An institutional- based prospective cohort study was conducted in health facilities in Gondar, northwest Ethiopia. Gondar is located in the northwest of Ethiopia, approximately 728 kilometers away from the capital of the country, Addis Ababa. In Gondar, there are 10 health facilities: one comprehensive specialized hospital, one primary hospital, and eight health centers. Four health facilities, including the University of Gondar Comprehensive Specialized Referral Hospital, Azezo Health Centre, Gondar/Poly Health Centre, and Maraki Health Centre, were selected based on their high patient flow. There were 6,042 adult WLWHIV who had registered for ART follow- up at health facilities in Gondar during the study period. The study was conducted from 1 September 2023 to 30 April 2024.

The study population was 1,043 WLWHIV attending the four health facilities with or without depressive symptoms during the study period who were screened at baseline assessment. The participants were included if they had attended for at least 6 months since starting ART treatment and had no history of ART non-adherence. Participants were included regardless of treatment regimens, viral load, CD4 count, and ART side effects. WLWHIV who met the above criteria were randomly selected and enrolled in the study if (a) they were aged 18 years or older, (b) had no plan to move out of the study area before 6 months, and (c) they were well enough to be interviewed, as judged by the interviewers.

The sample size was determined using the double population formula, with an exposed-to-non-exposed ratio of 1:1.1. The exposed group included individuals with depression who were ART non-adherent, while the non-exposed group consisted of non-depressed individuals who were also ART non-adherent. A 95% confidence level, 80% power, an odds ratio of 1.9 (40), and a 10% non-response rate were considered in the calculation. As a result, the total sample size for the prospective cohort study was 627, with 299 participants in the exposed group (depressed) and 328 in the non-exposed group (non-depressed).

A simple random sampling technique was used to select 299 from depressed and 328 from non-depressed study participants after collecting the baseline data. Reassess depressive symptoms and incidence of ART non-adherence at 3 months and 6 months. The sample size was allocated proportionally to each health facility after identifying the list of potential study subjects from the baseline data for each health facility ( Figure 1 ).

ART-non-adherence was defined as answering at least one yes for questions 1, 2, 3, or 5, missing the medication one or more times, or missing medication for 2 or more days, according to the Simplified Medication Adherence Questionnaire (SMAQ) (60).

Depression was defined as scoring 10 and above on the Patient Health Questionnaire (PHQ-9) (61).

Social support was defined by scores on the Oslo Social Support Scale-3 (OSSS-3) (poor was 3 to 8, moderate 9 to 11, and strong 12 to 14) (62).

Current substance use was defined as the use of alcohol, khat, and cigarettes by a WLWHIV in the previous 3 months according to the Alcohol, Smoking, and Substance Involvement Screening Test (ASSIST) (63).

Perceived stigma was assessed by 12 items related to HIV- related stigma, with a higher score indicating increased perceived stigma (64).

Gender- based violence was defined as scoring at least one for the 13 items on the World Health Organization’s Violence Against Women instrument. Physical violence was defined as at least one yes in the six items related to physical violence. Psychological violence was defined as at least one yes in four items related to psychological violence and sexual violence was defined as at least one yes for three items related to sexual violence (65).

Food insecurity was defined according to the Household Food Insecurity Access Scale (HFIAS), with food insecurity increasing as the score increases (66).

The SMAQ was used to collect the data. It is a self-administered questionnaire that consists of six items of which four are measured as “yes/no (dichotomous)”, one is Likert-type, and one has two options (60). The reliability and validity of the SMAQ has been studied in different countries (60, 67). The SMAQ was validated in Addis Ababa, Ethiopia among women living with HIV in a two- times measure as the case and control group. At baseline and at the follow-up study, the overall Cronbach’s alpha was 0.72. The concurrent validity of the six items of the SMAQ ranged from moderate to excellent positive correlation (68).

The PHQ-9 was used to assess depression. Participants were asked to rate the frequency of depressive symptoms experienced in the 2 weeks prior to data collection. The total score ranges from 0 to 27. The severity of depression is assessed using a four-point Likert scale: 0 = not at all, 1 = several days, 2 = more than half of the days, and 3 = nearly every day. The score interpretation is as follows: 0–4 points = no depression; 5–9 points = mild depression; 10–14 points = moderate depression; 15–19 points = moderately severe depression; and 20–27 points = severe depression (61). The women who scored 10 and above according to PHQ-9 were referred to the medical and psychiatry unit for further evaluation and counseling for depression.

The Oslo Social Support Scale-3 was used to measure social support for women living with HIV. The level of social support was classified as “poor social support” with a score of 3–8, “moderate social support” with 9–11, and “strong social support” with 12–14. The OSSS-3 consists of three items assessing the number of close intimates, perceived level of concern from others, and perceived ease of getting help from neighbors. The OSSS-3 has good convergent and predictive validity (69).

The risk of harmful substance use was assessed using the modified WHO ASSIST version 3.1 (63), which includes seven items for alcohol use, khat use, and tobacco product use. However, for this study, only the items related to current substance use were considered. This focus on current substance use ensures that the findings are directly relevant to participants’ present behaviors and their immediate impact on ART non-adherence, health outcomes, and psychological wellbeing. The specific items included were: (1) “Have you used any kind of alcohol in the last 3 months?” (Yes/No), (2) “Have you used khat in the last 3 months?” (Yes/No), and (3) “Have you used any tobacco products in the last 3 months?” (Yes/No) (63).

All eligible and consenting women living with HIV were recruited into the cohort. Women living with HIV who had at least 6 months of follow- up (299 with depression and 328 without depression) were randomly recruited from antiretroviral clinics. The data were collected at three time points: at baseline, 3 months, and 6 months. Data collectors (BSc in psychiatry nursing) conducted a face- to- face interview at each health facility with those who were willing to participate in this study and who fulfilled the criteria and were followed from 1 September 2023 to 30 April 2024. Depression, social support, HIV- related stigma, and medication adherence were measured at three time points. However, sociodemographic factors, clinical factors, substance- related factors, intimate partner violence, and food insecurity were collected at baseline.

Seven trained data collectors and three supervisors participated in the data collection process after 2 days of training. Data collectors had BSc degrees in psychiatry nursing. The supervisor had an MSc in mental health. The training aimed to help the data collectors and supervisors understand the content of the questionnaires, objectives, and ethical issues essential to the study. A pre-test was also conducted on 5% of the sample size at Kola Diba Health Centre to assess the reliability of the questionnaire. The reliability of the SMAQ was 0.71 (Cronbach’s alpha) in the pre-test. The ongoing quality of the data was closely monitored by supervisors and authors of the study every five days in face- to- face meetings and telephone conversations and any problems flagged by the study participants and data collectors were discussed and solved.

Data were checked for accuracy and completeness, coded and entered into Epi-Data version 3.1, and exported to STATA version 16 for analysis. Descriptive statistics were computed to illustrate the sociodemographic characteristics and clinical, substance, and psychosocial factors of the study participants and to summarize the distribution of the dependent and independent variables. Repeated measurements taken from the same individual over time are often correlated, reflecting within-subject dependencies. Independent variables that had a p-value <0.2 in the bi-variable analysis were exported to the multivariable logistic regression analysis. Thus, in the analysis, correlation within the response has to be accounted for to increase efficiency when estimating regression parameters. To overcome this autocorrelation, a generalized estimating equation (GEE) model was used (70, 71). The GEE method specifies how the average of a response variable of a subject changes with covariates while allowing for the correlation between repeated measurements of the same subject over time. To estimate the best fitting model, quasi-likelihood under the independent model criterion (QIC) for different correlation structures, such as independent, exchangeable, unstructured, and autoregressive, were employed (71, 72). The smallest QIC was found for the autoregressive correlation structure. Thus, it was chosen as the preferred model. The autoregressive correlation structure indicated that two observations taken closer in time within a subject tend to be more highly correlated than two observations taken far apart, which is also supported by theory. The autoregressive correlation structure is advised for use in time series data (73). The final odds ratio (OR) was computed using the “xtgee” command in STATA 16 with the link function (logit) and family (binomial). Time variant variables such as depression, social support, and HIV- related stigma were statistically significant for ART non-adherence. The relative risk ratio (RR) was analyzed to show the incidence of ART non-adherence in the 2^nd and 3^rd phases.

A flow chart of the recruitment of women living with HIV and the main outcome is presented in Figure 2 . A total of 627 WLWHIV were eligible to participate; four WLWHIV (0.6%) were excluded during the 2^nd and 3^rd phases of the follow-up period from the sample due to loss to follow- up (3) or transfer out (1). Finally, 625 WLWHIV in the 2^nd phase and 623 in the 3^rd phase were included in the analysis, with response rates of 99.7% and 99.4% respectively.

Of the total 1,043 potential participants, a total of 627 WLWHIV were enrolled in the cohort study, including 299 with depression and 328 with out depression. Of the enrolled participants, 222 (35.4%) were divorced, the majority were Orthodox Christians (89.8%), and 257 (41.0%) had no formal education. The majority of the study participants (422, 67.3%) had a low-income status, whereas 185(29.5%) were housewives ( Table 1 ).

Among the study participants, 73 (11.6%) were found to have a comorbid medical illness. Additionally, 9 (1.4%) participants reported having a mental illness other than depression (schizophrenia, bipolar disorder, or generalized anxiety disorder). This is because depression was specifically screened in the study and other mental health conditions were also identified and documented based on participant reports and document reviews ( Table 2 ).

Among the study participants, 349 (55.7%) had moderate social support, while 76 (12.1%) had poor support, with a mean score of 9.1 (SD = 2.2). HIV-related stigma was reported by 286 (45.6%), with a mean score of 24.2 (SD = 6.6).

Regarding substance use, 183 (29.2%) had a history of alcohol use in the previous 3 months, while tobacco use was minimal at 0.3% (2 participants). Gender-based psychological violence was reported by 130 (20.7%) women. Additionally, 213 (34.0%) participants had experienced food insecurity ( Table 3 ).

At baseline, 299 individuals were depressed and ART adherent, while 328 were not depressed and ART adherent. In the second phase (at 3 months), 2 individuals were lost to follow-up, 5 were no longer depressed, and 292 remained depressed. Among the depressed individuals, 39 were ART non-adherent, while among the non-depressed, only 1 was ART non-adherent. By the third phase (at 6 months), of the remaining 278 depressed individuals, 54 were ART non-adherent. Additionally, 17 individuals were no longer depressed, of whom 4 were ART non-adherent.

Among the non-exposed group, at 3 months and 6 months, 16 and 10 WLWHIV had depression, respectively, with no cases of lost follow-up ( Table 4 ).

During the second and third phase assessments, measurements were taken among 299 depressed WLWHIV and 328 non-depressed women. Among the total number of depressed women (including both exposed and non-exposed groups), 102 out of 596 (17.1%) had a history of ART non-adherence. In contrast, among the non-depressed WLWHIV (both from exposed and non-exposed groups), 51 out of 652 (7.8%) were ART non-adherent. Depressed WLWHIV were 2.2 times more likely to be ART non- adherent compared to their non-depressed counterparts [RR = 2.19, 95% CI (1.59, 3.00)] ( Table 5 ).

In the third- month measures of depression and ART non-adherence among WLWHIV, the crude odds ratio (COR) of ART non-adherence was 2.23 (95% CI: 1.31,3.83) in depressed WLWHIV compared to non-depressed WLWHIV. In the third phase (at 6 months), the COR of ART non-adherence was 2.66 (95% CI: 1.65, 4.29) in depressed WLWHIV compared to non-depressed WLWHIV ( Table 6 ).

In the bi-variable analysis, variables that had a p-value < 0.2 were depression, social support, and HIV- related stigma. Depression, social support, HIV- related stigma, and ART non-adherence were measured three times in a 3-month interval. Depressed women living with HIV were approximately 2 [adjusted odds ratio (AOR)=1.97, 95% CI (1.35, 2.87)] times more likely to be non-adherent than non-depressed WLWHIV. The WLWHIV who had poor social support were approximately 2 [AOR=2.15, 95% CI (1.05, 4.38) times more likely to be ART non- adherent than those who had strong social support. Women living with HIV who experienced HIV- related stigma were 1.56 times [AOR=1.56, 95% CI (1.09, 2.25)] more likely to be non-adherent than their counterparts ( Table 7 ).

In the current study, WLWHIV with depression were nearly twice as likely to be ART non- adherent compared to those without depression. This finding is supported by previous research, including a systematic review that reported higher non-adherence to ART among depressed WLWHIV (31), and studies have demonstrated that depression significantly worsens HAART adherence, thereby impacting HIV viral load control (15, 27). One study even found that depression tripled the risk of ART non-adherence (32). Additional research has identified depression as a key factor in suboptimal adherence (33, 74, 75) and a significant barrier to ART adherence (39), with depressed women being at a twofold higher risk of decreased ART uptake and increased treatment interruption compared to men (31). Possible reasons may include depressive symptoms, such as lack of concentration, and sleeping problems could result in forgetfulness, decreased energy, and negative thoughts about self and the effect of ART medication, which in turn could impact missing medical appointments and taking medication on time (76). Moreover, impaired memory, executive dysfunction, engagement in risk behaviors such as alcohol consumption (46), and feelings of hopelessness and apathy, which reduce motivation for self-care, further exacerbate ART non-adherence in this population (42, 77, 78).

Women living with HIV with poor social support were approximately twice as likely to be ART non- adherent compared to those with strong social support. This finding aligns with previous research indicating that robust social support enhances adherence, as evidenced by a study from Colombia (32), while another study revealed that low perceived social support is linked to suboptimal ART adherence (16). Moreover, evidence suggests that strong family cohesion and effective caregiver support contribute to improved ART adherence (79). Furthermore, poor social support has been associated with ART non-adherence in several studies (39, 48, 49). Additionally, inadequate social support can lead to challenges such as food insecurity, further compromising ART adherence (80). The underlying mechanism may be a lack of medication reminders and insufficient assistance, including transportation to the clinic and financial support for medication, which ultimately impact ART adherence (76, 80).

HIV is widely recognized as a highly stigmatized condition (45, 81, 82), and our study found that HIV- related stigma was positively associated with ART non-adherence. This observation is in line with previous research showing that internalized stigma can predict suboptimal ART adherence (16) and that HIV- related stigma is a significant factor in ART non-adherence (39, 48, 49). One possible explanation is that stigma serves as a major source of stress, leading to depression and anxiety, which in turn undermine the ability of WLWHIV to adhere to their ART regimen (46). Additionally, the negative self-perceptions and shame engendered by stigma may reduce motivation for maintaining treatment (83). Stigma experienced in both healthcare settings and the community can result in social isolation and decreased social support, critical elements for ART adherence. Moreover, to avoid discrimination, women may conceal their HIV status, leading to missed medical appointments and/or avoidance of taking medication in the presence of others (84). Discrimination and stigma impact health through complex pathways, including the induction of stress, internalization of negative stereotypes, physical assault, engagement in harmful coping behaviors such as substance abuse, and unequal access to essential health resources (85, 86).

By utilizing a longitudinal study, this study demonstrates the direct association between depression and ART non-adherence among women living with HIV. This study provides a new perspective to understand the relationships between depression, social support, HIV- related stigma, and ART non-adherence. The limitations of this study were as follows. Depressed women living with HIV were referred to the psychiatry clinic and medical unit, which may have affected ART adherence, and since depression was assessed by a screening tool and self-reporting rather than diagnostic interviews, it was difficult to make accurate depression diagnoses. The other limitation was that data on perceived HIV-related stigma, social support, and ART non-adherence were collected through self-report questionnaires, which may be influenced by social desirability bias. Participants may have provided responses they believed would be viewed positively by others, rather than reflecting their true thoughts or behaviors.

Depression appears to be the biggest risk factor for ART non-adherence in WLWHIV. Depression, poor social support, and HIV- related stigma are significant factors in ART non- adherence among WLWHIV. These findings suggest that interventions targeting depression, strengthening social support, and counseling for HIV- related stigma may be effective in improving ART adherence among women living with HIV. Therefore, a holistic and integrated approach is crucial for improving ART medication adherence. Health care providers have to be given the support needed to provide depression care and counseling for HIV- related stigma and strengthen social support for WLWHIV.

This study highlights the complex relationships between depression, social support, and stigma influencing ART adherence among WLWHIV in Gondar. It emphasizes the need for public health initiatives to address these factors, which are significant barriers to ART adherence. Policymakers, healthcare providers, and social service providers must prioritize these issues. Future research should focus on interventions to reduce depression and stigma while strengthening social support to improve ART adherence in Ethiopia and similar settings.