coded as corresponding to a virologic failure. Finally, the median time-to-events of interest will be estimated using each individual's: (1) release date (start date) and date corresponding to the first VL drawn in the community (event of interest, Aim 1c); and (2) date of first post-release VL (start date) and second post-release VL (event of interest, Aim 2c).

C.3.2. Construction of covariates:
Covariates will be included in all models to control for potential confounding and to study the exposures of interest alluded to in the Specific Aims (i.e., sex, race/ethnicity, age, history of addiction to a substance, number of health comorbidities, length of incarceration, number of prior incarcerations, supervised release, year of release, viral suppression at time of release, and release with TLC services). Specifically, covariates will include demographic variables such as age at time of release (categorical; 18-34, 35-54, or ≥55 years old), sex (binary), race/ethnicity (categorical; non-Hispanic white, non-Hispanic black, Hispanic, or other), marital status at time of DOC entry (binary; yes/no), education level at time of DOC entry (binary; less than high school vs. completed high school), and medical insurance status upon DOC entry (binary; yes/no) and at time of release (binary; yes/no). Health-specific variables will include HIV transmission route (categorical; heterosexual, homosexual, IDU, or not reported), ART status at time of release (binary; on ART vs. not), viral suppression status at time of release (binary; yes/no, based on whether a VL<400 copies/mL is reported by the DOC within the 3 months leading up to release), a co-morbidity rating (an ordinal 1-5 scale representing the number of non-HIV medical and psychiatric co-morbidities treated while in the DOC), an addiction severity score (an ordinal 1-5 scale representing the extent of one's addiction to a substance on intake, as determined by a trained DOC medical professional), and the extent of TLC services utilized (categorical; none, prerelease only, or post-release). Criminal justice variables will include criminal offense charge data (categorical; based on nature of crime), total number of previous incarcerations in one's lifetime (continuous), length of most recent incarceration (categorical; ≤5 days, 6-30 days, 31-90 days, or >90 days), year of release, and type of community release (categorical; unsupervised, parole, or probation). Because a single individual may experience more than one incarceration and consequently may have more than one post-release period over the course of the study (Figure 2), a "release number" variable will be created representing the sequential release period under analysis (categorical; e.g., an individual with three incarcerations during the study period may have up to three different post-release periods which will be chronologically assigned indicator values ranging from 1-3). A continuous "cumulative DOC time" variable will also be created based on the summed length of time each subject has spent in a correctional facility prior to the release period under analysis.

C.3.3. Overview of planned analyses:
For the subgroup analysis of PLH who are not reincarcerated during the study period and therefore have only one post-release period to analyze, standard logistic regression and Cox proportional hazards models can be used to examine the binary endpoints and the time-to-event endpoints of interest, respectively. Otherwise, a generalized estimating equations (GEE) model for binary data [45][46][47] and a modified Cox proportional hazards model known as the proportional hazards frailty model 48,49 will be used to analyze events of interest after release from each incarceration. These methods will allow us to account for the correlation (ρ) between a single subject's multiple post-release periods where it is possible for the event of interest to occur once during each period. For example, an individual incarcerated three times during the study period may have as many as three post-release periods during which he/she can be followed in the community with regard to post-release HIV treatment outcomes. For individuals with multiple release periods, each release period will be included as an observation and intrasubject correlation will be accounted for by treating each individual as a cluster.
For logistic regression and GEE models, individuals with at least one reported VL during the post-release time period of interest will be assigned a "linkage to care" outcome = 1, whereas individuals lacking a reported VL will be assigned an outcome = 0. Similarly, individuals fulfilling the two different definitions of retention in care will be assigned a "retention in care #1" outcome = 1 (if not fulfilled, outcome = 0) and "retention in care #2" outcome = 1 (if not fulfilled, outcome = 0). The secondary outcome of interest (virologic failure) will only be assessed for the subset of individuals who have a reported VL that can be evaluated; individuals with at least one reported VL≥400 copies/mL within the time frame of interest will be assigned a "virologic failure" outcome = 1, whereas individuals whose VLs are all <400 copies/mL within the time frame of interest will be assigned a "virologic failure" outcome = 0. Outcomes will be examined for each release period (j), clustered by individual (i). The most robust model will be selected based on minimization of the quasi-likelihood under the independence model criterion (QIC) for GEE. 50 Similarly, the proportional hazards frailty model for recurrent outcomes will be used to assess the timing of post-release VLs (per Aims 1c and 2c). Individuals not experiencing the outcome of interest during their follow-up time in the community will be right-censored on study end date (12/31/2015), date of reincarceration if applicable, or date of death if prior to the study end date. A model will be selected using minimization of the Bayesian Information Criterion (BIC). 51 For all study aims, descriptive analyses will first be conducted. Then, for the entire sample (≥1,125 estimated subjects with an anticipated total of 2,025 incarceration periods), GEE modeling will be used to examine the variables associated with VL monitoring within the previously described post-release time frames of interest (Aim 1a, 2a, 2b). For the group of individuals who have VLs reported within these time frames, potential predictors for having at least one measurement indicating virologic failure will also explored (Aim 1b, 2a, 2b). Finally, analyses using proportional hazards frailty models will be conducted to estimate time-to-first-VL (Aim 1c) and time between first and second VL (Aim 2c).

C.4. Strengths and limitations:
The key strength of this study lies in the use of existing data extracted from multiple databases, including an integrated correctional health database and a statewide HIV surveillance database. The innovative construction and use of a comprehensive database containing reliable correctional and public health data will enable all CJS-involved PLH in CT to be followed objectively in the community with minimal loss to follow-up. Due to the comprehensive nature of the data, extensive periods of time without a laboratory blood draw will be extremely informative as they will indicate a lack of engagement in HIV care as opposed to poor study retention. Specifically, for inmates who do not have any labs drawn post-release and were never reincarcerated, this study design allows these "missing" VLs to be used to as a study strength, as opposed to a limitation, because the lack of data indicates that these individuals are not successfully being linked to and retained in care.
This study's major limitation is the granularity of patient-level factors that contribute to poor HIV outcomes, including the DOC's absence of detailed substance use history and treatment data and HIV drug resistance profiles. Nevertheless, addiction severity scores on DOC intake will help determine which individuals are at risk for substance abuse relapse postrelease, and available pharmacy, clinical, and biological data will help predict which individuals are adhering to ART. Finally, the challenges inherent to linking multiple databases will be mitigated through collaboration with DPH database managers with extensive experience managing and linking data. Although these CT-specific findings may not be generalizable to US states with very different DOC HIV policies, this study takes significant steps toward informing much-needed health policy targeting at-risk, CJS-involved PLH. It also lays the groundwork for future studies using eHARS and other community-based data to assess and improve HIV outcomes in CJS-involved PLH and other vulnerable populations.

19.
Meyer JP, Cepeda J, Springer SA, Wu J, Trestman RL, Altice FL. HIV in people reincarcerated in Connecticut prisons and jails: an observational cohort study.

Protection of Confidentiality
We will adhere to all federal regulations regarding research involving human subjects. Prior to analysis, all health and criminal justice data will be approved for use by the institutional review boards at Yale University and the Connecticut Department of Public Health (CT DPH), and the Research Advisory Committee at the Connecticut Department of Correction (CT DOC).

Current status of applications for study approval from other institutional IRBs:
Yale HIC (including a prisoner representative) -approved (documentation included) CT Department of Correction Research Advisory Committee Application -under review

1.a Human Subjects Involvement, Characteristics, and Design
To be included in this study are all persons living with HIV (PLH) released from a CT jail or prison at least once during 2007-2015. The anticipated number of study subjects is 1,125 with approximately half of subjects experiencing more than once incarceration during the study followup period. Individuals included in the final analysis will meet the following criteria: 1) confirmed HIV seropositive; 2) incarcerated at least once within a CT DOC facility; 3) have at least one set of lab data available during incarceration; 4) have pharmacy data available during at least one incarceration period; and 5) able to be followed post-release for ≥6 months. Individuals who die during their first incarceration period or are a missing pre-release (within 30 days prior to release) laboratory blood draw will be excluded. In addition to having HIV, individuals may also have diagnosed or undiagnosed psychiatric illness, substance use disorders, and other medical comorbidities. All participants will be ≥18 years old, given what is known about the epidemiological distribution of HIV in CT.

1.b Sources of Materials
All data used in this study will originate from pre-existing CT DOC pharmacy records and custody and criminal offense charge data, as well as CT DPH TLC and HIV/AIDS surveillance data (eHARS), originally collected for non-research purposes. We will request that our collaborators at the DOC provide us with updated data spanning January 1, 2007 through December 31, 2015.
Using subjects' unique inmate ID numbers, our researcher Kelsey Loeliger will work with our collaborating eHARS data manager at DPH to link this updated DOC data with DPH eHARS and TLC data for all research subjects and producing a fully de-identified database for our research purposes.

1.c Potential Risks
We will not be administering surveys, conducting research-related laboratory blood draws, or directly interacting with study participants in any way. As there are no medical procedures or interventions in this protocol, there are no inherent risks to the study. Because the data used for analysis will be fully de-identified without constructing a codebook that could link personal identifiers and protected health information, a breach of confidentiality is highly unlikely. The measures we will take to minimize this risk to the individual subjects are further detailed below.

2.a. Recruitment and informed consent
Because this study proposes to use data that have already been collected, informed consent will not be obtained. Nevertheless, our plans for the use of these confidential data will first need to be approved by two separate institutional review boards and a research advisory committee as stated above.

2.b. Protection against risk
All data linkage involving the use of personal identifiers will be conducted internally within the DPH. Personal identifiers will only be accessible to select DPH staff members who have been pre-approved for access to these personal identifiers under a pre-existing DPH IRB application. After the DOC and DPH data have been merged, all subjects will be assigned a unique study code and unique identifiers will be removed before the merged database is shared with our Yale research team. The de-identified database will be kept on password protected and encrypted computers with secure server backup requiring multifactor authentication logons, maintained by Yale University IT. Data files will be stored on a limited access J drive, which is maintained by our Yale AIDS Program data manager and to which access is restricted by Yale University IT. Only research team members working directly on the study will have passwords to the database and servers. Additionally, all research personnel have completed NIH online training on the Protection of Human Subjects and Good Clinical Practice and/or Yale University Human Subjects Protection Training and HIPAA Privacy and Security Training.

Potential Benefits of the Proposed Research and Importance of the Knowledge to be Gained
While the risk to the individual is minimal due to multiple safeguards against loss of confidentiality, the potential benefit is significant. The knowledge to be gained holds the potential to guide considerable public health policy advances regarding the management of HIV in criminal justice populations. These advances are expected to benefit criminal justice-involved PLH in the near future, possibly including the same individuals that we are proposing to study.

Collaborating Sites
All collaborating sites have protocols in place that have been approved by their respective institutional review boards in order to protect the use of their privileged information. We have already received approval from the Yale Institutional Review Board, which includes a prisoner representative. Our application to CT DOC Research Advisory Committee is currently under review.
Informed Consent Forms -Not applicable due to the retrospective nature of the study.
Questionnaires -Not applicable due to the retrospective nature of the study.

HUMAN INVESTIGATION COMMITTEE (HIC) CONFIDENTIALITY PLEDGE
I recognize the importance of maintaining the confidentiality of identifiable health data collected by the Connecticut State Tumor Registry and other units of the Connecticut Department of Public Health (DPH), and of assuring the right to privacy of persons, facilities, and agencies which cooperate with the Registry and the DPH or participate in the DPH's data collection efforts. I also understand that the Registry and DPH are legally obligated to protect the privacy of public health information. I have read the Connecticut State Statute Sec. 19a-25 and Sections 19a-25-1 through 19a-25-4 of the Regulations of CT State Agencies concerning confidentiality of records concerning morbidity and mortality and have been advised that DPH can take necessary action if a breach of confidentiality occurs.
I therefore pledge that I will NOT divulge the identity of patients, physicians, facilities, or agencies included in data obtained from the DPH with HIC approval with anyone other than: • members of the Registry staff or other DPH unit from which the data are obtained or • persons who are also approved for access to the data by DPH HIC and have also signed a DPH confidentiality pledge