The Yield of a Screening Program
One factor that influences the feasibility of a screening program is the yield, i.e., the number of cases detected. This can be estimated from the positive predictive value.
Sensitivity and specificity are characteristics of the test and are only influenced by the test characteristics and the criterion of positivity that is selected. In contrast, the positive predictive value of a test, or the yield, is very dependent on the prevalence of the disease in the population being tested. The higher the prevalence of disease is in the population being screened, the higher the positive predictive values (and the yield). Consequently, the primary means of increasing the yield of a screening program is to target the test to groups of people who are at higher risk of developing the disease.
To illustrate the effect of prevalence on positive predictive value, consider the yield that would be obtained for HIV testing in three different settings. Serological testing for HIV is extremely sensitive (100%) and specific (99.5%), but the positive predictive value of HIV testing will vary markedly depending on the prevalence of pre-clinical disease in the population being tested. The examples below show how drastically the predicative value varies among three groups of test subjects.
These three scenarios all illustrate the consequences of HIV testing using a test that is 100% sensitive and 99.5% specific. All three show the effects of screening 100,000 subjects. The only thing that is different among these three populations is the prevalence of previously undiagnosed HIV.
Screening Program #1
The 1st scenario illustrates the yield if the screening program were conducted in female blood donors, in whom the prevalence of disease is only 0.01%. Even with 100% sensitivity and 95% specificity, the positive predictive value (yield) is only 1.9%.
Table - HIV Screening in a Population With HIV Prevalence of Female Blood Donors
|
|
Really HIV+ |
Really HIV- |
Row Totals |
|---|---|---|---|
|
Screen Test + |
10 |
510 |
520 |
|
Screen Test - |
0 |
99,480 |
99,480 |
|
Column Totals |
10 |
99,990 |
100,000 |
Prevalence is 10/100,000 = 0.01%
Positive predictive value = 10/520=0.019, or 1.9%
Screening Program #2
The 2nd scenario illustrates the yield if the screening program were conducted in males in a clinic for sexually transmitted infections, in whom the prevalence of disease is 4%. With the same sensitivity and specificity, the positive predictive value (yield) is 89%.
Table - HIV Screening in a Population of Males Visiting Clinics for Sexual Transmitted Diseases
|
|
Really HIV+ |
Really HIV- |
Row Totals |
|---|---|---|---|
|
Screen Test + |
4,000 |
480 |
4,480 |
|
Screen Test - |
0 |
95,520 |
95,520 |
|
Column Totals |
4,000 |
96,000 |
100,000 |
Prevalence in males visiting clinics for sexual transmitted disease = 4,000/100,000=0.04, or 4%
Positive predictive value = 4000/4480 = 0.83, or 83%
Screening Program #3
Table - HIV Screening in a Population of Intravenous Drug Users
|
|
Really HIV+ |
Really HIV- |
Row Totals |
|---|---|---|---|
|
Screen Test + |
20,000 |
400 |
20,400 |
|
Screen Test - |
0 |
79,600 |
79,600 |
|
|
20,000 |
80,000 |
100,000 |
Prevalence of HIV in these IV Drug Users = 20,000/100,000 = 0.20, or 20%
Positive predictive value = 20,000/20,400 = 0.98, or 98%
This 3rd scenario illustrates the yield if the screening program were conducted in users of intravenous drugs, in whom the prevalence of disease is 20%. With the same sensitivity and specificity, the positive predictive value (yield) is 98%.
What these three scenarios illustrate is that if you have limited resources for screening, and you want to get the most "bang for the buck," target a subset of the population that is likely to have a higher prevalence of disease, and don't screen subsets who are very unlikely to be diseased.
[From Richard M Hoffman, Frank D Gilliland, et al.: Prostate-specific antigen testing accuracy in community practice. BMC Family Practice 2002, 3:19]
"Methods: PSA testing results were compared with a reference standard of prostate biopsy. Subjects were 2,620 men 40 years and older undergoing (PSA) testing and biopsy from 1/1/95 through 12/31/98 in the Albuquerque, New Mexico metropolitan area. Diagnostic measures included the area under the receiver-operating characteristic curve, sensitivity, specificity, and likelihood ratios.
Results: Cancer was detected in 930 subjects (35%). The area under the ROC curve was 0.67 and the PSA cut point of 4 ng/ml had a sensitivity of 86% and a specificity of 33%."
Question: What was the positive predictive value in this study? Hint: You have to use the information provided to piece together the complete 2x2 table; then compute the PPV. See if you can do this before looking at the answer.
Optional
Dr. David Felson is a Professor of Medicine in the Boston University School of Medicine, and he teaches a course in Clinical Epidemiology at the BU School of Public Health. In the video below, he discusses serial and parallel diagnostic testing.
