Characterization and validation of diagnostic methods.
Diagnosis is defined as the determination of disease, but not as the determination of the signs and symptoms thereof. The use of modern diagnostic methodology in the clinic is hampered by cost considerations and by the still widespread belief that, e.g. caries lesions and periodontal breakdown are irreversible processes that need to be detected and treated invasively as early as possible, their measurement thus being irrelevant. Modern instrumental and quantitative methods allow early detection and introduction of noninvasive preventive measures to control the development of the disease. Such methods are also very beneficial in clinical research as they may describe the speed of progress or regress of disease. In epidemiology, such methods reduce the classical problem of calibration of observers. Repeatability, reproducibility, accuracy and validity are defined as method-characterizing quantities, for which examples are given. To express the validity of quantitative methods compared with a quantitative gold standard, the use of scatter plots and correlation and regression methods is suggested. Validation of a dichotomous method with a dichotomous gold standard in terms of sensitivity and specificity is discussed. To validate a quantitative method with a dichotomous gold standard, the receiver operating characteristic curve is suggested, with the requirement that the cutoff value should be determined in relation to the use of the method. However, preferably a quantitative method should not be reduced to a dichotomous one by using a cutoff value, but instead all available information should be used by the diagnostician. It is argued that the use of a secondary standard instead of the accepted gold standard usually leads to inadequate results, even when the validity of the secondary standard is known. Finally, it is argued that the choice of a gold standard is a matter of reasoning and weighing of arguments and not of following a prescribed procedure.