The reciprocal relationships between loneliness and antiretroviral therapy adherence in persons with HIV: a two-wave cross-lagged model

Abstract

Loneliness has been linked to medication-taking behaviors in the general population, yet its role in antiretroviral therapy (ART) adherence among persons with HIV (PWH) remains less understood. This study examined whether loneliness is a risk factor for, or consequence of, poor ART adherence. We conducted a secondary analysis of an observational study on racial disparities in health outcomes among PWH in the Southeastern United States. A cross-lag model tested reciprocal relationships between loneliness (PROMIS-Social Isolation Scale) and self-reported ART adherence across two time points (baseline and 6-month follow-up). Complete cases on both variables yielded a sample of 153 participants (mean age=46.3; 28.1% female; 64.7% Black). Adjusting for depressive symptoms, life stressors, HIV-related stigma and age, the results indicated that baseline loneliness predicted greater loneliness at follow-up (β=0.68, p<.001) and lower ART adherence (β=−0.28, p=.003). Baseline adherence predicted future adherence (β=0.36, p<.001) but not future loneliness (β=0.02, p=.73). Older age was positively associated with adherence (β=0.16, p=.02). The findings suggest that loneliness undermines ART adherence over time, highlighting the need for interventions in community and clinical settings that address loneliness to improve long-term HIV care outcomes.

Introduction

Loneliness has emerged as a significant public health concern in the United States, with a growing body of research demonstrating both high prevalence and its profound effects on mental and physical health (National Academies of Sciences, E. and Medicine, 2020; Office of the Surgeon, G., 2023). Defined as a subjective negative experience that stems from a discrepancy between one’s desired and actual social connection (Perlman & Peplau, 1981; Prohaska et al., 2020), loneliness is linked to detrimental health outcomes, including smoking, substance use, and premature mortality (Holt-Lunstad, 2021; Prohaska et al., 2020). Evidence also shows that loneliness is related to medication taking behaviors in nationally representative cohorts of community-dwelling adults in Canada (aged 45 years and older) (Vyas et al., 2020) and the United States (aged 65 years and older) (Kotwal et al., 2021).

Persons with HIV (PWH) often experience a compounded sense of loneliness due to the intersecting risks and burdens of their HIV diagnosis, as well as social withdrawal that may occur when living with multimorbidity (Austin-Keiller et al., 2023). While studies show that loneliness is related to poorer physical and mental health in PWH (Malika et al., 2023; Yoo-Jeong et al., 2022), the underlying mechanism by which it affects health remains elusive. The Evolutionary Theory of Loneliness (ETL) provides a useful framework for understanding the mechanisms through which loneliness affects health (Cacioppo et al., 2014; Hawkley & Cacioppo, 2010). ETL posits that loneliness can initiate a cascade of biological and behavioral changes detrimental to health. When loneliness is not addressed, it can affect decision-making processes, making it difficult for individuals to engage in proactive health behaviors, such as medication adherence.

Antiretroviral therapy (ART) adherence is an important health behavior for PWH. A recent cross-sectional analysis of PWH in Canada using the Ontario HIV Cohort Study found that higher loneliness increased the odds of ART non-adherence (Light et al., 2022). Despite this finding, less is known regarding the reciprocal relationships between loneliness and ART adherence among PWH. As such, we aimed to evaluate the reciprocal relationship between loneliness and ART adherence among PWH. We hypothesized that baseline loneliness will predict loneliness and lower ART adherence at follow-up.

Methods

Sample and data collection

This study was a secondary analysis of an observational study that examined the relationships between psychosocial factors, race/ethnicity, and health outcomes in PWH (Waldrop-Valverde et al., 2018). The parent study collected data at baseline (n = 701) and 6-months post-baseline (n = 602) from September 2012 to July 2016. Briefly, adults living with HIV (>18 years of age) were recruited from university-affiliated and public outpatient HIV clinics across metro-Atlanta, Georgia. Recruitment methods included distributing flyers and referrals from clinic staff. To be eligible for the parent study, participants had to have attended a minimum of one HIV-related medical appointment in the preceding 9 months at one of the recruiting clinics and must be prescribed to ART at the time of enrollment. Individuals were excluded if they were acutely intoxicated, had significant visual or hearing impairments, or intended to relocate from the metro-Atlanta region within the next 6 months.

Upon fulfilling the eligibility requirements, participants completed the informed consent process. Given the potential for low literacy levels and cognitive impairment among study participants, a consent post-test was administered after the informed consent document had been reviewed with a member of the research team. This post-test evaluated the participant’s understanding of the study’s procedures, study risks and benefits, as well as participants’ capacity to provide consent. If a participant did not pass the post-test within three attempts, they were excluded from the study. After obtainment of informed consent, participants completed two in-person study visits, initially at baseline (T1) and then six months after baseline (T2). Study questionnaires were administered audibly via an audio-computer-assisted self-interview (ACASI) system. Participants received compensation of $ $45 at T1 and $ $55 at T2. The parent study was approved by the University’s Institutional Review Board and the recruiting clinics prior to data collection.

Measures

ART Adherence was measured using sets of medication non-adherence questionnaires adapted from published literature (Gadkari & McHorney, 2012; Molloy et al., 2014; Wilson et al., 2007). All questions were answered on a binary scale (Yes, No) in reference to the prescription medications for managing the index disease, which in our case was ART for HIV, during the past six months: (1) did you ever forget to take the prescription medication; (2) did you ever run out of the prescription medication; (3) were you careless at times about taking the prescription medication; (4) skipped doses of medication because they did not think it was helping you; (5) skipped doses to make medication last longer; (6) took smaller doses to make medication last longer; (7) altered dose of medication to suit own needs; (8) stopped taking medication because you did not think it was helping you; (9) stopped taking medication because you felt better/ (10) worse; (11) took less medication than instructed because you felt better/(12) worse; (13) skipped medication because you felt better/ (14) worse. We dichotomized ART adherence as reporting “No” to all these non-adherence questions; any response with a “Yes” was classified as ART non-adherence.

Loneliness was assessed using the Patient-Reported Outcomes Measurement Information System (PROMIS)-Social Isolation (SI) Short Form v2.0 8a. The PROMIS-SI measure uses 8-items that are derived from social relationship scales and the UCLA-Loneliness Scale on a 5-point Likert scale (1 = “Never” to 5 = “Always”) that ask perception of isolation (Cyranowski et al., 2013). The items include: “I feel left out,” “I feel that people barely know me,” “I feel isolated from others,” “I feel that people are around me but not with me,” “I feel isolated even when I am not alone,” “I feel that people avoid talking to me,” “I feel detached from other people,” and “I feel like a stranger to those around me.” A total raw score ranging from 8 to 40 is calculated by summing the values of the response to each item. The total raw score was converted into a standardized T-score with a mean of 50 and a standard deviation (SD) of 10 (Cella et al., 2010). Cronbach’s alpha for our sample was 0.97 at T1 and 0.95 at T2.

Covariates

Sociodemographic characteristics.

The parent study developed a series of questions about the study participants’ sociodemographic characteristics and collected this information at the baseline interview. Sociodemographic characteristics included age, gender, race/ethnicity, marital status, sexual orientation, educational attainment, and socioeconomic status.

Depressive symptoms.

The Center for Epidemiologic Studies Depression Scale 10 (CESD-10) was used to assess depressive symptoms. The CESD-10 is a 10-item screening questionnaire derived from the full-length, 20-item scale, CESD (Andresen et al., 1994; Radloff, 1977). Participants responded to each item by rating the frequency of depressive symptoms in the previous week on a four-point Likert-type scale (0 = “Rarely or none of the time” to 3 = “All of the time”). The possible total range of response scores is 0–30, with scores greater than 10 indicating clinical depression. The CESD-10 is highly correlated with the original scale (r = .97, p < 001);( Zhang et al., 2012) and has adequate internal consistency in people with HIV (Cronbach α = 0.88) and without HIV (Björgvinsson et al., 2013). Cronbach alpha for participants in our sample was 0.94 at baseline.

HIV stigma.

The HIV Stigma Scale was used to measure internalized HIV stigma (Emlet, 2005). The HIV Stigma Scale consists of 13 items that assess three components of stigma: distancing, blaming, and discrimination. The following are example questions for each respective component: I avoided getting treatment because someone might find out about my illness, I felt blamed by others for my illness, and I feared that I might lose my job if someone found out about my illness. Each item is rated on a four-point Likert – type scale (1 = “Not at all” to 4 = “Often”), with a higher score indicating greater stigma. Cronbach alpha for participants in our study was 0.83 at baseline.

Everyday life challenges.

The 28-item Difficult Life Circumstances Scale was used to assess everyday life challenges (Barnard et al., 1989). Examples of life challenges captured by this scale included financial, housing, transportation, health, personal, child, and/or partner issues. Scale items are rated dichotomously (0 = “No” or 1 = “Yes”) with more responses of “Yes” representing a higher level of difficult life circumstances. Cronbach’s α for our sample was 0.97 at baseline.

Statistical analysis

Descriptive statistics were used to calculate the means and standard deviations of the continuous variables and the percentage and frequency of the categorical variables. Next, we calculated bivariate associations between baseline covariates, T1 and T2 ART adherence, and T1 and T2 loneliness using appropriate statistical techniques (e.g., chi-squared tests, t-tests). Finally, we used an asymptotic distribution-free function for measured variable of structural equation modeling to examine the cross-lag model between loneliness and ART adherence. Any covariates that had statistically significant associations with ART adherence or loneliness from bivariate associations were included as control variables in the cross-lag model. Structural equation modeling of saturated models, such as that tested in our study, requires sample sizes comparable to regression analysis (Wolf et al., 2013); therefore, we were adequately powered. All statistical analyses were performed using STATA/IC 14.2 with statistical significance at an alpha = 0.05.

Results

We used complete cases on the study variables across two time points, resulting in a total sample size of 153 participants. Demographics characteristics of the sample are reported in Table 1. The mean age was 46.3 (SD = 10.5) years. Most participants identified as Black/African American (64.7%), male gender (70.6%), had more than a high school education (62.1%), and reported being single (52.9%).

Table 1.

Sociodemographic characteristics of participants at baseline (N = 153).

Characteristic		n	%
Gender
Female		43	28.1
Male		108	70.6
Not reported		2	1.3
Marital status
Single/never married		81	52.9
Married/living with partner		39	25.5
Previously married (divorced, separated, widowed)		30	19.6
Other		3	2
Race/Ethnicity
White/non-Hispanic		41	26.8
Hispanic/Latinx		3	2
Black/African American		99	64.7
Other		10	6.6
Sexual orientation
Homosexual, gay, or lesbian		70	45.8
Heterosexual or straight		55	35.9
Bisexual		10	6.5
Other		12	7.8
Education
< High school		23	15
High school/GED		35	22.9
> High school		95	62.1
Characteristic	M ± SD	Minimum	Maximum
Age	46.25 ± 10.51	22	69
HIV-related stigma	1.86 ± 1.98	0	6
Depressive symptoms	7.09 ± 5.24	0	23
Difficult Life Circumstances	3.48 ± 2.97	0	17

Note. GED = graduate equivalency degree.

Table 2 presents descriptive statistics and bivariate relationship between loneliness and ART adherence at T1 and T2. Mean loneliness was 45.7 (SD = 11.1) at T1 and 45.4 (SD = 10.4) at T2. Up to 44% of participants (n = 67) at T1 and 52% (n = 80) at T2 were adherent to ART. Loneliness at T1 was significantly related to ART adherence at T1, t(151) = −2.59, p = .011, and at T2 t(151) = −4.89, p < .001. Loneliness at T2 was significantly related to ART adherence at T2, t(151) = −4.32, p < .001. ART adherence at T1 and T2 were significantly related to one another, X²(1, N = 153) = 30.64, p < .001.

Table 2.

Bivariate relationship between loneliness and art adherence at baseline (T1) and 6-month follow-up (T2).

Variable	M	SD	Range	1	2	3	4
1. Loneliness (T1)	45.66	11.13	43	-
2. Loneliness (T2)	45.44	10.43	43	r = .70***	-
	n	%	Min, Max	1	2	3	4
3. ART adherence (T1)	67	43.8	0, 1	t = −2.59**	t = −1.66	-
4. ART adherence (T2)	80	52.3	0, 1	t = − 4.89***	t = −4.32***	X2 = 30.64***	-

Notes. ART = antiretroviral therapy.

^*p < .05;

^**p < .01;

^***p < .001.

Of the baseline covariates considered, HIV-related stigma and depressive symptoms were significantly related with ART adherence at T1 (r = −.26, p < .001; r = −.17, p = .03; r = −.18, p = .02; respectively) and at T2 (r = −.27, p < .001; r = −.21, p = .01; r = −.22, p = .006; respectively). Similarly, difficult life circumstances, internalized HIV-related stigma, and depressive symptoms were significantly related with loneliness at T1 (r = .42, p < .001; r = .42, p < .001; r = .58, p < .001; respectively) and at T2 (r = .30, p < .001; r = .35, p < .001; r = .45, p < .001; respectively). Of the T1 sociodemographic characteristics, only age was significantly related to ART adherence at 6-months follow-up (r = .18, p = .023). Thus, these variables were adjusted for in the subsequent cross-lag model analysis.

In the adjusted cross-lag model (Figure 1), T1 loneliness was a statistically significant predictor of future loneliness (β = 0.68, p < .001) and future ART adherence (β = −0.28, p = .003). There was a statistically significant relationship between T1 loneliness and T1 ART adherence (β = −0.23, p = .0042). Time 1 ART adherence was a statistically significant predictor of later medication adherence (β = 0.36, p < .001) but not a significant predictor of later loneliness (β = 0.02, p = .731). Among the covariates that we considered, T1 age was the only significant predictor of future ART adherence (β = 0.16, p = .023). The model accounted for 52.5% and 30.1% of the total variances in loneliness and ART adherence, respectively, at 6-month follow-up.

Figure 1.

Standardized path coefficients (standard errors) for the adjusted cross-lag model. The model is controlled for T1 depressive symptoms, difficult life circumstances, HIV-related stigma, and age. Dashed and solid lines represent paths that are nonsignificant and significant, respectively. Two-headed arrows represent covariance between loneliness and ART adherence at baseline. *p < .05; **p < .01; ***p < .001. ART = antiretroviral therapy.

Discussion

A recent cohort study showed that higher loneliness increased the odds of ART non-adherence in PWH living in Canada (Light et al., 2022). The present study extended on this finding to test the reciprocal relationships between loneliness and ART adherence at two time points. Our results of the adjusted cross-lag model underscore the persistence of loneliness over time and its detrimental effect on future ART adherence. The strong predictive value of baseline loneliness on future loneliness highlights the chronic nature of loneliness, suggesting that without appropriate intervention, feelings of loneliness may persist. This finding is consistent with the ETL (Cacioppo et al., 2014; Hawkley & Cacioppo, 2010) and an existing body of literature that suggests the chronicity of loneliness (Mund et al., 2020) and its adverse effects on health (Lim et al., 2023).

Our results show an inverse relationship between baseline loneliness and future ART adherence, with higher loneliness at baseline significantly predicting future ART non-adherence, confirming our hypothesis that loneliness negatively affects ART adherence. Our findings align with the existing literature in the general population that documents the adverse effects of loneliness on health behaviors including medication-taking behaviors (Holt-Lunstad, 2021; Kotwal et al., 2021 Perlman & Peplau, 1981;; Prohaska et al., 2020; Vyas et al., 2020). Our finding is particularly significant given the importance of consistent ART adherence for viral suppression, indicating that interventions aimed at reducing loneliness may not only improve psychological well-being but also may improve ART adherence and, consequently, promote health in this population. Interestingly, while baseline ART adherence had a direct association with later ART adherence, it was not related to later loneliness. This finding indicates a unidirectional association of loneliness on ART adherence rather than a bidirectional relationship, emphasizing the need for targeted interventions to address loneliness independently of person’s level of ART adherence.

Extant literature suggests a range of interventions to reduce loneliness in the general population without HIV, but none exists for PWH with loneliness. Cognitive behavioral therapy and interventions targeting maladaptive social cognition have demonstrated moderate efficacy in reducing loneliness in the general population (Masi et al., 2011). Group-based programs, including structured social or educational sessions and group activities, have also shown consistent benefits, especially among older adults(Shekelle et al., 2024). These evidence-based strategies not only demonstrate broad utility across populations but also hold significant promise for future research targeting PWH who face persistent disparities in isolation and mental health. Future intervention research should focus on adapting and testing existing interventions and ways to optimize the intervention effects. Furthermore, future longitudinal research involving larger samples is needed to evaluate the persistence of these relationships beyond 6 months to allow for a more nuanced evaluation of these relationships with other key health outcomes such as HIV control.

Among the sociodemographic covariates considered, baseline age was the only significant factor related to future ART adherence, suggesting potential age-related differences in ART adherence. This aligns with existing evidence that consistently shows better medication adherence with increasing age in PWH (Barclay et al., 2007; Ghidei et al., 2013). This could be due to greater medication management skills, greater motivation for health-promoting behaviors, or stability in daily routines among older individuals, suggesting that younger PWH may require additional support to achieve optimal ART adherence (Ghidei et al., 2013).

Our findings must be interpreted in the context of several limitations. First, while this secondary analysis considered the temporal effects of loneliness at baseline on ART adherence at 6-months post-baseline, this analysis did not account for changes that may have occurred between baseline and 6-months post-baseline for all the variables of interest. This limitation prevents the explanation of causality among the variables. Additional longitudinal studies are necessary to establish the causal link among these variables and to assess what influences fluctuations in loneliness. Secondly, all questionnaires were obtained via self-report which may have introduced bias and reporting errors including social desirability or recall biases, despite having a good Cronbach’s alpha on each measure that was used. Future research should include objective measures of ART adherence. Third, while the model accounted for substantial variance in the outcome, due to the nature of secondary analysis, other possible unexamined factors that could explain more variance in the outcome were not assessed. Future research should explore other psychological, social, and structural determinants of loneliness and ART adherence to further elucidate these relationships and inform more comprehensive intervention targets. Finally, the findings may not be generalizable to the larger HIV population since participants were a sample of PWH recruited from HIV outpatient clinics in the metro-Atlanta area.

Despite the limitations, our findings demonstrate the critical role of loneliness in affecting ART adherence, highlighting the need for addressing psychological and social needs as part of comprehensive HIV care. Interventions aimed at reducing loneliness, particularly those tailored to meet the unique needs of individuals based on their age, could play a crucial role in improving ART adherence and overall health for PWH.

Data availability statement

The data that support our findings of this study are available from the corresponding author upon reasonable request.