Predictive Power Methods May Be More Helpful for Making a Diagnosis than Sensitivity and Specificity
Individual symptoms of child psychiatric disorders have rarely been systematically examined to determine which operate best as inclusion criteria (ruling in a disorder by their presence) and which as exclusion criteria (ruling out a disorder by their absence). The past use of sensitivity and specificity statistics to validate the criteria for different diagnoses has been misleading in certain situations. These problems are particularly salient in the classification of disruptive behavior disorders (including ADHD and CD) in DMS-IH-R, a polythetic diagnostic system in which any combination of a set ofsymptoms can be employed for diagnosis as long as the requisite number has been achieved. In this kind of "confirmatory" diagnostic system (the more symptoms, the more likely the diagnosis), symptoms that are most descriptive of a disorder (i.e., many of the children with the disorder have the symptom) may not be most efficient in the diagnosis of that disorder (i.e., many of the children with the symptoms have the disorder). In the case of attention-deficit hyperactivity disorder and conduct disorder, symptoms were identified that were associated with one specific diagnosis, but actually had greater implications for ruling out the other diagnosis. For example, the ADD symptom "doesn't listen" was found to be most useful as an exclusion criterion for a CD diagnosis. Alternatively, the polythetic DSM-III-R system is ideally suited to the application of predictive power methods. Efficient inclusion criteria can be determined from positive predictive power (PPP), and exclusion criteria can be determined by an examination of negative predictive power (NPP) rates. Item analysis of PPP and NPP rates can be used, for example, to establish the efficiency of symptoms in the differential diagnosis of ADD and CD. However, the DSM-III-R criteria for ADHD and CD address only inclusion criteria and make no provision for exclusion criteria. Until clinicians make accommodations for inclusion and exclusion criteria (i.e., until predictive power methods are employed), clinical choices for the child psychopharmacological treatment of disruptive behavior disorders will be based on sensitivity/specificity thinking and limited by our "confirmatory" diagnostic practices.
Digital Object Identifier (DOI)
Landau, Steven; Milich, Richard; and Widiger, Thomas A., "Predictive Power Methods May Be More Helpful for Making a Diagnosis than Sensitivity and Specificity" (1991). Psychology Faculty Publications. 108.