bias in estimates of hiv incidence based on the detuned assay: a proposed solution robert s remis,...
TRANSCRIPT
Bias in estimates of HIV Bias in estimates of HIV incidence based on the detuned incidence based on the detuned assay: assay: A proposed solutionA proposed solution
Robert S Remis, Robert WH Palmer, Robert S Remis, Robert WH Palmer,
Janet M RaboudJanet M Raboud
Department of Public Health Sciences, University of TorontoDepartment of Public Health Sciences, University of Toronto
Mount Sinai Hospital, Toronto, OntarioMount Sinai Hospital, Toronto, Ontario
STARHS satellite meetingSTARHS satellite meeting
Bangkok, Thailand, July 11, 2004Bangkok, Thailand, July 11, 2004
MOHLTC, Laboratories Branch, IMC – 2001
BackgroundBackground
• STARHS assay of HIV-positive specimens STARHS assay of HIV-positive specimens identifies recent infections identifies recent infections
• Used to calculate HIV incidence density, a Used to calculate HIV incidence density, a critical indicator usually difficult to critical indicator usually difficult to measuremeasure
• Numerator is discordant specimens; Numerator is discordant specimens; denominator is person-time from window denominator is person-time from window periodperiod
• ButBut analysis using diagnostic specimens analysis using diagnostic specimens may be subject to strong testing biasmay be subject to strong testing bias
MOHLTC, Laboratories Branch, IMC – 2001
Problem of biasProblem of bias
• In 2002, assessed sources, direction and strength of In 2002, assessed sources, direction and strength of bias with diagnostic specimens (bias with diagnostic specimens (Remis et al, XIV ICARemis et al, XIV ICA))
• For MSM, bias up to 7.3 fold, with plausible For MSM, bias up to 7.3 fold, with plausible parameter values yielding bias of 2-3 foldparameter values yielding bias of 2-3 fold
• Principal source of bias Principal source of bias “seroconversion effect”“seroconversion effect” i.e. i.e. increased likelihood of HIV testing following infection increased likelihood of HIV testing following infection due to seroconversion illness or to high risk due to seroconversion illness or to high risk exposureexposure
• Quantified as proportion of subjects who test within Quantified as proportion of subjects who test within 90 days after HIV infection (Psce)90 days after HIV infection (Psce)
MOHLTC, Laboratories Branch, IMC – 2001
Proposed solution #1Proposed solution #1
• Incidence density calculated using Incidence density calculated using STARHS assay with diagnostic specimens STARHS assay with diagnostic specimens must be interpreted with great cautionmust be interpreted with great caution
• Need to adjust calculated HIV incidence Need to adjust calculated HIV incidence taking into account bias due to Ptaking into account bias due to Pscesce
• Originally proposed studies to measure Originally proposed studies to measure knowledge of seroconversion illness and knowledge of seroconversion illness and assess likelihood of immediate HIV assess likelihood of immediate HIV testing under various scenariostesting under various scenarios
MOHLTC, Laboratories Branch, IMC – 2001
Not so fastNot so fast
• Studies take time and cost moneyStudies take time and cost money• Population studied may not be Population studied may not be
representative (may need many surveys to representative (may need many surveys to include different populations)include different populations)
• Questions about likely HIV testing Questions about likely HIV testing hypothetical; answers may not be validhypothetical; answers may not be valid
• No help with historical specimensNo help with historical specimens
MOHLTC, Laboratories Branch, IMC – 2001
Eureka!Eureka!
• HIV incidence calculated from STARHS HIV incidence calculated from STARHS assay at different window periods assay at different window periods provides empirical evidence of Psceprovides empirical evidence of Psce
• Slope of HIV incidence at different Slope of HIV incidence at different window periods is direct and quantitative window periods is direct and quantitative indicator of strength of Psceindicator of strength of Psce
MOHLTC, Laboratories Branch, IMC – 2001
Incidence calculated using different Incidence calculated using different window window periods with Vironostika assay, 2001periods with Vironostika assay, 2001
0.0
0.5
1.0
1.5
2.0
2.5
133 170 336
Window period (days)
Inci
dence
(per
100 p
ers
on-y
ears
) MSM
MSM-IDU
IDU
HR hetero
LR hetero
MOHLTC, Laboratories Branch, IMC – 2001
Determination of Psce and true Determination of Psce and true incidence using empirical dataincidence using empirical data
• Algebraic formula developed in 2002 Algebraic formula developed in 2002 expressed measured incidence density expressed measured incidence density as a function of true incidence density, as a function of true incidence density, PPscesce and HIV testing parameters and HIV testing parameters
MOHLTC, Laboratories Branch, IMC – 2001
Measured incidence as Measured incidence as function of Pfunction of Pscesce and true and true incidenceincidence
MOHLTC, Laboratories Branch, IMC – 2001
Measured incidence as Measured incidence as function of Pfunction of Pscesce and true and true incidenceincidence
Where:Where:
I’I’estest = measured incidence density = measured incidence density
N = study populationN = study population
TTobsobs = study period = study period
TTwinwin = detuned window period = detuned window period
IItruetrue = true incidence densit = true incidence densit
TTtesttest = mean inter-test interval = mean inter-test interval
PPscesce = proportion seroconverting <90 days after infection = proportion seroconverting <90 days after infection
MOHLTC, Laboratories Branch, IMC – 2001
Determination of Psce and true Determination of Psce and true incidence using empirical dataincidence using empirical data
• True incidence density is unknownTrue incidence density is unknown• Can determine value of Psce and true Can determine value of Psce and true
incidence density by varying values through incidence density by varying values through range to fit to measured incidence densityrange to fit to measured incidence density
• Repeated at discrete values of window Repeated at discrete values of window period and modeled incidence is fit to period and modeled incidence is fit to observed incidence by selecting values of observed incidence by selecting values of true incidence density and Psce that true incidence density and Psce that minimize the differenceminimize the difference
• Minimize sum of squares of difference Minimize sum of squares of difference (goodness-of-fit)(goodness-of-fit)
MOHLTC, Laboratories Branch, IMC – 2001
Determination of Psce and true Determination of Psce and true incidence using empirical dataincidence using empirical data
• Programmed software in APLProgrammed software in APL• Vary Psce from 0% to 50% in increments of Vary Psce from 0% to 50% in increments of
0.1%0.1%• Vary true HIV incidence from 0 to 20 per 100 Vary true HIV incidence from 0 to 20 per 100
person-years in increments of 0.01person-years in increments of 0.01• Program selects values of Psce and incidence Program selects values of Psce and incidence
for which sum of squares of difference for which sum of squares of difference between observed and modeled incidence is between observed and modeled incidence is lowestlowest
MOHLTC, Laboratories Branch, IMC – 2001
Psce, measured and true HIV incidence by Psce, measured and true HIV incidence by year and health region among MSM, 2001-year and health region among MSM, 2001-03 03
1.18
1.56
1.05
0.98
1.80
1.41
0.60
0.56
0.59
1.891.89
2.012.01
1.801.80
2.462.46
1.871.87
2.062.06
0.780.78
0.560.56
0.580.58
9.1%9.1%
4.6%4.6%
10.5%10.5%
21.1%21.1%
0.4%0.4%
6.1%6.1%
3.9%3.9%
0.0%0.0%
14.7%14.7%
TORONTOTORONTO
20012001
20022002
20032003
OTTAWAOTTAWA
20012001
20022002
20032003
OTHEROTHER
20012001
20022002
20032003
TrueTrue
incidenceincidence
MeasuredMeasured
incidenceincidencePscePsce
MOHLTC, Laboratories Branch, IMC – 2001
Crude and adjusted HIV incidence Crude and adjusted HIV incidence among MSM and IDU, Toronto, 1999-among MSM and IDU, Toronto, 1999-20032003
0.00
0.50
1.00
1.50
2.00
2.50
3.00
1999 2000 2001 2002 2003Period
HIV
in
cid
en
ce (
per
100 p
y)
MSMmeasMSMtrueIDUmeasIDUtrue
MOHLTC, Laboratories Branch, IMC – 2001
Crude and adjusted HIV incidence Crude and adjusted HIV incidence among MSM and IDU, Ottawa, 1999-among MSM and IDU, Ottawa, 1999-20032003
0.00
0.50
1.00
1.50
2.00
2.50
3.00
1999 2000 2001 2002 2003
Period
HIV
inci
dence
(p
er
100 p
y)
MSMmeasMSMtrueIDUmeasIDUtrue
MOHLTC, Laboratories Branch, IMC – 2001
Summary of findingsSummary of findingsAdjustment of HIV incidenceAdjustment of HIV incidence
• Goodness-of-fit approach allowed Goodness-of-fit approach allowed adjustment to remove testing biasadjustment to remove testing bias
• Modelled HIV incidence fit very well to Modelled HIV incidence fit very well to observed HIV incidenceobserved HIV incidence
• Data using specimens from diagnostic HIV Data using specimens from diagnostic HIV testing should be presented with both crude testing should be presented with both crude and adjusted values of HIV incidenceand adjusted values of HIV incidence
MOHLTC, Laboratories Branch, IMC – 2001
AcknowledgementsAcknowledgements
• Ontario Laboratory Enhancement Study Ontario Laboratory Enhancement Study fundingfunding• Ontario HIV Treatment NetworkOntario HIV Treatment Network• Centre for Infectious Disease Prevention Centre for Infectious Disease Prevention
and Control, Health Canada and Control, Health Canada • Neil Hershfield developed custom software Neil Hershfield developed custom software
to adjust HIV incidenceto adjust HIV incidence