The document analyzes the relationship between income and hypertension in South Africa using national survey data. It finds that wealthier men are more likely to be hypertensive, while there is no income gradient for women. It also finds no evidence of an income gradient in unawareness of hypertensive status. The high rate of unawareness among hypertensive South Africans points to missed opportunities for screening. Improving screening for hypertension is an urgent priority given its high prevalence in South Africa.

1. Richer but poorer in health?
The income gradient in chronic
conditions: evidence from South Africa
R Thomas, R Burger, K Hauck
Katharina Hauck
Senior Lecturer in Health Economics
Department of Infectious Disease Epidemiology, School of Public Health
Imperial College London

2. Introduction and Motivation
Relationship between income and health
• Grossman model predicts health is a normal good
• Empirical evidence for high income countries
• Empirical evidence for communicable diseases in low- and middle-
income countries (LMICs)
• But: Conflicting empirical evidence for non-communicable diseases in
LMICs
• Some studies find prevalence concentrated among the poor
• Others find prevalence concentrated among the rich
• Others find no relationship between wealth and chronic diseases

3. Introduction and Motivation
Relationship between income and healthcare
•Grossman model predicts health care is a normal good
•Empirical evidence for all countries
•Visits to healthcare facilities provide opportunity to screen for chronic
conditions
 Richer individuals should have greater levels of awareness
Confirmed by empirical evidence for high income countries
• If rich are less likely to suffer chronic conditions, greater awareness
among them will aggravate inequalities in health
• If rich are more likely to suffer chronic conditions, greater awareness
will ameliorate inequalities in health

4. Research question
Complex relation between income, prevalence of chronic conditions and
awareness
Research objectives:
Investigate existence and direction of income gradients in…
1. Prevalence of hypertension, and
2. Unawareness of hypertensive status in South Africa
Why hypertension?
•Prevalence is very high in LMICs: 78% among South Africans of ages 50
and above! (Lloyd-Sherlock et al. 2014)
•Suited for studying awareness
 Asymptomatic but inexpensive to screen for in primary care

6. Background: Income gradient in hypertension in LMICs
• Double-burden of infectious and non-communicable diseases in LMICs
• Poor evidence on income gradient in prevalence of chronic conditions
» Prevalence concentrated among the poor (Hosseinpoor et al. 2012,
Murphy et al. 2013, Lloyd-Sherlock 2014)
» Prevalence concentrated among the rich (Zhao et al. 2013, Gaziano
et al. 2010)
» No relationship between wealth and chronic diseases (Lei et al.
2012, Witoelar et al. 2009, Vellakkal et al. 2013, Case et al. 2004)
• Conflicting evidence on income gradient in prevalence of hypertension
for LMICs: Why?

7. Background: Income gradient in prevalence
• Grossman model predictions (1972):
• Rich individuals demand more health because they have higher
returns to health capital
But:
• Rich individuals can also more easily afford unhealthy diets and
transport (Zhao 2013)
• Counteracts the higher returns to health capital
• Richer individuals have a more varied diet
• But also consume more processed food high in fat and sugar
(Hosseinpoor et al. 2012)
• Richer individuals are more likely to have access to transport and work
in sedentary occupations

8. Background: Income gradient in unawareness
Lack of evidence on income gradient in awareness of hypertensive status
in LMICs
• Grossman model predictions (1972):
» Individuals make health investments in each time period
» Individuals with higher returns to health capital make greater
investments in their health
» Need to be well informed about their health status
» Positive relationship between income and awareness
• Empirical evidence from high-income countries
• Johnston et al. 2009:
• Negative income gradient in hypertension prevalence for England
• Less false-reporting amongst richer and more educated individuals

9. Background: Income gradient in unawareness
Cawley and Choi (2015):
• Higher income associated with more accurate reporting of a
number of conditions, including hypertension
• Not fully explained by gradient in healthcare utilization
Chatterji et al. (2012):
• Racial/ethnic disparities in awareness of chronic disease in USA
Poor awareness reinforces negative income gradient in hypertension in
high-income countries
Richer and better educated individuals are less likely to suffer from
chronic disease, but if they do, they are more likely to be aware of
their condition

10. Background: Income gradient in unawareness in LMICs
Empirical evidence from LMICs
•Chow et al. (2013):
• Multinational study of 400,000 individuals in 17 countries
• Better education associated with greater awareness in low- but
not middle- or high-income countries
•Lloyd-Sherlock (2014):
• WHO’s study of Global Ageing and Adult Health (SAGE) from 6
low- and middle-income countries
• Greater wealth and education associated with better awareness
in some but not all countries

11. Background: Income gradient in unawareness in LMICs
Empirical evidence from LMICs
•Case et al. (2004):
• Data from 200 households in Khayelitsha township in South Africa
• Richer individuals more likely to be on hypertensive medication
• No income gradient in hypertension prevalence
• Greater awareness among the rich?
•Zhao et al. (2013):
• Higher prevalence of hypertension among richer individuals in
China
• Upon receiving diagnosis, richer individuals reduced fat intake
more
• Awareness seems to ameliorate the positive income gradient in
hypertension

13. Data
South Africa’s National Income Dynamics Survey (NIDS)
• National Household Panel Survey
• 4 waves (in 2008, 2010, 2012, and 2016)
• High attrition in 2010
• This study uses 2008 and 2012 as pooled cross section
• Individuals above the age of 18
• Estimation sample 12,493 (2008) and 16,391 (2012)
• Socio-economic information
• Objective measures of height, weight, waist circumference, blood
pressure and pulse at each wave
• Self-reported mainly chronic health conditions

14. Data: Descriptive Statistics

15. Data: Descriptive Statistics (continued)

16. Data: Prevalence of hypertension

17. Data: Self-reported and measured hypertension

18. Data: Self-reported and measured hypertension by
quintiles of household income

19. Methods: Income gradient in hypertension prevalence
• Finite mixture model to estimate income gradient in hypertension
prevalence (Deb et al. 2011; Conway and Deb 2005)
• Pooled data from 2008 and 2012 waves
• Separate models for men and women
SBPi: measured systolic blood pressure
LINCi: annual household income (in natural log)
Zi: vector of individual specific characteristics (education, age, race,
married, smoker, alcohol, waist circumference)
Xi: vector of household level characteristics (number of children, number
of adults, urban or rural location)
Also included are wave and province dummies

20. Methods: Income gradient in hypertension prevalence
• Ordinary Least Squares estimate of α is average effect of income
across sub-groups within sample
• Finite Mixture Model (FMM) represents heterogeneity in sample by
using a small number of latent classes
• Each class represents ‘types’ of individuals
• C-group FMM model:
• j = 1 … C
• πj: proportions of classes C
• f j(SBPi|.): j-th density

21. Methods: Income gradient in hypertension prevalence
• We apply mixture of normal distributions
• Component distributions:
• We estimate posterior probabilities for belonging into each class
• Sampling weights and robust standard errors clustered at individual
level

22. Methods: Income gradient in hypertension unawareness
• We only observe unawareness for those who are hypertensive
• Sample selection problem
• Censored Bivariate Probit Model (CBPM)
• Following Johnson et al. (2009) misreporting of hypertension in
England
1st equation:
2nd equation:
Zi and Xi : vectors of socio-economic characteristics
ε1i and ε2i are bivariate normally distributed with covariance ρ

23. Methods: Income gradient in hypertension unawareness
• CBPM requires valid exclusion restrictions
» Must determine the probability of having high BP
» but not directly affect the probability of being unaware
• Two measured variables:
1. Heart rate
2. Waist circumference
• Having high waist circumference may make individuals more likely to
seek healthcare and being aware of high BP
• But we control for healthcare visits

24. Results: Income gradient in hypertension – OLS and FMM
for systolic blood pressure
Men in component 2 have a 0.8
mmHg increase in SBP for a
one log-point increase in
income Women in component 2 have a 7.8
mmHg reduction in SBP if they
completed secondary schooling
Models 3 and 4 control
for employment

25. Results: Income gradient in hypertension – OLS and FMM
for systolic blood pressure

26. Results: Income gradient in hypertension – OLS and FMM
for systolic blood pressure
The employed in component 2
have higher SBP

27. Results: Income gradient in hypertension – determinants
of being in component 2 for SBP

28. Results: Income gradient in hypertension – alternative
specifications
• Separate OLS models for employed and unemployed
• Positive income gradient only for the employed
• Additional control for stress with a ‘depression symptoms index’
in the model for the employed
• Reduces the coefficient on income to 0.6 (p<0.05)
• Coefficient on depression index is significant and positive

29. Results: Income gradient in hypertension – comparison
with previous findings
• WHO’s SAGE study in elders in South Africa: higher education
associated with lower hypertension prevalence (Lloyd-Sherlock et al.
2014)
• Elderly sample
• Not separate models by men and women
• No income gradient in hypertension prevalence in deprived
Kayelitsha township in Western Cape (Case et al. 2004)
• But not a random sample of the South African population
• Positive income gradient in hypertensive medication

30. Results: Income gradient in hypertension unawareness
Unconditional marginal effects from a censored bivariate probit selection model
Model 2 controls for
education, and model 3
for lifestyle and
employment
No statistically significant effect of
income on probability of being
unaware
Having had no or private recent healthcare
increases probability of being unaware,
in comparison to public healthcare
consultation

31. Results: Income gradient in hypertension unawareness
Education has not impact on unawareness

32. Results: Income gradient in hypertension unawareness –
alternative specifications
• Controls for stress with a ‘depression symptoms index’
• Controls for other chronic and infectious diseases
• Alternative thresholds for hypertension (SBP ≥ 150; DBP ≥ 95, and
SBP ≥ 160; DBP ≥ 100)
• Using only heart rate as instrument
The finding of no income gradient in unawareness remains unchanged

33. Results: Income gradient in hypertension unawareness –
comparison with previous findings
• WHO’s SAGE study in elders in South Africa: secondary education
not associated with greater awareness (Lloyd-Sherlock et al. 2014)
• Chow et al. 2013:
• Unawareness associated with income in low- but not middle- or
high-income countries
• Study of 400,000 individuals from 17 countries
• Johnson et al. (2009):
• A degree level qualification reduces false negative reporting of
hypertensive status by 7%
• Men and non-White are more likely to be unaware: confirmed for
LMICs and high income countries by Lloyd-Sherlock et al. 2014,
Johnston et al. 2009, Chow et al. 2013, Lei et al. 2012

34. Results: Income gradient in hypertension unawareness –
comparison with previous findings
• Higher BMI and diabetes associated with greater awareness (Lloyd-
Sherlock et al. 2014, Johnston et al. 2009, Lei et al. 2012)
• Recent healthcare visits associated with greater awareness in
Indonesia (Sohn 2015)
• Private patients less likely to be aware than public patients in the
case of tuberculosis in South Africa (Van Wyk et al. 2011)

35. Limitations
• BP readings may be incorrect
• ‘white coat’ syndrome
• Hopefully randomly distributed across sample
• Intentional misreporting of chronic health conditions
• Social desirability bias
• The better educated more influenced by this? (Cawley and Choi
2015)
• Misreporting of income and wealth

36. Conclusions
• Study investigates existence and direction of income gradients in
hypertension prevalence and awareness in South Africa
• Identify subpopulations with distinct characteristics
• to analyse income gradient in hypertension
• Finite mixture model
• Richer individuals more likely to be aware of hypertensive status?
• theoretical prediction
• adjusting for censoring in awareness
• Censored bivariate probit model

37. Conclusions
• Wealthier men more likely to be hypertensive
• Among younger and White or Asian men
• No income gradient for women
• Unawareness is a major problem in South Africa
• 56% of hypertensive are unaware
• No evidence of income gradient in unawareness
• Large number of missed opportunities for screening in primary care
• In particular in private healthcare
• Unawareness aggravated by South Africa’s fragmented health
system
• Improving and expanding screening for hypertension urgent priority

38. Conclusions
Thanks!!
Paper is under review and comments are highly welcome!
For references, please request a copy of the paper
Ranjeeta Thomas: r.thomas@imperial.ac.uk
Katharina Hauck: k.hauck@imperial.ac.uk