Correlates of psychological intimate partner violence with HIV care outcomes on patients in HIV care

This cross-sectional study included PLWH from two large collaborations in the U.S. including Puerto Rico to ensure demographic, geographic, and clinical diversity, with PLWH from geographically dispersed clinical care settings to enhance generalizability and study settings focused on substance using PLWH to ensure inclusion of vulnerable populations. Centers for AIDS Research Network of Integrated Clinical Sites (CNICS) (https://​www.​uab.​edu/​cnics/​) is a cohort study of PLWH in clinical care at eight sites across the US from 1995 to present [34]. Seek, Test, Treat, Retain (STTR) is a consortium initiated by the National Institute on Drug Abuse [35, 36] to address research questions related to HIV care continuum outcomes among vulnerable particularly substance using populations.

From CNICS, we included PLWH representing six CNICS sites: the 1917 Clinic at University of Alabama-Birmingham; Case Western Reserve University in Cleveland, OH; Fenway Community Health-Boston, MA; Owen Clinic at University of California at San Diego; University of North Carolina; and Madison Clinic at Harborview Medical Center/University of Washington-Seattle. IPV data collection from these patients began in 2016. From the STTR study consortium, we included PLWH from two studies: Proyecto PACTo (Proveyendo Acceso a Cuidado y Tratamiento) [37], and Project RETAIN [38].

Proyecto PACTo is a clinical trial which evaluated effectiveness of the “Enhanced HIV Care Access and Retention Intervention” in achieving HIV virologic suppression among substance using PLWH in Puerto Rico that ran from 2013 to 2014. Project RETAIN is a clinical trial done to evaluate the efficacy of an integrated “Retention Clinic” in achieving virologic suppression among cocaine/crack-using PLWH in Florida and Georgia that ran from 2013 to 2015.

For both the STTR (PACTO/RETAIN) and CNICS groups, our analyses included PLWH who were 18 years of age or older and completed a self-administered patient-reported outcome (PRO) measure querying IPV. For PACTO/RETAIN, we included PLWH who answered IPV questions at their baseline visit. For CNICS, we included PLWH that had been administered the IPV measure in their most recent PRO assessment at the beginning of their routine care visit. Table 1 shows full inclusion and exclusion criteria across consortia and networks.

Both CNICS and STTR have data repositories that harmonize and integrate demographic, clinical, laboratory, and other data such as patient-reported measures including IPV [36, 39].

CNICS utilizes a brief validated measure, the IPV-4 [40], that is inclusive of physical IPV, sexual IPV, and two forms of psy-IPV (see Table 2): controlling behavior by an intimate partner and fearfulness of a partner in the past year. The IPV-4, initiated in 2016, is administered to PLWH within CNICS clinics on-site prior to their clinic visit as part of routine care. The PACTO and RETAIN studies used the second two items of a lifetime IPV measure known as STaT [an acronym for ‘slapped, threatened, and throw (things)’], as well as an item querying controlling behavior that was used in development of the STaT measure [41]; these items were adapted to reflect a 6 month recall period (Table 2). Data from items regarding controlling behavior were harmonizable across CNICS and PACTO/RETAIN (in Table 2, CNICS item 1 harmonizes with PACTO/RETAIN item 3), as were data regarding threatening behavior (in Table 2, CNICS item 2 harmonizes with PACTO/RETAIN items 1 and 2); these two areas comprise our dimensions of psy-IPV. Psy-IPV was considered reported if a person answered “yes” to any of these items.

All participants were informed that responses to study measures are kept confidential within their respective sites and that data used for research purposes is de-identified. In CNICS, patients are informed, prior to answering questions, that their providers will see their responses. Patients in all studies were given the option of not answering any questions. PLWH in CNICS indicating any type of violence in the IPV-4 prompt a pager alert for an on-site social worker to check-in with the patient, same-day, on-site during their clinic visit, at which point they are evaluated and provided with resources if needed. For the PACTO and RETAIN studies, indication of IPV prompted an assessment in real time with the patient by a licensed psychologist who was part of the study team.

We examined the association between psy-IPV and the following risk behaviors and symptoms: depressive symptoms, ART adherence, current use of methamphetamine/crystal, cocaine/crack, illicit opioids, marijuana, and alcohol, as well as current binge alcohol use. Depressive symptoms was defined as a score of ≥10 on the PHQ-9 [42, 43] in CNICS, and a score of ≥16 on the Center for Epidemiological Studies-Depression (CES-D-20) [44] measure in PACTO and RETAIN. Adherence to ART was measured as a percentage of HIV medication taken over the past 30 days [45]. Current alcohol use was defined as a score of ≥1 as measured by the AUDIT-C [46] and current binge drinking was defined as having had 5 or more drinks on one occasion using the AUDIT-C. Drug use was identified using the ASSIST [47] which includes methamphetamines, illicit opioids, cocaine/crack, sedatives, stimulants, hallucinogens, inhalants, and marijuana. We examined associations between psy-IPV and clinical outcomes including HIV care cascade steps, specifically ART use, detectable viral load, and CD4 count, as well as self-reported ART adherence.

We performed statistical analysis using STATA version 14.2. We used logistic regression for analyses with binary variables and linear regression for analyses with continuous variables, to compare those who did and did not indicate psy-IPV. Observations were weighted by inverse probability of treatment weights (IPTW) to reduce the ratio of adjustment factors to outcomes in models; IPTW using propensity scores is a good alternative to control for confounding when there are seven or fewer events per confounder [48]. We generated propensity scores using logistic regression based on the model by Hernan et al. (2000) [49]. IPTWs were calculated using two different propensity score models to estimate the propensity of indicating psy-IPV: (1) Propensity of indicating psy-IPV was estimated adjusting for a limited set of covariates including age, site, and race/ethnicity; (2) Propensity of indicating psy-IPV was estimated by adjusting for a larger set of covariates including age, site, race/ethnicity, current substance use (methamphetamine, cocaine/crack, illicit opioids, marijuana, alcohol and binge alcohol use) and depressive symptoms. Age was modeled linearly. Propensity score density plots were created for both scores to assess adequacy of overlap between participants that reported psy-IPV and those that did not report psy-IPV. We performed sensitivity analyses in CNICS with propensity scores that also adjusted for physical IPV. We examined demographic characteristics among those with and without psy-IPV including age, sex, race/ethnicity, and sexual orientation. We examined psy-IPV and risk behaviors and symptoms as well as clinical characteristics including HIV care cascade steps. These associations were assessed using the limited IPTW, and associations except depressive symptoms and substance use were also assessed using the more complex IPTW. When age was compared between those with and without psy-IPV, the model was not weighted and simply adjusted for study and race since age was in both IPTWs. We examined the association of psy-IPV and HIV care cascade variables exclusively in the CNICS cohort, due to smaller sizes of the other studies. We note that due to discordant recall periods between the IPV measures administered by study sites, we did not perform formal mediation analyses.

A list of all statistical models that were run can be found in Supplemental Table 1.

A total of 5950 PLWH were included in this study, with 564 (10%) indicating recent psy- IPV; CNICS contributed the largest number of participants (5195), followed by PACTO (408), and RETAIN (347) (Table 3). Of the 5950 participants, 18% were female sex at birth, and the mean age was 47 years. One percent of participants were transgender. Race and ethnicity varied across studies, with CNICS having 45% white, PACTO having 99% Hispanic, and RETAIN having 82% black participants. Depressive symptoms (25%) and substance use (up to 62%) were commonly reported. In CNICS, which assessed physical/sexual and psy-IPV, 50% of those indicating psy-IPV did not report physical/sexual IPV. PACTO/RETAIN did not have data on physical or sexual IPV within a harmonizable time window (asked lifetime only), hence, psy-IPV independent of physical/sexual IPV could not be assessed for those sites.

Propensity score density plots (Supplemental Figs. 1 and 2) showed good overlap between participants reporting psy-IPV and those not reporting psy-IPV, confirming that IPTW is an appropriate method to control for confounding (Table 4; see Supplemental Tables 2, 3, and 4 for breakdowns by study site). Those reporting psy-IPV were on average 3 years younger (95% CI [2, 4], p-value < 0.001). In models using the limited IPTW, psy-IPV was significantly associated with HIV outcomes including lower odds of ART use (0.73 [0.55,0.97], p = 0.03), lower ART adherence (− 4.2 [− 5.9,-2.4], p < 0.001), and higher odds of having a detectable VL (1.43 [1.15,1.78], p = 0.001). Psy-IPV was also associated with higher odds of depressive symptoms (2.63 [2.18,3.18], p < 0.001) and greater odds of methamphetamine, cocaine/crack, illicit opioid, and marijuana use, with methamphetamine use having the largest odds ratio (2.98 [2.30,3.87], p < 0.001). Female sex at birth, alcohol use, binge alcohol use, sexual orientation, and CD4 count were not significantly associated with psy-IPV.

Using the full IPTW, which additionally adjusted for depressive symptoms and substance use, effect sizes for psy-IPV were uniformly smaller and psy-IPV was no longer associated with most outcomes (e.g. ART use, CD4 count) except for ART adherence (− 1.5 [− 3.1,-0.2], p = 0.047). After including depressive symptoms and substance abuse in the propensity score, female sex was significant for psy-IPV in PACTO and RETAIN (Table 4, Supplement Tables 2, 3 and 4).

In CNICS, 50% of patients who indicated psy-IPV did not also indicate physical or sexual IPV. Our sensitivity analysis on the effects of psy-IPV adjusting for physical and sexual IPV in CNICS (Table 5) showed that when using a propensity score that includes age, site, race/ethnicity, and physical/sexual IPV, that psy-IPV was associated with having a viral load > 400 [1.82 (1.20,2.76), p = 0.005], depressive symptoms [3.21 (2.42,4.27), p < 0.001], lower odds of using ART [0.52 (0.32,0.87),0.01], and higher odds of methamphetamine/crystal, cocaine/crack, and illicit opioid use [2.25 (1.57,3.24), p < 0.001; 2.01 (1.25,3.22),p = 0.004; 2.06 (1.08,3.95),p = 0.03, respectively]. Additionally, after adjusting for site, race/ethnicity, and physical/sexual IPV, those who indicated psy-IPV were on average 1.8 years younger (− 3.1,-0.6), p = 0.005 than those who did not. When adding substance use and depressive symptoms to the propensity score analysis, these associations were no longer present.

In analyses that stratified by viral load, psy-IPV was significantly associated with depressive symptoms and methamphetamine/crystal use regardless of viral load [depressive symptoms 2.96 (2.37, 3.71), p < 0.001 undetectable, 1.81 (1.19,2.74),p = 0.005 detectable; methamphetamine/crystal use [2.96 (2.20,4.00),p < 0.001 undetectable, 3.00(1.75,5.14),p < 0.001 detectable], whereas psy-IPV was significantly associated with cocaine/crack [2.14(1.57, 2.93),p < 0.001], illicit opioid [2.19(1.42,3.35),p < 0.001], and marijuana use [1.55(1.24,1.94),p < 0.001] only in those with undetectable viral loads (Table 6).

The multi-site nature of this study yielded a demographically and geographically diverse sample of study participants.

Study data was heavily weighted towards CNICS (87%). Also, since data for physical and sexual IPV was not harmonizable across both consortia, we were only able to examine outcomes of psy-IPV in the context of known presence or absence of physical and sexual violence for PLWH in CNICS. The fact that participating PLWH in the study were at least somewhat engaged in care may have introduced selection bias, as those out of care may experience greater psychological abuse and more adverse health outcomes. We note that while all three study populations are care-based groups, study criteria varied, as did the IPV assessment including the platform and setting, which, while increasing generalizability, may influence interpretability of the results. Finally, due to the cross-sectional nature of the study, we were not able to determine the directionality of associations such as between substance use and psy-IPV.