Validation of non-western blot based algorithms for HIV testing

Validation of non-western blot
based algorithms for HIV testing
Susan Best
NRL, Melbourne, Australia
Outline
Reasons for validating non-WB algorithms
Brief how-to
What we can do to minimise false positive
classifications
Types of Tests - Serology
Confirmation using immunoblot
Disadvantages
Advantages
Reactivity to each
Subjective reading
protein can be visualised
Indeterminate reactions
Auxiliary reactions
Labour intensive
IgG control
Expertise
Well established
$$$
interpretation criteria
Increased cost with
Stage in disease
small numbers
Seroconversion
CDC algorithm for HIV diagnosis
Figure 1. Laboratory Testing Algorithm for the Diagnosis of HIV Infection[CDC HIV Testing]
WHO HIV testing strategies
Prevalence > 5%
Prevalence < 5%
http://www.who.int/diagnostics_laboratory/evaluations/hiv/131107_hiv_assays17_final.pdf
Positive predictive value
The likelihood of a sample identified as
reactive by a test being truly POSITIVE for
the analyte in question
Prevalence and predictive value
Example: sensitivity 100%; specificity 99.0%
Population: 100,000
Prevalence
(%)
TRUE
Positives
FALSE
Positives
Ratio of
True:False*
PPV (%)
20
20,000
800
25:1
96.1
5
5,000
950
5:1
84.0
0.5
500
995
1:2
33.4
0.1
100
999
1:10
5.0
0.01
10
1000
1:100
1.0
0.002
2
1000
1:500
0.2
* Rounded
PPV of sequential testing
Assay 1
Assay 2
Assays 1 & 2
(in sequence)
Sensitivity
100%
100%
Specificity
99%
99%
Prevalence
PPV
PPV
PPV
0.2%
16.6%
16.6%
95.2%
2.0%
67.1%
67.1%
99.5%
20.0%
96.1%
96.1%
99.9%
Example: common false reactivity
Test 1: 100% sensitive; 99% specific
Test 2: 100% sensitive: 99% specific
Prevalence: 0.5%
Test 1 and 2 share 1% common false reactivity
# tested
# reactive
# true
reactive
#false
reactive
Test 1
10,000
150
50
100
Test 2
150
51
50
1
• Using a combination of 2 tests with these characteristics means
that, in 10,000 tests, 1 result will be reported incorrectly as
positive
• PPV of algorithm: 98%
Validating non-WB algorithms
PRINCIPLES
Conduct centrally
Choose from well pedigreed tests
Independently assessed
WHO pre-qualified; USAID list; widely used and cited;
Screening assay: choose high sensitivity test
Supplemental assays: different antigen sources;
good specificity
Validating non-WB algorithms (cont’d)
More assays than required
Test negative and false positive samples
in each
For significance, numbers could be large
Compare the results for common false
reactivity
Validating non-WB algorithms (cont’d)
True negative sample
Test 1
Falsely reactive
Test 2
R
Incorrect
result
Test 3
N
Test 4
Test 5
N
R
Incorrect
result
False Positive Reactivity
Sample ID
Screening
test
z369911
A/1058/08
False Postiive
z370966
False Positive
z370967
False Positive
z372341
False Positive
z371067
False Positive
V512
Negative
V583
Test A
Test B
Test C
Test D
Test E
Negative
Positive
Negative
Negative
Negative
Positive
Negative
Negative
Positive
Negative
Negative
Negative
Negative
Negative
Positive
Positive
Positive
Negative
Negative
Negative
Negative
Negative
Positive
Positive
Negative
Negative
Negative
Negative
Negative
Negative
Negative
Negative
Positive
Negative
V860
Negative
Negative
Negative
Negative
Positive
Negative
z371780
Negative
Negative
Negative
Negative
Negative
Positive
Validating non-WB algorithms
True negative sample
CONSIDERED:
 Not practical
Test 1
Falsely reactive
Too expensive
3
Test 4
Test 2
Too
time Test
consuming
Test 5
R
R
Incorrect
result
Incorrect
result
Unnecessary
Risk of false positive
misclassification
MINIMISED RISK
Increasing
risk
Stringent IVD regulatory
environment
Independent assessment
of assays
Validated algorithms
Functioning QMS in
testing sites
Trained operators
HIGH RISK
Weak / no IVD regulation
Weak / no criteria for
rejection of assays
Many assays available
Weak / no understanding
of quality at test sites
Insufficient operator
training
Algorithms for under-regulated
environments
Choose widely used, published algorithms
Similar prevalence; similar region
Use tests in same sequence
National / regional false positive sample banks
Monitor
E.g. Pacific algorithm: Determine / Unigold / Insti
All reactives, ~ 250 negatives confirmed at NRL
Results Obtained on Ongoing
Referred Samples
Reactive Samples
HIV Status
Determine (+)
UniGold (+)
Insti (+)
NRL (+)
HIV -
54
0
6
0
HIV +
8
8
8
8
Negative Samples
267 negative in Determine negative in gold standard algorithm at NRL
Conclusion
- No false results reported from the algorithm. Algorithm delivered
correct results when used in the field
- Three assay algorithm avoided false positive results
- Two assay algorithm would produce significant false positive reports
Summary
Risk of false positive reports from unvalidated
algorithms
Not restricted to HIV
Degree of risk is multi-factorial
Experimental validation, although best
practice, is resource intensive
Regulate; choose high quality tests; train;
monitor