HIV, hypertension and diabetes care and all-cause mortality in rural South Africa in the HIV antiretroviral therapy era: a longitudinal cohort study
•,,,,,,,,,,.
...
Abstract
Introduction South Africa is in the midst of rapid epidemiological transition from extremely high HIV and tuberculosis (TB) mortality to one characterised increasingly non-communicable disease-related deaths. However, longitudinal data linking modifiable risk factors and disease care indices to mortality in the country are extremely rare, and a prerequisite to appropriately prioritise health system responses.
Methods Individuals in the Africa Health Research Institute Southern Demographic Health Surveillance area were invited to health fairs to collect data on modifiable risk factors and HIV, TB, diabetes and hypertension disease status and control. Individuals are then followed longitudinally through routine surveillance to detect deaths. We fit Cox proportional hazards models and estimated population-attributable fractions (PAFs) to identify modifiable risk factors and disease control indicators associated with all-cause mortality.
Results A total of 18 041 individuals completed health screening and were followed for 114 692 person-years. Men had higher mortality rates than women across all age bands. The median follow-up time was 3.3 years (IQR: 3.0–3.5 years). For men, communicable diseases accounted for a higher PAF of mortality (PAF=13.7% for HIV and 8.3% for TB) than non-communicable diseases (6.6% for hypertension and 1.9% for diabetes). By contrast, despite extremely high HIV prevalence, non-communicable diseases with comorbid communicable disease accounted for the greatest share of deaths. In both sexes, having a chronic disease with poor control was most predictive of all-cause mortality. For example, among men, compared with those without each condition, adjusted HRs of all-cause mortality for people with uncontrolled disease were 3.47 (95% CI 2.10 to 5.72) for HIV, 1.52 (95% CI 1.05 to 2.20) for hypertension and 2.34 (95% CI 1.75 to 6.79) for diabetes. Among women, these same ratios were 5.32 (95% CI 3.54 to 7.99) for HIV, 1.73 (95% CI 1.31 to 2.28) for hypertension and 3.11 (95% CI 2.02 to 4.77) for diabetes.
Discussion Poor control of chronic, treatable diseases predicts all-cause mortality in rural South Africa in the HIV antiretroviral therapy era. Health system strengthening to improve chronic disease and multimorbidity care should be prioritised.
What is already known on this topic
Communicable diseases have been responsible for the lion share of deaths in South Africa over the past two decades. However, since approximately 2015, there has been a shift in the leading causes of death in the country, particularly non-communicable diseases, such as diabetes, stroke and heart disease, that are now responsible for greater than 50% of deaths nationwide. However, data on modifiable risk factors for mortality, and particularly how the most prevalent communicable (ie, HIV) and non-communicable diseases (ie, hypertension and diabetes) and their care contribute to all-cause mortality in South Africa, are unknown.
What this study adds
Poor control of diagnosed chronic diseases such as HIV, hypertension and diabetes is strongly predictive of all-cause mortality in rural South Africa. These conditions, alone and in combination, now appear to be attributable to a majority of deaths in the region.
How this study might affect research, practice or policy
Health system strengthening to improve chronic disease control is urgently needed in South Africa. In particular, interventions and programmes that integrate care and/or address the multimorbidity epidemic will be needed to emerge beyond the successful, but siloed HIV care programme.
Introduction
Following the national rollout of antiretroviral therapy (ART) programmes, South Africa has observed a rapid decline in HIV and tuberculosis (TB) mortality rates.1 2 According to death reporting data, non-communicable diseases (NCDs) are now estimated to be responsible for over 50% of deaths in the country, with most occurring before the age of 65 years.3–5
In the midst of this shifting disease burden, there is a growing need to understand the chief risk factors for morbidity and mortality in the current era. For example, large longitudinal cohort studies from high-income countries, such as the Framingham Heart Study and the Multi-Ethnic Study of Atherosclerosis, helped identify how smoking and hypertension have become the predominant reversible risk factors for death in the USA.6 Modelling methods have been used to demonstrate the increasing contributions of NCDs to morbidity and mortality in sub-Saharan Africa (SSA).7–9 However, these exercises rely on assumptions about the transportability of risk prediction data from areas with rich data, such as the global North, to areas where such relationships are much less established.10 11 In SSA, prospective studies that include risk factor data and prospectively observed mortality remain extremely scarce. Such data are urgently needed to better determine which conditions and behaviours are most predictive of mortality in the current ART era, and ultimately improve health system planning and optimise efficiency in the context of constrained health resources.12
We leveraged two research platforms—Vukuzazi, a large population-based study that collected comprehensive risk factor data, and the Africa Health Research Institute (AHRI) Health and Demographic Surveillance System (HDSS) which collects death data on this same population13–15—to identify risk factors for all-cause mortality among adolescents and adults in rural South Africa. We hypothesised that a combination of communicable and cardiovascular risk factors would be associated with all-cause mortality in the midst of the converging HIV, TB and NCD epidemics in the region.
Methods
Study population and data collection
Since 2000, AHRI (https://www.ahri.org/) has operated an HDSS platform in a population of approximately 140 000 individuals (approximately 20 000 households) in rural KwaZulu-Natal. Households are visited thrice annually to collect data on births, deaths, migrations and household asset ownership. Full details of the AHRI HDSS cohort have been described previously.14
From May 2018 through March 2020, AHRI conducted a population-wide health assessment among individuals living in the HDSS area (the Vukuzazi Study). Full details and results of the study have been published previously.13 15 In brief, all residents 15 years and older within the HDSS area were visited at home and invited to health screening. At the screening, individuals completed a questionnaire on medical history and current engagement in care, as well as questions on alcohol intake, smoking behaviour and employment. Participants had anthropometric measurements and blood pressure measured according to the WHO Steps protocol.16 Blood was collected for HIV testing by ELISA, with reflex CD4 count and HIV-1 RNA viral load for those testing positive, as well as diabetes screening by haemoglobin A1c. Finally, we conducted TB symptoms screening, chest X-ray and sputum collection, among anyone with a positive symptom screen or abnormal chest X-ray, for Gene Xpert testing and mycobacterial culture.
Study definitions
Our primary outcome of interest was all-cause mortality, which is captured through thrice annual household visits conducted as part of the AHRI demographic surveillance. During each visit, all deaths since the previous visit are recorded, including those of non-resident household members. All deaths are verified by a home-based follow-up verbal autopsy interview. We used surveillance data from May 2018 (ie, the start of the Vukuzazi Study) through October 2022, when this analysis was conducted, to ascertain mortality.
Our primary exposures of interest included the following: (1) smoking behaviour, categorised as never, former and current smoking; (2) alcohol intake, categorised as never, former drinking but not in the past 12 months and drinking within the past 12 months; (3) body mass index (BMI), categorised as <18.5, 18.5–25, 25–29.9 and ≥30 kg/m2; (4) hypertension disease status, categorised as (a) no hypertension, (b) hypertension not in care as defined by an elevated blood pressure (>140 mm Hg systolic or >90 mm Hg diastolic) with no clinic engagement for the past 6 months, (c) hypertension in care but not controlled as defined by elevated blood pressure and reporting clinic engagement in the past 6 months, or (d) controlled hypertension as defined by a blood pressure <140 mm Hg systolic and <90 mm Hg diastolic and on antihypertensive treatment; (5) diabetes disease status, categorised as (a) no diabetes, (b) diabetes not in care as defined by glucose intolerance (haemoglobin A1c >6.5) with no clinic engagement for the past 6 months, (c) diabetes in care but not controlled as defined by glucose intolerance and reporting clinic engagement in the past 6 months, or (d) controlled diabetes (A1c <6.5) and on diabetes treatment; (6) HIV disease status, categorised as (a) HIV negative (negative HIV ELISA), (b) HIV ELISA positive and undiagnosed if they are unaware of their diagnosis, (c) HIV ELISA positive and uncontrolled if they are aware of their disease but with a viral load (VL) >40 copies/mL, and (d) controlled if they are HIV ELISA positive with a VL <40 copies/mL; (7) TB status, categorised as (a) no history of TB, (b) prior TB in people reporting prior TB or with a chest X-ray suggestive of prior TB, or (c) current TB in those with a positive TB culture or Gene Xpert test.
Statistical analyses
We first summarised the characteristics of the cohort using frequencies and proportions. We then estimated the crude incidence of all-cause mortality with person-time defined from the date the individual underwent a health screen in the Vukuzazi Study, until the earliest date of death, date of migration out of the AHRI demographic surveillance area or date last seen in the surveillance area. Mortality estimates were made both for the entire cohort, then for each sociodemographic, anthropometric and clinical disease stage.
We next used Cox proportional hazards regression to estimate relative hazards of mortality for each behavioural and clinical exposure of interest. All models were adjusted for age, as the analysis timescale of the Cox model. Because we expected risk factors for mortality to differ between men and women, all analyses were a priori stratified by sex. In the initial model, we adjusted for age only. In a second set of models, we additionally adjusted for economic activity (employed, unemployed or not in the labour force) and household socioeconomic status (SES). Household SES was constructed using principal component analysis based on ownership of household assets and characteristics such as access to piped water, type of toilet, electricity and type of cooking fuel.17 To provide population-wide estimates, all analyses were weighted to account for non-participation in the Vukuzazi Study, with weights calculated as the inverse probability of study participation in strata defined by age group and sex. Lastly, to ascertain relative contributions of disease states to all-cause mortality, we fit a Cox proportional hazards model to each individual condition and combinations of those conditions, adjusted for age, employment, SES and stratified by sex. From these models, we used the estimated HRs and proportion of person-years to calculate the population-attributable fraction (PAF) as follows:
Where p is the proportion of person-years for a particular condition or combination of conditions, and HR is the HR for that condition adjusted for employment and SES and stratified by sex. All analyses were done in STATA V.17.0 (Stata Corp).
Patient and public involvement
This analysis was done without patient involvement. Patients were not invited to comment on the study design, develop patient-relevant outcomes, or interpret the results, writing or editing of this document for readability or accuracy. Dissemination of study results will be made to participants and the HDSS community through road shows lead by the AHRI Community Engagement Unit.
Results
A total of 36 097 individuals were eligible for enrolment in the parent Vukuzazi Health Fair Study, of whom approximately 75% were contacted and accepted the invitation to visit the mobile camp. Among those who accepted the invitation (n=26 795), approximately one-quarter did not come, resulting in a total of 18 041 individuals who enrolled in the Vukuzazi Programme. Characteristics of those who participated and those who did not are presented in online supplemental table 1. The analytical cohort consisted of 18 041 participants (table 1).
Table 1
|
Cohort characteristics, n=18 041
The median age was 37 years (IQR 23–56 years) and most participants were female (n=12 229, 68%). Approximately half of participants were not in the labour force (n=8042, 45%), with 22% (n=3888) unemployed. Approximately one in five men (20%, n=1177) and very few women (1%, n=124) reported current smoking, while 24% (n=1418) of men and 4% (n=548) of women reported drinking in the past 12 months.
Individuals were observed for a median of 3.3 years (IQR 3.0–3.5 years) from the time of their participation in the Vukuzazi Health Fair until death or data extraction in October 2022, comprising a total of 114 692 person-years of observation. The crude all-cause mortality rate was 11.2 per 1000 person-years (95% CI 10.4 to 12.1). Crude mortality was higher among men than women overall (12.4, 95% CI 11.0 to 14.0 vs 10.4, 95% CI 9.5 to 11.4, p<0.001) and across all age strata (table 2).
Table 2
|
All-cause mortality rate by sex and age group
In unadjusted models, poor disease control was associated with increased all-cause mortality among both men and women, and for both communicable diseases and NCDs (figure 1). In models restricted to men and adjusted for age, employment and household SES, the presence of both communicable diseases (ie, HIV) and NCDs (ie, diabetes and hypertension) was persistently associated with greater all-cause mortality (table 3 and figure 2). For example, we found an increased hazard of mortality for men at all stages of the HIV cascade of care compared with the HIV uninfected, but having uncontrolled HIV disease (adjusted HR (aHR)=3.47, 95% CI 2.10 to 5.72) carried much higher mortality than those who were virologically suppressed (aHR 1.37; 95% CI 0.98 to 1.91) or those who had not yet been linked to HIV care (aHR 2.38, 95% CI 1.23 to 2.59). Similarly, for NCDs, uncontrolled disease was most strongly associated with increased risk of all-cause mortality. For example, men in care but with uncontrolled diabetes had over three times the hazard of mortality compared with those without diabetes (aHR 3.45, 95% CI 1.75 to 6.79), whereas men with uncontrolled hypertension had a 50% increased risk of all-cause mortality compared with those without high blood pressure (aHR 1.52, 95% CI 1.05 to 2.20). There was an increased hazard of all-cause mortality in men who were underweight (BMI <18.5 kg/m2) compared with those with normal BMI (18.5–24.9 kg/m2; aHR 2.22, 95% CI 1.47 to 3.36). There was no evidence that mortality differed between men with normal versus overweight BMI. Although mortality among obese men was lower than those with normal BMI, there was no evidence of a significant difference. Neither self-reported smoking, self-reported tobacco use, nor TB disease stage was associated with mortality among men.
Kaplan-Meier survival plots demonstrating crude survival for men (top row) and women (bottom row) by disease status and disease control for hypertension (A and E), HIV (B and F), diabetes (C and G) and tuberculosis (TB) (D and H).
Table 3
|
Cox proportional hazards models to estimate the association between sociodemographics, communicable and non-communicable diseases with all-cause mortality
HR of mortality across the spectrum of disease stage and control for obesity, hypertension (HTN), diabetes mellitus (DM), HIV and tuberculosis (TB) in rural KwaZulu-Natal, South Africa. Plots for each section represent a separate model adjusted for age, economic activity and relative wealth index. VL, viral load.
Similarly, among women, there was increased all-cause mortality for both presence and poor control of communicable diseases and NCDs (table 3 and figure 1). For example, women with undiagnosed HIV (aHR 1.91, 95% CI 0.96 to 3.79) and controlled HIV (aHR 1.40, 95% CI 1.09 to 1.80) had increased hazard of mortality, but the greatest risk of death was found in those who were in HIV care but had a detectable VL (aHR 5.32, 95% CI 3.54 to 7.99). These relationships were similar for the NCD conditions, and generally more pronounced than what was seen in men, with a significantly increased risk of mortality at all stages of the diabetes and hypertension cascades of care, compared with women without these NCD conditions. Participants who are female, with either prior (aHR 1.35, 95% CI 1.07 to 1.69) and active TB (aHR 2.90, 95% CI 1.61 to 5.20), had significantly increased risk of all-cause mortality. Like men, women had an increased hazard of mortality with a low BMI (<18.5 kg/m2), but they also had reduced risk of all-cause mortality with both overweight (BMI 25–30 kg/m2, aHR 0.72, 95% CI 0.55 to 0.96) and obesity (ie, BMI >30 kg/m2, aHR 0.76, 95% CI 0.60 to 0.96), compared with those with a BMI 18.5–25 kg/m2. Finally, although few women reported alcohol use, intake in the last 12 months (aHR 1.74, 95% CI 1.09 to 2.76)and prior to that (aHR 2.39, 95% CI 1.06-5.38) was associated with an increased risk of death.
PAF estimates reinforced that NCDs and communicable diseases, either alone or in combination, increase the risk of mortality in rural South Africa (table 4 and figure 3). Among both sexes, HIV and TB remain predominant risk factors of mortality.
Table 4
|
Population-attributable fraction (PAF) for communicable and non-communicable diseases alone and in combination
Population-attributable fraction contributions for communicable (CDs) and non-communicable diseases (NCDs) to all-cause mortality in rural KwaZulu-Natal by the presence of a single condition or combination of conditions. DM, diabetes mellitus; HTN, hypertension; TB, tuberculosis.
Discussion
In rural South Africa, after decades of an outsize burden of HIV and TB as primary contributors to mortality, we observed a mix of both communicable diseases and NCDs as predictors of all-cause mortality. Whereas HIV, and particularly HIV with detectable viraemia among those in care, remains a major risk factor for death, we found that in the ART era, people with uncontrolled diabetes and hypertension also carry a substantially increased risk of death. For example, men and women who are in care for diabetes but have poor disease control had greater than three and five times the hazard of death, respectively, compared with those without diabetes. Although not as pronounced, we also found 50–70% increased risk of all-cause mortality in men and women in care with uncontrolled hypertension. These data reinforce the importance of rapidly and efficiently diversifying the healthcare system in South Africa, beyond the well-established HIV and TB care programmes, to also include the rapidly emerging epidemic of NCDs that are contributing to mortality. A particular area of need is strengthening chronic disease care programmes for those with diagnosed but uncontrolled hypertension and diabetes.
Few studies have prospectively explored addressable risk factors for all-cause mortality in rural South Africa in the ART era. Those that have done so also found contributions of both NCDs and communicable diseases to death. For example, a recent study of approximately 4000 older adults in Mpumalanga province found an increased risk of all-cause mortality among individuals with multimorbid NCD conditions,18 and that those with both HIV and NCDs had higher risk of mortality than those with NCD multimorbidity alone. This increasing burden of NCDs as contributors to mortality is also evident in South African cause of death reporting. Whereas communicable diseases accounted for approximately 50% of causes of deaths in the country through the end of 2009, NCDs are now the leading reported cause of death in the country and account for approximately 59% of deaths as of 2018, compared with 29% of deaths due to communicable diseases as of 2018, the last year for which data are publicly available.19
Data on the increasing burden of NCDs on mortality in South Africa are somewhat contrasting to conclusions drawn by modelling studies, which continue to show communicable diseases as the leading causes of morbidity in South Africa.20 21 There are putative reasons for the discrepancies between these data sources. First, modelling studies often focus on outcomes that account for duration and quality of life deficits, such as disability-adjusted life years (DALYs), which will be expected to give more weight to conditions such as HIV that generally cause morbidity and mortality earlier in the life span. By contrast, our study and cause of death data are focused more strictly on mortality, for which NCDs and communicable diseases appear to be converging in terms of risk. Second, modelling studies such as the Global Burden of Disease (GBD) Study implement modelled estimates for death and disability based on risk factor estimates. Such models necessarily rely on assumptions about mortality that are derived from alternate populations, due to the scarcity of data that link risk factor data with mortality. For example, the GBD South Africa Study ranks high BMI as the third leading cause of age-standardised, risk-attributable DALY lost. However, in both our current study and prior work, we have demonstrated that traditionally defined overweight (ie, BMI 25–30 kg/m2) and obesity (BMI 30–35 kg/m2) are associated with a lower risk of age-adjusted all-cause mortality.22 Ultimately, to better elucidate the contributions of health conditions and other risk factors on morbidity, disability and mortality, studies like ours that link risk factor data to morbidity and mortality, but more broadly reflect general populations, will be needed.
The burgeoning data in this area highlight the need for an urgent redesign of the health system to increase resources and infrastructure for NCD prevention and management in the region. Our PAF data suggest that NCDs are now a major risk factor for mortality in those with and without HIV, such that NCD prevention and control programmes will have important health effects in both populations. Whereas the need to expand NCD control infrastructure in South Africa is recognised by the public health community,23 and official strategic plans dating back to 2013,24 empirical evidence on how best to integrate NCD care is sparse.25 Many such programmes have failed to improve process measures and metabolic outcomes in practice, and those with preliminary promise have yet to report sustained improvements in downstream outcomes at scale, such as cardiovascular events or mortality.25–28 A specific challenge is that, despite the rapidly increasing burden of NCDs29 and estimate needs for billions of US dollars in annual funding,30 there has been a relative lack of donor funding for NCDs, which remains less than 1% of global funding into health.31
Finally, we also demonstrate a near doubling in all-cause mortality among men compared with women across the life span. These results are in accordance with longstanding effects of reduced healthcare access, poorer maintenance in care and worse outcomes for men in the region, which have been particularly evident in the HIV and TB care systems.32–34 As health systems in the region increasingly undertake approaches to better manage chronic disease care for NCDs, particularly attention to improving engagement and retention of men in the healthcare system should be a parallel aim.35
This study should be interpreted in the context of several limitations and considerations. First, the study data collection period included multiple waves of the COVID-19 pandemic in South Africa. As such, reported mortality rates should be interpreted in light of reported increases in mortality during this period.36 In particular, relative hazards of mortality due to chronic, uncontrolled infections might have been higher than expected in the absence of the pandemic.37 38 Second, although our analysis is strengthened by the pairing of risk factor data with prospectively captured mortality data, we cannot definitively assign causality to our estimates of mortality. Our study was also imbalanced by sex. We have used the presence of a population census as our sampling frame to mitigate this effect with an inverse probability of sampling weighting. Nonetheless, this method does not account for bias from non-random missing not accounted for by age and sex weighting. Finally, our estimates should be taken to indicate relatively short-term mortality risks with a median of 3.3 years of observation per individual. We are persistently updating our mortality data and plan to analyse risk factors for 5-year and 10-year mortality once those data have become available.
Conclusion
In summary, in a population-based longitudinal cohort study in rural South Africa in which communicable disease and NCD control data were paired with prospective all-cause mortality data, we found poor disease control to be the factor most strongly associated with mortality. These data reinforce the urgency to strengthen health systems for chronic disease care management in the region. Ultimately, there is a particular need for interventions which improve HIV, hypertension and diabetes disease control among those diagnosed but with suboptimal indices of care.
