No answer should be given here, because one would have to start with philosophical discussions of the cognitive process ... and continue with the principles of medical action. The terms illness, diagnosis, diagnostics, test, diagnostic process etc. should be defined. - The biometrician can make things easy for him: he sees the diagnostic measure as a means, an a priori probability for the correctness of the presumption that a patient suffers from an illness, in a (possible) higher a posteriori probability transform.
In order to describe the diagnosis, it is interpreted as a sequence of binary individual decisions. These individual distinctions use diagnostic tests that are supposed to decide between two conditions: disease present / not present. Accordingly, the test result is a yes / no statement: sick (= positive) / not sick (= negative). For tests with quantitative results, such as B. for laboratory values, the conversion into such a binary statement takes place with a cut-off value (
Cut-off point).
From this, a four-field table can be created that compares the patient's true condition (reference standard, gold standard) and the test result (examined test = index test):
On this table, the ratios of the individual cells to the sums can be formed in columns and rows.
First we can give the prevalence: D + / ([D +] + [D-]) = (TP + TN) / (TP + FP + TN + FN).
Diagnostic quality is always reported as a pair of two values: {sensitivity, specificity}, or {PPV, NPV} or {DLR +, DLR-}.
Summary measures that only use one value (e.g. sum of sensitivity and specificity, Youden index, efficiency, ...) do not and do not adequately reflect the quality of a test not suitableto describe the diagnostic goodness. If only a number is reported (eg only the negative predictive value, or an "accuracy of 95%", etc., this information is not sufficient to assess the quality of a diagnostic test. I personally go in such cases assume that an unfavorable property of the test should be kept secret.
The sensitivity determines the proportion of the patients who are correctly recognized as positive in all patients TP / D + = TP / (TP + FN), the specificity the proportion of the patients who are recognized as really negative in the non-patients TN / D- = TN / (TN + FP ). The grades can also be formulated statistically as conditional probabilities: the sensitivity is the conditional probability of having a really positive test result given the disease [this is noted: P (T + | D +)], the specificity is the conditional probability of being present given a really negative test result, the non-disease P (T- | D-). Sensitivity and specificity are the quantities that developers and manufacturers can use when evaluating their diagnostic tests. Instead of sensitivity, the correct-positive rate TPF = sensitivity and false-positive rate = 1-specificity are also sometimes given. Since sensitivity and specificity are determined within the columns of the table reported above, they do not depend on the prevalence.
The predictive values (line-by-line consideration), on the other hand, consider the probabilities that the patient actually has the condition that the test indicates (positive predictive value PPV: TP / (TP + FP), negative predictive value NPV: TN / (TN + FN) also formulate the predictive values statistically as conditional probabilities: the PPV is the conditional probability of a disease being given a positive test result P (D + | T +) [note the reversed order of T + and D + compared to sensitivity], the NPV is that conditional probability for the existence of a non-disease given a negative test result P (D- | D-). The predicted values thus describe the view of the doctor (or the patient) to whom the test result is available: I have a positive test result: how high is it Probability that I am actually ill? The PPV answers this.Physician and patient can test the test result with the PPV or the NPV assess its relevance.