Behavioral Modeling and its Association with PrEP and ART Use in Ugandan HIV-Serodifferent Couples

Integrating Pre-Exposure Prophylaxis (PrEP) delivery into Antiretroviral Therapy (ART) programs bridges the Human Immunodeficiency Virus (HIV) prevention gap for HIV-serodifferent couples prior to the partner living with HIV achieving viral suppression. Behavioral modeling is one mechanism that could explain health-related behavior among couples, including those using antiretroviral medications, but few tools exist to measure the extent to which behavior is modeled. Using a longitudinal observational design nested within a cluster randomized trial, this study examined the factor structure and assessed the internal consistency of a novel 24-item, four-point Likert-type scale to measure behavioral modeling and the association of behavioral modeling with medication-taking behaviors among heterosexual, cis-gender HIV-serodifferent couples. In 149 couples enrolled for research, a five-factor model provided the best statistical and conceptual fit, including attention to partner behavior, collective action, role modeling, motivation, and relationship quality. Behavioral modeling was associated with medication-taking behaviors among members of serodifferent couples. Partner modeling of ART/PrEP taking could be an important target for assessment and intervention in HIV prevention programs for HIV serodifferent couples. Supplementary Information The online version contains supplementary material available at 10.1007/s10461-024-04286-2.


Introduction
In the context of Human Immunodeficiency Virus (HIV) treatment, adherence to antiretroviral medication is an important health behavior that facilitates viral suppression, the driver of better health outcomes and prevention of onward HIV transmission.The sexual partnership within HIV-serodifferent couples, where one partner is living with HIV and the other is HIV-negative, puts the HIV-negative partner at high risk for HIV acquisition when adequate strategies for HIV prevention and treatment are not in place [1,2].Programs for HIV-serodifferent couples focus on supporting HIV treatment for the partner living with HIV and time-limited pre-exposure prophylaxis (PrEP) using antiretroviral medication for HIV-negative partners, when HIV viremia may not be suppressed in the partner living with HIV [3][4][5].When PrEP is integrated into antiretroviral treatment (ART) programs for HIV-serodifferent couples, it bridges the immediate HIV prevention gap prior to viral suppression in the partner living with HIV and has been demonstrated to have 95% effectiveness in reducing HIV incidence [6].A randomized trial of oral antiretroviral therapy for HIVserodiscordant heterosexual couples in Kenya and Uganda found 67% and 75% reduction in the incidence of HIV with tenofovir and combination tenofovir-emtricitabine respectively [7].The HIV-negative partners chose to take PrEP to reduce the HIV transmission anxiety, improve the relationship, and together build shared protection against HIV as a couple [8].
Behavioral modeling, or observational learning, refers to the process of individuals learning and acquiring new behaviors from watching others [9].Observational learning has four processes, including attention, retention, production, and motivation.Individuals first direct their attention to the modeled behaviors or interactions (attention) and remember them (retention), which is dependent on individuals' intellectual capacities.Individuals then perform the modeled behavior when the circumstances and physical skills allow them to (production).Motivation is the internal factor that drives the production of the modeled behavior, depending on the expectations about costs and benefits [7].Within HIV-serodifferent couples using antiretrovirals for prevention and treatment, behavioral modeling is one mechanism that could explain each partner's medication adherence behavior.Specifically, partners living with HIV could build self-efficacy for taking HIV medications by observing their HIV-negative partners manage medication side effects, adherence, and HIV-related stigma.The perceived benefits on their health, relationship quality, and shared future with their partners may also motivate ART adherence [10].Observing the partners taking ART could reinforce PrEP adherence among HIV-negative partners in return, creating a positive feedback loop within the partners.
To our knowledge, no measure has been developed yet to capture whether heterosexual HIV-serodifferent couples exhibit behavioral modeling in their antiretroviral use (PrEP for prevention, ART for treatment).The current study aimed to (1) use factor analysis to identify the underlying factors that the items on a questionnaire are measuring and examine the internal consistency of a new behavioral modeling scale; (2) examine the association between behavioral modeling with medicationtaking behaviors among HIV-serodifferent couples; and (3) explore the association between behavioral modeling with viral load levels among partners with HIV and Tenofovir (TFV) drug levels among HIV-negative partners.

Study Design and Participants
This study is a longitudinal observational study that was nested within the Partners PrEP Program, a stepped wedge cluster randomized trial evaluating the integration of PrEP delivery into ART programs (#NCT03586128) 11] in Kampala and Wakiso, Uganda from June 2018 to December 2021.Heterosexual sero-different couples were recruited into the program through couples-based HIV testing and received access to standard of care for HIV care after the intervention was launched at their clinic.
Participant inclusion criteria were as follows.For both members of the couple:  [12].
Programmatic data were abstracted from patient medical charts including dates when PrEP and ART were dispensed and HIV viral load results from testing that occurred approximately 3-6 months after ART initiation in alignment with national guidelines.Additionally, couples were invited to enroll for supplemental research procedures that included interviewer-administered questionnaires on every quarter for about 24 months to ascertain data on demographics, sexual behavior, modeled behavior, and self-reported PrEP and ART adherence.A dried blood spot was also collected from HIV-negative partners in the research cohort and a random sample of 10% were selected for quantification of tenofovir (TFV)-diphosphate levels via validated liquid chromatography/tandem mass spectrometry methods [13].

PrEP Delivery Program
The 12 ART clinics received training on PrEP delivery, ongoing technical assistance visits from the training team, and PrEP commodities before the program was launched.The PrEP training protocol was adapted from the national PrEP training curriculum in Uganda, with adjustments to tailor the service to HIV-serodifferent couples.The technical assistance visits were conducted by the training team who were experienced in PrEP delivery and clinical management.Once clinics entered the PrEP delivery stage, trained staff offered PrEP to the HIV-negative members of recently diagnosed HIV-serodifferent couples [14].HIV-negative partners were counseled to use PrEP for at least 6 months after their partners started ART and supported to discontinue when the partner sustained ART use and there were no other potential sources of exposure (e.g., other partners).

Behavioral Modeling Scale
Through discussions and initial pilot testing with a group of clinicians, ART clients and their partners, and research staff, we developed a 28-item questionnaire with a four-point Likerttype response scale to measure behavioral modeling.The items were created based on the observational learning concepts of the Social Cognitive Theory [7].Nineteen items mapped to the four processes of observational learning: attention (e.g., "I often see my partner taking PrEP/ART"), retention (e.g., "I remind my partner to take PrEP"), production (e.g., "I take my ART at the same time my partner takes PrEP/ART"), and motivation (e.g., "When my partner takes PrEP/ART I have hope for our future together") [9].Additionally, we added 9 items to capture the relationship quality between the partners (e.g., "I feel close to my partner").These items were based on previous research that suggested a strong association between relationship satisfaction and PrEP adherence, and that role modeling is more effective when it comes from someone the observer feels close to and values [15].Four items were removed due to poor face validity.Face validity refers to whether a question appears to measure the construct it aims to measure [16].For example, one item that was removed was "I have people I can talk to about my relationship when I need support", which was more about social support than behavioral modeling.
The final scale contained 24 items.The items were identical for both partners within a couple, with "PrEP" or "ART" used respectively for the HIV-negative partners and partners living with HIV.The response options ranged from (0) "Strongly disagree" to (3) "Strongly agree".The scale was scored by computing a total raw score ranging from 0 to 72.Separate total scores were computed once subscales were identified.We computed total scores for both HIV-negative partners and the partners living with HIV to capture each partner's self-report of the other partners' behaviors that could have been observed.

Self-Reported ART/PrEP-Taking Behavior
Medication -aking behaviors were measured using three self-reported items adapted from Wilson et al. (1) whether the partner has taken ART/PrEP since the last clinic visit (yes/no); (2) how well the client took their ART/PrEP medication as directed in the past month (0 = very poor to 5 = excellent), and (3) how often the client took their ART/PrEP medication in the past month (0 = not at all to 3 = every day).We performed linear transformation on the item responses for the second and the third item to compose the adherence score (with a range 0-100), with higher scores indicating better adherence [17].

Statistical Analysis
All analyses were conducted in R [18].Demographic characteristics of the sample and medication-taking behaviors across three time points (months 1, 3, and 6 after enrollment) were presented and compared between HIV-negative and -positive partners using Chi-square tests for categorical variables and ANOVA for continuous variables.We conducted an exploratory factor analysis (EFA) to examine the structure of the behavioral modeling scale, using data from the first follow-up visit (e.g., 1 month after enrollment).The rationale for using factor analysis was based on this being a novel scale without previous exploration of its structure in this specific cultural and study setting.We used the promax factor rotation method that allows factor correlations and results in simple and interpretable factor structure [19].The final factor structure was determined after an evaluation of eigenvalues, residual variances, chi-square, Bayesian Information Criterion (BIC), Comparative Fit Index (CFI), Root Mean Square Error of Approximation (RMSEA), and variance accounted for by different factor structures [20].The internal consistency of the scale is indicated by Cronbach's α.The data from both HIV-positive and negative partners were combined to conduct EFA and examine the internal consistency of the scale.

Exploratory Factor Analysis
The Kaiser-Meyer-Olkin index of sampling adequacy for the behavioral modeling scale was 0.92, suggesting the data were suitable for factor analysis.Eigenvalues from factor analysis with promax rotation suggested a 5-factor model would be the most explanatory, with the first five components having an eigenvalue larger than one and the five factors accounting for 66% of the total variance (Table 2).As the number of factors increased from three to seven, Chisquare statistics and RMSEA both decreased, while CFI increased, suggesting improvement in model performance.Model fit index BIC decreased initially as the number of factors increased from three to five, suggesting improvement in model performance.BIC started to increase after the factor number exceeded five, suggesting that the model performance started to decrease.The fit statistics together supported the five-factor model.Additionally, the five-factor model provided a meaningful interpretation of the factor structure based on the factor loadings (Table 3).The five factors were: (1) attention to partner behavior, (2) collective action, (3) partner as a role model, (4) motivation, and (5) relationship quality.The mean and SD of factor correlations were presented in Table 4.When we split the dataset by HIV status, the factor structure was replicated in both subgroups (Supplement Tables 1, 2 and 3).
We examined the association between a participant's behavioral modeling scale and subscales and their own selfreported PrEP/ART adherence score across three time points using generalized estimating equations (GEE) to account for repeated measurements.The regression coefficients and 95% confidence interval (CI) from the GEE model indicated the strength of the associations and the uncertainty of the estimates.We used logistic regression to examine the associations between the behavioral modeling scale and (a) outcome of viral suppression (HIV RNA < 1000 copies/ml) at 6 months post ART initiation and, in separate models, (b) outcome of TFV levels associated with high adherence (TFV > 700 fmol/punch) [21].The odds ratio and 95%CI indicated the strength and precision of observed associations.Since we had only one time point with viral load and TVF outcome, behavioral model scale scores were averaged across month 1, 3, and 6 for this analysis.We also examined associations between behavioral modeling scale score and continuous TFV levels and viral load outcomes using linear regression.For all analyses, adjusted models were generated including participant age (continuous) and sex (male, female) as a priori determined confounding factors.

Ethics
The protocol was approved by ethics committees at the Uganda National HIV/AIDS Research Committee (ARC 194), the Uganda National Council for Science and Technology (HS 2381), and the University of Washington Human Subjects Division (STUDY00000320).The Kampala Capital City Authority and Wakiso District gave permission for data to be abstracted from clinic medical records.All participants for research procedures underwent an informed consent process and provided written consent in their preferred language.

Descriptive Characteristics
In total, both members of 149 serodifferent couples enrolled for the research cohort at baseline, with 141 retained at month 1, 3 and 139 at month 6.More than half of the partners living with HIV were female (64%).The median age of the participants was 28 years (IQR 9.8).Based on selfreport adherence score, the partners living with HIV were more adherent to ART than their partners were to PrEP, across all three follow-up visits (Table 1).12 and 13).

Discussion
To our knowledge, this is the first study that developed a scale to measure dyadic behavioral modeling within HIVserodifferent couples.We found that modeled ART-taking behavior was associated with PrEP use and that modeled PrEP-taking behavior was associated with ART use.This association was bidirectional irrespective of sex.The role model subscale was positively associated with self-reported adherence to ART and PrEP.
The final factor structure of the scale (attention, collective action, motivation, relationship, and role model) mapped well onto the four processes of observational learning.Notably, we grouped the items that were intended to measure retention (item 13-15) and production (item [16][17] and named this factor "collective action", representing the joint efforts of both partners to support each other in medication adherence.Collective action is a construct of unique value in the context of measuring modeled behavior within a couple.While social cognitive theory recognizes individuals' ability to actively construct their environment to achieve their individual goals, it also emphasizes collective action, where a group of individuals work together to enact change that can benefit the group [9].The collective action within the couple, such as reminding each other to take medication, serves the goals of maintaining optimal health Compared to partners living with HIV, HIV-negative partners reported lower scores on relationship subscale (t(280) = -2.99,p = 0.005) and collective action (t(280) = -2.27,p = 0.026) at 6-month follow-up visit (Table 5).
When averaged across three follow-up visits, neither the behavioral modeling scale composite score, nor subscale scores were associated with either dichotomized or continuous parameterizations of viral load or TFV levels at the month 6 follow-up visit (Tables 8, 9, 10 and 11).The results were similar when limited to data from the 6-month time point.Self-reported ART adherence mapped to viral load data with 91.0% of people who took ART every day having Additionally, the measure included a "partner as a role model" factor that reflected a partner's medication adherence behavior.Although this factor is composed of only two items, the role model subscale consistently predicted the adherence score of ART/PrEP and had the largest coefficient among all subscales.This is consistent with the literature on role modeling and behavior change.A study using peers as role models who were similar to participants along multiple dimensions (e.g., sex, age, occupation) found that the peer modeling intervention increased participants' physical fitness, compared to the group that only received general health information [23].A review of peer-based interventions on HIV-related outcomes found that peer-based interventions increased linkage and retention to HIV care [24].The improved health behaviors might be due to the positive effect of exposure to peer models on participant confidence and knowledge.The behaviors of the partner might contribute to the establishment of a norm within the couple that is mutually reinforcing the adherent behavior [25].
When applied in clinical settings and research, our novel behavioral modeling scale can be used to outline intervention targets across five dimensions (motivation, relationship, for both partners and preserving the long-term relationship.Prior work also suggested that the shared experience of taking antiretroviral medication together which often involves mutual reminders, and providing emotional and material support to each other, has been documented to draw members of the couples closer to each other and prevent their separation, contributing to the success of integrated delivery of ART and PrEP to HIV serodifferent couples [22].share protection that motivated them to adopt HIV prevention strategies coupled with viral control [8].HIV-serodifferent couples are uniquely positioned where the sustained HIV viremic control through ART and maximizing prevention benefit of PrEP can be attained through collective attention, collective action, role modeling) for members of serodifferent couples to improve medication adherence and HIV prevention and treatment outcomes.This is consistent with the perspectives of sero-different couples where the participants reported factors such as relationship quality and The adjusted models controlled for age, gender 1 3 for the specific context of serodifferent couples to capture dimensions of dyadic behavioral modeling that are associated with adherence behavior in HIV prevention and treatment.However, this measure has the potential to be adapted and generalized to the measurement of dyadic behavioral modeling effect in other health-related research.Further studies are needed to examine the psychometric properties of this measure in other health behavior settings and cultural contexts.medication-taking efforts from both partners.Intervention programs that strategically target the different dimensions of behavioral modeling between partners can create a sense of collective action and promote optimal adherence to antiretrovirals of both partners.This measure was developed This study has several limitations.First, we were not able to utilize all three self-reported adherence items when calculating the adherence scores due to the adaption of the first item in this study.Future studies may use the original adherence items to test the predictive effects of this behavioral modeling.Second, despite the concordance between selfreported adherence data and biological outcomes (viral suppression and TFV level), our data are insufficiently powered to determine whether the scale and subscales are associated with objective adherence measures, viral suppression and TFV.Future studies can collect a larger sample of biological outcomes and examine whether behavioral modeling has differential predictive effects for self-reported versus biological outcomes.

Conclusions
Our novel behavioral modeling measure successfully assessed five dimensions of modeled behavior (i.e., attention, motivation, relationship quality, collective action, and role modeling), which are components of observational learning.Furthermore, our results suggested the behavioral modeling measure was associated with self-reported medication-adherence behavior among members of HIVserodifferent couples.With further validation, this measure could be applied in clinical and research settings to outline modifiable behavioral change within partners using the five dimensions of behavioral modeling of medication taking.Furthermore, the results suggest that further interventions with HIV-serodifferent couples could think beyond role modeling, and include dimensions such as collective action and relationship quality between couples, to maximize adherence to antiretrovirals and protection against HIV transmission.

( 1 )
Age ≥ 18 years; (2) Ability and willingness to provide informed consent; (3) Being sexually active with each other; (4) Willingness to engage with the clinic system collaboratively.For the HIV-positive members (index participants): (1) Diagnosis as HIV-positive per the national HIV testing algorithm; (2) Recent identification as a member of an HIV serodiscordant couple; (3) Absence of current enrollment in any HIV treatment clinical trial.For the HIV-negative members (partner participants): (1) Diagnosis as HIV-negative using the national HIV testing algorithm; (2) Recent identification within an HIV serodiscordant relationship; (3) Absence of current enrollment in any HIV treatment clinical trial; (4) Non-usage of PrEP at present; (5) Eligibility for PrEP as defined by the WHO or Ugandan national guidelines

Table 1
Baseline characteristics of HIV-serodifferent couples1 3suppressed viremia, a greater frequency than those who reported gaps in taking ART.Similarly, self-reported PrEP use was consistent with TFV level data, with 66.7% of people who took PrEP every day having detectable TFV levels, which was more frequent than people who self-reported low adherence to PrEP (Tables

Table 2
Fit statistics of explorative factor analysis models a T-test b Chi-square Test c Viral suppression definition: HIV RNA < 1000 copies/ml d TFV detectable: TFV > 700 fmol/punch.Available data for TFV (n = 32)

Table 3
Factor loadings of fivefactor model

Table 4
a T-test 1 3

Table 6
Associations between behavioral modeling scores and self-reported ART adherence score living with HIV) Each row is a separate model.All models were GEE.The main predictor of each model is listed in the table rows, the outcome of all models was self-reported ART adherence score a

Table 7
GEE models: behavioral modeling and self-reported PrEP adherence score (HIV-negative partners) Each row is a separate model.The main predictor of each model is listed in the tablebThe adjusted models controlled for age and gender a

Table 8
Associations between modeling behavior and HIV viral load (continuous) Each row is a separate model.The main predictor of each model is listed in the table a b The adjusted models controlled for age, gender

Table 9
Association between modeling behavior and viral suppression (binary, HIV RNA < 1000 copies/ml)

Table 10
Association between modeling behavior and TFV diphosphate levels (continuous)

Table 11
Association between modeling behavior and TFV detection (binary, TFV > 700 fmol/punch) Each row is a separate model.The main predictor of each model is listed in the table a bThe adjusted models controlled for age, gender

Table 12
Concordance between ART adherence and viral suppression (HIV RNA < 1000 copies/ml) How often the client took their ART medication in the past month (Took ART often)