CALLM: Enhancing Clinical Interview Analysis through Data Augmentation with Large Language Models

The global prevalence of mental health disorders is increasing, leading to a significant economic burden estimated in trillions of dollars. In automated mental health diagnosis, the scarcity and imbalance of clinical data pose considerable challenges for researchers, limiting the effectiveness of machine learning algorithms. To cope with this issue, this paper aims to introduce a novel clinical transcript data augmentation framework by leveraging large language models (CALLM). The framework follows a "patient-doctor role-playing" intuition to generate realistic synthetic data. In addition, our study introduces a unique "Textbook-Assignment-Application" (T-A-A) partitioning approach to offer a systematic means of crafting synthetic clinical interview datasets. Concurrently, we have also developed a "Response-Reason" prompt engineering paradigm to generate highly authentic and diagnostically valuable transcripts. By leveraging a fine-tuned DistilBERT model on the E-DAIC PTSD dataset, we achieved a balanced accuracy of 0.77, an F1-score of 0.70, and an AUC of 0.78 during test set evaluations, which showcase robust adaptability in both Zero-Shot Learning (ZSL) and Few-Shot Learning (FSL) scenarios. We further compare the CALLM framework with other data augmentation methods and PTSD diagnostic works and demonstrates consistent improvements. Compared to conventional data collection methods, our synthetic dataset not only demonstrates superior performance but also incurs less than 1% of the associated costs.