Modelling determinants, impact, and space–time risk of age-specific mortality in rural South Africa: integrating methods to enhance policy relevance

Figures & data

Fig. 1.  Maps showing the location of the Agincourt HDSS 41.

Table 1. Demographic profile of study sample by age group, 1992–2008

Characteristic	Infants (<1 years)	Children (1–4 years)	Children (5–14 years)	Adults (15–49 years)	Adults (50–64 years)	Overall^b
Denominator (person-years)	28,470	116,729	311,387	549,030	70,864	1,110,166
Female (%)	16,030 (50.4)	20,838 (50.3)	34,770 (50.8)	99,994 (56.6)	4,062 (59.5)	576,680 (51.9)
South African (%)	20,382 (64.1)	25,848 (62.3)	43,870 (65.2)	66,926 (66.9)	4,292 (62.9)	117,448 (64.1)
Deaths (% of overall deaths)	826 (9.1)	669 (7.4)	289 (3.2)	3,798 (42.0)	1,337 (14.8)	9,035 (100)
Mean age at death (standard deviation)	126.2 days (112.7)	2.1 years (0.9)	10.0 years (2.9)	34.6 years (8.5)	57.3 years (4.5)	41.8 years (26.8)
Median age at death (interquartile range)	99 days (16–217)	1.9 years (1.4–2.6)	10.0 years (7.5–12.5)	34.8 years (28.4–41.5)	57.0 years (53.5–61.4)	40.7 years (24.2–63.8)
Mortality rate^a	29.0	5.7	0.9	6.9	18.9	8.1

^aPer 1,000 person-years.

^bIncludes 5–14 and 65+ age groups.

Table 2. Top five causes-of-death by age group, 1992–2008

Rank	Infants (<1)	No. (percent)	Children (1–4)	No. (percent)	Adults (15–49)	No. (percent)	Adults (50–64)	No. (percent)
1	HIV/TB	132 (16.6)	HIV/TB	192 (29.5)	HIV/TB	1,545 (43)	HIV/TB	317 (24.7)
2	Diarrhoea or malnutrition	104 (13.1)	Diarrhoea or malnutrition	164 (25.2)	Assault	164 (4.6)	Vascular	159 (12.4)
3	ARI or pneumonia	103 (13)	External	39 (6)	Transport accident	134 (3.7)	Neoplasms	88 (6.9)
4	Perinatal condition^a	93 (11.7)	ARI or pneumonia	27 (4.2)	Vascular	109 (3)	Digestive	54 (4.2)
5	Congenital	25 (3.1)	Congenital	13 (2)	Neoplasms	106 (2.9)	Suicide	35 (2.7)

^aBased on ICD-10 main cause-of-death only, that is, P00–P96 (not inclusive of date of birth and date of death timing).

Fig. 2.  Age-specific mortality rates by year, 1992–2008.

Table 3. All-cause multivariable risk factor analyses for age-specific mortality using Bayesian modelling

	Infants (<1)	Children (1–4)^a	Young adults (15–49)	Older adults (50–64)
Factors	IRR (95% BCI)^b	OR (95% BCI)^b	HR (95% BCI)^b	HR (95% BCI)^c
Temporal
Year (continuous)	1.25 (1.07, 1.58)^c	1.05 (1.04, 1.07)^f	–	–
By period
1992–1997	–	–	1	1
1998–2003	–	–	2.61 (2.42, 2.83)^c	2.59 (2.16, 3.08)^c
2004–2008	–	–	2.90 (2.52, 3.32)^c	4.07 (2.85, 5.60)^c
Proximate individual- and household-level determinants
Increasing age in years	n/a	0.51 (0.46, 0.56)^b	1.06 (1.06, 1.07)^c	1.04 (1.02, 1.06)^c
Winter season	n/a	1.33 (1.12, 1.57)^b	–	–
Male gender	1.10 (0.83, 1.41)^c	e	1.47 (1.38, 1.56)^c	3.03 (2.64, 3.48)^c
Mozambican (maternal for infants and children)	e	1.12 (0.90, 1.38)^b	0.94 (0.87, 1.02)^c	0.59 (0.50, 0.70)^c
Maternal death (in first year for infants or 1–4 for children)
Not due to HIV/TB	6.01 (3.18, 11.37)^c	5.45 (2.99, 8.75)^b	n/a	n/a
Due to HIV/TB	30.78 (12.13, 78.11)^c	15.11 (8.39, 24.54)^b	n/a	n/a
Migrant: ≥6 months outside site per year (maternal status for infants and children)	0.71 (0.48, 1.02)^c	e	1.17 (1.09, 1.25)^c	1.23 (1.04, 1.44)^c
Paternal death (prior birth to within first year for infants or 1–4 for children)
Not due to HIV/TB	e, f	1.51 (0.75, 2.54)^b,f	n/a	n/a
Due to HIV/TB	e, f	2.19 (0.95, 4.06)^b,f	n/a	n/a
Tertiary education (maternal for infants and children)	d	d	–	–
Secondary or higher for adult models	n/a	n/a	0.98 (0.91,1.05)^c	0.56 (0.40, 0.77)^c
Cumulative other household deaths	7.24 (5.70, 9.28)^c	g	–	–
Number of other household deaths	g	g	–	–
None	g	g	1	e
1–4	g	g	1.17 (1.09, 1.25)^c	e
5+	g	g	1.81 (1.39, 2.30)^c	e
Household head:
Male	e	0.78 (0.60, 1.02)^b	0.53 (0.50, 0.57)^c	0.40 (0.35, 0.45)^c
Death	e, f	D	4.52 (4.16, 4.91)^c	6.74 (5.70, 7.92)^c
≥40 years of age	1.44 (1.08, 1.90)^c	E	0.62 (0.58, 0.66)^c	0.15 (0.13, 0.17)^c
Mozambican	e	1.71 (1.40, 2.09)^b	e	e
Infant and child specific factors
Parity	e	E	n/a	n/a
Death of previous sibling	1.56 (0.75, 2.90)^b	1.63 (1.08, 2.45)^b	n/a	n/a
Never breastfed	203.7 (113.9, 342.4)^c,e	e	n/a	n/a
Increasing birth weight	d	d	n/a	n/a
Other child born less than one year prior	8.07 (1.33, 28.02)^c	g	n/a	n/a
Four or more children in the household	g	1.44 (1.13, 1.80)^b	n/a	n/a
Socio-economic determinants
Household SES (MCA) quintile for infant and child models; tertiale for adult models
Most poor	e	g	1	1
Poor	e	g	0.76 (0.69, 0.84)^c	0.78 (0.58, 1.05)^c
Least poor	e	g	0.65 (0.58, 0.71)^c	0.78 (0.58, 1.03)^c
Unknown	e	g	1.48 (1.26, 1.71)^c	1.01 (0.75, 1.35)^c
Village level determinants
Proportion of village deaths due HIV/TB (≥17%)^h (proxy for high prevalence)	–	–	1.92 (1.40, 2.69)^c	e
Average migrant months per year per village individual	–	–	0.95 (0.87, 1.07)^c	e
Environmental or geographic factors
Distance to nearest health facility (>6 km)	e	e	5.34 (3.11, 9.98)^c	1.20 (0.17, 8.63)^c,e
Climatic proxy (elevation in metres)	e	e	0.998 (0.997, 0.998)^f	0.998 (0.997, 0.999)^f
Additional model parameters
Individual unstructured variation (σu ^2)	–	0.04 (0.01, 0.11)	0.11 (0.08, 0.14)	0.16 (0.10, 0.27)
Spatially structured variation (σϕ ^2)	0.08 (0.03, 0.23)	0.23 (0.10, 0.48)	2.21 (1.09, 4.11)	0.25 (0.11, 0.56)
Autoregressive structured temporal variation (σα ^2)	0.60 (0.27, 1.57)	0.29 (0.12, 0.68)	–	–
Dispersion parameter (r) for negative binomial model	0.03 (0.025, 0.035)	–	–	–
Shape parameter (Weibull)	–	–	1.93 (1.86, 1.99)	1.92 (1.83, 2.01)
Deviance Information Criterion (DIC)	8,103	8,612	49,576	10,193

^aResults for child model published previously in (45).

^bMeasure of association based on analytical data structure and model type for given age-group; incidence rate ratio (IRR), odds ratio (OR); hazard ratio (HR); BCI stands for Bayesian credibility interval (similar in interpretation to a conventional 95% confidence interval).

^cSignificant in bivariate analysis, estimate following multivariable adjustment.

^dSignificant (at 10% level) in bivariate analysis but substantial missing data so not included in final multivariable model.

^eNot significant at 10% level in bivariate analysis.

^fSmall numbers (hence wide BCI's).

^gSignificant at bivariate level but not run in Bayesian multivariable framework due to co-linearity with another predictor.

^hMedian split.

Fig. 3.  Geographic risk of adult mortality adjusting for distance to nearest health facility at prediction point using a univariate Bayesian spatial kriging model that includes distance to health facility as the risk factor (white and black scale represents lowest and highest risk, respectively).

Fig. 4.  Maps of all-cause mortality risk by age group within the Agincourt sub-district based on baseline Poisson model without covariates.

Kahn K, Tollman SM, Collinson MA, Clark SJ, Twine R, Clark BD, et al.. Research into health, population and social transitions in rural South Africa: data and methods of the Agincourt Health and Demographic Surveillance System. Scand J Public Health Suppl. 2007; 69: 8–20.10.3402/gha.v6i0.19239.

 Google Scholar

Sartorius B, Kahn K, Vounatsou P, Collinson M, Tollman S. Survived infancy but still vulnerable: spatial-temporal trends and risk factors for child (1–4) mortality in rural South Africa (Agincourt), 1992–2007. Geospat Health. 2011; 5: 285–95.

 Google Scholar