Study design and setting

We conducted a facility-based cross-sectional survey in three ART centres within Tamale Metropolis, namely Tamale Teaching Hospital (TTH), Tamale West Hospital (TWH), and Tamale Central Hospital (TCH). These facilities function as crucial referral points for the five northern regions of Ghana and neighbouring countries such as Togo, Ivory Coast, and Burkina Faso [17].

Study population

Eligibility criteria included PLHIV, aged 18 years or more, receiving ART at TTH, TWH, or TCH for at least six months, consenting to participate, and being in good physical/mental health, as determined through objective observation/interaction and a basic physical assessment. PLHIV experiencing critical clinical conditions during data collection or with speech or hearing impairments were excluded.

Sampling size determination

To estimate the desired effect and ensure the study was adequately powered, we calculated the sample size using the Cochran formula [18]:

where z = reliability coefficient, d = margin of error, p = approximate percentage, q = 1 – p.

This calculation accounted for a reliability coefficient of 1.96, a 95% confidence interval (95% CI), and a margin of error of 5%, assuming an approximate percentage of 44.6% for PLHIV [8]. The estimated sample size was calculated using the following approach:

The resulting sample size was 380 participants. To address potential non-response, a 10% adjustment was applied (10% of 380 = 38), increasing the required sample size to 418 participants.

Sampling techniques and procedures

Three ART facilities (TTH, TCH and TWH) were purposively selected because they are the major referral hospitals within the Tamale Metropolis. Participants were then allocated to each facility using proportional allocation based on the formula: n_i = (N_i/N) × n, where n_i is the number of participants sampled from facility I, N_i is the total number of eligible individuals at facility I, N is the total eligible population across all three sites, and n is the total sample size. This approach yielded 200 participants from TTH, 118 participants from TCH and 100 participants from TWH. Within each facility, convenience sampling was then employed to recruit participants who met the eligibility criteria. If a chosen individual declined to participate, another eligible participant was recruited in their place to ensure the target sample size was maintained.

Data collection

We adapted a questionnaire from various studies [8, 17, 19–23] and pre-tested with a convenience sample of 3 PLHIV at the TCH to ensure clarity and relevance. Topics included ART adherence (i.e., missed none, < 3 doses, 3–12 doses, over 12 doses), socio-demographics, clinical and health system factors with scaled responses (i.e., “Strongly Agree,” “Agree,” “Disagree,” “Strongly Disagree”).

Structured interviews were conducted by seven trained enumerators using a standardized approach. Each interview lasted a maximum of 45 minutes and took place between 27/11/23 and 20/02/24. To ensure consistency and minimize interviewer-related variability, all enumerators received the same training and followed a standard interview protocol. Enumerators were proficient in both English and Dagbani and translated questions as needed for participants who preferred Dagbani. Interviews were conducted in private rooms within each facility to guarantee privacy. Written informed consent was obtained from all participants, and confidentiality was maintained by assigning each participant a unique identification code; no personally identifiable information was collected. Data reliability and completeness were checked daily, and data were entered and coded using Microsoft Excel.

Analysis

Data were transferred to SPSS version 21 for analysis. The primary outcome variable was self-reported ART adherence, with a binary categorisation of ‘good adherence’ if participants reported taking at least 95% (i.e., missed none or < 3 doses) of their prescribed medication in the past 30 days or ‘poor adherence’ if they reported taking less than 95% (i.e., missing 3+ doses) of their prescribed medication in the past 30 days.

Independent variables were clinical or health system-related. These were recoded to facilitate binary logistic regression, with “Strongly Agree” and “Agree” taken as Agree, while “Strongly Disagree” and “Disagree” taken as Disagree. Descriptive statistics, including frequencies and percentages, were calculated for all variables to present a comprehensive summary of the dataset. To identify factors significantly associated with ART adherence, bivariate logistic regression was initially conducted. Variables achieving a significance level of 0.05 in the bivariate analysis were subsequently included in the multivariable logistic regression model to control for potential confounding factors. The variables included in the multivariable model were: fatigue, too many pills, pill odour, auditory hallucination, pill size, cost of accessing ART, social support, interrupted ART supply and cost of treating comorbidities. These variables were selected based on their significance in the bivariate analysis.

To ensure the validity of the responses related to variables influencing ART adherence, a pre-assessment of the internal consistency of the survey instrument was conducted using Cronbach’s Alpha analysis. This method assessed the reliability of participants’ responses across the 31 survey items. A Cronbach’s Alpha coefficient of 0.8 was obtained, demonstrating good internal consistency. This step, conducted before analyzing the data, confirmed the reliability of the instrument in capturing meaningful responses about ART adherence.

Participant characteristics

Table 1 highlights the demographic profile of the 418 PLHIV surveyed, showing a predominance of females (73.2%; 95% CI: 68.7%–77.3%). The majority were between the ages of 30–49 years (56.5%; 95% CI: 51.7%–61.1%), married or cohabiting (64.4%; 95% CI: 59.5%–68.4%), Muslim (71.5%; 95% CI: 67.0%–75.7%), and of Dagomba ethnicity (64.6%; 95% CI: 59.9%–69.1%), reflecting the local socio-cultural landscape. Educational attainment was low, with about one-third (35.4%; 95% CI: 30.9%–40.1%) having no formal education, whilst the majority (73%; 95% CI: 68.5%–77.2%) lived in urban areas.

ART adherence

Table 2 shows that among the 418 PLHIV surveyed, 93.1% (95% CI: 90.3%–95.2%) demonstrated good adherence (defined as ≥ 95% of prescribed doses taken in the past 30 days). The highest adherence was recorded at TTH (95.5%), compared to TWH (89%) and TCH (92.4%).

Potential clinical factors affecting ART adherence

Table 3 shows that while 69% (95% CI: 64.6%–73.4%) of participants expressed preference for injectable ART, this finding does not directly predict adherence and requires further investigation. Most (57.9%; 95% CI: 53.3%–62.6%) also reported adhering to their ART regimen despite experiencing side effects.

Potential health system factors affecting ART adherence

Table 4 shows that most participants reported health system factors positively. For example, 91% (95% CI: 87.8%–93.4%) reported favourable health-worker influence, 95% (95% CI: 92.5%–96.8%) reported that health workers emphasized the importance of medication adherence, and 70% (95% CI: 65.5%–74.2%) received refill reminders. A little over 96% (95% CI: 92.5%–96.8%) reported not experiencing any harsh treatment, discrimination, or abuse at healthcare facilities. Most indicated ART services were always available and accessible (98%; 95% CI: 96.1%–99.1%), were confident in facility confidentiality practices (90.4%), and experienced financial (16.3%) and geographical (35.4%) barriers. Notably, only 22% reported receiving social support from NGOs, welfare units, or volunteers. The highest negative response was 55%, who indicated that increasing patient numbers made it more crowded to access ART at their facility, hence impacting their adherence to therapy.

Clinical and health system-related determinants of adherence to ART

After adjusting for potential confounding variables using multivariable logistic regression analysis, several factors, as shown in Table 5, remained significantly associated with adherence to ART. These included both clinical and health system-related determinants. Clinically, the absence of post-pill fatigue and the lack of concern regarding pill size were positively associated with adherence. From a health system perspective, reduced financial burden, uninterrupted ART supply, and access to strong social support, whether from NGOs, government programs, or community networks, also significantly enhanced adherence.

Specifically, participants who did not experience post-pill fatigue had 91% lower odds of non-adherence compared to those who did experience fatigue (AOR = 0.09; 95% CI: 0.02–0.37). In other words, fatigue was a strong risk factor for suboptimal adherence. Similarly, individuals who reported no issues with pill size were 3.71 times more likely to adhere than those who found pill size problematic (AOR = 3.71; 95% CI: 1.23–11.18). For financial costs, participants who did not report high financial burden were 73% less likely to be non-adherent (AOR = 0.27; 95% CI: 0.10–0.73), indicating that cost is a barrier to optimal adherence. Moreover, ART supply was associated with a nearly eightfold increase in the odds of optimal adherence (AOR = 7.76; 95% CI: 1.02–59.30), the wide confidence interval for uninterrupted ART supply reflects the relatively small number of participants who experienced supply interruptions (n = 34). This estimate should therefore be interpreted with caution. Finally, strong social support was associated with a more than six-fold increase in the odds of optimal adherence (AOR = 6.62; 95% CI: 1.18–37.21). The combination of these five variables explained 16.4% of the variance in ART adherence (Nagelkerke R² = 0.164). The model demonstrated good fit to the data, as indicated by a non-significant Pearson Chi-Square test (χ² = 29.51, df = 22, p = 0.131) and Deviance statistic (χ² = 24.45, df = 22, p = 0.324), suggesting that the predicted values from the model do not significantly differ from the observed values, supporting the model’s adequacy in explaining ART adherence behavior.

Limitations

This study offers important insights into ART adherence in Northern Ghana. The high response rate and consistent data collection procedures also enhance the reliability of the findings. Nonetheless, several limitations should be acknowledged. First, the study excluded individuals in their first six months of ART initiation, those who were critically ill, co-infected with tuberculosis, or with speech and hearing impairments. As a result, some unique adherence challenges experienced by these groups may not have been captured. Future studies should broaden inclusion criteria to reflect a more diverse patient population and ensure greater inclusivity.

Second, the cross-sectional nature of the study limits the ability to draw causal inferences between the identified factors and ART adherence. To better understand temporal dynamics and causal pathways, longitudinal and interventional studies are recommended. Third, although the study employed a consecutive sampling method, a non-probabilistic approach that limits generalizability, we mitigated potential selection bias by recruiting participants systematically as they presented at ART clinics. Additionally, data collection across multiple health facilities with diverse demographics helped to improve representativeness and reduce the likelihood of sampling skew. Fourth, adherence was measured by self-report. Self-reported adherence is subject to recall bias and social desirability bias. Future studies should consider objective measures such as pill counts, electronic drug monitors, or pharmacy refill records to complement self-reports. Fifth, the pretest sample for questionnaire refinement was small (n = 3), which may have limited our ability to detect all ambiguous or confusing items. Sixth, variables for the multivariable model were selected based on bivariate statistical significance (p < 0.05). While this approach is common in exploratory crosssectional studies, it may have excluded variables that are clinically important but did not reach statistical significance due to limited power or small subgroup sizes. Future studies should consider alternative variable selection methods (e.g., inclusion of all a priori clinically relevant variables) to reduce the risk of residual confounding. Because of the cross-sectional design, we cannot infer causality. All reported associations reflect statistical relationships at a single point in time and should not be interpreted as causal effects. Longitudinal or interventional studies are needed to establish directionality.

Despite these limitations, the study offers valuable baseline data and contextual evidence that can inform policy and programming efforts to improve ART adherence in similar low-resource settings.

In conclusion, this study identified a 93% ART adherence rate among PLHIV which falls short of the 95% global benchmark. Clinical factors promoting adherence included the absence of fatigue and concerns related to pill size, while health system related factors promoting adherence included reduced cost of access, consistent ART supply, and good social support. The Ghana AIDS Commission and its implementing partners are urged to strengthen community-based social support networks, expand ART distribution points, and develop targeted educational initiatives to improve therapy adherence and contribute to achieving epidemic control.