Study design and settings

An explanatory mixed-methods design was used. Quantitative data were collected and analysed, and findings guided qualitative data sampling, collection and analysis.27 28 Quantitative data were extracted from client medical records while qualitative data used the positive deviance approach to identify unique practices of PLHIV (individuals or groups) who were enrolled in CCLADs. Through joint displays, quantitative and qualitative data were integrated to gain new insights.

The study was conducted in Uganda’s East Central region, which comprises 12 districts (5 urban and 7 rural). ‘Urban’ was defined as districts in Uganda that have city, municipality, town council or town board status.29 30 In 2017, the region reported a viral suppression rate of 48%, which was the lowest in the country. Additionally, it exhibited relatively high HIV prevalence rates among adults aged 15–64, reaching 4.7%.31 All 12 districts were included in the study. Up to 134 accredited public and private-not-for-profit facilities provided ART services to approximately 85 000 PLHIV in the region in 2019.32 Decentralised HIV services are primarily delivered by these health facilities for free.33 CCLADs were first introduced by private-not-for-profit facilities as a pilot programme in 2013. In 2017, Uganda’s Ministry of Health (MOH) adopted guidelines for differentiated care, which allowed for implementation of CCLADs in all health facilities.13

Quantitative part

An analytical cross-sectional study design was used.

Target population and selection criteria

The target population consisted of PLHIV receiving ART refills from CCLADs in the region. The study population included PLHIV aged 18 years or older, who were enrolled in CCLADs between 1 January 2013 and 31 December 2020. Newly enrolled clients in CCLADs (less than 2 months) and those who had not received their first ART refill in the group were excluded.

Sample size

Since the CCLAD model was relatively new, we collected medical records of all eligible PLHIV in CCLADs within the region who met our criteria to ensure maximum precision.

Variables

A data extraction tool was created using three health facility source documents: the electronic medical records system,34 ,35–37 the client’s ART card13 and the CCLAD monitoring form. Using the tool, these variables were extracted for each individual: client initials and unique number, age, sex, occupation, village/district, health facility, CCLAD group code, adherence status, viral load, clinical stage, regimen change, the status of PLHIV in CCLADs, date of enrolment/attrition from CCLADs and reasons for entry or attrition, and CCLAD group code.

Treatment outcomes and definitions

The primary outcome was retention in CCLADs—defined as the period from joining the CCLAD to leaving it or censoring (end of follow-up or death). Individuals in CCLADs were followed up from the date they joined the CCLADs (between 1 January 2013 and 31 December, 2020). They were required to have been part of the CCLAD for a minimum of 2 months and to have received at least one ART refill within CCLADs. A secondary outcome was attrition—was defined as the act of a PLHIV leaving their initial CCLAD temporarily or permanently. In this study, attrition was classified into two categories—switching or withdrawing. Switching occurred when a PLHIV temporarily returned to facility-based care for monitoring or joined a new CCLAD group. In the withdrawing group, there were PLHIV who moved to alternative ART models, stopped receiving ART or were lost to follow-up. PLHIV who were lost to follow-up were defined as those who missed ART pick-up for at least 90 days but had not died or transferred out.38

Other treatment outcomes included were the latest recorded clinical stage, viral load and adherence. Viral load results and adherence status data were collected for each year from 2017 to 2020. Virological suppression was defined as less than 1000 copies/mL.13 In our study, viral load results ranging between 1 and 1000 copies/mL were categorised as ‘suppressed with detectable viral load’. PLHIV with zero copies/mL were assigned a ‘suppressed with an undetectable viral load’ status. Annual viral load analyses aimed to identify PLHIV who experienced virological failure throughout the study period. Adherence assessed and recorded at each ART refill visit using pill counts was adopted. Adherence was defined as ‘good’ if ≥95%, ‘fair’ if 85%–94% and ‘poor’ if <84%.38 See online supplemental file 1 for more variable definitions.

Data collection

Data were collected at 65 health facilities using clients’ files and the routine electronic monitoring system between 18 April 2021 and 30 May 2021. An electronic version of the data extraction tool (designed in the KoBo toolbox) was created, accessible offline on a mobile phone or laptop. Six experienced research assistants (four men and two women) were identified and trained (half-day) on the questionnaire content and the ethical procedures. A hard copy and electronic version of the data extraction tool were pretested against eight client files from one health facility. Pretest feedback was used to revise the tool.

Data analyses

Clients’ sociodemographic, clinical and group characteristics were summarised by descriptive statistics. Retention was assessed as an open cohort, where individuals entered the cohort throughout the follow-up period. Prior to assessing CCLAD retention and attrition, missing data (WHO clinical stage, group size, patient occupation, regimen change in CCLADs, history of virological failure) were imputed by multiple imputation using chained equations (MICE).39 40 MICE approach imputed missing values using observed values for a given individual and the relations observed in the data for other participants. Using ‘mi impute’ command, we set the number of imputations to 50 ensuring this number slightly exceeded the percentage of missing data for the variable with the highest proportion of missing data (43%). Retention proportions were estimated using the Kaplan-Meier method.41 We also analysed retention by cohort (pre-2018 cohort, 2018 cohort, 2019 cohort and 2020 cohort) to identify retention differences over time.

For attrition, the univariable and multivariable Cox proportional hazards regressions were run. The multivariable model aimed to provide a comprehensive understanding of the relationship between risk factors and attrition from CCLADs. Potential confounders such as age, sex, WHO clinical stage and level of care were identified from the literature42–44 and included in the model alongside other predictors. Multicollinearity was assessed using the variance inflation factor (VIF), with a threshold of <5 considered acceptable. The WHO clinical stage exhibited a higher VIF and was excluded from the final set of variables lowering the mean VIF to 1.88 from 7.97.

Proportional hazards assumption was assessed using Schoenfeld residuals before model fitting while Cox-Snell residuals (online supplemental file 2) evaluated the goodness-of-fit.45 Backward elimination method was then employed, initially fitting the model with all variables except WHO stage. Non-significant variables (p>0.05) were sequentially removed, starting with those with the highest p values. We compared this model with one including only significant variables at the univariable level and there were no differences in the final variables significant at p<0.05. Time-varying effects for virological failure were incorporated, and an interaction term for gender and CCLAD composition was included (interaction term was non-significant). Clustering in the variables was accounted for by including a cluster-robust SE in the multivariable model to estimate the coefficient’s SE, thus addressing potential residual correlation within clusters. All statistical analyses were performed with Stata/IC V.13.1. Strengthening the Reporting of Observational Studies in Epidemiology checklist for reporting quantitative data was followed (online supplemental file 7).

Qualitative part

Individual in-depth interviews (IDIs) and focus group discussions (FGDs) were used.

Study participants and recruitment

Four districts with high attrition of PLHIV from initial CCLADs were identified using historical data collected in the quantitative part. Within the four districts, nine health facilities were selected where attritions were highest. Participants for IDIs and FGDs were recruited from the nine health facilities based on these criteria: (1) those who were enrolled or had previously belonged in CCLADs in districts experiencing high attrition and (2) those who were alive and receiving ART in CCLADs or returned to the facility model within the high-attrition districts. PLHIV attending IDIs included those attending CCLAD activities or who had returned to the facility model. For the FGDs, only clients attending CCLAD activities were selected. The FGDs were used to validate findings from IDIs. Figure 1 shows detailed criteria for selecting participants for IDIs and FGDs, including positive deviants.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Selection criteria for target sites and positive deviants in the study. CCLADs, community client-led antiretroviral distribution; FGDs, focus group discussions; PLHIV, people living with HIV.

Sample size

As part of the purposeful sampling method, the positive deviance comparison strategy was used. The strategy involves comparing persons and groups that have discovered solutions with peers where the problem persists.46 47 In the study, 50 of 1702 PLHIV enrolled in CCLADs in the 4 high attrition districts were selected (figure 1). In total, 22 PLHIV participated in IDIs and 28 in the FGDs. The total number selected for IDIs and FGDs matched the recommended sample to reach saturation in qualitative research.48 49

Data collection and qualitative aspects assessed

A semistructured questionnaire, topic guide and observation checklist were created using information from various studies.17 47 Information was collected on age, sex, group size and composition, meeting frequency, duration and activities, as well as individual responsibilities. Other variables assessed were respect, confidentiality, consensus and cooperation, and communication channels. A bilingual translator translated the tools into Lusoga, the local language. An expert committee of bilingual professionals reviewed the translated tools to evaluate linguistic and cultural appropriateness. These tools were then tested with representative participants for comprehension, clarity and cultural relevance. Translated tools were back-translated into English, ensuring alignment with the cultural context and linguistic preferences of the target population. Four qualitative interviewers pretested the tools at two health facilities with four PLHIV and two CCLAD groups and incorporated feedback. To recruit participants for IDIs, research assistants worked with health providers within ART clinics to identify PLHIV in CCLADs and made telephone calls to explain the objectives of the study. FGD participants were invited through their group leaders. Each research assistant interviewed clients individually during IDIs at the health facility (45–60 min). In FGDs, one research assistant observed non-verbal behaviours and summarised key discussion points while the other facilitated the discussion. FGDs took place at the health facility and lasted up to 90 min. Some FGDs included PLHIV from the same CCLAD group to observe their non-verbal behaviours and interactions. Data saturation was achieved in both IDIs and FGDs.

All IDIs and FGDs were audio recorded, and field notes were taken. Occasionally, research assistants used translators where people spoke a different language. PLHIV who participated in the study received monetary compensation for their time and travel expenses.

Data analyses

Audio recordings were transcribed into English by native speakers and transcripts were double-checked for accuracy. A list of notes taken during the interviews was also compiled and compared with responses from the interviews. Three authors performed a thematic analysis of the qualitative responses. Open coding was used to identify common and uncommon practices.50 A common practice was defined as a behaviour or practice that supported long-term HIV retention and was commonly practised by both positive deviants and the comparison group. Uncommon practices were defined as positive, unique and demonstrably successful behaviours or practices common among positive deviants, rather than the comparison group. The final list of codes was generated after comparing and harmonising terms used by the three coders. A checklist of Consolidated Criteria for Reporting Qualitative Studies was followed (online supplemental file 8).51

Data integration

This study approached integration from the three levels proposed by Fetters et al.27 Integration refers to a conscious effort to combine quantitative and qualitative research approaches to achieve a better understanding of the topic.52 Table 1 summarises the levels of integration used in the study.

Patient and public involvement

We engaged various stakeholders including representatives from projects supporting HIV services, and health authorities in refining our research question and methods. Health facility focal persons, CCLAD leaders and members contributed insights to refine our study tools, including the questionnaire and discussion guides. CCLAD leaders facilitated site selection and patient recruitment for FGDs. All CCLAD members including leaders contributed to data interpretation, particularly identifying positive deviant practices. Our dissemination targets CCLAD leaders, facility focal persons and key stakeholders, including hospitals, at facility and district levels, and the wider research community.