Depression and Anxiety Prediction Using Deep Language Models and Transfer Learning
This work addresses digital screening for behavioral health conditions, offering a tool for providers, but it is incremental as it applies existing deep learning methods to a new dataset.
The study tackled the problem of detecting depression and anxiety from conversational speech using deep language models, achieving AUC scores ranging from 0.79 to 0.86 for binary classification across different conditions and co-occurrence scenarios.
Digital screening and monitoring applications can aid providers in the management of behavioral health conditions. We explore deep language models for detecting depression, anxiety, and their co-occurrence from conversational speech collected during 16k user interactions with an application. Labels come from PHQ-8 and GAD-7 results also collected by the application. We find that results for binary classification range from 0.86 to 0.79 AUC, depending on condition and co-occurrence. Best performance is achieved when a user has either both or neither condition, and we show that this result is not attributable to data skew. Finally, we find evidence suggesting that underlying word sequence cues may be more salient for depression than for anxiety.