Looking for a needle in a haystack: A machine-learning study of cognitive-behavioral constructs and symptoms for diagnosing single and comorbid anxiety disorders

ADRIAN YORIS; ALVARO DELEGLISE; GUIDO PASCARIELLO; JOSEFINA PÉREZ DEL CERRO; MARCELO CETKOVICH; PATRICIO DONELLY-KEHOE; FERNANDO TORRENTE

Journal of Anxiety Disorders

Elsevier

Año: 2023

0887-6185

Anxiety disorders are prevalent and comorbid conditions, hindering diagnosis decisions.The assessment of anxiety commonly involves measuring symptom severity or utilizingcognitive-behavioral therapy (CBT) constructs through self-report questionnaires. Whileboth approaches have their limitations, combining the measurement of symptom severityand psychopathological constructs may offer a more comprehensive understanding ofanxiety psychopathology. A total of 392 outpatients with anxiety disorders took part in acomprehensive clinical evaluation, which involved a structured interview (DSM´s SCIDI)and a battery of self-report questionnaires assessing anxiety symptoms and cognitivebehavioraltherapy (CBT) constructs. The analysis focused on social anxiety disorder(SAD), panic disorder (PD), obsessive-compulsive disorder (OCD), generalized anxietydisorder (GAD), as well as comorbid conditions. To examine the predictability ofsymptoms and constructs for diagnosing individual and comorbid conditions, supervisedclassification (Random Forest) and regression analyses (Elastic Nets) were employed.Results indicated that SAD and PD achieved high test accuracy rates of 87% and 83%respectively, while accuracy rates for OCD, GAD, and comorbidity were below 72%.Among the anxiety symptoms, those related to social interactions were the mostinformative feature for SAD, whereas the CBT construct of agoraphobic physicalconcerns was the most informative feature for PD. Regression analyses revealedsignificant associations between constructs and diagnosis for the entire sample, exceptfor SAD. These findings highlight the importance of using self-report questionnaires thatmeasure both anxiety symptoms and CBT constructs as valuable diagnostic tools forSAD and PD respectively. The implications of these findings for clinical practice arediscussed.