SYMPTOM COUNTS AND DIAGNOSTIC ALGORITHMS AS MEASURES OF 5 COMMON PSYCHIATRIC-DISORDERS IN CHILDREN

Instruments of psychiatric evaluation have followed two very different methodological traditions in combining symptom data. Researchers in t he clinical tradition have designed diagnostic computer algorithms, wh ile others, working in the psychometric tradition, have constructed ad ditive scales. Dichotomized scales can also be used as case-identifica tion instruments. This paper uses a common set of symptom data to comp are these two approaches in a sample of children drawn from the genera l population of Puerto Rico. Board-qualified child psychiatrists admin istered the Diagnostic Interview Schedule for Children (DISC), directl y recording the responses of both the child and one parent. They also made diagnoses based on their own clinical judgment. For each of five common childhood disorders, the DISC responses were used to compute a diagnostic algorithm and a simple count of its constituent symptoms. T he clinical diagnosis was used as a standard of reference in comparing the algorithms and the symptom counts. Both the clinical diagnoses an d the algorithms were based on DSM-III criteria. Each symptom count wa s found to have a unidimensional structure and to constitute an intern ally consistent scale. For three disorders, sensitivity and specificit y of the symptom count, when dichotomized at an appropriate threshold, were almost identical to the algorithm's sensitivity and specificity in identifying clinically diagnosed cases. For the remaining two disor ders, the algorithm already took the form of a symptom count with a th reshold, so the count was equally valid by definition. A psychometric approach to improving diagnostic algorithms is suggested. If, however, efforts to improve an algorithm do not produce anything demonstrably better than a dichotomized count of its symptoms, it will be hard to j ustify rejecting the parsimony of a simple count.