This is the first large-scale longitudinal study examining the pathways from household chronic illness to child abuse in the developing world through multiple mediation analysis. AIDS-affected households, showed higher levels of physical and emotional abuse compared to healthy households while households affected by other chronic illness had lower abuse prevalence rates.
There are no research findings to date that explain the difference in abuse risk between families affected by AIDS and those affected by other chronic illness. It is possible that families affected by AIDS experience additional and different stress factors to families affected by other chronic illness. These could be fear of death and severe symptomology [
59], AIDS-related stigma [
60] and lower quality of life [
61]. Furthermore, families affected by chronic illness in this study suffered from diseases with more straightforward treatment options and lower perceived stigma such as diabetes or high blood pressure. Existing studies investigating parenting in families with chronic illness either focused on AIDS-affected or cancer-affected families [
19,
21], both illnesses with high levels of stigma and perceived shorter life expectancy. Results from previous studies might therefore be more applicable to AIDS-affected families than to those affected by other chronic illness. Future research could valuably explore linkages and differences between these factors.
Direct and indirect effects of household chronic illness on physical and emotional abuse victimisation were found. In particular, direct and indirect effects were observed for household AIDS-illness showing increased risk for abuse victimization for children in AIDS-ill families through poverty and disability. Direct and indirect effects were also found for households with other chronic illness, surprisingly, showing reduced risk of abuse and poverty for children in other ill households. However, an increased risk for severe disability taht increased risk for child abuse victimisation was also observed. The findings of this study therefore extend previous research from South Africa which found direct associations between physical and emotional abuse victimisation and household AIDS-illness but not with other chronic illnesses [
13,
23] and the findings partially correspond with this.
Poverty was also an important mediator of the relationship between AIDS-illness and physical and emotional abuse. Unexpectedly, lower levels of poverty were a protective mediator of the relationship between other chronic illness and physical and emotional abuse. This study therefore corroborates current evidence that found that households affected by other chronic illness in South Africa appear to have a lower risk for poverty compared to healthy and AIDS-affected ones [
62]. The lower risk for poverty in households affected by other chronic illness in this current study decreased the risk for child abuse victimisation in the mediation models. However, poverty as a factor itself remained clearly linked to an increased risk of child abuse.
Differences in poverty risk between households affected by AIDS and other chronic illness could be attributed to differences in the age groups between the ill household members. Chronic illnesses measured (i.e. high blood pressure, diabetes) may be more likely to appear in older age people who are entitled to a state pension in South Africa [
63]. Of the children in households with chronic illness in this study, 26.6% reported being cared for by their grandparents compared to 13.8% in AIDS-affected households. State pensions have been shown to reduce household poverty as they are spread across all members of a household [
64,
65].
The role of poverty is consistent with research linking AIDS-affected households with high levels of deprivation [
66] as AIDS-illness increases household poverty [
67] through inability to work, medical expenses and excessive funeral costs in case of AIDS-death [
68]. On the other hand, poorer households are at higher risk for HIV infection and this can set up a vicious cycle [
69].
Considerable differences were found in the pathways to abuse depending on the abuse outcome, the child’s gender and the illness status of the family. Differences in maltreatment according to the child’s gender and to family illness were expected due to previous studies suggesting differences in risk between these groups [
13,
70]. However, no previous research has investigated pathways from illness to abuse in a similar fashion before, and speculation about these differences in results would go beyond the scope of the data. Thorough future research is needed to corroborate these findings and examine possible reasons for these differences. If these persist in future studies, there may be implications for policy makers and practitioners in focussing interventions.
Limitations and future research
This study had a number of limitations. First, less than two-thirds of the South African population know their HIV status, which makes self-reporting of HIV status unreliable [
71]. This study was therefore not able to identify households with HIV+, but asymptomatic members. However, the verbal autopsy to identify AIDS-illness has been successfully used in previous studies with good reliability [
15,
23]. Furthermore, identifying only households with AIDS sequelae allows for a fuller understanding of this subgroup of individuals. Second, no scales for child abuse victimisation have been validated for use in South Africa. However, all scales were successfully used in prior studies and showed good reliability in this sample [
13,
23].
Third, the study was carried out in randomly sampled areas with 30%+ HIV prevalence. Results are therefore not generalisable across the South African child population but give a good indication of risks for children in low-income areas with high HIV prevalence. Fourth, this study measured the risk of abuse in families affected by chronic illness, however, it should be noted that the ill person and the person abusing the child may not be one and the same. However, the results clearly indicate that household illness increases the risk for child abuse victimisation through poverty and extent of disability.
Fifth, referrals to child protective services at baseline could have potentially influenced the results and levels of abuse at follow-up. Unfortunately, social services in South Africa are overburdened and understaffed and rarely able to respond to referrals in a timely manner [
72]. Only a tiny number of children referred at baseline (<3%) had been contacted by the appropriate services by follow-up and impact of baseline referrals on results is therefore unlikely. Sixth, the study measured child abuse committed by an adult within the child’s network but investigated mediation between household factors. The perpetrator could, therefore be an adult outside the child’s home i.e. a teacher. In this study, 74.6% of all physically and emotionally abusive acts were carried out within the child’s home, with parents and relatives as the perpetrators [
73], suggesting that the observed effects on physical and emotional abuse are primarily associated with events occurring within the household.
Seventh, the data presented cannot determine causality and this study was therefore not able to determine whether living in a family affected by chronic illness causes an increased risk for child abuse victimisation. Longitudinal observational designs allow for controlling of baseline confounders and identification of correlate directionality because the hypothesised risk factors precede the outcome [
34]. They are therefore superior to cross-sectional studies where temporality cannot be determined. This study established that baseline chronic illness has an effect on risk for child abuse victimisation at follow-up. Temporal order could not be established for the mediation analyses as these were only cross-sectional due to only two time points collected. However, cross-sectional mediation analyses can be used for theory generation and development, with the understanding that the hypothesis arising from these analyses will then have to be verified in longitudinal data [
74].
Eighth, there is a strong likelihood of unmeasured confounding in this study as suggested by the low values in R2. Even though models adjusted for potential confounding variables reported by children, caregiver related confounders such as mental health or substance use could not be accounted for. Due to the design of the study, unmeasured confounding cannot be ruled out.
Finally, the study used child self-report with interviewer-guided questionnaires. Opinions differ whether children are reliable informants regarding disability and illness within the household. However, previous research has used the verbal autopsy and disability measures successfully [
15] and has shown that children often carry out caring tasks within the home that allow them to witness physical ability and symptomology of ill household members [
75]. Furthermore, a recent study investigating inter-rater reliability between adult-child dyads using the verbal autopsy tool found concordant reporting of adult HIV status to be 72% and no significant association between concordance and child age [
76].
Interviewer presence during surveying may have increased the likelihood of under-reporting, in particular of socially undesirable events such as child abuse. Computer assisted interviewing may increase reporting of stigmatized events or behaviours in some cases [
77]. However, it may not be suitable for all settings, such as the very rural ones in which parts of this study were conducted and where participants may be intimidated by the opportunity to use a computer to answer questions [
78,
79]. The advantage of the system used in this study is that it allowed for more detailed answers and a very good interviewer-participant relationship, which facilitated follow-up. Future work is needed to examine other potential factors, such as parental risk factors of mental health and substance abuse [
12,
80,
81] and predictors of multiple abuse victimisation.