Data was collected from April to July, 2019, in health facilities in Jinja that offer paediatric ART services. Jinja is a district in the south-eastern part of Uganda with a population of about 471,242 people (12). At the district level, the public health system in Uganda is decentralized with Village Health Teams making up the lowest level, followed by Health Centre II facilities (HCII’s), HCIII’s, HCIV’s and the district and regional hospitals. Data was collected from Jinja Regional Referral Hospital, 2 Non-Government Organization-run health facilities, the 5 HCIV’s and 9 HCIII’s. These health facilities all provide paediatric ART services at no charge.
A cross-sectional study design was used. Quantitative data was collected through a health facility-based survey and medical record review.
The study population was children under five years living with HIV who were receiving ART services from health facilities in Jinja, and had been on ART for at least three months, as recorded in their medical records. The respondents were the caregivers with whom the children presented to the health facilities on the days of data collection. Only children who had lived with their caregiver for at least one month were included in the study. These were sampled consecutively as they were presented for care and, for caregivers with more than one eligible child, the first child to be presented was included. Eligible children whose caregivers declined to give consent to participate were excluded from the study.
Data collection, variables and their measurement
An interviewer-administered electronic questionnaire was used to collect data from eligible caregivers using KoBo Toolbox kit (13) installed on mobile phones. The questionnaire was provided in English and also translated into Lusoga, the language predominantly spoken in Jinja. Data was extracted from medical records of children whose caregivers had been interviewed using an electronic data abstraction form. The study tools had been pretested on five children in a health facility in a neighbouring district. Data was collected by a male and female research assistant, who had training in Nursing and Social Sciences, respectively. Both had experience in data collection and were fluent in both Lusoga and English. They were also trained in the study procedures and ethical conduct of research before data collection commenced.
The outcome variable, non-adherence to ART, was measured using a Visual Analogue Scale (VAS) and the Proportion of Days Covered (PDC). The VAS was based on the modified Medication Adherence Self-Report Inventory questionnaire which includes two VAS items: one part measures how much of the prescribed medicine is taken (VAS-dose) while the second part measures how much of it is taken within 2 hours of the recommended time (VAS-timing), both measured over 30 days. (14) Through medical record review, concurrent adherence to all medicines in the ART regimen for each child was also determined by calculating the PDC, the proportion of days over a 90-day period when the child had supply of all drugs in their ART regimen. Adherence was categorized at the 95% cut-off on the VAS-dose, VAS-timing and the PDC measures, and only children with adherence levels of 95% or higher on all three measures were considered adherent.
Data was also collected on the exposure variables: the child, drug regimen, health system and caregiver factors. The child factors were age, gender, place of delivery, school/day-care attendance, illness in the past two weeks and clinical stage on ART initiation, which were measured using the questionnaire. The child’s most recent viral load reading before the date of data collection was recorded from their medical records. Data on drug regimen factors including the drug combination the child was on, the experience of side effects and drug regimen changes since ART initiation was obtained from medical records. Health system factors were measured using the questionnaire. These included: the type of health facility, the physician-patient relationship which was measured using 13 items from the Patient Communication Behaviours Scale (15) and the caregiver satisfaction with the quality of service at the child’s ART clinic, assessed using a 9-item scale adopted from Ivy, Ng (16).
The caregiver factors were assessed through the questionnaire and they included: age, sex, highest education level attained, marital status, place of residence, HIV status, relation to the child, main source of financial support, beliefs about medicine, alcohol use and household food security. The caregiver’s beliefs about medicine were assessed using 10 questions from the 18-item Beliefs About Medicine questionnaire (BMQ) (17). These included 5 questions assessing beliefs about necessity of the medicine (BMQ-Necessity scale), and 5 questions assessing concerns about the medicine’s negative effects (BMQ-Concern scale). Statements in the BMQ were scored on a Likert-type scale that ranged from 1, for strongly disagree to 5, for strongly agree. The maximum score on both the BMQ-Necessity and BMQ- Concern scales was 25, and using a cut-off of 12.5, caregiver beliefs about medicine could be classified as “Accepting”, “Ambivalent”, “Indifferent” or “Skeptical”. (18, 19). Household food security was measured using the Household Food Insecurity Access Scale (HFIAS), which categorized households as “Food secure”, “Mildly food insecure”, “Moderately food insecure” or “Severely food insecure” (20). Caregiver alcohol use was assessed over a period of 30 days before the date of data collection using the question, “How often have you had an alcoholic drink in the last 30 days?”, with responses, “Never”, “Once a month”, “2 or 3 times a month”, “Once or twice a week”, “3 or 4 times a week”, “ Nearly every day” and “Daily”.
The formula for sample size determination of cross-sectional studies (21) was used to estimate the number of children to sample, using an estimated prevalence of ART non-adherence of 21% (7), a precision of 5%, and adjusted for a finite population of 305, the number of children active on ART in Jinja as of September, 2018. The final sample size was then adjusted for non-response of 20% to give a minimum sample size of 168 children.
Data was analysed using STATA version 14 (StataCorp LP, TX, USA). Means with standard deviations were used to summarize normally distributed continuous variables while medians with interquartile ranges were used for continuous variables that were non-normally distributed. Frequencies and their corresponding percentages were used for the categorical variables. The outcome variable, non-adherence measured by the combined VAS and PDC measures, was presented as a proportion.
Modified Poisson Regression was used for bivariate and multivariable analysis and associations were measured using Prevalence Ratios (PRs). Exposure variables with p-values less than 0.1 at bivariate analysis were considered for inclusion into the multivariable analysis model. Multi-collinearity between the candidate variables was examined and a correlation coefficient greater than 0.4 was considered high collinearity. Stepwise elimination was used to build the multivariable model, where elimination and addition of variables was based on the Akaike Information Criteria and variables specified as important from previous literature. A p-value less than 0.05 was considered statistically significant for all analyses.