Can large language models simulate patients with mental health problems? Meet Patient-Ψ: a new patient simulation framework for cognitive behavioral therapy (CBT) training

Mental illness is a critical global public health problem, with one in eight people affected and many lacking access to adequate treatment. A significant challenge in training mental health professionals is the disconnect between formal education and real-world interactions with patients. A study surveyed 12 experts to address this issue, revealing that traditional role-playing games are often unrealistic and insufficient to prepare trainees. Leveraging advances in LLMs like ChatGPT, the researchers propose using LLMs to create simulated patients. This approach aims to improve the fidelity and effectiveness of training, helping to bridge the gap between current training methods and the complexities of real-world therapy.

Researchers from Carnegie Mellon, Princeton, Pittsburgh, and Stanford have developed PATIENT-Ψ, an innovative patient simulation framework for training in cognitive behavioral therapy (CBT). PATIENT-Ψ uses cognitive models based on CBT principles and integrates them with LLMs to simulate patients in therapy. The interactive training framework, PATIENT-Ψ-TRAINER, allows trainees to practice training cognitive models through simulated therapy sessions. A study involving 33 experts and trainees showed that PATIENT-Ψ closely mimics real patients and improves trainees’ skills and confidence beyond traditional methods. PATIENT-Ψ-TRAINER provides personalized training with diverse patient models and conversational styles, offering significant advantages over existing training approaches.

The development of PATIENT-Ψ is closely related to the recent application of LLMs in fields such as psychology, education, and computational social sciences. Unlike existing research on LLMs in cognitive behavioral therapy (CBT), which is primarily concerned with detecting cognitive distortions and reframing negative thoughts, PATIENT-Ψ focuses on creating realistic, interactive simulations for training professional in CBT. This work aligns with studies using LLMs for training in communication skills, emotion management, and clinical diagnosis. PATIENT-Ψ uniquely bases its simulations on CBT principles, incorporates feedback mechanisms, and evaluates effectiveness with mental health professionals and trainees.

The construction of PATIENT-Ψ involves the integration of cognitive models from CBT with LLM to simulate realistic interactions with patients. PATIENT-Ψ-CM, a dataset of 106 cognitive models created by clinical psychologists, serves as the basis. These models cover various scenarios and conversation styles, reflecting various emotional and cognitive states. PATIENT-Ψ-TRAINER is an interactive framework for mental health trainees to practice formulating cognitive models through simulated therapy sessions. Trainees engage with PATIENT-Ψ, develop cognitive models, and receive feedback by comparing their formulations with the original models, facilitating effective, self-guided skill development.

The realism of PATIENT-Ψ was evaluated by experts who compared it to traditional methods and a GPT-4 baseline. PATIENT-Ψ significantly outperformed both, closely mimicking real patients across emotional states, conversational styles, and maladaptive thoughts. Experts highlighted its ability to provide a realistic challenge in information extraction, in contrast to the overly reactive GPT-4 baseline. Additionally, PATIENT-Ψ accurately reflects the components of its cognitive model, providing trainees with high-quality feedback. PATIENT-Ψ-TRAINER was rated as more effective in skill development and confidence building, making it a superior training tool for preparing mental health professionals.

PATIENT-Ψ is an innovative simulated patient that fuses cognitive models with an LLM to replicate real-world patient interactions. PATIENT-Ψ-TRAINER facilitates role-play therapy sessions, where trainees formulate underlying cognitive models. User studies with mental health experts and trainees confirm the fidelity of PATIENT-Ψ and the effectiveness of PATIENT-Ψ-TRAINER, outperforming traditional methods and a GPT-4 baseline. The framework holds promise for transforming mental health professional training and can be adapted to broader therapeutic paradigms. However, the study acknowledges limitations in its current evaluation, suggesting that future research could include longitudinal RCTs to objectively measure skill improvements.

Check Paper. All the credit for this research goes to the researchers of this project. Also, don’t forget to follow us on Twitter.

Join our Telegram channel And LinkedIn Groops.

If you like our work, you will love our bulletin..

Don’t forget to join our 45,000+ ML subreddit

Sana Hassan, a consulting intern at Marktechpost and a dual degree student at IIT Madras, is passionate about applying technology and AI to address real-world challenges. With a keen interest in solving practical problems, he brings a fresh perspective to the intersection of AI and real-world solutions.