Estimates of HIV incidence
based on the detuned assay
may be strongly biased
Robert S. Remis, Robert W.H. Palmer, Janet Raboud
Ontario HIV Epidemiologic Monitoring Unit
Department of Public Health Sciences and
Division of Infectious Diseases, University of Toronto
14th International Conference on AIDS
Barcelona, Spain, July 7-12, 2002
Abstract MoPeC3457
Laboratories Branch, MOHLTC, IMC – 2002
Background
• Detuned assay an epidemiologist’s
“philosopher’s stone”?
• Transforms prevalence into incidence
• Incidence = #S/LS / Tests * WP
Laboratories Branch, MOHLTC, IMC – 2002
Background
• Incidence by detuned assay in prevalence studies may be
biased;
• Bias compounded when sera from diagnostic testing;
• Three possible sources of bias:
1. persons who test not representative
2. HIV testing frequency may vary by HIV risk
3. persons who experience isolated high risk exposures or
seroconversion illness more likely to test immediately
following
Laboratories Branch, MOHLTC, IMC – 2002
Methods
• Examined bias in incidence as a function of :
1) incidence density, 2) inter-test interval, 3) window period of
detuned assay and 4) study duration
• Parameter values:
• incidence: literature review, cohort studies in Montreal and
Vancouver, epidemiologic modelling of HIV among MSM in
Ontario
• inter-test interval: #tests MSM in Ontario
• window period: OT Vironostika assay
• study duration: likely scenarios
• Output was bias as expressed by “observed” incidence density
compared to “true” incidence density (ratio of Iest / Itrue)
Laboratories Branch, MOHLTC, IMC – 2002
Methods
Determination incorporated:
• Inter-test interval a function of incidence density:
• five strata with scaled incidence density
• inter-test interval varied inversely with incidence density
• Seroconversion illness effect (SCE)
• increased probability of HIV testing due to episodic high
risk and symptoms of primary HIV infection
• defined as additional proportion of newly infected persons
testing <90 days following infection
• values compared of initial outputs of model with proportion
HIV tests from MSM discordant, 1999-2001
Laboratories Branch, MOHLTC, IMC – 2002
Methods – Parameter values used
Units
Incidence density (Itrue)
/100 py
Base case
Minimum
Maximum
1.0
0.25
5.0
Inter-test interval,
mean (Ttest)
days
1,095
365
1,825
Window period,
detuned (Twin)
days
133
7
336
Study duration (Tobs)
days
365
90
1,095
Base case
SCE
(proportion <= 90 days)
Ttest - Itrue interaction
Laboratories Branch, MOHLTC, IMC – 2002
Medium
High
Proportion
0.00
0.20
0.40
Slope
Non
Low
High
Bias (Iest /Itrue) as a function of Twin–Itrue interaction
and SCE varied through range of incidence
4.0
No Ttest-Itrue, SCE = 0.0
No Ttest-Itrue, SCE = 0.2
3.5
No Ttest-Itrue, SCE = 0.4
Low Ttest-Itrue, SCE = 0.0
Low Ttest-Itrue, SCE = 0.2
Bias
3.0
2.5
Low Ttest-Itrue, SCE = 0.4
High Ttest-Itrue, SCE = 0.0
High Ttest-Itrue, SCE = 0.2
2.0
High Ttest-Itrue, SCE = 0.4
1.5
1.0
0.5
Incidence (per 100 person-years)
Laboratories Branch, MOHLTC, IMC – 2002
5.00
4.75
4.50
4.25
4.00
3.75
3.50
3.25
3.00
2.75
2.50
2.25
2.00
1.75
1.50
1.25
1.00
0.75
0.50
0.25
0.0
Bias (Iest /Itrue) as a function of Twin–Itrue interaction
and SCE varied through range of inter-test
interval
No Ttest-Itrue, SCE = 0.0
No Ttest-Itrue, SCE = 0.2
5.0
No Ttest-Itrue, SCE = 0.4
Low Ttest-Itrue, SCE = 0.0
Low Ttest-Itrue, SCE = 0.2
4.0
Low Ttest-Itrue, SCE = 0.4
High Ttest-Itrue, SCE = 0.0
High Ttest-Itrue, SCE = 0.2
3.0
High Ttest-Itrue, SCE = 0.4
2.0
Inter-test interval (days)
Laboratories Branch, MOHLTC, IMC – 2002
1,825
1,748
1,671
1,595
1,518
1,441
1,364
1,287
1,210
1,133
1,057
980
903
826
749
672
596
519
0.0
442
1.0
365
Bias
6.0
Bias (Iest /Itrue) as a function of Twin–Itrue interaction
and SCE varied through range of window period
6.0
No Ttest-Itrue, SCE = 0.0
No Ttest-Itrue, SCE = 0.2
No Ttest-Itrue, SCE = 0.4
Low Ttest-Itrue, SCE = 0.0
Low Ttest-Itrue, SCE = 0.2
4.0
Low Ttest-Itrue, SCE = 0.4
High Ttest-Itrue, SCE = 0.0
High Ttest-Itrue, SCE = 0.2
3.0
High Ttest-Itrue, SCE = 0.4
2.0
Window period (days)
Laboratories Branch, MOHLTC, IMC – 2002
336
322
308
294
280
266
252
238
224
210
196
182
168
154
140
126
112
98
0.0
84
1.0
70
Bias
5.0
Bias (Iest /Itrue) as a function of Twin–Itrue interaction
and SCE varied through range of study duration
4.0
No Ttest-Itrue, SCE = 0.0
No Ttest-Itrue, SCE = 0.2
3.5
No Ttest-Itrue, SCE = 0.4
Low Ttest-Itrue, SCE = 0.0
Low Ttest-Itrue, SCE = 0.2
2.5
Low Ttest-Itrue, SCE = 0.4
High Ttest-Itrue, SCE = 0.0
High Ttest-Itrue, SCE = 0.2
2.0
High Ttest-Itrue, SCE = 0.4
1.5
1.0
Study duration (days)
Laboratories Branch, MOHLTC, IMC – 2002
1,095
1,042
989
936
883
831
778
725
672
619
566
513
460
407
355
302
249
196
0.0
143
0.5
90
Bias
3.0
Mean values of bias by SCE in presence of no, low and
high interaction between inter-test interval and incidence
density
Ttest – Itrue = None
Parameter
values
SCE
0.00
0.20
0.40
Minimum
1.0
1.1
1.3
Mean
1.0
2.1
3.2
Maximum
1.0
4.0
6.9
Minimum
1.3
1.4
1.6
Mean
1.3
2.4
3.5
Maximum
1.3
4.2
7.2
Minimum
1.4
1.5
1.6
Mean
1.4
2.5
3.6
Maximum
1.4
4.3
7.3
Ttest - Itrue = Low
Ttest - Itrue = High
Laboratories Branch, MOHLTC, IMC – 2002
Summary of observations
• No bias in absence of SCE and Ttest-Itrue interaction
• SCE imore important source of bias than Ttest-Itrue
interaction
• Biases as high as 7.3 under certain circumstances
(high SCE and Ttest-Itrue interaction)
• Biases may still be considerable (e.g. 2-3) at plausible
parameter values
Laboratories Branch, MOHLTC, IMC – 2002
Interpretation
• Model used plausible values for MSM in Ontario
• Good data not yet available for several parameters,
especially Ttest-Itrue interaction, inter-test interval and
SCE
• Biases may be different in other study populations;
however, range of values probably spans most other
situations
• Did not investigate bias for different values of Tsce
Laboratories Branch, MOHLTC, IMC – 2002
Conclusions
• Incidence density by detuned assay can be seriously
overestimated;
• Biases as high as 7-fold in some situations,
considerable even at plausible values of parameters;
• Bias mostly from SCE also if persons at increased HIV
risk test more frequently; both likely to occur among
MSM and possibly in other at-risk populations;
• Incidence estimates using detuned assay must be
interpreted with considerable caution;
• Techniques to “standardise” incidence estimates from
the detuned assay under development and could be
extremely useful.
Laboratories Branch, MOHLTC, IMC – 2002
Acknowledgements
• Instructional Media Centre, Laboratories Branch,
Ontario Ministry of Health and Long-Term Care for
preparation of the poster;
• Frank McGee, coordinator, AIDS Bureau, Ontario
Ministry of Health and Long-Term Care for core
funding;
• Carol Major, formerly of the Laboratories Branch,
Ontario Ministry of Health and Long-Term Care for
advice and collaboration.
Laboratories Branch, MOHLTC, IMC – 2002