Freely Generated Word Responses Analyzed With Artificial Intelligence Predict Self-Reported Symptoms of Depression, Anxiety, and Worry

Background. Self-reported language-based assessments, that are based on freely generated word responses and analyzed with artificial intelligence, is a potential complement to currently used methods in identifying mental health issues. In a previous study, this approach demonstrated higher, or competitive, validity and reliability as compared with the total score of the state-of-the-art rating scales. This study investigates to what extent this approach capture symptom-based items in rating scales targeting depression and anxiety. To add value to clinical practices the semantic measures approach needs to capture cognitive, behavioral and physiological symptoms associated with mental health aspects described in diagnostic criteria e.g., the Diagnostic and Statistical Manual of Mental Disorders (DSM- 5). We study this by investigating whether the semantic information that participants generates contain information of all, or some, of the criteria that defines depression and anxiety in clinical practices. Method. Participants (N=411) described their mental health with freely generated words and rating scales relating to depression and worry/anxiety. Word responses were quantified and analyzed using natural language processing and machine learning. Results. The semantic measures correlated significantly with the individual items connected to the DSM 5 diagnostic criteria of depression (Pearson’s r = .22 - .51, p<.001) and worry (anxiety rating scale: Pearson’s r = .29 - .44, p<.001; worry rating scale: Pearson’s r = .35 - .44, p<.001) for respective rating scales. Conclusion. The semantic measures correlated significantly with the individual items of depression and worry for respective rating scales. The results indicate that the semantic measures approach significantly predict all self-reported criteria and that items measuring cognitive aspect yielded higher predictability than behavioral items. The valence aspect of the semantic representation appears to carry a lot information, potentially capturing a more general negative feeling and a help seeking behavior common for depression and anxiety. Together these results support that semantic measures may be apt to complement other methods in measuring mental health in clinical settings.