Objectives This study assessed children's awareness for adult HIV-associated symptoms and illnesses using a verbal assessment tool by analysing inter-rater reliability between adult–child dyads. This study also evaluated sociodemographic and household characteristics associated with child awareness of adult symptomatic HIV.
Methods A cross-sectional survey using a representative community sample of adult–child dyads (N=2477 dyads) was conducted in KwaZulu-Natal, South Africa. Analyses focused on a subsample (n=673 adult–child dyads) who completed verbal assessment interviews for symptomatic HIV. We used an existing validated verbal autopsy approach, originally designed to determine AIDS-related deaths by adult proxy reporters. We adapted this approach for use by child proxy reporters for reporting on HIV-associated symptoms and illnesses among living adults. Analyses assessed whether children could reliably report on adult HIV-associated symptoms and illnesses and adult provisional HIV status.
Results Adult–child pairs concurred above the 65th percentile for 9 of the 10 HIV-associated symptoms and illnesses with sensitivities ranging from 10% to 100% and specificities ranging from 20% to 100%. Concordant reporting between adult–child dyads for the adult's provisional HIV status was 72% (sensitivity=68%, specificity=73%). Children were more likely to reliably match adult's reports of provisional HIV status when they lived in households with more household members, and households with more robust socioeconomic indicators including access to potable water, food security and television.
Conclusions Children demonstrate awareness of HIV-associated symptoms and illnesses experienced by adults in their household. Children in households with greater socioeconomic resources and more household members were more likely to reliably report on the adult's provisional HIV status.
Statistics from Altmetric.com
An estimated 6.3 million (6–6.5 million) people are living with HIV in South Africa.1 Due to the magnitude of the epidemic, home-based care programmes have rapidly expanded in generalised endemic setting to meet the needs of individuals with HIV. Consequently, provision of care for HIV-infected adults is increasingly provided by non-medical professionals such as children in the home. Poverty and lack of support services are major factors contributing to children providing care to HIV-sick adults.2 We have a growing understanding of the possible challenges, risks and benefits associated with children serving as ‘young carers’. These young carers describe the challenges of limited knowledge and skills for care tasks.3 They describe negative consequences including difficulty concentrating on school-related tasks, attendance and performance;4 increased burden of domestic chores due to adult illness;5 and psychological distress.6 However, they also report positive outcomes including the perception that care has contributed to their maturation,7 positive self-image and development of resilience strategies such as building of social networks.8
Both quantitative and qualitative studies document involvement of children in intimate care tasks including bathing and dressing HIV-sick adults.9–11 As such, it is not surprising that children might directly observe symptoms of HIV illness. However, few studies explicitly examine children's awareness of adult HIV illness.12 No studies, to our knowledge, systematically assess the accuracy of children's awareness of adult HIV-associated symptoms and illnesses in their household. Although affordable HIV testing is now widely available, knowledge of how accurately children report on HIV-associated symptoms and illnesses using a standardised tool would be helpful in cases where adults cannot be directly assessed (eg, when the adult is absent or not included in studies). Additionally, developing understanding of whether children are aware of adult HIV-associated symptoms and illnesses, and evaluating accuracy of children's awareness, would significantly enhance understanding of children's outcomes. Such knowledge could contribute to our understanding of why some young carers report differential outcomes (either positive or negative) compared with other young carers. Such knowledge could better contextualise individual and family dynamics of HIV illness including choice of strategies children use to support sick adults, implementation of protective strategies to diminish HIV infection risk, choice of who young carers turn to for support and how family members interact as the HIV-ill adult becomes more symptomatic.
As such, this study set out with two objectives: (1) to measure child awareness for symptomatic HIV using a verbal assessment tool and (2) to evaluate factors associated with reliability in children's reports. To develop a better understanding of whether children were aware of HIV-associated symptoms and illnesses when involved in caregiving, or direct observation of sick adults living in their household, we used a standardised multiquestion tool to assess symptomatic HIV. This tool was based off an existing validated verbal autopsy tool used throughout sub-Saharan Africa to identify AIDS-related deaths, and adapted in this study to assess HIV-associated symptoms and illnesses.13 We term this adapted tool the verbal assessment tool for symptomatic HIV. To develop an understanding of whether children could accurately report on adult HIV-associated symptoms and illnesses using the verbal assessment tool, we analysed rates of concordance between adult–child dyads on adult HIV-associated symptoms and illnesses and provisional HIV status.
The study was conducted in South Africa, the country with the largest HIV epidemic in the world.14 The study focused on the province with the highest HIV prevalence: KwaZulu-Natal Province, South Africa. Within this province, data collection focused on an urban and rural site, selected based on HIV antenatal prevalence of 30% or higher.15 The urban site was a township in the eThekwini municipality; the rural site was located in the uMhlabuyalingana municipality. All data collection occurred between August 2009 and December 2010.
Within each study site, geographical information systems (GIS) was used to map census enumeration size areas. Random sampling of geographical areas representing census enumeration size areas was conducted to reduce site selection bias. Within each census enumeration size area, each household was visited to determine eligibility. Eligible adults were primary caregivers of children in the household, 18 years of age or older, and resident in the household at least four nights per week. The primary adult caring for the child was defined as the individual responsible for the majority of day-to-day active care. The adult could be related to the child in any way (biological parents, aunts, grandparents, non-relatives). Eligible children were 10–17 years of age, had to reside in the same household for at least four nights per week and concur that the adult interviewed was their caregiver. If more than one eligible adult or child was identified in the household, one was randomly invited for participation. In each household, one adult and one child were interviewed. In total, N=2477 adult–child dyads were eligible and included in the study.
Adults and children were interviewed separately by trained survey enumerators. Respondents completed face-to-face interviews lasting 45–60 min in the language of their choice. All questions were shown to respondents and were read aloud by the interviewer. All respondents were provided with detailed information and services relating to HIV testing, care and support; mental health and substance use services; government services including agencies for accessing social welfare grants; child welfare; and local community organisations providing more general support and services to families. Individuals needing more immediate care were offered referral letters, and with permission from respondents, the study team assisted respondents in linkage to appropriate services.
Creating an HIV verbal assessment tool based on a verbal autopsy measure
The measure used in this paper was based off of a verbal autopsy approach. Verbal autopsy approaches have been used throughout sub-Saharan Africa where vital registration systems may not accurately capture the cause of death.16–18 Verbal autopsy approaches obtain information about a deceased person's illness based on the principle that each disease category has a distinct pattern of symptoms that can be recognised, recalled and reported accurately by medical personnel or laypeople.19 Typically, this approach begins with interviewing a relative, non-relative (eg, neighbour), primary caregiver or non-primary caregiver (eg, community health worker) who act as a proxy when answering questions about the deceased individual's health. During this interview, the proxy reporter describes symptoms observed prior to death of the individual. Then, symptoms reports are reviewed to determine the cause of death.
The measure used in this study—which we term the verbal assessment tool for symptomatic HIV—was based off of an existing validated verbal autopsy measure from two studies by Lopman et al,20 ,21 which inquired about 10 signs and symptoms of HIV illness including weight loss, wasting, jaundice, herpes zoster, sores or abscesses, oral candidiasis, acute respiratory tract infection, vaginal tumours, tuberculosis (TB) and diarrhoea. Lopman et al's20 2006 study, conducted in Zimbabwe, found a sensitivity of 66% and a specificity of 76% for predicting death due to AIDS when using a rule-based algorithm to minimise false positives. Lopman et al's21 subsequent 2010 study used sites in Tanzania and Zimbabwe to retest the algorithm. This study found a sensitivity of 79% and specificity of 79% from Zimbabwe phase 1, sensitivity of 83% and specificity of 75% from Zimbabwe phase 2 and sensitivity of 75% and specificity of 74% from Tanzania for deaths in 15–44 year olds.
In our study, the verbal assessment tool for symptomatic HIV was translated and back-translated from English into isiZulu and piloted to ensure coherency and cultural validity. Data were collected by a team of trained research assistants fluent in isiZulu. Each medical term was accompanied by a layman's description, which was composed with the assistance of the South African nurse who worked directly with HIV-infected individuals and extensively pretested. Questions on the child version of the verbal assessment tool were adapted from the original verbal assessment tool to use more child-friendly language, as demonstrated in table 1, and were pre-piloted with the study's Teen Advisory Group of 14 AIDS-affected children.
All statistical analyses were conducted using STATA V.10.22 Our analyses focused on a subset of the main study sample, consisting of n=673 children and n=673 adults who answered questions on the verbal assessment tool for symptomatic HIV. Since our analyses focused on measuring child awareness and accuracy on adult HIV-associated symptoms and illnesses through concurrency assessment, the following criteria had to be met to be included in analyses of the subsample: (1) adult self-reported on illness, and answered the verbal assessment tool for their own HIV-associated symptoms and illnesses; and (2) child reported on adult HIV-associated symptoms and illnesses using the verbal assessment tool.
We assessed child awareness and accuracy of adult HIV-associated symptoms and illnesses by examining concurrency between adult–child dyads for each HIV symptom and illness. Then, an adult provisional HIV status was made by tallying responses on symptoms and illnesses reports based on the following algorithm calculation: (1) any combination of three or more symptoms or illnesses of acute respiratory tract infection, jaundice, sores/abscesses, vaginal cancer, wasting, weight loss, constant diarrhoea, herpes zoster, oral candidiasis and TB; or (2) any two symptoms or illnesses that are hyper-indicative of HIV/AIDS, which included constant diarrhoea, herpes zoster, oral candidiasis or TB. Bivariate statistical tests were conducted to assess whether differences in children's sociodemographic and household characteristics might explain differential accuracy in reporting. We also conducted logistic regression analysis to examine which sociodemographic variables were predictors of concurrency between adult–child dyads on adult provisional HIV status.
We achieved a high response rate of 99.9% for the urban site and 99.7% for the rural site. Response rate was based off of eligible households (both adult and children had to meet inclusion criteria for a household to be deemed eligible) and consenting and assenting dyads. Adult respondents were primarily women (94%), first-language isiZulu-speaking (95%), from the urban study site (62%) and had a mean age of 44 years old. Child respondents also primarily spoke isiZulu (96%). Approximately half of the child respondents were women (53%), and the mean age for the children was 14 years old. More child respondents lived in the urban study site (62%), lived in formal housing (68%) defined as buildings made with brick, concrete, and so in, and had access to potable water (93%) defined as water sourced from a house or community tap opposed to non-potable water, which is sourced from a river or stream. Slightly more children lived in households that were food secure (68%) defined as being ‘never hungry’, had a household size of six people or smaller (68%), had access to radio (76%) and access to TV (68%).
Results summarising child awareness of adult HIV-associated symptoms and illnesses using the verbal assessment tool are found in table 2. We assumed that adults would be aware of their own symptoms and illnesses. Adult–child pairs concurred above the 65th percentile for 9 of the 10 HIV-associated symptoms and illnesses with sensitivities ranging from 10% to 100% and specificities ranging from 20% to 100%. Vaginal cancer had the highest concurrency (99%), sensitivity (100%) and specificity (100%) while constant diarrhoea had the lowest concurrency (29%) and specificity (20%) but a relatively high sensitivity (80%). The dyad agreement on adult provisional HIV status can be found in table 2 as well. In total, 484 adult–child dyads concurred on adult provisional HIV status, resulting in an overall concurrency of 72% (sensitivity=68%, specificity=73%).
We describe whether sociodemographic and household characteristics, age, gender, site location, household type, potable water, household food security, household size, access to radio and access to TV, of children respondents were significantly related to adult–child concurrence in table 3.
We performed a multivariate logistic regression of n=670 child respondents (3 children omitted due to missing data for household size) assessing significance of sociodemographic and household characteristics to correctly matching adult provisional HIV status (table 4). Children were more likely to accurately report adult provisional HIV status if they lived in households of larger size by nearly a factor and a half (OR=1.49, 95% CI 1.0 to 2.2, p<0.05). Children were more likely to accurately report adult provisional HIV status if they lived in a household with more robust socioeconomic indicators including access to potable water (OR=0.49, 95% CI 0.25 to 0.95, p<0.04), a television (OR 1.7, 95% CI 1.04 to 2.76, p<0.04) and food security (OR=0.36, 95% CI 0.25 to 0.53, p<0.001).
This study systematically assessed child awareness of adult HIV-associated symptoms and illnesses using a verbal assessment tool. Concurrency rates showed that for six symptoms and illnesses—vaginal cancer, jaundice, oral candidiasis, herpes zoster, TB and wasting—children concurred with the adult self-report at rates between 75 and 100%. Although vaginal cancer had almost perfect concordance, the low number of adult cases (n=5) and high number of non-applicable responses (n=410) suggest we should interpret with caution. For three symptoms and illnesses—sores/abscesses, weight loss and respiratory tract infection—concurrency rates ranged between 50% and 75%. Finally, only one symptom—diarrhoea for three or more days—had a concurrency rate below 50%. The low concurrency in reports for diarrhoea may relate to the difficulty of children accurately recalling the number of days another was affected by diarrhoea, especially since this symptom of HIV may be less likely to be observed than the other symptoms and illnesses included in the assessment tool. A 2010 study conducted in Kenya highlighted this issue of poor recall for diarrhoea finding that diarrhoeal recall should not extend back >3 days for children and >4 days for adults to achieve 80% accuracy.23 Therefore, it may have proved difficult for child respondents to remember a consecutive ≥3-day event that had happened in the past. Sensitivity analysis for the symptoms and illnesses varied widely from 10% for jaundice to 100% for vaginal cancer. Specificity analysis proved mostly better outcomes with the majority of symptom ranging from 70% to 100%. Constant diarrhoea was an exception with specificity at 20%. Along with the low number of cases for vaginal cancer, its perfect sensitivity and specificity may be due to its ‘diagnostic’ characteristic, namely that a person must be told by a medical professional (or told by someone privy to this information) that they have cancer. The variations in sensitivity and specificity may be partly explained by how ‘observable’ a symptom or illness is on the human body. The adult provisional HIV status had a concurrency of 72%, sensitivity of 68% and specificity of 73%. These findings were consistent with outcomes for the individual symptom or illness analyses. Children’s ability to reliably match adult provisional HIV status was associated with socioeconomic indicators including access to potable water, food security and access to TV. This may be related to factors that influence the spread of information regarding HIV and public health, such as the ability to understand information provided in English, access to books, TV and media, and the number of HIV campaigns. Children's ability to reliably match adult provisional HIV status was also associated with households with above median household size, a possible proxy for household crowding where children would be in frequent observable contact with the ill adult.
Our study benefited from a large sample size and a high response rate (>99%). We also recognise several study limitations. For example, we assume the adults are familiar with their own symptoms and illnesses and can accurately report on them. This assumption is the basis of our comparison between child and adult reports of HIV-associated symptoms and illnesses and adult provisional HIV status. Another limitation includes the possibility that our results may not generalise to other contexts and that concordance rates may differ when the adult being assessed is a man as the vast majority of adults in our sample were women. Other South African studies have shown that men are less likely to report HIV status compared with women,24 and this difference may affect behaviour in the home in such a way that the child's ability to observe adult HIV-associated symptoms and illnesses is also affected. Finally, our logistic regression may suffer from unmeasured confounding. Despite these limitations, we believe our findings are novel due to our study design of measuring child awareness of adult HIV-associated symptoms and illnesses using a verbal assessment tool modelled after a validated verbal autopsy measure. Overall, findings reveal that children were seemingly aware of a broad range of HIV-associated symptoms and illnesses affecting adults in their household, as demonstrated by high concurrency with adult self-reports of HIV-associated symptoms and illnesses. Children's awareness of the adult HIV-associated symptoms and illnesses may be a result of children's increasing involvement in the provision of medical and intimate care for HIV-positive adults within their household, especially in HIV-endemic settings.6 ,10 ,25
HIV-associated symptom and illness awareness is important to study in contexts where care of HIV-ill adults in the household becomes more and more common. Such knowledge contributes to the dearth of current research on culturally constructed forms of biological knowledge available to children in this context.26 Moreover, assessing the accuracy of children's reports on adult HIV-associated symptoms and illnesses can provide important information on a range of outcomes for both young carers and family members. For example, infection control can be a challenge in home-based care settings.27 Findings from a large multicommunity sample demonstrated a 13-fold increase in severe TB symptomology associated with family with AIDS and children's provision of medical care, exacerbated by socioeconomic vulnerability.28 Also, a study in Kenya found that most caregivers (85%) were unaware of the risks they are exposed to when handling patients with AIDS-related infections and many caregivers (65%) used their bare hands when handling body fluids of patients.29 Awareness and accuracy of HIV-associated symptoms and illnesses might help explain why some young carers implement protective procedures to prevent against infection while others do not.
Adult–child pairs concurred above the 65th percentile for 9 of the 10 HIV-associated symptoms and illnesses and at 72% for adult provisional HIV status.
Children were more likely to reliably match adult's reports of adult provisional HIV status when they lived in households with more robust socioeconomic indicators.
Children were more likely to reliably match adult's reports of adult provisional HIV status in households with above median household size (>6 people).
We would like to acknowledge the important contribution of our entire Young Carers KwaZulu-Natal research team and our local NGO partner in the rural site, Tholulwazi Uzivikele, for this research. We would also like to thank the children and their families who welcomed us into their homes.
Handling editor Jackie A Cassell
Contributors LC and CK conceptualised the study, and contributed to data collection, analysis and writing. EB, DO and MM contributed to data analysis and writing. MM designed tuberculosis symptom screening and contributed to writing. All authors have read and approved this text.
Funding This study was funded by the Health Economics and HIV and AIDS Research Division (HEARD) at the University of Kwazulu-Natal, the Economic and Social Research Council (UK), the National Research Foundation (SA), the National Department of Social Development (South Africa), the Claude Leon Foundation and the John Fell Fund. Support to CK for analysis and writing was generously funded by K01 MH096646 and L30 MH098313 (PI: CK), and R24HD077976. Support to DO was generously funded by NIH grant R24HD077976. Support to MM was generously funded by the Discovery Foundation Academic Fellowship. Support to LC was generously funded by the European Research Council under the European Union's Seventh Framework Programme (FP7/2007–2013) / ERC grant agreement n°313421 and the Nuffield Foundation.
Competing interests None.
Patient consent Adults provided voluntary informed consent for their own participation and for the interviewer to seek the assent of the child under their care. Children provided assent for participation in the study.
Ethics approval Ethical review committees at Oxford University, the University of KwaZulu-Natal, the government Department of Health in KwaZulu-Natal and the Department of Basic Education in KwaZulu-Natal approved research protocols.
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.