**Background:** Reducing preventable child mortality remains a global priority, with most under-5 deaths concentrated in sub-Saharan Africa. Household structure and composition are known to influence child mortality, but limited longitudinal evidence exists from HIV-endemic areas, particularly regarding how these associations may change with the introduction and widespread availability of antiretroviral therapy (ART). This study aimed to investigate the relationship between a child's risk of dying and their household's structure and composition using comparative, longitudinal data from two demographic surveillance sites (DSS) in rural South Africa from 2000 to 2015, covering periods before and after ART became widely available.
**Methods:** The study used household census data from two DSSs: the Agincourt Health and socio-Demographic Surveillance System (AHDSS) in Mpumalanga (primary ethnic group amaShangaan, ART available from 2008) and the Africa Health Research Institute (AHRI) in KwaZulu-Natal (primarily Zulu-speaking, ART available from 2004). Data were organized as person-months for 101,105 children under 5 years (0–59 months), with 3,603 deaths recorded between 2000 and 2015. Household structure was categorized into seven types: (1) both parents and no other kin (nuclear); (2) one or both parents and grandmother (vertical); (3) one or both parents and aunts/uncles (horizontal); (4) one or both parents, grandmother and aunts/uncles (vertical and horizontal); (5) no parents but any kin present; (6) mother only, no kin present; and (7) other (e.g., lone father). Kin presence indicators included grandmother, counts of aunts, uncles, and older siblings (aged 5+), and parental co-residence (both, one, or no parents). Multilevel discrete time event history analysis with relative risk regression models was used, including controls for child sex, age, multiple birth, mother's age at birth, household size, site, and time period. InterVA-5 was used to assign causes of death from verbal autopsy data. Sensitivity analyses included household socioeconomic status (SES) and household head gender.
**Key Results:** Mortality risks differed significantly by co-residence patterns. Children in nuclear households with both parents had the lowest risk of dying compared with all other household types. For both DSSs, the period of ART availability lowered the probability of dying across all household types. At AHDSS, children in mother-only and 'other' household types had elevated mortality risk even in the ART period. The kin presence model revealed that associations with kin were moderated by parental status. Having older siblings lowered the probability of dying only for children in a household with both parents (relative risk ratio [RRR] = 0.736, 95% CI 0.633 to 0.855). Having other adult kin present in two-parent households resulted in a higher probability of dying. Only in the later ART period was there evidence that older adult kin lowered the probability of dying for children in single-parent households (RRR = 0.753, 95% CI 0.664 to 0.853). The distribution of household structures differed between sites: at AHDSS, about 40% of households were nuclear compared with a declining proportion at AHRI (33% in 2000 to 15% in 2015). Living with one parent was increasingly common at both sites but more so at AHRI (58% in 2000 to 78% in 2015). Living with both parents remained relatively stable at AHDSS (44% in 2000 to 37% in 2015) but declined at AHRI (34% in 2000 to 15% in 2015). The share of deaths due to HIV/AIDS and tuberculosis declined in 2008–2015 compared with 2000–2007 at both sites, while respiratory infection deaths increased. Sensitivity analyses including household SES and household head gender confirmed the main findings remained robust.
**Clinical Implications:** This study provides comparative, longitudinal evidence that household structure and kin composition are important determinants of child mortality in HIV-endemic rural South Africa. The protective effect of both parents highlights the importance of supporting two-parent households, while the finding that older adult kin only became protective for single-parent households in the ART era suggests that widespread ART availability may enable kin to provide caretaking roles that were previously compromised by HIV-related illness and mortality. The heterogeneity between sites underscores the need for context-specific interventions. Formative research is needed to understand the role of other household kin in promoting child well-being, particularly in one-parent households that are increasingly prevalent. These findings have implications for policies and programs aimed at reducing preventable child mortality in settings undergoing rapid demographic, epidemiological, and social transitions.