Understanding psychiatric illness through natural language processing (UNDERPIN): Rationale, design, and methodology

Area of Science:

• Psychiatry
• Computational Linguistics
• Neuroscience

Background:

• Current psychiatric diagnoses rely on subjective observations, leading to potential inconsistencies.
• Objective biomarkers for psychiatric disorders are lacking in clinical practice.
• Linguistic features in speech show potential as indicators for certain psychiatric conditions.

Purpose of the Study:

• To create a large dataset of Japanese speech data for psychiatric and neurocognitive disorders.
• To quantify linguistic features associated with these disorders using natural language processing.
• To develop objective and user-friendly biomarkers for diagnosis and severity assessment.

Main Methods:

• Multi-center prospective study design.
• Inclusion criteria based on DSM-5/ICD-11 for major depressive disorder, bipolar disorder, schizophrenia, anxiety disorder, and neurocognitive disorders.
• Speech data collected via interviews (free conversation, picture description, storytelling) from >400 participants, with clinical rating scales for severity assessment.

Main Results:

• Collected over 1000 speech records from >400 participants as of September 2022.
• Data represents one of the largest speech datasets in this field.
• Addressing challenges in inter-rater reliability and Japanese linguistic peculiarities.

Conclusions:

• The study's large dataset and novel methodology offer potential for developing objective biomarkers.
• Findings could significantly enhance the objectivity and ease of diagnosing and assessing psychiatric disorders.
• Future applications may extend to various fields requiring objective patient assessment.