| |
In addition to estimating
a patient's pre-test probability of disease, it is important to
estimate the sensitivity and specificity of diagnostic tests.
Ideally, any study of the accuracy of a diagnostic test should
be prospective, with each patient following a standard protocol.
Steps in estimating
the sensitivity and specificity of diagnostic tests are as follows:
Identify the study population - The population should
contain all those who need the index test (the test that is
the object of study). In a day-to-day clinical setting, the
reasons for obtaining the index test will vary from patient
to patient. Physicians will differ in their threshold for obtaining
the test, and the study population is often heterogeneous and
difficult to characterize so that the reader of an article can
decide if the results apply to the next patient in her practice.
Perform the index test - The authors must use a well-standardized
procedure to perform the index test and must describe the procedure
carefully. The person who interprets the index test should follow
written criteria for classifying the results. Several individuals
should interpret the results independently and resolve any differences
in interpretation by discussing each such case. The person who
interprets the index test should not know the results of the
gold standard test, lest the results of the gold standard test
influence the interpretation in the case of an equivocal result
on the index test.
Perform the gold standard test - The gold standard
test should indicate the true state of the patient. Because
the gold standard test is typically costly, painful, and risky,
a normal result on the index test often influences the clinician
to forgo the gold standard test. Each person who interprets
the gold standard test should adhere to written criteria when
classifying the results. Several individuals should interpret
the results independently and resolve any differences in interpretation.
The person who interprets the gold standard test should not
know the results of the index test.
Avoid verification bias - The biggest source of error
in studies of diagnostic tests is the natural tendency to avoid
doing a costly, painful, risky gold standard test in those who
have a negative result on the index test. This selection bias
reduces the number of patients with negative results on the
index test and leads to underestimating the true-negative and
false-negative rate of the test. The best way to avoid verification
bias is to expand the definition of the gold standard test to
include careful clinical follow-up to detect diseased patients
who had a negative index test result.
Record the results - The usual way to express the results
of a study is a 2x2 table (see table below). The columns in
the table represent the results of the gold standard test. The
rows represent the results of the index test. In this 2.x2 table,
p[Result if diseased] = 0.30 (30/100) and p[Result if no disease]
= 0.1 (30/300).
| Table
1.7.1: Hypothetical typical 2 by 2 table in a study
of test performance
|
|---|
| Test
result
| No.
Patients with disease present
| No.
Patients with disease absent
| Totals
|
|---|
|
No. Patients withabnormal test result
| 30
(true-positives) | 30
(false-positives) | 60 |
|---|
| No.
Patients withnormal test result
| 70
(false-negatives) | 270
(true-negatives) | 340
|
|---|
| Totals
| 100
| 300
| 400 |
|---|
|
|