PatientPsi

PATIENT-Ψ: Using Large Language Models to Simulate Patients for Training Mental Health Professionals

Venue: EMNLP 2024 (Main Conference)
Paper: ACL Anthology

Overview

PatientPsi creates realistic patient simulations based on cognitive behavioral therapy (CBT) principles. It models patients with specific cognitive distortions, core beliefs, and automatic thoughts.

Key Features

• Cognitive Model: Implements CBT cognitive model (Relevant History, Core Beliefs, Intermediate Beliefs, Coping Strategies...)
• Conversational Styles: Defines 6 conversational styles (Plain, Upset, Verbose, Reserved, Tangent, Pleasing)
• CBT-Grounded: Uses CBT profiles as a ground truth to evaluate trainees/therapists' ability.

How It Works

1. Profile Loading: Loads a patient persona with CBT-relevant background (history, beliefs, coping strategies, situation, emotions, automatic thoughts).
2. CBT-Grounded Prompt: Builds a system prompt that uses the CBT cognitive conceptualization to guide responses.
3. Style Conditioning: Applies a configurable conversational style (e.g., upset/verbose/reserved/tangent/pleasing) to shape tone and disclosure patterns.
4. Contextual Dialogue: Keeps multi-turn conversation history so later replies reflect prior turns and gradually reveal deeper concerns.

Usage

CLI

patienthub simulate client=patientPsi

Python

from patienthub.clients import get_client

client = get_client(agent_name="patientPsi", lang='en')

response = client.generate_response("How have you been feeling lately?")
print(response)

Configuration

Option	Description	Default
prompt_path	Path to prompt file	data/prompts/client/patientPsi.yaml
data_path	Path to character file	data/characters/patientPsi.json
data_idx	Character index	0
patient_type	Behavior pattern	"plain"

Character Data Format

The shipped data/characters/patientPsi.json contains the single illustrative example ("Abe") released by the PATIENT-Ψ authors, across three situations (id 1-1, 1-2, 1-3) that share the same beliefs. The full annotated Patient-Ψ-CM dataset is not redistributed here.

Accessing the full dataset

To get access to the Patient-Ψ-CM dataset, please fill out this form. See the authors' repository at ruiyiw/patient-psi for more details.

{
  "name": "Abe",
  "id": "1-1",
  "type": ["plain", "verbose", "tangent", "upset", "reserved", "pleasing"],
  "history": "Father leaves family when Abe is 11 years old. He never sees him again. Mom is overburdened, criticizes when he can't meet her unrealistic expectations. Precipitants to current disorder: Abe struggles and then loses his job and undergoes divorce.",
  "helpless_belief": ["I am incompetent.", "I am a failure."],
  "unlovable_belief": [],
  "worthless_belief": [],
  "intermediate_belief": "It's important to be responsible, competent, reliable and helpful. It's important to work hard and be productive.\n[during depression]\n(1) If I avoid challenges, I'll be okay, but if I try to do hard things I'll fail. (2) If I avoid asking for help, my incompetence won't show but if I do ask for help, people will see how incompetent I am.",
  "coping_strategies": "Avoids asking for help and avoids challenges.",
  "situation": "Thinking about bills.",
  "auto_thought": "What if I run out of money?",
  "emotion": ["anxious, worried, fearful, scared, tense"],
  "behavior": "Continues to sit on couch; ruminates about his failures."
}

The three core-belief fields (helpless_belief, unlovable_belief, worthless_belief) are lists and may be empty. Newer profiles embed depression-phase beliefs inside intermediate_belief under a [during depression] marker; the optional intermediate_belief_depression field is supported for backward compatibility.