Burden of mortality linked to community-nominated priorities in rural South Africa

Background: Community knowledge is a critical input for relevant health programmes and strategies. How community perceptions of risk reflect the burden of mortality is poorly understood. Objective: To determine the burden of mortality reflecting community-nominated health risk factors in rural South Africa, where a complex health transition is underway. Methods: Three discussion groups (total 48 participants) representing a cross-section of the community nominated health priorities through a Participatory Action Research process. A secondary analysis of Verbal Autopsy (VA) data was performed for deaths in the same community from 1993 to 2015 (n = 14,430). Using population attributable fractions (PAFs) extracted from Global Burden of Disease data for South Africa, deaths were categorised as ‘attributable at least in part’ to community-nominated risk factors if the PAF of the risk factor to the cause of death was >0. We also calculated ‘reducible mortality fractions’ (RMFs), defined as the proportions of each and all community-nominated risk factor(s) relative to all possible risk factors for deaths in the population  . Results: Three risk factors were nominated as the most important health concerns locally: alcohol abuse, drug abuse, and lack of safe water. Of all causes of deaths 1993–2015, over 77% (n = 11,143) were attributable at least in part to at least one community-nominated risk factor. Causes of attributable deaths, at least in part, to alcohol abuse were most common (52.6%, n = 7,591), followed by drug abuse (29.3%, n = 4,223), and lack of safe water (11.4%, n = 1,652). In terms of the RMF, alcohol use contributed the largest percentage of all possible risk factors leading to death (13.6%), then lack of safe water (7.0%), and drug abuse (1.3%)     . Conclusion: A substantial proportion of deaths are linked to community-nominated risk factors. Community knowledge is a critical input to understand local health risks. The study was located in South Africa, as part of the VAPAR (Verbal Autopsy with Participatory Action Research, www.vapar.org) programme in which community stakeholders participate in identifying and collectively addressing health challenges in cooperative learning partnerships with the authorities [15]. South Africa is an upper-middle-income country with 66 years of life expectancy in 2019 [16]. There is substantial, entrenched inequality in South Africa in terms of socioeconomic and health status, and access to health services, resulting in a deeply uneven distribution of ill-health and diseases [17]. The study was progressed within the Agincourt Health and Socio-Demographic Surveillance System (HDSS), located in Mpumalanga province, close to the Mozambican border in northeast South Africa [18]. The province is relatively poor and rural, with high unemployment, limited water, sanitation, and electricity services [18,19]. Reflecting on the national situation, the disease burden in the study area is a combination of non-communicable diseases (NCDs), infectious diseases, maternal and child-related disorders with considerably high rate of road accidents and external causes of deaths [20–23], often described as a ‘quadruple burden’ of disease [24]. Age-adjusted HIV prevalence is considerably high in the study area: 19% among men and 26% among women [25]. Community-nominated health risk factors were determined by three community discussion groups (total 48 participants) representing rural villages across the Agincourt HDSS. We progressed a Participatory Action Research (PAR) process to identify and address local health concerns. PAR transforms the roles of passive research subjects into active co-researchers and changes agents through collective analysis, taking, and evaluating action and learning from action [26]. We re-engaged participants involved in earlier research across three villages in the Agincourt HDSS [27]. Villages were selected to vary by distance to health facilities and levels of child-headed households, and participants represented a cross-section of the community (traditional healers, community and religious leaders, community health volunteers and family members). In each village, we held an introductory workshop in which participants nominated a range of issues, collectively validating and prioritising them using ranking and voting. Participants also directed expansion of the participant base to include perspectives that may otherwise be excluded. Each village nominated the highest priority risk factor, hereafter considered as community-nominated risk factor(s). After the nomination, new participants were recruited and worked together, through a series of workshops, sharing, and systematising experiences to build consensus on the problem’s identified, and locally acceptable actions to address them. A total of 16 workshops were held in the common local language xiTsonga. Throughout, participants were supported to assume ownership and control of the process. These elements are described elsewhere [28–31]. Longitudinal VA data from 1993 to 2015, for which period the data was available for this analysis, were used to ascertain the probable cause of death of individuals living in the HDSS based upon results from the InterVA-5 algorithm. InterVA-5 assigns each death to up to three cause(s) and the likelihood of that cause [14]. In this analysis, we used the first and most probable cause of death and excluded all causes of death, where the most probable likelihood was <50%. Using population attributable fractions (PAF) for South Africa extracted from Global Burden of Disease (GBD) data version 2019 [32], we estimated whether each VA classified cause of death was ‘attributable at least in part’ to the community-nominated risk factors. We also calculated the ‘reducible mortality fraction’ (RMF), defined as the relative proportions of the community-nominated risk factors contributing to all possible risk factors contributing to deaths. The GBD extraction was performed between 18.1.2021 and 21.1.2021. PAF is an estimate of the reduction in population mortality if an exposure to a certain risk factor is minimised, as previously described by the WHO [33]. GBD provides PAFs for multiple risk factors for each disease and enables the contribution of individual risk factors to be estimated for any disease. The GBD risk factor hierarchy is based on four levels [34]. Level 1 risk factors are environmental and occupational, behavioural, or metabolic; Level 2 consists of 20 risks or clusters of risks, such as air pollution, tobacco, alcohol use, dietary risks, hearing, vision, and intellectual disabilities; Level 3 contains 52 risks or clusters of risks; and Level 4 includes 69 specific risk factors. The causality between these risk-outcome estimates have been previously established [34]. Sex (binary), age at death (continuous), cause of death (CoD) and year of death (continuous) were obtained from the VA data. The main outcomes were (1) the burden of mortality that was attributable, at least in part, in Agincourt to each community-nominated risk factor and (2) the total RMF due to each community-nominated risk factor. Diseases attributable at least in part to community-nominated risk factors were identified from the GBD data tool, and PAFs from South Africa for all diseases between 1993 and 2015 were extracted covering the same period that VA data were available for this analysis. PAFs were extracted for all ages and both sexes. From these PAFs across all years, the median was used. Diseases for which PAFs were extracted were mapped onto corresponding VA cause of death codes (supplementary material 1). The mapping was based on the WHO Verbal Autopsy Manual’s classification of VA causes of deaths and their corresponding ICD-10 codes. These steps are illustrated in Figure 1. Flowchart of the study phases. (PAR = participatory action research, GBD = global burden of disease, PAF = population attributable fraction, VA = verbal autopsy). The total number (and %) of deaths occurring between 1993 and 2015 that were attributable at least in part to each community-nominated risk factor and the number of risk factors that each death was attributable at least in part to were described. Death was categorised as attributable at least in part to a community-nominated risk factor(s) if PAF for the cause of death was >0. If a death was assigned as Indeterminate, the PAF was assumed to be 0. To quantify the RMF: the relative proportion of each and all community-nominated risk factor(s) to all possible risk factors contributing to all deaths between 1993 and 2015, we first summed the PAFs of all possible risk factors to every death in the population (Figure 2). Second, we summed the PAFs of each and all community-nominated risk factor(s) to every death. Third, we divided the summed PAFs of each and all community-nominated risk factor(s) with the summed PAFs of all possible risk factors (example calculation is contained in Supplementary material 3). It is possible for PAFs to add up to >100% for any death [35]. However, our aim was not to produce absolute numbers, but rather to ascertain the relative contributions of community-nominated risk factors to deaths in the population. The formula for ‘reducible mortality fraction’: the relative proportion of the PAFs of all risk factors that were due to each and all community-nominated risk factor(s) (example calculation is contained in supplementary material 3). (PAF1 = the PAFs for a community-nominated risk factor for each cause of death multiplied by the number deaths due to that cause, PAF2 = the PAFs for all risk factors for each cause of death multiplied by the number deaths due to that cause, n = 14,430 = the total number of deaths in the population between 1993 and 2015). Results are presented for all community-nominated risk factors and each separately for the whole population and then disaggregated by sex, age category, and mortality category. All deaths and deaths from each risk factor were described according to seven age groups (neonatal (<28 days), infant (28 days–12 months), 1–5 years, 5–15 years, adult (15–50 years), mid-ager (50–65 years) and >65 years), five cause categories as categorised by VA (infectious and parasitic diseases; non-communicable diseases; pregnancy, childbirth, and puerperium-related disorders; neonatal and external causes of death, indeterminate), and over time. Categorical data were described as n (%); continuous data were described as mean (SD) where normally distributed or median (IQR) where not. The analysis was done using SPSS version 25. The research was a secondary analysis of VA data from Agincourt HDSS, which has been previously approved by the Committee for Research on Human Subjects at the University of Witwatersrand (Nos. M960720 & M110138). Consent (informed consent at individual and household level as well as community consent from traditional leaders) was secured at the start of surveillance in 1992 and is reaffirmed regularly. The principle of informed consent and right to refusal or withdrawal was fully respected.