5.22 Improving the Predictive Value of Self-Report in Adult ADHD Diagnosis

Abstract

A limited number of studies have assessed the predictive value of self-report questionnaires and items in diagnosing and/or screening for adult ADHD. Recently, for example, a shortened version of the World Health Organization ADHD Screening Scale was developed by selecting items that, taken together, are most predictive of diagnosis. Here, we aim to compare the diagnostic value of multiple self-report and other-report questionnaires commonly used in adult ADHD evaluation and draw on all of these questionnaires to identify a set of items that are mutually most predictive. Analyses drew on a large cohort of adult patients evaluated for ADHD within the Duke University Health System between 2008 and 2019. Each patient completed a subset of the following self-report questionnaires as part of their evaluation: the Barkley Adult ADHD Rating Scale (BAARS) self- and other-report; the Conners’ Adult ADHD Rating Scale (CAARS) self- and other-report; the Conners’ Adult ADHD Diagnostic Interview for DSM-IV Part 1 (CAADID-1); and an additional ADHD screening scale developed at Duke University (DSS). Diagnoses were based on the consensus of a team of ADHD specialists. Predictive models were developed based on: 1) each self-report scale in isolation; and 2) all self-report scales, taken together. Model selection and evaluation were conducted using nested cross-validation, and performance was evaluated using the area under the receiver operating characteristic curve (AUROC). Among the individual questionnaires, prediction performance was highest for the DSS self-report (AUROC = 0.80) followed by the DSS other-report (AUROC = 0.75), CAARS self-report (AUROC = 0.68), BAARS other-report (AUROC = 0.66), and CAADID-1 (AUROC = 0.66). However, combining all available items led to an average prediction performance of AUROC = 0.65 and revealed that the CAADID-1 and CAARS self-report were most predictive across the entire cohort. Moreover, the predictive values of individual items were ranked. Self- and other-report questionnaires used in adult ADHD evaluation vary substantially in their predictive value, and the predictive value of a set of items selected from multiple scales is projected to be higher than that of any individual questionnaire.