Statistics Is for Providing Quality Criteria for Diagnostic Tests, General Remarks
Not the trials, but the diagnostic tests are the heart of evidence-based medicine.
The STARD group launched in 2003 quality criteria for diagnostic tests (STARD = standards for reporting diagnostic accuracy).
Clin Chem 2003; 49: 7-18
The STARD Statement for Reporting Studies of Diagnostic Accuracy: Explanation and Elaboration
Patrick M. Bossuyt1,a, Johannes B. Reitsmal, David E. Bruns2, 3 et al
Department of Clinical Epidemiology and Biostatistics, Academic Medical Center—University of Amsterdam, PO Box 22700, 1100 DE Amsterdam, The Netherlands.
The main characteristics of diagnostic tests can be described with little words. Some synonyms are also given:
- (1) valid=accurate,
- (2) reproducible=reliable,
- (3) precise=robust=containing a small spread in the data.
Intervention-studies are usually
- - well paid for,
- - published high impact journals,
- - providing excellent career perspectives for investigators
However, the evaluation of diagnostic tests/research is usually
- - not well paid for,
- - difficult to publish,
- - giving a poor career perspective,
- - post hoc, ergo propter hoc, performed in a sloppy way.
Yet, intervention studies are impossible without diagnostic tests. And diagnostic tests are the only real basis of evidence-based medicine. Young investigators are often requested to test diagnostic tests. How to do so?
First, assess validity.
A test that shows who has the disease and who has not is valid.
Second, assess reproducibility.
A test that produces the same result the second time is reproducible (= reliable). Third, assess precision.
A test that produces little spread in the outcome data is precise (= robust). Diagnostic tests are, traditionally, classified as
(1) qualitative (the “yes/no” tests or binary tests).
An example is “an elevated erythrocyte sedimentation rate above 32 mm for diagnosis of pneumonia”.
(2) quantitative (outcome values have a continuous character).
An example is “the echographical cardiac output measurements around 5 liter/ min”.
A table of methods for assessing high quality diagnostic tests is given underneath.
ROC = receiver operating characteristic, SD = standard deviation, SE = standard error, a = intercept of linear regression, b = regression coefficient of linear regression