Modelling the demographic impact of HIV/AIDS
Joubert Ferreira (President ASSA)
Wim Els (Executive Director)
David Schneider (Convenor AIDS Committee)
Rob Dorrington (member AIDS Committee)
Overview
The ASSA AIDS Committee and the suite of models
Features of the model and calibration The fit to the provinces Models vs surveys Comparison of ASSA2001 prototype with
the HSRC results by sex and age
The ASSA suite of models
www.assa.org.za/aidsmodel.asp
ASSA AIDS Committee
Set up in 1987 Objective: To assist the actuarial profession and
society in assessing and addressing the impact of the AIDS epidemic in South Africa
Membership: Over 20 members split roughly 50/50 between Cape and Gauteng, with one person (the present convenor, David Schneider) working in Botswana
ASSA AIDS Committee
Some of the current projects: ASSA2001 Professional guidance notes Economic impact of HIV/AIDS CPD, including AIDS impact consulting Data, including life assurance, pathology lab, and blood
transfusion data PR Urban-Rural model Impact on medical schemes
History of the ASSA model
Doyle-Metropolitan model (c1990) ASSA500 (c1995) ASSA600 (c1998) The ASSA2000 suite (2001)
- ASSA2000 lite
- ASSA2000 full
- Aggregate of application to the provinces (2002)
Additional models
Other models:- urban-rural (not released)
- multi-state select population model- interventions model (not released)
Add-ons (not released)
- orphans (maternal, paternal and dual)
- numbers by stages of infection
Features of the model A heterosexual behavioural cohort
component projection model Population divided by risk by:
Age (young, adult, old) ‘behaviour’(PRO, STD, RSK, NOT) ‘previous social disadvantage’ (population group) Geographic (province)
Sex activity Source of partner, probability of transmission,
number of new partners p.a., number of contacts per partner, condom usage, no sex between population groups and no sex between provinces
Diagram 1: A schematic diagram of the ASSA600 Aids Model
Adu
lt (1
4 -
59)
Old
(60+
)
HIV- Young HIV+ Young
NOT RSK STD PRO
Increasing sexual mobility
Increasing risk of HIV infection
HIV- Old HIV+ Old
Dea
ths
Normal Deaths AIDS Deaths
Imported HIV
Migrants (0-59)
Migrants (Aged 60+)
HIV- Births HIV+ Births
You
ng (0
-
13)
The fitting process - calibration
Set as many of the parameters/assumptions from independent estimates (% STD, probability of transmission, condom usage, age of male partners, the median term to survival of adults and children, impact of HIV on fertility, all non-HIV demographic
assumptions) Set some other assumptions (which are not particularly
important) by reasonable guesses (e.g. relative fertility, and risk
groups of migrants) The remaining assumptions are set in order to replicate
known data of the prevalence or impact of the epidemic such as the antenatal prevalence and the mortality figures - calibration (e.g. the mixing of risk groups, sex activity, no. of
partners, age of partners)
Calibration targets Prevalence levels
- *Antenatal – overall prevalence by province and population group over time
- *Antenatal – prevalence by age over time- Ratio of antenatal to national by age- HSRC prevalence by sex and age
Deaths- *Population register – overall by sex, age and over
time- Cause of Death – proportion AIDS in adults by sex
and age- Cause of Death – proportion AIDS in children by age- Cause of Death – ratio of male to female by age over
time
Calibration targets(not yet available)
Census- Numbers by sex and age
- Mortality rates by age and sex (and province?)- orphanhood
- CEB/CS
- deaths in household
Calibration: antenatal vs model - African
0%
5%
10%
15%
20%
25%
30%
35%
1985 1987 1989 1991 1993 1995 1997 1999 2001 2003
ANC(model)
ANCsurvey
ANCsurvey(adjusted)
Calibration: antenatal vs model - Coloured
0%
5%
10%
15%
20%
25%
30%
35%
1985 1987 1989 1991 1993 1995 1997 1999 2001 2003
ANC(model)
ANCsurvey
Calibration: antenatal vs model - Indian
0%
5%
10%
15%
20%
25%
30%
35%
1985 1987 1989 1991 1993 1995 1997 1999 2001 2003
ANC(model)
ANCsurvey
ANCTarget
Calibration: antenatal vs model - White
0%
5%
10%
15%
20%
25%
30%
35%
1985 1987 1989 1991 1993 1995 1997 1999 2001 2003
ANC(model)
ANCsurvey
ANCTarget
National calibration:antenatal vs model
0%
5%
10%
15%
20%
25%
30%
35%
1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005
ANC(model)
ANC survey
ASSA2001
ANC survey(adjusted)
Projected vs actual curve of deaths - males
0
5000
10000
15000
20000
25000
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85+
Proj1999
1996
1997
1998
1999
Projected vs actual curve of deaths - females
0
2000
4000
6000
8000
10000
12000
14000
16000
18000
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85+
Proj1999
1996
1997
1998
1999
Eastern Cape
Eastern Cape
0.0%
5.0%
10.0%
15.0%
20.0%
25.0%
30.0%
35.0%
19
90
19
92
19
94
19
96
19
98
20
00
20
02
20
04
20
06
20
08
20
10
Pe
rce
nta
ge
Model
anc prevalence
adjusted for bias
Free State
Free State
0.0%
5.0%
10.0%
15.0%
20.0%
25.0%
30.0%
35.0%
40.0%
19
90
19
92
19
94
19
96
19
98
20
00
20
02
20
04
20
06
20
08
20
10
Pe
rce
nta
ge Model
anc prevalence
adjusted for bias
Gauteng
Gauteng
0.0%
5.0%
10.0%
15.0%
20.0%
25.0%
30.0%
35.0%
1990
1992
1994
1996
1998
2000
2002
2004
2006
2008
2010
Pe
rce
nta
ge Model
anc prevalence
adjusted for bias
KwaZulu-Natal
KwaZulu-Natal
0.0%
5.0%
10.0%
15.0%
20.0%
25.0%
30.0%
35.0%
40.0%
45.0%
1990
1992
1994
1996
1998
2000
2002
2004
2006
2008
2010
Pe
rce
nta
ge Model
anc prevalence
adjusted for bias
Limpopo
Limpopo
0.0%
5.0%
10.0%
15.0%
20.0%
25.0%
30.0%
19
90
19
92
19
94
19
96
19
98
20
00
20
02
20
04
20
06
20
08
20
10
Pe
rce
nta
ge
Model
anc prevalence
adjusted for bias
Mpumalanga
Mpumalanga
0.0%
5.0%
10.0%
15.0%
20.0%
25.0%
30.0%
35.0%
40.0%
1990
1992
1994
1996
1998
2000
2002
2004
2006
2008
2010
Pe
rce
nta
ge Model
anc prevalence
adjusted for bias
Northern Cape
Northern Cape
0.0%
5.0%
10.0%
15.0%
20.0%
25.0%
19
90
19
92
19
94
19
96
19
98
20
00
20
02
20
04
20
06
20
08
20
10
Pe
rce
nta
ge
Model
anc prevalence
adjusted for bias
North West
North West
0.0%
5.0%
10.0%
15.0%
20.0%
25.0%
30.0%
35.0%
1990
1992
1994
1996
1998
2000
2002
2004
2006
2008
2010
Pe
rce
nta
ge Model
anc prevalence
adjusted for bias
Western Cape
Western Cape
0.0%
2.0%
4.0%
6.0%
8.0%
10.0%
12.0%
14.0%
19
90
19
92
19
94
19
96
19
98
20
00
20
02
20
04
20
06
20
08
20
10
Pe
rce
nta
ge
Model
anc prevalence
adjusted for bias
Models vs surveys
ASSA involved in modelling, not surveying Modelling involves creating a tool that tries to simulate
reality in a way that is consistent with empirical data Modelling does not produce empirical data, but rather an
interpretation of, and extrapolation from, empirical data Conclusions to be drawn from models are limited to the
extent that modelling involves a great many simplifications and assumptions
However, to the extent that models attempt to tie together data from many sources, with some sort of consistency, they can give useful indications of errors (random or otherwise) in surveys
HSRC survey - limitations
Invaluable piece of research – particularly if prepared to share with other researchers
Potential for bias high non-response By design excludes some high-risk populations
(prisons, military and hospitals), by default others (e.g. truck drivers, and those not part of permanent homes, criminals, etc)
Use of retired nurses to ask about sexual behaviour Wide confidence intervals Unwillingness to share (even questionnaires)
Prevalence by province (all women 15-49)
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
WC EC NC FS KZ NW GT MP LP SA
ASSA2000
HSRC
Male population prevalence vs HSRC
0%
5%
10%
15%
20%
25%
30%
35%
2-14
15-1
9
20-2
4
25-2
9
30-3
4
35-3
9
40-4
4
45-4
9
50-5
4
55+
ASSA2001
HSRC
Female population prevalence vs HSRC
0%
5%
10%
15%
20%
25%
30%
35%
40%
45%2-
14
15-1
9
20-2
4
25-2
9
30-3
4
35-3
9
40-4
4
45-4
9
50-5
4
55+
ASSA2001
HSRC