Despite significant progress made in HIV prevention, care, and treatment in the past decade, HIV remains a leading cause of death among adolescents and young adults, ages 15–24 years, in sub-Saharan Africa [1]. Compared to adults, these youth in the region demonstrate lower levels of antiretroviral therapy (ART) adherence and viral suppression [2]. A national survey in Zambia found that only 34.3% of young people living with HIV ages 15–24 years had achieved viral suppression, compared to 79.0% of older adults ages 45–59 years [3].
Violence is also a leading cause of death among youth [1], and levels of violence against youth are among the highest in sub-Saharan Africa compared to other regions [4, 5]. In Zambia, 43% of female and 34% of male adolescents ages 13–17 experience past-year physical, emotional, or sexual violence [6]. Among young women aged 20–24 years, over one-third experience past-year physical violence and one-tenth experience past-year sexual violence [7].
Researchers are increasingly recognizing violence as a barrier to ART adherence and viral suppression among people living with HIV (primarily women) [8, 9], including in sub-Saharan Africa [10]. Threats or acts of violence from a controlling intimate partner can directly affect a woman’s ability to access the clinic for ART or adhere to their medication [11]. Violence victimization is also associated with greater likelihood of psychological distress, depression, and alcohol use [12], which are barriers to adherence among adults [13] and may thus prevent viral suppression and exacerbate risk of onward transmission.
Despite the growing literature on violence victimization and ART adherence/viral suppression among adult women, only three studies were identified among youth in sub-Saharan Africa [14,15,16]. These studies found associations between ART non-adherence and: violence exposure at home in Malawi [16] and violence from multiple perpetrators in the Eastern Cape, South Africa [14], among both male and female adolescents; and physical or sexual intimate partner violence (IPV) against female adolescents in Soweto, South Africa [15]. The study from the Eastern Cape found that non-adherence to ART increased with exposure to multiple types of victimization [14], echoing results from a study among perinatally-infected adolescents in the U.S. which found associations between higher levels of violence exposure and both unsuppressed viral load (> 400 copies/mL) and a CD4 of less than 25% [17].
A detailed assessment of the association between multiple forms of violence and HIV outcomes among youth in sub-Saharan Africa is critical to developing a holistic understanding of this public health problem and developing appropriate prevention and response efforts. Beyond looking at any experience of violence, we must understand the cumulative effects of violence on HIV outcomes, since exposure to multiple forms of violence (versus a single form) is associated with greater negative health outcomes [18]. Investigations into the unique contributions of specific forms of violence—e.g., the type of victimization (physical violence, psychological abuse, or forced sex) and perpetrator of violence—on HIV outcomes are needed to shed light on whether approaches to HIV care should be tailored to the type or perpetrator of violence. A study among HIV-positive women in Zambia, for instance, found that experiences of sexual and emotional IPV had stronger associations with ART adherence than physical IPV [19]. Furthermore, the association between violence victimization and VL failure may differ based on a youth’s sex or age group, given that violence exposure has shown differential effects for male and female youth [17, 20] and since young adulthood encompasses multiple developmental stages [21]. Strengthening the literature in these areas is particularly important since youth are undergoing cognitive, psychosocial, emotional, and social changes [21]; hence, we cannot assume that associations observed among adult women apply to youth, especially males.
Using data from adolescents and young adults living with HIV in Ndola, Zambia, we examined associations between viral load (VL) failure and past-year exposure to violence, including any victimization, cumulative victimization (i.e. frequency of violence and polyvictimization), types of victimization (physical, psychological, sexual), and perpetrators of violence. In line with the existing literature [18, 22, 23], we hypothesized that we would observe stronger associations with VL failure among youth who experience any violence, a higher frequency of any violence, and multiple types of violence, compared to those who experience no violence. We also investigated whether we would observe stronger associations with VL failure depending on the type or perpetrator of violence. Finally, we examined the presence of statistical interaction to determine whether any associations observed would differ according to the youth’s sex or age group.
Methods
Theoretical Approach
In taking a holistic approach to our analyses, we considered exposure to violence across multiple contexts, drawing on Kaufman’s socio-ecological framework [24]. Researchers have advocated for the use of socio-ecological frameworks in studies of violence [25] and HIV [24, 26], both for developing a deeper understanding of these multi-faceted health issues and for designing appropriate interventions. We focused our analyses on the individual and interpersonal levels, and considered the interpersonal (e.g. homes), institutional (e.g. clinics/schools), and structural (e.g. Zambian law) levels in formulating our study implications.
Sample and Procedures
Analyses used cross-sectional baseline data from Project YES! (Youth Engaging for Success), a randomized controlled trial among youth living with HIV attending four clinics in Ndola, Zambia [27, 28]. The trial compared an intervention and comparison group to assess the effects of a peer-mentoring intervention on youths’ VL suppression (< 1000 copies/mL), ART treatment adherence (gap of 48 or more consecutive hours), and internalized/self-stigma [28]. Youth were consecutively sampled if they were: (a) aged 15–24 years, (b) aware of their HIV status, (c) on ART for 6 months or more, (d) a speaker of English or Bemba, and (e) available for study activities over 18 months (detailed elsewhere [28]).
In accordance with Zambian law, written informed consent was obtained from all participants age 18 and older [29]. For minors (ages 15–17 years), parental/caregiver permission and participant assent were obtained [29]. Participants completed baseline surveys between December 2017 to May 2018 in English or Bemba during face-to-face interviews, using Magpi software on tablet computers. Participants who reported experiences of severe violence or suicidal ideation were referred to designated healthcare providers at each clinic, according to the study’s safety protocol. Participants underwent blood draws for HIV-1 RNA viral load testing using the Qiagene QiAmp viral RNA mini kit (QIAGEN, Germany). Study teams also collected participants’ ART start dates from their medical records.
Measures
Viral Load
Youth with a VL test of ≥ 1000 copies of HIV-RNA/mL were categorized as having VL failure, in line with consolidated guidelines on HIV treatment and prevention from the Zambian Ministry of Health and the World Health Organization (WHO) [30, 31].
Violence Victimization
Violence victimization was measured using items from the International Society for the Prevention of Child Abuse and Neglect Screening Tool-Child Instrument (ICAST-C) [32] and the WHO Multi-Country Study on Women’s Health and Domestic Violence (WHO MCS) [33]. Items assessed past-year experiences of physical violence (7 items), psychological abuse (6 items), and sexual violence (4 items) (Supplement 1). Items measuring physical violence were distinguished by severity level (three items for moderate, four items for severe violence) [33]. The act’s frequency in the past year was queried (never, once, a few times, many times), and 12 possible perpetrator types could be selected: romantic partner, parent/caregiver, other family member, friend/peer, stranger, school staff member, employer, health care worker, neighbor, religious leader, military/police, or someone else the youth knows. Three items assessing sexual violence were removed since they lacked clarity on whether the act was consensual [34]. Measures were translated into Bemba and the full instrument piloted among youth in Ndola for appropriateness.
Any victimization: Youth were classified as having experienced any victimization if reporting one or more behavioral acts of past-year violence (physical violence, psychological abuse, or forced sex) versus no acts.
Frequency of any victimization: A continuous measure was generated to offer insight into the accumulation of harm [35]; the frequency of any victimization was assessed by summing frequency scores across the 14 measures of violence (score range: 0-no frequency to 42-high frequency).
Polyvictimization: A categorical variable was generated for polyvictimization by grouping youth according to their experience of zero, one, or two or more types of past-year violence (physical violence, psychological abuse, or forced sex).
Types of victimization: Three measures assessed the specific types of violence experienced. A severity-times-frequency measure of physical violence was generated by multiplying the severity level (moderate-1, severe-2) by the frequency (never-0, once-1, a few times-2, many times-3) for each of the seven items and summing the scores across items (score range: 0-no severity/frequency to 42-high severity-times-frequency). This approach was modeled on the severity-times-frequency measure developed for the Conflict Tactics Scale [36]. The frequency of psychological abuse was assessed by summing frequency scores across the six items (score range: 0-no frequency to 18-high frequency). Forced sex was assessed as a binary variable (any versus no reports), given the small sample reporting this act.
Perpetrators of violence: Binary variables were generated for both any versus no reported physical violence and any versus no psychological abuse from the following perpetrators: parent/caregiver, other family member, romantic partner, and friend/peer. We distinguished perpetrators by the type of violence—i.e. physical violence and psychological abuse—for a more nuanced look at these forms of victimization. Associations for the remaining perpetrator types or for any perpetrator of forced sex were not assessed due to sparse data.
Covariates
Covariates were considered if potentially associated with violence victimization and VL failure, and not on the causal pathway between the two. Socio-demographic characteristics included the youth’s age (categorized as 15–19 or 20–24 years), sex, completion of primary school (yes or no), and orphan hood status (none, single orphan, or double orphan). HIV measures included the self-reported mode of HIV acquisition (from parents, through sex, or another way/don’t know/refused) and length of time on ART (6 months to 3 years, 3 to 6 years, or 6 + years). Study clinic was also included as a covariate.
Analysis
Descriptive analyses were performed to estimate the proportion reporting VL failure, past-year violence, and the covariates of interest. Chi-square tests were used to assess differences in proportions by VL failure for all variables. Categorical measures of violence were generated from continuous measures based on locally weighted scatterplot smoothing (lowess) plots of the association between the variable and VL failure. We used this approach to make the models more robust against violations of the linearity assumption. We also conducted exploratory analyses to assess the overlap between the forms of violence experienced.
We built six logistic regression models to obtain crude and adjusted odds ratios (ORs), 95% confidence intervals (CIs), and p values (Wald tests) for the association between VL failure and: any victimization (binary) (Model 1); the frequency of any victimization (categorical) (Model 2); polyvictimization (categorical) (Model 3); the types of victimization, including severity-times-frequency of physical violence (categorical), frequency of psychological abuse (categorical), and forced sex (binary) (Model 4); the perpetrator of physical violence (indicator variables for each type) (Model 5); and the perpetrator of psychological abuse (indicator variables for each type) (Model 6). In all models, the reference group for the violence variable(s) consisted of those who had not experienced the form of violence being assessed. When exploring associations for the types of victimization (Model 4) and perpetrators of violence (Models 5 and 6), we included all variables assessing the violence type/perpetrator in adjusted models, alongside covariates, to determine whether any particular violence variable would show a stronger association with VL failure than the others. Missing item values were imputed as the referent, including completion of primary school (n = 1, 0.3% of sample) and time on ART (n = 3, 1.1% of sample).
All covariates were deemed theoretically important and therefore considered as candidates for inclusion in the six adjusted models. For each model, backwards elimination was used, where covariates were retained in adjusted models if reaching a significance level of 0.10 or if the covariate substantially influenced the OR of the main association of interest (+/− 10%) upon removal. All adjusted models included the youth’s sex and age, considered a priori covariates, and the study clinic as a fixed effect to account for the lack of independence of observations. Potential collinearity between any pairs of variables was examined using variance inflation factors. Hosmer–Lemeshow goodness of fit tests were conducted to assess the fit each model to the data. The final candidate multivariate models were extended to include an interaction term between the violence variable(s) and the youth’s sex and age group (15–19 versus 20–24 years), respectively. In post-hoc analyses, we stratified estimates by sex. Analyses were conducted in Stata 14 [37].
Ethics
Study procedures aligned with the WHO ethical and safety recommendations [38], including: using broad terms to describe the research to youths’ caregivers in case the caregiver was perpetrating violence; addressing ethical considerations for violence research in the study staff training; minimizing under-reporting by avoiding judgmental or stigmatizing interpretation of youths’ experiences; and establishing a safety protocol to support violence victims. Ethical approval was obtained from the Johns Hopkins Bloomberg School of Public Health Review Board and the Zambian ERES Converge ethics review board. The research was reviewed and approved by the Zambian Ministry of Health through the National Health Research Authority.