Case control studies

Report
Principles of diagnostic sampling
– the “ bead game”
Pilot training school in PRRS diagnostics, 2012
Zagreb, Croatia
Dr. Derald Holtkamp
September 25, 2012
Iowa State University College of Veterinary Medicine
Exercises
1. Detecting one or more positive pigs in an expected
NEGATIVE population
–
Role of sample size and target prevalence
2. False positive test results in a truly negative
population
3. Estimating Sensitivity (SE) & Specificity (SP) by testing
"known status" samples (if time permits)
Iowa State University College of Veterinary Medicine
Exercises
1. Detecting one or more positive pigs in an expected
NEGATIVE population
–
Role of sample size and target prevalence
2. False positive test results in a truly negative
population
3. Estimating Sensitivity (SE) & Specificity (SP) by testing
"known status" samples
Iowa State University College of Veterinary Medicine
What you need to know to determine
the sample size needed to find at least
one positive animal in a herd
1. Target prevalence (TP)
2. Confidence level (CL)
3. Population size (PS)
Sample size needed = f(TP, CL, PS)
– Cannon RM, Roe RT. Livestock disease surveys: A field manual
for veterinarians. AGPS, Canberra, 1982.
Iowa State University College of Veterinary Medicine
Table of sample sizes needed based on
Cannon and Roe
Population Size (Detecting One or More Positives)
Prevalence Confidence
Estimate
Level
>1%
70%
80%
90%
95%
99%
>10%
70%
80%
90%
95%
99%
>25%
70%
80%
90%
95%
99%
>50%
70%
80%
90%
95%
99%
100
71
81
91
96
100
12
16
21
26
37
6
7
9
11
16
3
4
5
6
8
Iowa State University College of Veterinary Medicine
1000
114
149
206
259
369
13
17
23
30
44
6
7
10
12
17
3
4
5
6
8
4000
120
158
224
289
434
13
17
23
30
45
6
7
10
12
18
3
4
5
6
8
10000
121
160
228
295
449
13
17
23
30
45
6
7
10
12
18
3
4
5
6
8
Exercises
1. Detecting one or more positive pigs in an expected
NEGATIVE population
–
Role of sample size and target prevalence
2. False positive test results in a truly negative
population
3. Estimating Sensitivity (SE) & Specificity (SP) by testing
"known status" samples
Iowa State University College of Veterinary Medicine
Slide courtesy of D. Polson
Iowa State University College of Veterinary Medicine
Exercises
1. Detecting one or more positive pigs in an expected
NEGATIVE population
–
Role of sample size and target prevalence
2. False positive test results in a truly negative
population
3. Estimating Sensitivity (SE) & Specificity (SP) by testing
"known status" samples
Iowa State University College of Veterinary Medicine
Properties and Definitions of Diagnostic Tests
Test Positive
Test Negative
TRUE Positive
TRUE Negative
a
(true positive)
c
(false negative)
b
(false positive)
d
(true negative)
Sensitivity = a / ( a + c )
False Negatives = c / ( a + c )
Specificity = d / ( b + d )
False Positives = b / ( b + d )
Positive Predictive Value = a / ( a + b )
Negative Predictive Value = d / ( c + d )
Accuracy = ( a + d ) / ( a + b + c + d )
True Prevalence = ( a + c ) / ( a + b + c + d )
Assay Prevalence = ( a + b ) / ( a + b + c + d )
Prevalence Estimate = a / ( a + b + c + d )
Likelihood Ratio of a Positive Test = Sensitivity / ( 1 - Specificity )
Likelihood Ratio of a Negative Test = ( 1 - Sensitivity ) / Specificity
Sensitivity is the percent of TRUE positives that are TEST positives.
Specificity is the percent of TRUE negatives that are TEST negatives.
False Positive Rate is the percent of FALSE positives that are TEST positives.
False Negative Rate is the percent of FALSE negatives that are TEST negatives.
Positive predictive value (PPV) is the probability of a test positive being a true positive.
Negative predictive value (NPV) is the probability of a test negative being a true negative.
Accuracy is the percent of all test results that agree with the true status of the sample.
True Prevalence is the percent of all samples that are true positives.
Prevalence Estimate is the percent of all samples that are tested positive.
Likelihood Ratio of a Positive Test is the odds of a positive test result in a true positive sample.
Likelihood Ratio of a Negative Test is the odds of a negative test result in a true negative
sample.
Sensitivity and specificity are absolute properties of a test. PPV and NPV are relative, and vary
depending on the true prevalence of positive samples. As the prevalence decreases, PPV
decreases and NPV increases.
Iowa State University College of Veterinary Medicine
Acknowledgements
• Dale Polson, Boehringer Ingelheim Vetmedica Inc., U.S
Iowa State University College of Veterinary Medicine

similar documents