Overall and cause-specific mortality and associated risk factors among middle-aged and older South Africans: findings from the health and ageing in Africa: a longitudinal study of an INDEPTH community in rural South Africa (HAALSI)

Abstract

Background

Assessing mortality levels, causes of death and associated risk factors is critical for strengthening health systems and guiding targeted health interventions. We examined overall and cause-specific mortality and their associated risk factors among middle-aged and older adults in a rural, rapidly transitioning South African setting.

Methods

Data were drawn from the Health and Ageing in Africa: A Longitudinal Study of an INDEPTH Community (HAALSI) cohort. Risk factors for mortality from 2014/2015 to 2021 were analysed using Cox-Proportional Hazards and Fine and Gray competing-risk models.

Results

The 5059 adults aged 40 years and older at enrolment in the HAALSI cohort in 2014/2015 experienced 1116 (22.1%) deaths over 28 955 person-years of follow-up by 2021. Overall, mortality was higher in men (48.3 deaths per 1000 person-years) than women (30.7 deaths per 1000 person-years). Risk of death increased with age and was significantly associated with male sex, lower education, history of smoking or alcohol use, poor self-rated health, underweight, limitations in activities of daily living, slower walk speed, lower grip strength and histories of hypertension, diabetes, stroke, tuberculosis and unsuppressed Human Immunodeficiency Virus (HIV) infection. Cardiovascular diseases were the leading causes of death (30.3%) in both men (25.9%) and women (35.8%), followed by other infectious diseases (23.8%), neoplasms (14.4%), HIV/AIDS and tuberculosis (10.1%).

Conclusion

These findings highlight the dual mortality burden from non-communicable and infectious diseases among older adults in rural South Africa. Integrated interventions targeting non-communicable diseases, infectious diseases and physical function are urgently needed to reduce premature mortality and improve health outcomes.

Key points

• There is persistent dual mortality burden from non-communicable and infectious diseases.

• Existing need for interventions targeting non-communicable disease prevention, physical function and infectious disease control.

• Cardiovascular diseases were the leading causes of death in both men and women.

Introduction

The risk of mortality in a population is influenced by a complex interaction of a range of personal and contextual factors including age, sex, lifestyle behaviours, living and working conditions, environmental exposure, the quality of the health care system and indicators of socioeconomic status such as education, marital status and wealth [1–10]. These factors either reduce or increase the susceptibility of individuals to various diseases and subsequent mortality independently or jointly. For example, (i) poor sanitation and water supply increase susceptibility of individuals to diarrhoea, cholera and other water and sanitation-related diseases; (ii) cigarette smoking increases the susceptibility of individuals to lung cancer; and (iii) physical inactivity increases individuals’ risk of developing chronic diseases such as heart disease, hypertension, type 2 diabetes, stroke and cancers [4].

Since 2005, South Africa has experienced reductions in overall mortality and an increase in life expectancy at birth due to the reductions in HIV/AIDS-related mortality owing to widespread availability of antiretroviral treatment [11, 12]. As a result, the number and proportion of middle-aged and older adults have increased, bringing a rising prevalence of non-communicable diseases (NCDs) alongside HIV/AIDS [13]. In turn, there has been a corresponding increase in the mortality burden from NCDs alongside the residual burden from HIV/AIDS and tuberculosis (TB), other communicable diseases, nutritional deficiencies, violence and injuries [14–24].

Assessing and characterising mortality levels and cause of death profiles and their risk factors in different settings and time periods are critical for identifying locally relevant elements for strengthening of health and social systems and determining the focus of health services. However, data on mortality patterns and risk factors among middle-aged and older adults in rural sub-Saharan African (SSA) settings remain scarce. This study investigates the salience of demographic, socioeconomic, self-reported and objectively measured health-related factors on overall and cause-specific mortality in a rural South African setting.

Methods

Study design and setting

This prospective cohort study was conducted in the Agincourt sub-district of Bushbuckridge municipality in Mpumalanga Province, South Africa. The study area, characterised by poor public infrastructure and services, lies in the northeastern part of South Africa adjacent to southern Mozambique. It covers 31 contiguous villages spread over 450 km² underpinned by the Agincourt Health and Demographic Surveillance System (HDSS) [25, 26]. The population in the area is largely Tsonga-speaking and about a third of it are former Mozambican refugees and their descendants [25, 26]. High rates of unemployment and labour migration and reliance on income from remittances are some of the key characteristics of the population [27].

Data source

The study utilises data on self-reported and objectively measured health-related factors from the baseline wave of the HAALSI study [28] and mortality and cause of death information from the Agincourt HDSS.

The HAALSI study was established between November 2014 and November 2015 as a population-representative cohort of 5059 individuals (2345 men and 2714 women) aged ≥40 years in the Agincourt HDSS to describe biological, social and economic determinants and consequences of health and ageing in rural South Africa [28]. It collected household data on consumption, income and assets alongside individual-level sociodemographic factors, health and health behaviours, physical and cognitive function, anthropometric measurements and biomarkers using point-of-care testing and dried blood spots. Most measures are harmonised with international ageing studies to allow comparisons of risk factors for health and well-being in rural South Africa with other settings [28].

Mortality and cause of death information was prospectively collected through annual updates of vital status of the whole Agincourt HDSS population. Verbal autopsy (VA) interviews were conducted with caregivers of deceased individuals to elicit signs and symptoms of the illness or injury prior to their death. The interviews were conducted using WHO-VA instruments [29].

Variables and measurements

The socio-demographic variables in the study include age, sex, marital status, level of education, wealth index and religion. Health-related risk factors include smoking status, alcohol consumption, body mass index (BMI), HIV status, antiretroviral treatment (ART) usage, viral suppression status, grip strength, walk time, number of activities of daily living (ADL) limitations, self-rated health and history of TB, hypertension, diabetes, stroke, heart attack, health failure and angina collected in HAALSI baseline wave in 2014/2015 [28]. Information on the history of various diseases were self-reported by the participants by answering if they were ever diagnosed with the disease. We also analysed objective measurements of blood pressure (BP) and blood glucose taken during the HAALSI study. Based on South African guidelines [30], we divided BP measurements into the following categories: hypotension (SBP < 90/DBP < 60 mmHg), normal BP (SBP 90–129 and DBP 60–79 mmHg), elevated BP (SBP 130–139/DBP 80–89 mmHg), grade I hypertension (SBP 140–159/DBP 90–99 mmHg), grade II hypertension (SBP 160–179/DBP 100–109 mmHg) and grade III hypertension (SBP ≥ 180/DBP ≥ 110 mmHg). Diabetes mellitus was defined as a random blood glucose (RBG) ≥ 11.1 mmol/L or fasting blood glucose (FBG) ≥ 7.7 mmol/L or reporting to be taking diabetes medications. BMI, HIV status, viral suppression status, ART usage, grip strength and walk time were also objectively measured. ADL limitations were assessed by asking participants if they had difficulty walking across the room, dressing up, bathing, eating, getting in or out of bed and using the toilet. We categorised the responses into four groups: None; 1; 2–3; and 4+. For self-rated health, participants were asked to rate their health today as very good, good, moderate, bad or very bad.

Assigning causes of death

InterVA-5 probabilistic model (version 5.1) [31] was used to assign probable causes of death. For each death, the InterVA-5 model assigns up to three likely causes of death with associated likelihoods. Consistent with a previously reported approach [32], we assigned the cause of death with the highest likelihood as the most probable cause. An indeterminate cause is assigned when the VA information is insufficient or unavailable. Compared to physician coding, the InterVA-5 model assigns causes in a standardised manner that is faster and more consistent for assessing changes over time and across settings. The causes of death derived from InterVA-5 have been found to not substantially differ from those generated by physician coding [31]. The causes of death from the InterVA-5 model in this HAALSI cohort have also shown 84.5% agreement (kappa = 0.81) with those derived using the Insilico-VA model. See Appendices 1 and 2 for the overall and sex-specific level of agreement between the InterVA-5 and Insilico-VA models.

Statistical analysis

We computed mortality rates using survival analysis techniques for right censored data. We computed exposure time from the date of enrolment in the HAALSI study to the date of death or outmigration, or the latest follow-up date in 2021. Factors associated with all-cause mortality were examined using Cox Proportional Hazard (Cox-PH) regression models with time since HAALSI baseline measurements as the start of follow-up. The Cox-PH model is a semi-parametric model, where the baseline hazard () has no functional form. Hazards are defined as the probability of dying in the next instance , given that one survived until time and the model equation is given by , where is the hazard rate at time , is the baseline hazard, are the covariates and are regression coefficients [33]. Three incremental models were fitted: model 1—unadjusted; model 2—adjusted for age; and model 3—adjusted for age and sex. The effects of the covariates on overall mortality risk are measured using hazard ratios.

For cause-specific mortality analysis, the causes of death generated by the InterVA-5 model are grouped into eight broad categories: (i) HIV/AIDS and TB; (ii) other infectious diseases; (iii) cardiovascular diseases (CVDs); (iv) other NCDs; (v) neoplasms; (vi) metabolic diseases; (vii) external causes; and (viii) indeterminate causes if the cause of death could not be determined. Thereafter, the Fine and Gray competing risk model with sub-distribution hazards is used to assess the effects of the various covariates. The Fine and Gray competing risk model is analogous to the Cox-PH model but accounts for deaths due to other causes when considering deaths attributable to a specific cause [34]. In the Fine and Gray competing risk model, instead of non-informative censoring of other deaths, there is informative censoring of deaths due to other causes [35].

All analyses have been performed using STATA version 17 [36], and the statistical significance level is set at α = 0.05.

Results

Characteristics of study population

We included 5059 participants who completed the baseline wave of the HAALSI study. Most participants were aged 50–59 years [n = 1389 (27.5%)] or 60–69 years [n = 1310 (25.9%)] and slightly more than half [n = 2712 (53.6%)] were women.

Overall mortality

As of December 2021, there were 1116 deaths over 28 955 person-years of follow-up (38.5 deaths per 1000 person-years). Overall, men had a higher mortality rate (MR) (48.3 deaths per 1000 person-years) compared to women (MR: 30.7 deaths per 1000 person-years). As shown in Table 1 and Appendix 3, mortality rates increase with age for both men and women, with survivorship substantially higher for women in each age group.

Table 1.

All-cause mortality rates by characteristics of the HAALSI study participants at baseline.

Variable	Categories	Total (N)	No of deaths,  n (col %)	Person-years (per 1000)	Mortality rate (per 1000 py)	95% CI
Age group (years)	40–49	867	52 (4.7)	5.3	9.9	7.5–13.0
	50–59	1389	183 (16.4)	8.3	22.0	19.0–25.4
	60–69	1310	285 (25.5)	7.7	37.1	33.1–41.7
	70–79	911	291 (26.1)	5.0	58.4	52.0–65.5
	80+	582	305 (27.3)	2.7	113.2	101.2–126.6
Sex	Men	2347	625 (56.0)	12.9	48.3	44.7–52.3
	Women	2712	491 (44.0)	16.0	30.7	28.1–33.5
Education	No formal	2306	649 (58.2)	12.7	51.0	47.2–55.1
	Primary	1716	344 (30.8)	10.0	34.3	30.8–38.1
	Some sec	574	75 (6.7)	3.4	21.9	17.4–27.4
	Sec or more	446	42 (3.8)	2.7	15.7	11.6–21.2
Religion	None	888	289 (25.9)	4.7	61.5	54.8–69.0
	Christian	3781	735 (65.9)	22.1	33.3	31.0–35.8
	Traditional	378	91 (8.2)	2.1	43.0	35.0–52.8
	Islamic and others	8	0 (0.0)	0.1	0.0	–
Marital status	Never married	290	56 (5.0)	1.6	34.9	26.8–45.3
	Separated/divorced	650	133 (11.9)	3.7	36.4	30.7–43.1
	Widowed	1540	423 (37.9)	8.6	49.3	44.9–54.3
	Currently married	2575	503 (45.1)	15.1	33.3	30.5–36.4
Wealth index	Q1 (Lowest)	1046	254 (22.8)	5.8	43.5	38.5–49.2
	Q2	1001	220 (19.7)	5.7	38.7	33.9–44.2
	Q3	991	241 (21.6)	5.6	42.7	37.6–48.5
	Q4	1007	198 (17.7)	5.9	33.8	29.4–38.9
	Q5 (Highest)	1014	203 (18.2)	5.9	34.2	29.8–39.3
Smoking	Yes	1081	303 (27.2)	5.9	51.5	46.1–57.7
	No	3973	813 (72.8)	23.0	35.3	32.9–37.8
Alcohol	Yes	2253	607 (54.4)	12.5	48.5	44.8–52.5
	No	2803	509 (45.6)	16.4	31.0	28.4–33.8
HIV	Negative	3512	807 (72.3)	20.2	39.9	37.3–42.8
	Positive	1048	183 (16.4)	6.1	30.1	26.1–34.8
HIV, ART and viral loada	Negative	3512	807 (72.3)	20.2	39.9	37.3–42.8
	Positive no ART	386	65 (5.8)	2.2	29.0	22.7–37.0
	On ART above threshold	114	31 (2.8)	0.6	49.6	34.9–70.5
	On ART below threshold	548	87 (7.8)	3.2	27.2	22.0–33.5
Ever diagnosed with TB	Yes	446	121 (10.8)	2.4	49.5	41.4–59.2
	No	4606	991 (88.8)	26.5	37.4	35.2–39.8
BMIb	Normal	1719	375 (33.6)	1.3	77.2	63.5–93.8
	Underweight	258	101 (9.1)	9.9	38.0	34.3–42.1
	Overweight	1328	231 (20.7)	7.9	29.4	25.8–33.4
	Obese	1384	236 (21.1)	8.4	28.1	24.8–32.0
Hypertensionc (Self-reported)	Yes	2118	556 (49.9)	12.0	46.5	42.8–50.5
	No	2937	559 (50.1)	17.0	32.9	30.3–35.8
Hypertensiond (Objectively measured)	SBP < 90/DBP < 60	106	62 (5.9)	0.4	138.7	108.2–177.9
	SBP 90–129 & DBP 60–79	1426e	284 (26.8)	8.2	34.4	30.7–38.7
	SBP 130–139/DBP 80–89	1177	200 (18.9)	6.9	29.1	25.3–33.4
	SBP 140–159/DBP 90–99	1233	253 (23.9)	7.2	35.1	31.1–39.7
	SBP 160–179/DBP 100–109	616	152 (14.4)	3.5	43.1	36.8–50.5
	SBP ≥ 180/DBP ≥ 110	337	108 (10.2)	1.9	57.6	47.7–69.6
Diabetesf	Yes	372	118 (10.6)	2.0	57.6	48.1–69.0
	No	4254	896 (80.3)	24.6	36.5	34.1–38.9
Stroke	Yes	149	45 (4.0)	0.8	57.2	42.7–76.6
	No	4907	1070 (95.9)	28.2	38.0	35.8–40.4
Heart attack	Yes	21	6 (0.5)	0.1	55.8	25.1–124.2
	No	5035	1109 (99.4)	28.8	38.5	36.3–40.8
Heart failure	Yes	33	10 (0.9)	0.2	55.1	29.6–102.4
	No	5023	1105 (99.0)	28.8	38.8	36.2–40.8
Angina	Yes	29	29 (2.6)	0.7	43.1	30.0–62.1
	No	1085	1085 (97.2)	28.3	38.4	36.2–40.7
Grip Strength	Q1 (Weakest)	1488	421 (37.7)	8.3	50.9	46.3–56.1
	Q2	1489	278 (24.9)	8.8	31.5	28.0–35.4
	Q3 (Strongest)	1481	217 (19.4)	8.8	24.6	21.6–28.1
Walk time	Q1 (Shortest)	1720	264 (23.7)	10.3	25.7	22.8–29.0
	Q2	1496	264 (23.7)	8.9	29.7	26.4–33.5
	Q3 (Longest)	1466	419 (37.3)	8.1	51.6	46.9–56.7
No. of ADL Limitationsg	None	4603	899 (80.6)	26.9	33.4	31.3–35.7
	1	202	86 (7.7)	1.0	87.5	70.8–108.0
	2–3	155	68 (6.1)	0.7	94.3	74.3–119.6
	4+	99	63 (5.6)	0.4	171.6	134.1–219.7
Self-reported health	Very good	1021	139 (12.5)	6.1	22.7	19.2–26.8
	Good	2421	443 (39.7)	14.2	31.3	28.5–34.3
	Moderate	650	183 (16.4)	3.7	50.2	43.4–58.0
	Bad	862	296 (26.5)	4.6	64.6	57.7–72.4
	Very bad	102	54 (4.8)	0.4	129.9	99.5–169.6

NB: Percentages not summing up to 100% for some of the variables due to missing data.

^aViral load below threshold was defined as <400 copies/ml; viral load above threshold was defined as ≥400 copies/ml

^bNormal BMI: 18.5 to 24.9 kg/m²; underweight: <18.5 kg/m²; overweight: 25 to 29.9 kg/m^b; Obese: ≥30 kg/m²

^cSelf-report of hypertension diagnosis

^dHypertension: systolic blood pressure (SBP) ≥ 140 mm/Hg and/or diastolic blood pressure (DBP) ≥90 mmHg, or on hypertension medications

^e30.1% (n = 429) of those with controlled BP self-reported hypertension diagnosis

^fDiabetes: rapid blood glucose (RBG) ≥ 11.1 mmol/L or fasting blood glucose (FBG) ≥ 7.7 mmol/L or self-reported on diabetes medications

^gADL: activity of daily living, including difficulty walking across the room, difficulty dressing up, difficulty bathing, difficulty eating, difficulty getting in or out of bed and difficulty using the toilet.

Mortality rates were highest among individuals with no formal education (MR: 51.0; 95% CI: 47.2–55.1), no religious affiliation (MR: 61.5; 95% CI: 54.8–69.0), who were widowed (MR: 49.3; 95% CI: 44.9–54.3), with history of smoking (MR: 54.5; 95% CI: 46.1–57.7) or alcohol use (MR: 48.5; 95% CI: 44.8–52.5). Mortality was also higher in HIV-positive individuals on ART with high viral load (MR: 49.6; 95% CI: 34.9–70.5) followed by HIV-negative individuals (MR: 39.9; 95% CI: 37.3–42.8). Mortality rates were higher for participants who self-reported any of the chronic conditions. Mortality rate was the highest among participants with hypotension (SBP <90/DBP <60 mmHg) (MR: 138.7; 95% CI: 108.2–177.9) and increased with increasing blood pressure levels reaching a peak among those with DBP >110 mmHg or SBP >180 mmHg (MR: 57.6; 95% CI: 47.7–69.6). Further results on mortality by hypertension and treatment status are reported in Appendix 4. Mortality rate was highest among participants with weakest grip strength (MR: 50.9; 95% CI: 46.3–56.1) and slowest pace (MR: 51.6; 95% CI: 46.9–56.7), and increased with worsening levels of self-reported health (MR = 129.9 for very bad self-rated health) and with increasing number of ADL limitations (MR = 171.62 for 4+ ADL limitations) (Table 1). Annual mortality rates (Figure 1) show an overall decrease in age-standardised mortality from 2015 to 2019 and a substantial increase in 2020 and 2021, likely due to the added mortality burden associated with the COVID-19 pandemic.

Figure 1.

Age standardised mortality rates by sex from 2015 to 2021.

Risk factors for overall mortality

Factors significantly associated with overall mortality from unadjusted models are presented in Appendix 5.

Age and sex-adjusted factors associated with overall mortality

Figure 2 presents the factors associated with overall mortality adjusted for sex and age groups. Currently married (RR: 0.50; 95% CI: 0.38–0.67; P < .001), divorced (RR: 0.71; 95% CI: 0.52–0.98; P = .040) and widowed (RR: 0.69; 95% CI: 0.51–0.93; P = .017) adults were less likely to die than those who were never married. Compared to no religion, Christianity (RR: 0.69; 95% CI: 0.60–0.80; P < .001) and African tradition (RR: 0.64; 95% CI: 0.50–0.81; P < .001) were associated with lower risk of mortality. Participants who reported ever smoking (RR: 1.27; 95% CI: 1.09–1.48; P = .002) had a 27% higher risk of mortality. The risk of mortality increased with increasing severity of self-reported health. Participants who reported very bad self-rated health had 3.8 times the risk of death than those with very good self-rated health (RR: 3.84; 95% CI: 2.79–5.28; P < .001). Compared to participants with normal BMI, mortality risk was twice as high among those underweight (RR: 1.87; 95% CI: 1.50–2.34; P < .001) and lower among those overweight (RR: 0.83; 95% CI: 0.70–0.98; P = .024). Mortality risk increased with increasing number of ADL limitations. Participants with four or more limitations had four times higher risk of dying compared to those who reported no limitation (95% CI: 3.08–5.16; P < .001). The risk of mortality was lower among participants with the strongest grip strength (RR: 0.51; 95% CI: 0.42–0.61; P < .001). Participants with hypotension (SBP <90/DBP <60) (RR: 2.66; 95% CI: 2.01–3.51; P < .001) and those with grade III hypertension (DBP >110 or SBP >180) (RR: 1.28; 95% CI: 1.03–1.60; P = .028) were associated with a higher risk of mortality compared to those with normal BP. Diabetes increased risk of mortality by 69% (95% CI: 1.44–2.00; P < .001), history of stroke by 35% (95% CI: 1.00–1.81; P = .053) and history of TB by 35% (95% CI: 1.11–1.63; P = .002). HIV-positive participants on ART and with viral load above 400 copies/ml had a two times higher risk of mortality compared to those HIV-negative (RR: 1.99; 95% CI: 1.38–2.87; P < .001).

Figure 2.

Factors associated with overall mortality—sex and age-adjusted. Normal BMI: 18.5 to 24.9 kg/m²; underweight: <18.5 kg/m²; overweight: 25 to 29.9 kg/m²; obese: ≥30 kg/m². ART, antiretroviral therapy; SBP, systolic blood pressure; DBP, diastolic blood pressure.

Cause-specific mortality fractions and risk

Of the 1116 deaths, 1064 (95.3%) had an assigned cause, while 52 (4.7%) were classified as indeterminate. Based on the VA interview dates, 53.8% of the interviews were conducted within 1 and 6 months after death, 33.2% were conducted between 7 and 12 months and 13.0% just after 1 year.

CVDs were the leading cause of death for both men (25.9%, n = 162) and women (35.8%, n = 176), accounting for 30.3% of all deaths (n = 338; MR: 11.7), followed by other infectious diseases [n = 265 (23.8%); MR: 9.2], neoplasms [n = 161 (14.4%); MR: 5.6] and HIV/AIDS/TB [n = 113 (10.1%); MR: 3.90] (Table 2). For both sexes, CVD and metabolic mortality increased with age, while HIV/AIDS/TB mortality declined with age, as shown in Appendix 6. Women had higher HIV/TB cause-specific mortality fraction (CSMF) at ages 40–59 years, slightly higher CVD CSMF across all ages and higher early-age NCD CSMF likely due to obstetric causes (Appendix 7).

Table 2.

Distribution of deaths and mortality risk by broad cause of death categories.

Cause of death categories	All		Men		Women		Rate per 1000 person-years
	n	%	n	%	n	%
HIV/AIDS and TB	113	10.1	69	11.0	44	9.0	3.9
Other infectious diseasesa	265	23.8	150	24.0	115	23.4	9.2
Neoplasmsb	161	14.4	109	17.4	52	10.6	5.6
Metabolicc	45	4.0	24	3.9	21	4.3	1.6
Cardiovasculard	338	30.3	162	25.9	176	35.8	11.7
Other NCDSe	99	8.9	56	9.0	43	8.8	3.4
External causef	43	3.8	30	4.8	13	2.6	1.5
Indeterminateg	52	4.7	25	4.0	27	5.5	1.8
Total	1116	100	625	100	491	100	38.5

^aOther infectious diseases include: sepsis, acute respiratory infections including pneumonia, diarrhoeal diseases, malaria, meningitis and encephalitis.

^bNeoplasms include: oral, digestive, respiratory, breast, reproductive and other unspecified neoplasms.

^cMetabolic causes include: severe malnutrition and diabetes mellitus.

^dCardiovascular causes include: acute cardiac disease, stroke and other unspecified cardiac diseases.

^eOther NCDs include: asthma, COPD, liver cirrhosis, renal failure, epilepsy, obstetric non-infectious causes including abortion-related death.

^fExternal causes include: road traffic accidents, accidental falls, accidental drowning and submersion, accidental exposure to fire and flame, accidental poisoning and assault.

^gIndeterminate: the cause of death could not be determined.

Competing cause-specific mortality risks by age groups and sex

Stratified by sex, men had higher competing mortality risk from HIV/AIDS, other NCDs and external causes, while women had higher competing risk of dying from infectious diseases, cardiometabolic diseases and neoplasms. These differences are more pronounced in the older age groups (Figure 3).

Figure 3.

Fine and Gray competing cause-specific mortality risks by age groups and sex. *Cardiometabolic diseases include a combination of cardiovascular and metabolic diseases mostly diabetes mellitus. Other infectious diseases include: sepsis, acute respiratory infections including pneumonia, diarrhoeal diseases, malaria, meningitis and encephalitis. Neoplasms include: oral, digestive, respiratory, breast, reproductive and other unspecified neoplasms. Metabolic causes include: severe malnutrition and diabetes mellitus. Cardiovascular causes include: acute cardiac disease, stroke and other unspecified cardiac diseases. Other NCDs include: asthma, chronic obstructive pulmonary disease (COPD), liver cirrhosis, renal failure, epilepsy, obstetric non-infectious causes including abortion-related death. External causes include: road traffic accidents, accidental falls, accidental drowning and submersion, accidental exposure to fire and flame, accidental poisoning and assault.

Risk factors associated with cause-specific mortality

Across all causes of death examined, several consistent risk patterns emerged. Men generally were at a higher mortality risk than women, particularly for HIV/AIDS and TB, neoplasms and NCDs, while women had higher mortality risk from other infectious diseases. Being unmarried (separated, divorced or never married) was repeatedly associated with higher mortality, as was poor self-rated health. Higher socioeconomic status increased the risk of HIV/TB mortality and higher education was associated with lower mortality risk from other infectious diseases but paradoxically higher neoplasm-related mortality. Underweight status was associated with higher mortality risks from HIV/TB, neoplasms and metabolic diseases, while obesity showed increased risk for other infectious diseases and NCDs. Functional capacity markers such as strong grip strength, faster walking speed and absence of ADL limitations were associated with reduced mortality across multiple causes. HIV-positive individuals with uncontrolled viral load faced markedly higher mortality risk across several categories, whereas those virally suppressed on ART had reduced CVD mortality compared to their HIV-negative counterparts. Hypotension was associated with higher mortality from all causes except external and indeterminate causes. Additional cause-specific predictors of mortality included grade III hypertension, diabetes, smoking and prior stroke or heart attack, for CVD and metabolic deaths, while alcohol use and TB history increased the risk for infectious and NCD-related mortality. See Appendix 8 of the supplementary data section for the full details of the risk factors associated with cause-specific mortality.

Discussion

This study examined how self-reported and objectively measured demographic, socioeconomic and health-related factors affect all-cause and cause-specific mortality among middle-aged and older adults in a rural South African setting using longitudinal data from the HAALSI cohort. Over 6–7 years, 22.1% of participants died (38.5 deaths/1000 person-years), with spikes in mortality in 2020–2021 coinciding with the COVID-19 pandemic. All-cause mortality was higher among men and increased with age. Key predictors included older age, male sex, lower education level, no religion, smoking, alcohol use, poor self-rated health, underweight status, disability, slow walking, lower grip strength and history of hypertension, diabetes, stroke, tuberculosis and unsuppressed HIV infection. CVD, neoplasms, HIV/TB and other infectious diseases were the leading causes of death.

Mortality rates exceeded those reported in other African studies [37, 38] and the European Survey of Health, Ageing and Retirement in Europe (SHARE) cohort, but were lower than in the US Health and Retirement Study (HRS) [39]. The higher mortality among men aligns with well-established male–female health-survival paradox, where women report higher morbidity and poor quality of life yet live longer than men [40, 41]. This reflects differences in risk exposures, disease severity and health-seeking behaviour, as well as availability of social protection for women [41–45].

Age remained a stronger predictor of mortality, particularly for CVD and metabolic causes, while HIV/TB mortality declined with age. Women had increased mortality risk from neoplasms, metabolic diseases and CVD with age, while men had a higher risk of mortality from other infectious diseases, neoplasms, CVD and other NCDs. These findings are consistent with the growing impact of NCDs as common contributors to mortality in ageing African population [46], alongside persistent infectious diseases threats [33]. This is likely driven by physiological ageing processes that lead to immune decline and increased inflammation [47, 48].

Our findings showed a complex relationship between HIV status, viral load suppression and mortality. We observed lower CVD mortality among virally suppressed HIV-positive individuals compared to HIV-negative participants. This might be explained by the HIV advantage on controlling hypertension [49]. In SSA, HIV care programmes provide increased opportunities for screening and treatment of NCDs among people living with HIV compared to the general population [50]. Some studies in SSA have reported similar or even lower prevalence of hypertension, cardiovascular disease and frailty among treated HIV-positive individuals compared with HIV-negative individuals. This may potentially reflect increased monitoring, treatment adherence and preventive care within HIV treatment programmes [51–53]. The observed lower CVD mortality among virally suppressed HIV-positive adults may reflect differences in healthcare access and risk factor control rather than a direct protective effect of HIV infection itself.

Like previous findings, behavioural risks such as alcohol use and smoking contributed to increased risk of mortality [54–57]. Higher mortality trends among men could be associated with binge alcohol drinking [58, 59]. Additionally, the finding that both low and uncontrolled blood pressure, diabetes and prior stroke were linked to elevated mortality is consistent with existing evidence that places these conditions as the leading risk factors for mortality globally [60–63].

Poor self-rated health strongly predicted mortality from all causes. Self-rated health is used as a predictor of clinical outcomes in older adults outside of clinical or performance-based assessments [64, 65]. Reporting poor self-rated health may likely reflect underlying disease burden and elevated risk of mortality [66]. Functional limitations and weak grip strength, known to increase with ageing [64, 67, 68], were also associated with higher mortality, reinforcing the role of physical function as a predictor of survival among ageing populations [53, 69, 70]. Preventive measures aiming to reduce or delay disability in older ages and individualised physical activity could reduce this heightened mortality risk [71].

Limitations of our study include missing data and potential recall bias from delays in some of the VA interviews due to follow-up challenges and difficulties in locating suitable next of kin shortly after death. Although mortality substantially increased in 2020 and 2021, likely due to the COVID-19 pandemic, the exact proportion of deaths attributable to COVID-19 is unknown. This is because VA approaches for assigning causes of death do not include COVID-19 as a cause. In addition, variables that were self-reported may be affected by social desirability and reporting bias. Underreporting in the HAALSI cohort and the Agincourt HDSS in general can be attributed to limited awareness of certain chronic conditions such as heart attack among participants. However, women, who are known to be better at seeking health care, may be more aware of their conditions than men. Future studies should, therefore, incorporate more objective measures to improve accuracy.

Overall, mortality is rising among ageing adults in a rapidly transitioning rural South African population. While the contribution of CVD is predominant, that of HIV and other infectious diseases remain persistent, creating a double burden of disease alongside multimorbidity. Socioeconomic disparities, unhealthy behaviours and disability are key drivers of this burden. Therefore, in addition to sustaining control of HIV/AIDS, there is a need for interventions targeting disability and continued efforts to tackle infectious diseases among older adults in South Africa.

Declaration of Conflicts of Interest

None declared.

Declaration of Sources of Funding

The HAALSI study was supported by the National Institute on Ageing at the National Institutes of Health (grant number P01 AG041710) and is nested within the MRC/Wits Rural Public Health & Health Transitions (Agincourt) Research Unit which is supported by the Wellcome Trust (Grant numbers 058893/Z/99/A, 069683/Z/02/Z, 085477/Z/08/Z, 085477/B/08/Z); the University of the Witwatersrand; the South African Medical Research Council; and the South African Department of Science and Technology (via the South African Population Research Infrastructure Network). The funders had no role in the design, execution, analysis and interpretation of data or writing of the study.

Research Data Transparency and Availability

The full HAALSI baseline survey data is available in an open-access repository at https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/F5YHML. Mortality and cause of death data will be made available upon reasonable request.