Methods for establishing the extent of HIV epidemics and trends in prevalence

Geoff Garnett

Imperial College London

UNAIDS methods for developing countries: Models

• Earlier model start time and current prevalence fitted to a gamma curve

• Epidemic Projection Package (EPP) – Generalised epidemic – fits prevalence trends to ANC data (later adjusted for national surveys)

• Workbooks – Concentrated & low level epidemics – numbers and prevalence in specific groups

• Spectrum – Calculates ages specific incidence and impact on the basis of overall prevalence trends.

• Short term incidence model

Risk behaviours

HIV incidence

HIV prevalence

AIDS

Deaths

Primary prevention

Secondary prevention

Tertiary prevention

Measuring epidemiological impact of prevention

Behavioural surveys

Surveillance with novel tests / modelling

Sentinel surveillance / population based

surveys

Case reports

Registration / census / surveys

HIV Prevalence data

Generalised epidemics

ANC data

Household based surveys

Fitted in EPP to generate trends

Derived incidence

Spectrum – age structuredMortality, orphanhood.

Concentrated epidemics

ANC data minus estimated SWs & IDUs

IDUs, MSM, SWs, clients

Size estimates

Associated prevalence estimates

Upper and lower bounds over time

Workbooks spreadsheet

Sources of uncertainty in HIV/AIDS estimates for generalised epidemics

Estimate

Source Uncertainty

Relationship of adult prevalence to prevalence among pregnant women

Survival of infected adults

Epidemic curve

National coverage of sentinel surveillance

Adult HIV prevalence, new infections and AIDS mortality

New infections in children

Probability of mother to child transmission

Child AIDS deaths and HIV

prevalence

Child survival (AIDS and other causes

The natural course of incidence and prevalence of a local HIV epidemic over time

Time (years)

0

5

10

15

20

25

0 10 20 30 40 50

Perc

ent I

ncid

ence

/Pre

vale

nce

Prevalence HIV

Incidence HIV infection

Incidence AIDS deaths

Rt=R0>1 Rt<1Rt=1

Interested in current incidence – but even if a validated test available would require an order of magnitude increase in sample sizes.

1980 1990 2000 20100

10

20

30

40

50

60

Year

HIV

pre

va

len

ce

(%)

Data – HIV prevalence in ANC clinics in urban and semi-urban Zimbabwe

1990 1994 1998 2002 2006-40

-20

0

20

40

Year

Ye

ar-o

n-y

ear

ch

ang

e (

%)

Median Change

Change in specific clinic

Models fitted to urban ANC HIV prevalence trends in ZimbabweSample importance resampling (sample 4,000,000 resample 8,000)

1980 1985 1990 1995 2000 2005 20100

10

20

30

40

50

60

Year

HIV

pre

va

len

ce

(%)

No behaviour change

Behaviour change

Bayes Factor 58Likelihood ratio test p< 0.0001

1980 1985 1990 1995 2000 2005 20100

2

4

6

8

10

Year

HIV

inci

de

nce

(p

er

10

0p

yar)

Incidence trends associated with model fits – urban Zimbabwe

HIV

inci

denc

e (p

er 1

00 p

er y

ear)

10

8

6

4

2

0

Year

Mode

2.5%

97.5%

1990 2000 20101980

1980 1985 1990 1995 2000 2005 20100

5

10

15

20

25

30

35

40

Year

HIV

pre

vale

nce

(%)

Urban Rwanda

1980 1985 1990 1995 2000 2005 20100

5

10

15

20

Year

HIV

pre

vale

nce

(%)

Rural Rwanda

1980 1985 1990 1995 2000 2005 20100

5

10

15

20

25

30

35

40

45

Year

HIV

pre

vale

nce

(%)

Rural Zimbabwe

1980 1985 1990 1995 2000 2005 201005

10152025303540455055

Year

HIV

pre

vale

nce

(%)

Urban Zimbabwe

1980 1985 1990 1995 2000 2005 20100

5

10

15

Year

HIV

pre

vale

nce

(%)

Urban Niger

a b c

d e f

g h

HIV Prevalence data

Generalised epidemics

ANC data

Household based surveys

Fitted in EPP to generate trends

Derived incidence

Spectrum – age structuredMortality, orphanhood.

Concentrated epidemics

ANC data minus estimated SWs & IDUs

IDUs, MSM, SWs, clients

Size estimates

Associated prevalence estimates

Upper and lower bounds over time

Workbooks spreadsheet

HIV prevalence among at-risk groups

Population size of at-risk groups

Survival of infected adults

Adult new infections and AIDS mortality

Age and sex distribution of HIV prevalence in at-risk groups

Impact of HIV on fertility

Female age-specific fertility rate in at-risk groups New infections in children

Child AIDS deaths & HIV prevalence

Child survival (AIDS-related & other causes)

Rates of entering and leaving at-risk groups

Adult HIV prevalence

Coverage of sentinel surveillance system

Probability of mother-to-child transmission

Sources of uncertainty in concentrated HIV epidemics

Concentrated epidemics - Workbooks

• ‘Estimates’ upper and lower bounds for risk group size and for prevalence in risk groups – local expert based.

• Calculates a non-overlapping number of infections at a point in time

• If multiple time points fits a simple curve (either a single or double logistic curve)

Adult population

Entire population

At riskChildren of +ves

MSM IDUs

Sex Workers

Partners of those with risk

Immigrants from high prevalence

states

Populations at risk in Europe (need to avoid double counting)

Sources of information on the extent of HIV spread

HIV prevalence surveys

General population High risk group

Estimate of size ofhigh risk group

Case reports (HIV/AIDS/deaths)

Risk behaviour data

Prediction of future HIV trends

Need consistent sources of information and sampling over time to explore trends

Estimates of size of high risk groups:

Counting/mapping

Capture/recaptureMultiplier

Contact tracing/snowball sampling

Population in HIV prevalence survey needs to match population for which size is estimated

Prevalence in population based surveys

MSM in capital city club ≠ men who frequently have sex with men ≠ men who occasionally have sex with men ≠ men who have ever had sex with men

Tipping point R0=1

Increasing contacts, transmission likelihood, duration

Stab

le H

IV P

reva

lenc

e

Increased heterogeneity

What do we expect the long term prevalence of HIV to be? When can we expect new outbreaks?

Routes into healthcare

Undiagnosed

Diagnosed at VCT

Attends ANC

Never diagnosed

Presents at clinic when develops severe symptoms

Enters Health-care system

Referred

Diagnosed at VCT

Referred

0

1000

2000

3000

4000

5000

6000

7000

8000

9000

1999 2000 2001 2002 2003 2004 2005 2006 2007 2008

Nu

mb

er

of

HIV

an

d A

IDS

dia

gn

oses a

nd

d

eath

s

Year of HIV or AIDS diagnosis or death

HIV diagnoses

AIDS diagnoses

Deaths

New HIV and AIDS diagnoses in the UK, and deaths among HIV infected individuals: 1999 -2008 (HPA)

Estimated late diagnosis of HIV infection by prevention group, UK: 2007 (HPA)

CD4 cell count <200 per mm3 within three months of diagnosis among adults

19%

42%

36%

30% 31%

0%

10%

20%

30%

40%

50%

MSM Heterosexual men

Heterosexualwomen

Injecting drugusers

Overall

Per

cen

tag

e d

iag

no

sed

lat

e

Total n= 2,679 1,434 2,180 152 7,649

Figure 1The BED response function relationship between probability of sample being classified as recent by BED test and time since HIV-infection. The first 2 years (shaded area) is informed by observational data (see text), and the pattern over the remaining time is uncertain and three hypothetical scenarios are constructed.

0%

20%

40%

60%

80%

100%

0 2 4 6 8 10 12 14 16 18 20

Sam

ples

iden

tifie

d as

BED

-rec

ent

(%)

Time since infection (years)

Scenario I Scenario II Scenario III

False positives (1-specificity)

True positives(sensitivity)

150 days

Scenario III

Scenario IIScenario I

Proportions of infections of at least one year that are miss-classified by the BED test for six African countries over time (ages 15-49), using BED response scenario B (increasing proportion false positive)Incidence based on spectrum model fits to EPP prevalence trends

Kenya

LesothoMozabique

Uganda

ZambiaNigeria

Median CD4 count at diagnosis by prevention group: UK (1998-2007)

Data on pregnancy status only from 2000

0

50

100

150

200

250

300

350

400

450

1998 1999 2000 2001 2002 2003 2004 2005 2006 2007

Me

dia

n C

D4

Co

un

t (c

ells

/mm

3 )

MSM Heterosexual Men Pregnant Women

Non-pregnant women Injecting Drug Use All Women

0 5 10 15 200

100

200

300

400

500

600

700

800

900

Time since infection

CD

4 c

ou

nt (

nu

mb

er/

mic

rolit

re)

MeanIndividual simulation

Modelled decline in CD4+ cells – 5 realisations from 1000

Sources of information on the extent of HIV spread

HIV prevalence surveys

General population High risk group

Estimate of size ofhigh risk group

Case reports (HIV/AIDS/deaths)

Risk behaviour data

Prediction of future HIV trends

Variable and diverse data sources can be combined in a modelling framework with consistent relationships

between behaviours, incidence, prevalence, CD4 counts, opportunistic infections and death

Sources of information on the extent of HIV spread

HIV prevalence Case reports (HIV/AIDS/deaths)With CD4 counts

Risk behaviour

data

Models linking epidemiological processesOutputs compared with observation where available

Screening, testing and care patterns