The TEEWA study was a cross-sectional case-control study conducted in Thailand from 2011 to 2014 that aimed to investigate the overall living conditions and needs of ALPHIV.
The study included all ALPHIV – defined as the “cases” – aged 12–19 and receiving ART in one of the 19 participating hospitals throughout Thailand (11 in Northern Thailand, which is the region most affected by HIV, three in the north-east, four in the central part of the country, including Bangkok, and one in the south).
For each ALPHIV living in a family setting, one control, presumed uninfected, was randomly selected from the general population and individually matched for sex, age, and district of residence. The matching procedure was the following: for every hospital, we selected the village located in the same district where most of the surveyed ALPHIV lived with their families. In the health centre of this village, we extracted, from the computerized file of the whole population for the village, the list of adolescents of the same age and sex as the ALPHIV interviewed and then randomly selected the controls, one control per case.
ALPHIV living in institutions followed-up in the same participating hospitals were also included. They were living in eight institutions (four in the north, two in the north-east, one in the central part of the country, and one in the south). Six of these institutions were hosting both ALPHIV and known HIV-uninfected adolescents (both HIV-exposed uninfected (HEU) or non-HIV exposed children). Since a matching procedure (by institution, sex, and age) was not feasible due to the small number of children in each institution, the control group was composed of all the uninfected adolescents living in the same institution.
Each adolescent completed a self-administrated questionnaire, providing information on his/her everyday life. To prevent unintended HIV disclosure, no reference to HIV or AIDS was made in these questionnaires. The adolescents’ life history was reconstructed from structured face-to-face interviews with their primary caregivers or institution staff. Also, a medical form was completed by the hospital staff based on the ALPHIV medical records. For the control group, the adolescents’ self-administrated questionnaire was similar to that of the cases as well as the questionnaire for the caregivers except for questions related to HIV, which were omitted. The questionnaires have been already published elsewhere (20).
Definition of school trajectory disruption
The main outcome was a school trajectory disruption defined as a composite outcome: 1) a delay of 1 or more years compared to the age expected for the school grade; or 2) a dropout before the end of compulsory education at Grade 9; or 3) an enrolment in a non-formal education programme.
To generate this composite outcome, we used the type of education (formal vs. non-formal, based on information obtained from the caregivers) and two binary variables obtained from the adolescent questionnaire: academic delay (≥ 1 year) derived from age and current grade for adolescents attending school at the time of the survey, or early school dropout derived from age and grade at the time of school termination for the others. Thus, a binary variable allows us to distinguish disrupted school trajectory from normal school progression.
Variables obtained from the adolescent questionnaire were: sex, age (used as continuous variables), as well as variables related to their school life, such as current school attendance (yes, no), attendance in extra-curricular programmes (yes, no), friends at school (none or few vs. many), school life enjoyment (yes, no), self-reported academic performance (very poor, poor, fair, good, or excellent), history of hospitalizations (yes, no), absenteeism for medical reasons (rarely, sometimes, regularly, or for a long time), and higher education aspirations (yes, no, don’t know).
Other information was obtained from the caregiver questionnaire, such as ethnic origin (ethnic minority vs. Thai), orphan status (one or both parents deceased, both parents alive), type of caregiver (parent, grandparent, more distant relative or guardian), caregiver’s level of education (secondary school and above, primary school, never attended school), perception of the household’s financial situation (fair, good, or very good vs. difficult or very difficult), type of living area (rural, urban), type of school (public or not), any history of school grade repetition (yes, no), perception of neurocognitive difficulties experienced by the adolescents (yes, no), and knowledge of stigmatization experienced by the adolescents at school (no, yes, don’t know). However, information obtained from the institution’s staff was less detailed about the adolescents’ life history.
Finally, for ALPHIV, additional variables were obtained from the medical file: height (converted in height-for-age z-score (HAZ) using WHO child growth standards with stunting defined by HAZ < -2 SD), age at HIV diagnosis (< 7.5 years, ≥ 7.5 years), age at ART initiation (< 9 years, ≥ 9 years), ART type (NNRTI-based ART vs. PI-based ART or other), most recent CD4 count (<20%, ≥ 20%), and HIV viral load (< 1000 cp/mL, ≥ 1000 cp/mL).
To compare cases and their matched controls living in family settings, McNemar’s paired test for categorical variables was used. To compare cases and controls living in institutions (not matched), the Chi-squared test was used for categorical variables. Non-parametric tests were used for continuous variables: Wilcoxon’s signed-rank test for matched samples and the Mann–Whitney–Wilcoxon test for independent samples.
For the analysis of factors associated with disrupted school trajectory, we performed bivariate and multivariable logistic regressions. To select the variables included in the multivariable logistic regressions, a conceptual framework was developed for potential pathways leading to disrupted school trajectories, based on a literature review (10–14,16,21–26) and on our research hypotheses (additional file 1).
Among all adolescents surveyed, the main explanatory variable of interest was HIV-infection. Models were adjusted for the following factors: sex, age, living circumstances, and hospitalization history. We also estimated the proportion of school trajectory disruptions attributable to HIV (attributable fraction, or AF) using an approach developed by Bruzzi et al. (27). The estimate was adjusted for potential confounders using our main logistic model.
Then the analysis focused on ALPHIV to investigate the effect of HIV-specific factors on schooling. Models were adjusted for age at ART initiation, ART type, history of hospitalization and comorbidities such as neurocognitive difficulties or growth delay defined as HAZ < -2.
Finally, the analysis was restricted to ALPHIV living in family settings, a group for whom an additional set of contextual variables was available, such as ethnicity, type of caregiver, caregiver’s educational level, household’s financial situation, type of living area, school type, as well as stigmatization related to HIV infection.
We used multivariate imputations by chained equations to manage missing data (R “MICE” package) before fitting models. Data were analysed using R software version 3.5.3.
We also performed two sensitivity analyses using different definitions of school trajectory disruption: 1) academic delay of 2 or more years or early school dropout; 2) age–grade delay as proposed by Psacharopoulos & Yang (28) using the following school-for-age (SAGE) formula:
where S represents the number of completed school years, A the current age or age at school dropout, and E the upper age limit for primary-school admission (7 years in Thailand). For early dropout, i.e. before Grade 9, the score was calculated using the upper limit of compulsory education (age 15 in Thailand). A binary variable was created, the age-grade delay, defined as a score below 100 or attendance in a non-formal education programme.
For these two sensitivity analyses, we conducted logistic regressions adjusting for the same independent variables as in the main analysis.
Finally, we analysed the main outcome, excluding the HIV-exposed uninfected controls (HEU) to rule out the possible confounding role of parental HIV.