Data are from Pact’s community-based, USAID-funded Kizazi Kipya Project in Tanzania (2016–2021). The project aims to increase the uptake of HIV and other health and social services by OVC and their household members. Community Case Workers (CCWs) collected the data from caregivers’ self-reports during beneficiary enrollment using the project’s Screening and Enrollment, and Family and Child Asset Assessment (FCAA) tools. Beneficiaries are enrolled into the project if their household meets one or more of the 14 enrollment criteria that cover household vulnerabilities related to HIV: household is headed by a child (under age 18 years), household is headed by an elderly caregiver (age 60 years or older), household cares for at least one single or double orphan, caregiver is chronically ill and unable to meet his/her children’s basic needs, caregiver is a drug user, caregiver or an adolescent age 10-19 years in the household is a sex worker, at least one adolescent girl age 10-19 years in the household is sexually active, adolescent girl age 10-19 years in the household is pregnant or has a child of her own, at least one household member is HIV-positive, at least one child in the household has tuberculosis, at least one child in the household is severely malnourished, at least one child in the household has been or is being abused or at risk of abuse, at least one child in the household is living and/or working on the streets, and at least one child in the household is working in mines. These criteria are equally applied for all implementation areas and age groups.
Data for this study originated from 18 regions of Tanzania where the USAID Kizazi Kipya project had implemented enrollment activities in 2017: Dar es Salaam, Dodoma, Geita, Iringa, Kagera, Katavi, Mbeya, Mjini Magharibi, Morogoro, Mtwara, Mwanza, Njombe, Pwani, Rukwa, Ruvuma, Singida, Tabora, and Tanga. Of these regions, Mjini Magharibi has very low adult HIV prevalence (0.6%), while Njombe has the highest in the country (11.4%) according to the recent Tanzania HIV Impact Survey (9). A total of 67 district councils (48 rural and 19 urban) considered high HIV burden from the 18 regions were included in this study.
The study population encompassed 39,578 OVC who were enrolled in the USAID Kizazi Kipya Project from January to March 2017, and had complete information on their HIV status and their caregivers’ characteristics. In the context of the U.S. President’s Emergency Plan for AIDS Relief (PEPFAR), an OVC is a child, ages 0-17 years, who is either orphaned (i.e. lost one or both parents to HIV/AIDS) or made more vulnerable because of HIV/AIDS (47). For programming purposes, the USAID Kizazi Kipya Project extends the OVC age to 19 to include all adolescents (48). Therefore, OVC included in this study were aged 0–19 years. The majority of the OVC were under age 15 years (n = 27,935, 70.6%).
USAID Kizazi Kipya defines a caregiver as a guardian with the greatest responsibility for the daily care and rearing of one or more OVC in a single household. A caregiver is not necessarily a biological parent. Only one caregiver per OVC was included in this study: the person identified as having primary responsibility for caring for the child, i.e., the primary caregiver. The youngest caregiver in this study was aged 18 years. References to caregiver in this manuscript denote each child’s singular, primary caregiver.
The study design constituted a cross-sectional secondary analysis of the existing FCAA data, as described above. The data were collected once during beneficiary screening and enrollment.
OVC HIV status as reported by the caregiver was the outcome or dependent variable and was measured through the two categories of negative and positive. For computational purposes, the variable was organized as follows:
(See Formula 1 in the Supplementary Files)
The main independent variable for this study was sex of the caregiver, measured through the two categories of male and female. Other independent variables included OVC sex, OVC age (in years), OVC nutritional status, caregiver age (in years), HIV status of the caregiver, education of the caregiver, family size, whether some or all the family members are covered by health insurance, whether the caregiver is physically or mentally disabled, household wealth quintile, and type of residence (rural or urban). Rural residence included all those living in district councils, whereas those living in township, municipal or city councils were considered as urban residents.
Family size (i.e., number of people living in the same household) was divided into three categories: households with 2–3 members, households with 4–13 members, and households with 14 or more members. This was based on an explorative analysis of OVC HIV prevalence by family size as a discrete variable. Families with similar prevalence were grouped together, thus the categories. The smallest household had two occupants – the OVC and his/her caregiver.
Nutritional status was assessed using mid-upper arm circumference (MUAC) measuring tapes. MUAC is recommended for community-based screening of acute malnutrition (49). Interpretation of the readings was guided by the standard definitions of the colors, whereby the person being assessed is nourished if the reading falls in the tape’s green zone, the person being assessed is moderately undernourished if the reading falls in the yellow zone, and the person being assessed is severely undernourished if the reading falls in the red zone (50).
Wealth quintile was constructed using principal component analysis (PCA) of household assets to determine household socio-economic status (51). Five wealth quintiles were formed, ranging from the lowest quintile (Q1) for the poorest households, to the highest quintile (Q5) for the most well-off households. Family-owned assets included in the PCA were dwelling materials (brick, concrete, cement, aluminium, other), livestock (chicken, goats, cows, other), transportation assets (bicycle, motorcycle/moped, tractor, motor vehicle, other), and productive assets (sewing machine, television, couch/sofa, cooking gas, hair dryer, radio, refrigerator, blender, oven, other).
Data analysis was conducted using Stata version 14.0 statistical software. Exploratory analysis was conducted through one-way tabulations to obtain distributional features of the respondents in each variable. Cross-tabulation of OVC HIV status by each of the independent variables was conducted to assess the variability of OVC HIV prevalence by levels of each of the independent variables. The Chi-Square (χ2) test was used to assess the degree of association between OVC HIV status and each of the independent variables.
Multivariate analysis was conducted using a random-effects logistic regression model due to the hierarchical or clustered structure of the data (52). The usual assumption of independence of the observations did not hold because two or more OVC who have the same caregiver, or who reside in the same household may be correlated. Thus, a multilevel model, which recognizes these data hierarchies and allows for residual components at each level in the hierarchy, was used (53). This choice was based on the assumptions that OVC from the same household and caregiver are dependent in their behavioral, physical, or mental characteristics because they share the same social, health, and economic resources, as well as risks available at the household level. This is likely to exert a related influence in their social life and health outcomes.
Five multivariate models were constructed. The first model encompassed the entire study population of 39,578 OVC ages 0–19 years. The remaining models broke down the study population by age group: the second model was for 5,217 OVC in the age group 0–4 years, the third model was for 10,457 OVC in the age group 5–9 years, the fourth model was for 12,261 OVC in the age group 10–14 years, and the fifth model was for 11,643 OVC in the age group 15–19 years. The stratification of the multivariate analysis by OVC age offered a deeper examination, interpretation and comparisons of the patterns and concentration of the association between caregiver sex and OVC HIV status across different bands of the OVC age.
All statistical inferences were made at the conventional significance level of 5% (α = 0.05), whereby any association corresponding with a p-value less than 0.05 was considered statistically significant.
Some key variables, such as whether the caregiver was the child’s biological parent, were not available in the data. Recall bias was possible during data collection because all information (except for nutritional status, which was measured) was self-reported, though findings suggest that the effect may be minimal because results are comparable with existing biomedical and clinical studies. Since this study was cross-sectional in design, temporality cannot be established, which precludes drawing causal inferences from these findings.