Finding the best questions for measuring AIDS mortality
using verbal autopsy:A validation study in
Kisesa, Tanzania and
Manicaland, Zimbabwe
Ben Lopman
Co-authorsManicaland Cohort• Jennifer Smith• Godwin Chawira • Simon Gregson
Kisesa Cohort• Adrian Cook• Yusufu Kumogola• Milalu Ndege• Basia Zaba • Mark Urassa• Raphael Isingo
WHO• Ties Boerma
Objective•To develop a classification
system of AIDS/ non-AIDS deaths using VA data validated against HIV testing that performs consistent in–Place–Time
VA for AIDS mortality
• HIV/AIDS is the leading cause of death among young adults
• Hospital records and vital registration of deaths are inadequate for AIDS – bias, underreporting and stigma.
• Improving measurement of AIDS mortality is urgent
• Monitoring the success of programmes relies on accurate measurement of AIDS deaths
KisesaWard (Magu DSS)
HIV Prev1994/1995: 6.0%2000/2001: 8.3%
Manicaland HIV/STD Prevention Study
HIV Prev1998/2000: 23.0%2001/2003: 20.5%
Mortality surveillance and VA tool
• The study teams identified deaths through the use of checklists of all individuals interviewed at baseline
• Nurse conducted interview with primary caregiver• VA developed in Kisesa, used in that site until
2002. After which, a verbal autopsy questionnaire based on the INDEPTH standard was used – This questionnaire lacked a number of questions related
to opportunistic infections often seen in AIDS patients
• A nearly identical version was used in Manicaland in both Round 1 and Round 2.
Gold Standard
• Gold standard of AIDS deaths. An individual who was – a) HIV positive at previous test – b) not accident/injury– c) not direct obstetric deaths
3945 to 59Test
15815 to 44TestK isesa
3845 to 59TestRound 2
18115 to 44TestRound 2
5145 to 59TestRound 1
8815 to 44TestRound 1
23715 to 44TrainRound 1Manicaland
Total with HIV test and VA (n)Age range
3945 to 59Test
15815 to 44TestK isesa
3845 to 59TestRound 2
18115 to 44TestRound 2
5145 to 59TestRound 1
8815 to 44TestRound 1
23715 to 44TrainRound 1Manicaland
Total with HIV test and VA (n)Age range
Training and testing datasets
AIDS deaths (gold standard)
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
15 to 44 15 to 44 45 to 59 15 to 44 45 to 59 15 to 44 45 to 59
Train Test Test Test Test Test Test
Round 1 Round 1 Round 1 Round 2 Round 2
Manicaland Kisesa
Rule-based algorithm• On TRAIN data• Calculate LR for all signs &
symptoms • Based on SPECIFICITY
1. Weight loss
2. …
.
WeightlossVaginal tumours
Wasting
Zoster
Abscesses
ARTIJaundice
Oral candidiasis
Diarrhoea
Recent TB
0.00
0.25
0.50
0.75
1.00
Sen
sitiv
ity
0.00 0.25 0.50 0.75 1.001 - Specificity
Area under ROC curve = 0.7697
.
WeightlossVaginal tumours
Wasting
Zoster
Abscesses
ARTIJaundice
Oral candidiasis
Diarrhoea
Recent TB
0.00
0.25
0.50
0.75
1.00
Sen
sitiv
ity
0.00 0.25 0.50 0.75 1.001 - Specificity
Area under ROC curve = 0.7697
Specificity: 15 to 44 year
80
85
90
95
100W
eigh
t los
s
Her
pes
zost
er
Jaun
dice
Vag
inal
tum
ours
Was
ting
AR
TI
Abs
cess
es o
r so
res
Ora
l can
didi
asis
Dia
rrho
ea
Rec
ent T
B
Manicaland
Kisesa
Sensitivity: 15 – 44 year
0
10
20
30
40
50W
eigh
t los
s
Her
pes
zost
er
Jaun
dice
Vag
inal
tum
ours
Was
ting
AR
TI
Abs
cess
es o
r so
res
Ora
l can
didi
asis
Dia
rrho
ea
Rec
ent T
B
Manicaland
Kisesa
15 to 44 years (9 Signs/Symptoms)
0%
25%
50%
75%
100%
Round OneTest
Round Tw o 1994 – 2002
Manicaland Kisesa
0%
25%
50%
75%
100%
Round OneTest
Round Tw o 1994 – 2002
Manicaland Kisesa
Sensitivity Specificity
Manicaland Kisesa Manicaland Kisesa
0%
25%
50%
75%
100%
Round OneTest
Round Tw o 1994 – 2002
Manicaland Kisesa
0%
25%
50%
75%
100%
Round OneTest
Round Tw o 1994 – 2002
Manicaland Kisesa
45 to 59 years (9 Signs/Symptoms)
Sensitivity Specificity
Manicaland Kisesa Manicaland Kisesa
0%
25%
50%
75%
100%
Round Tw o 1994 – 2002
Manicaland Kisesa
0%
25%
50%
75%
100%
Round Tw o 1994 – 2002
Manicaland Kisesa
15 to 44 years, INDEPTH variables
(5 Signs/Symptoms)Sensitivity Specificity
Manicaland Kisesa Manicaland Kisesa
Using INDEPTH questionsKisesa 1994-2002
40
45
50
55
60
65
70
75
80
85
5 av
ailab
le sy
mtp
oms
Dys
ente
ry
Fev
er w
conv
ulsion
s
Fev
er w
unc
onsc
iousn
ess
TB re
porte
d
TB sy
mpt
oms
HIV
/AID
S repo
rted
Weig
ht lo
ss w
ith d
ry/cr
acke
d sk
in
Weig
ht lo
ss w
ith o
edem
a
Fev
er w
pale
ness
and
shive
rs
Chr
onic
diarrh
oea
Fev
er w
coug
h an
d pa
in br
eath
ing
SenSpec
Using INDEPTH questionsManicaland R2
40
45
50
55
60
65
70
75
80
85
5 av
ailab
le sy
mtp
oms
Dys
ente
ry
Weig
ht lo
ss w
ith o
edem
a
TB re
porte
d
Fev
er w
conv
ulsion
s
Fev
er w
pale
ness
and
shive
rs
HIV
/AID
S repo
rted
Fev
er w
unc
onsc
iousn
ess
Fev
er w
coug
h an
d pa
in br
eath
ing
Chr
onic
diarrh
oea
Weig
ht lo
ss w
ith d
ry/cr
acke
d sk
in
TB sy
mpt
oms
SenSpec
AIDS CSMF, Correcting for misclassification
Estimate True
Manica R1 76% 74%
Manica R2 88% 76%
Kisesa 51% 53%
Estimating HIV prevalence
Estimate (95% CI)
True
Manica R1 24% (18-31)
23%
Conclusions
• Developed a set of criteria using VA that consistently measures AIDS mortality
• Algorithm performs consistently in these settings of variable HIV prevalence
• Only reliable for adults under age 45– OK in Manicaland, fewer female AIDS deaths
in older adults in Kisesa• Not subject to clinical biases• Can estimate HIV mortality in
populations lacking serosurveillance
INDEPTH and WHO VA
• Either of these widely used VA tools would face some limitations surveying AIDS mortality with the proposed criteria– Highly predictive and common
• Herpes zoster, oral candidiasis, abscesses/sores
– Highly predictive and rare• Vaginal tumors
Acknowledgements
• Funding– Health Metrics Network– Wellcome Trust– Netherlands Government
• People of Kisesa Ward, Tanzania and Manicaland Province, Zimbabwe– Especially Kin and Caregivers