Development of a computerized adaptive testing for ADHD using Bayesian networks: An attempt at classification

Quantifying the severity of neurological disorders, such as attention-deficit/hyperactivity disorder (ADHD), is encouraged in practice. To explore the methodological properties of data-driven classification, this study aimed to develop computerized adaptive testing (CAT) based on Bayesian networks to classify ADHD symptom severity for individual subjects efficiently. The Current Symptoms Scale was used to build an ADHD symptom severity classifier in a sample of 892 college students, and the CAT making use of information gains from conditional probabilities of the Bayesian network was evaluated through a simulation study. The results showed that distributing 8 out of 18 items on average to an individual subject would result in 85% classification accuracy, supporting the adaption of surveying ADHD to digital platforms demanding much-shortened scales. The severity classifier based on Bayesian networks can assist diagnostic tasks via a data-driven mechanism. Further, integrating CAT into the classifier can shorten the scale while maintain acceptable measurement quality. The ADHD symptoms were used in the paper; however, the proposed method applies to other psychological symptoms as well. The impact and limitations of the CAT for ADHD were discussed, and several suggestions for future research were provided.