The study population consisted of PLHIV receiving ART and routinely followed at the participating military health facilities. Eligible participants were aged 18 years or older, had been receiving ART for at least 12 months, were receiving care at one of the 35 study sites, and had provided informed consent. Individuals who were unable to provide their views or who declined to participate were excluded. The minimum sample size was calculated using the formula for estimating a proportion with finite population correction [6]:

where: n = sample size; N = total population size; Z = 1.96 for a 95% confidence level; d = 0.05 for a 5% margin of error.

The estimated proportion (p) of patients considered non-adherent was 12.7% [7]. The sample was proportionally allocated across the 35 participating military health facilities according to the number of PLHIV actively followed at each site, yielding a minimum sample of 194 participants after rounding. Eligible patients were then recruited consecutively during routine visits until the required sample size was reached at each site. Table 1 summarizes the sampling data.

Data were collected through individual interviews using a structured electronic questionnaire. Developed using CSPro version 8 in collaboration with the Military Health Information Management Service, the questionnaire comprised 115 questions organized into four sections: sociodemographic characteristics; interactions with healthcare services regarding HIV testing and medical follow-up; experiences of stigma and discrimination; and treatment and medication adherence.

Table 1. Active patient population and minimum sample size by study site.

The questionnaire was administered using Android smartphones or tablets. Biological data required for the study, particularly the most recent Viral Load (VL) result, were obtained from participants’ medical records. Data collection was conducted by community-based workers from two non-governmental organizations partnering with the Armed Forces. A total of 35 data collectors received two training sessions covering data collection procedures, administration of the electronic questionnaire, ethical principles, confidentiality, and respect for participants’ privacy. Data collection was supervised and technical support was provided by the Military Health Information Management Service in collaboration with supervisors from the partner non-governmental organizations to ensure consistent implementation of study procedures across all sites.

ART adherence was assessed using a prespecified composite criterion integrating self-reported medication-taking behaviour and virological status. This approach was chosen to capture complementary behavioural and biological dimensions of adherence rather than relying on either measure alone. Self-reported adherence was assessed using a three-item questionnaire adapted from the CASE Adherence Index [8]. The adapted questionnaire assessed, over the previous 12 months, the frequency of missed doses, difficulty remembering to take ART, and doses taken more than six hours later than scheduled. Responses were assigned prespecified scores, which were summed to give a total score ranging from 0 to 12. A score ≥ 10 indicated good self-reported adherence, whereas a score < 10 indicated suboptimal adherence. Virological status was included as the objective component of the composite criterion. It was determined from VL result obtained within the previous 12 months. In accordance with World Health Organization (WHO) recommendations, a VL ≤ 50 copies/ml was considered undetectable, whereas a VL > 50 copies/ml was considered detectable [9]. Participants were classified as adherent when they simultaneously had good self-reported adherence and an undetectable VL. Those who did not meet both criteria were classified as non-adherent. This composite classification was used as the dependent variable in the analysis of factors associated with ART non-adherence.

HIV-related stigma was assessed using items adapted from the PLHIV Stigma Index 2.0, an instrument developed to document different manifestations of stigma and discrimination by PLHIV [10]. The questionnaire explored experiences of stigma and discrimination within social and family environments and healthcare settings, as well as manifestations of internalised stigma. The latter included self-exclusion behaviours and negative perceptions related to HIV status, such as shame, guilt, feelings of worthlessness, and concealment of HIV status. Each manifestation was analysed separately as a categorical variable according to the questionnaire response options. For opinion based items related to internalised stigma, responses were grouped into two categories, “yes/agree” and “no/disagree”. Associations between each stigma-related manifestation and ART adherence were then examined separately.

Data were analysed using R version 4.5.0. Categorical variables were summarized as frequencies and percentages. Continuous variables were summarized using means and standard deviations, as well as medians and interquartile ranges, where appropriate. The final analytical dataset was complete for all study variables. “No response” and “Not applicable” were predefined response categories and were retained in descriptive analyses but excluded from the corresponding bivariate analyses. Consequently, the number of participants included in each analysis is reported where applicable. No data imputation was performed. Associations between ART adherence and categorical variables were assessed using Pearson’s chi-squared test (χ²) or Fisher’s exact test with Monte Carlo simulation (20,000 replicates) for tables with small cell counts. Continuous variables were compared using the Wilcoxon-Mann-Whitney test when parametric assumptions were not met. Effect estimates were reported with 95% confidence intervals (CIs): Cramér’s V for categorical variables, odds ratios (ORs) for 2 × 2 tables, and the absolute rank-biserial correlation coefficient for ART duration. Bootstrap CIs were used for Cramér’s V and the rank-biserial correlation coefficient. The Benjamini-Hochberg procedure was applied within families of related variables to control the false discovery rate. Findings that were statistically significant before but not after correction were considered exploratory. All tests were two-sided, with statistical significance set at 5% (α = 0.05). Analyses of factors associated with ART adherence were bivariate and unadjusted and were intended to identify crude associations rather than independent predictors. Accordingly, no inference was made regarding independent effects or causal relationships.

The study protocol was approved by the Directorate of Military Health and Social Services of Côte d’Ivoire, acting as the local ethics committee. All participants provided informed consent before enrolment. Data were anonymized and handled confidentially throughout the study.

Of the 243 interview records collected, 30 (12.3%) were excluded following eligibility and data quality checks, leaving 213 participants for analysis. Reasons for exclusion included age < 18 years (n = 1; 0.4%), ART duration < 12 months (n = 5; 2.1%), and missing ART initiation date (n = 24; 9.9%). The mean age was 46.9 ± 11.8 years, ranging from 19 to 79 years. The male to female sex ratio was 0.6. Regarding educational attainment, 37.6% had completed secondary education, 26.3% primary education, 8.0% higher education, and 3.8% vocational training, while 24.4% had no formal education. Nearly one-third were unemployed, whereas 35.7% were self-employed or engaged in entrepreneurial activities. Table 2 summarizes the sociodemographic characteristics of the study population.

Table 2. Sociodemographic characteristics of study participants.

Among the 213 PLHIV, 71.8% were classified as adherent and 28.2% as non-adherent according to the composite adherence criterion. When the two components of the composite criterion were considered separately, 176 participants (82.6%) had good self-reported adherence, while 173 (81.2%) had an undetectable VL (≤50 copies/ml). The mean self-reported adherence score was 10.26 ± 1.89, with a median of 11 out of 12. The reported reasons for missed doses or irregular ART intake are presented in Figure 1.

Figure 1. Reported reasons for missed doses or irregular ART intake.

3.3.1. Sociodemographic and Socioeconomic Factors

None of the sociodemographic characteristics examined was significantly associated with ART adherence (p > 0.05). Membership in a support group was also not associated with adherence (p = 0.175). In contrast, adherence decreased with increasing difficulty in meeting basic needs, from 80.3% among participants reporting no difficulty to 55.6% among those reporting difficulties some of the time and 41.7% among those reporting difficulties most of the time (BH-adjusted p < 0.001; V = 0.288, 95% CI 0.156 - 0.428). The circumstances of HIV testing were also associated with adherence (BH-adjusted p = 0.004; V = 0.255, 95% CI 0.181 - 0.383), although some categories included small numbers of participants.

3.3.2. Healthcare and Service Access-Related Factors

Access to support services was associated with higher adherence in the unadjusted analysis (78.2% versus 63.8%; p = 0.021; V = 0.158, 95% CI 0.025 - 0.294), but this association did not remain significant after Benjamini-Hochberg correction (BH-adjusted p = 0.074). Similarly, participants reporting issues related to the confidentiality or management of their medical records had lower adherence (60.8% versus 77.7%; p = 0.009; V = 0.179, 95% CI 0.041 - 0.314), but the association did not remain significant after correction (BH-adjusted p = 0.064). These findings were therefore considered exploratory. No associations were found with access to treatment, information, testing, prevention, or counselling services.

3.3.3. HIV-Related Stigma

None of the experiences of stigma or discrimination in healthcare settings was significantly associated with ART adherence (p > 0.05). In contrast, several dimensions reflecting the perceived impact of HIV status were associated with adherence, including stress management (BH-adjusted p = 0.031; V = 0.201, 95% CI 0.081 - 0.354), ability to find love (BH-adjusted p = 0.014; V = 0.251, 95% CI 0.113 - 0.410), achievement of personal goals (BH-adjusted p = 0.031; V = 0.197, 95% CI 0.070 - 0.352), contribution to the community (BH-adjusted p = 0.024; V = 0.222, 95% CI 0.088 - 0.378), and religious practice (BH-adjusted p = 0.024; V = 0.210, 95% CI 0.084 - 0.369). Among manifestations of internalised stigma, feelings of guilt and worthlessness were associated with lower adherence in the unadjusted analyses but did not remain significant after correction (BH-adjusted p = 0.103 and 0.058, respectively) and were therefore considered exploratory.

3.3.4. Treatment-Related Factors

The median duration of ART was longer among adherent than non-adherent participants (92.2 [58.0 - 164.7] versus 70.9 [48.9 - 105.8] months; p = 0.017), although the effect size was small (|r| = 0.211, 95% CI 0.047 - 0.374; N = 212). Concomitant treatment for a chronic condition was associated with higher adherence (95.8% versus 68.6%; p = 0.005; V = 0.191, 95% CI 0.108 - 0.262; N = 212). In contrast, the use of traditional medicines or herbal remedies was not significantly associated with adherence (63.0% versus 72.9%; p = 0.284; V = 0.074, 95% CI 0.004 - 0.217; N = 208). Factors and psychosocial dimensions related to HIV status that were associated with adherence are presented in Table 3 and Table 4.

Table 3. Bivariate analysis of factors associated with ART adherence among PLHIV.

Table 4. Psychosocial dimensions related to HIV status associated with ART adherence.

^‡Fisher’s exact test with Monte Carlo simulation; ^§Median [Q1 - Q3], Wilcoxon-Mann-Whitney test; V = Cramer’s V; |r| = absolute value of the rank-biserial correlation coefficient.

Our findings support the multifactorial nature of ART adherence. It is influenced by circumstances that extend beyond individual sociodemographic characteristics. The overall pattern points to the relevance of the material, psychosocial, and treatment context in which PLHIV manage long-term therapy. This interpretation is consistent with evidence from sub-Saharan Africa showing that adherence reflects the interplay of individual, social, economic, psychosocial, treatment-related, and healthcare factors [4][12]. The variation in these contextual influences may also help explain why the sociodemographic correlates of non-adherence have varied across contexts [12].

Socioeconomic vulnerability emerged more clearly in our study, with adherence decreasing progressively as difficulties in meeting basic needs increased. Similar patterns have been reported across sub-Saharan Africa, where food insecurity, financial constraints, and transportation costs have been identified as barriers to sustained ART adherence [4][12]. The Stigma Index 2.0, conducted in the Democratic Republic of Congo (DRC), also highlighted this vulnerability, with 80% of participants reporting difficulties meeting their basic needs [13]. However, Ouner et al. found no association between food security and adherence in Ghana (OR = 0.69; 95% CI 0.25 - 1.83) [14]. Differences in the definitions and measures of socioeconomic vulnerability could partly explain these divergent results.

The healthcare environment may nevertheless influence how patients engage with long-term treatment. Although the association between access to support services and adherence did not persist after correction for multiple comparisons, the observed pattern warrants further investigation rather than firm interpretation. In the AFRICOS study, Mbah et al. found that participation in a support group was not associated with improved ART adherence or viral suppression among 1959 PLHIV receiving ART [15]. Similarly, Somi et al. found no significant association between satisfaction with HIV care and viral suppression [16]. Taken together, these findings suggest that access to support services, group participation, or perceived satisfaction with care may not alone be sufficient to promote adherence. The quality, accessibility, and responsiveness of support to patients’ needs may also be relevant.

The confidentiality of medical information deserves particular attention in our context. Reported problems concerning the management or confidentiality of medical records were associated with lower adherence in the unadjusted analysis. However, this association did not remain statistically significant after correction for multiple comparisons and should therefore be interpreted as exploratory. The Stigma Index 2.0 Côte d’Ivoire provides relevant contextual information for interpreting this exploratory finding. Among the PLHIV interviewed, 78% felt that the confidentiality of their medical records was guaranteed, while 20% were uncertain and 2% thought it was not. Fear that their HIV status would be discovered by third parties or disclosed without consent by healthcare professionals was also among the reasons given for delaying or interrupting care [17]. In this context, confidence in the protection of HIV status may contribute to engagement in care. This issue could be particularly relevant in an institutional environment that accommodates both civilians and members of the defence and security forces, although this hypothesis was not specifically evaluated in our study.

The relationship between HIV-related stigma and ART adherence appears to be complex and may depend on the dimension of stigma considered. Our findings suggest that the psychosocial experience of living with HIV may be more relevant to adherence than specific manifestations of stigma considered in isolation. This distinction is important because stigma may affect treatment behaviour indirectly through its effects on emotional well-being, interpersonal relationships, social participation, and the ability to pursue personal aspirations, rather than solely through overt experiences of discrimination. This interpretation is particularly relevant in the Ivorian context. The Stigma Index 2.0 in Côte d’Ivoire documented shame (32.2%), guilt (28.5%), and feelings of worthlessness (15.8%), as well as broader psychosocial effects among PLHIV [17]. Comparable manifestations have also been reported in the DRC [13]. These descriptive data illustrate the multiple ways in which HIV may affect psychosocial well-being, although they do not establish a relationship with adherence. Evidence regarding such a relationship is itself heterogeneous. In the AFRICOS study, experienced stigma was associated with lower adherence (OR = 0.67; 95% CI 0.56 - 0.80) and lower viral suppression [5]. Conversely, Ouner et al. reported an association between stronger internalised stigma and better adherence (OR = 1.08; 95% CI 1.01 - 1.15) [14]. These apparently contrasting findings suggest that different dimensions of stigma should not be considered interchangeable and that their relationship with adherence may vary according to coping mechanisms, social support, and sociocultural context.

Treatment-related factors yielded more nuanced findings. Longer ART duration was associated with better adherence, which may reflect the progressive integration of treatment into daily routines. Conversely, sustained adherence may itself contribute to longer retention in care, making the direction of this relationship difficult to determine. The cross-sectional design does not allow these possibilities to be distinguished. Reported reasons for missed doses or irregular ART intake were dominated by fear that someone might discover the participant’s HIV status, followed by fear of being treated poorly by healthcare professionals and forgetting to take treatment for no specific reason. Psychological or emotional factors and difficulties related to distance or access to the treatment centre were reported less frequently. These findings illustrate the interplay between stigma-related, psychosocial, and practical barriers to ART adherence described across sub-Saharan Africa [4]. The relationship with concomitant treatment for a chronic condition may also reflect greater familiarity with long-term medication use or more frequent contact with healthcare services, both of which could facilitate the incorporation of ART into daily routines. However, this interpretation remains tentative given the small number of participants receiving such treatment. Furthermore, the lack of association between the use of traditional medicines or herbal remedies and adherence contrasts with the meta-analysis by Heestermans et al., which reported a higher risk of non-adherence among users of these treatments (OR = 2.65; 95% CI 1.18 - 5.95), although substantial heterogeneity was observed across studies [12]. These discrepancies further emphasize the context-dependent nature of factors associated with ART adherence.

This study has several strengths. The use of a composite endpoint combining self-reported adherence and virological suppression allowed for the consideration of both behavioural and biological dimensions of adherence. The simultaneous assessment of individual, socioeconomic, treatment-related, care-related, stigma-related, and psychosocial factors also provided a multidimensional perspective on non-adherence. However, several limitations should be noted. The cross-sectional nature of the study did not allow for the assessment of the temporality or causality of the observed associations. Self-reported data were susceptible to recall bias and social desirability bias, while the small sample sizes in some categories limited the precision of certain estimates. Although the Benjamini-Hochberg procedure was applied within families of related variables to control the false discovery rate, the number of comparisons warrants cautious interpretation, particularly for associations that did not remain statistically significant after correction. Finally, the absence of multivariable adjustment is an important limitation, as residual confounding cannot be excluded. Consequently, the observed associations cannot be interpreted as independent effects, and the findings should primarily be regarded as hypothesis-generating.