Reading between the Lines: Leveraging Large Language Models for Global Dementia and Depression Assessment from Clinical Interviews

2026-06-16 · Source: Computation and Language · Field: Health & Wellbeing — Artificial Intelligence & Machine Learning, Mental Health & Psychological Support, Clinical Care & Medical Practice · Depth: Expert, quick

Summary

A study investigated open-weight Large Language Models (LLMs) for assessing dementia and depression severity from speech samples. Researchers utilized interviews with 154 German-speaking subjects, introducing an observer-based Global Depression Scale (GDS-D) aligned with the Global Deterioration Scale (GDS). Three LLMs—Mistral 3.1, DeepHermes, and Qwen3—were evaluated in zero-shot prediction and LLM-based feature extraction for Support Vector Regression, using both human and pause-enriched transcripts. Results showed LLMs effectively predict depression severity in zero-shot settings, achieving a best Mean Absolute Error (MAE) of 0.60. Dementia assessment significantly improved with structured feature extraction, reaching a best MAE of 0.78 and reducing errors by up to 35% compared to zero-shot baselines. The competitive performance of pause-enriched transcripts suggests the viability of fully automatic screening pipelines for differential neuropsychiatric assessment.

Key takeaway

For Machine Learning Engineers developing diagnostic tools, you should consider integrating open-weight LLMs for neuropsychiatric assessment. Implement structured feature extraction with LLMs like Mistral 3.1 for dementia severity prediction, as it significantly reduces errors. For depression, zero-shot LLM approaches are highly effective. Furthermore, explore using pause-enriched transcripts to build fully automatic, cost-efficient screening pipelines, enhancing accessibility and scalability of early diagnosis.

Key insights

LLMs can effectively assess neuropsychiatric disorder severity from speech, enabling automated screening.

Principles

• Differential diagnosis benefits from structured LLM features.
• Pause-enriched transcripts are viable for automated assessment.
• Zero-shot LLMs excel in depression severity prediction.

Method

The method involves comparing zero-shot LLM predictions with LLM-based feature extraction for Support Vector Regression, using human and pause-enriched speech transcripts from standardized interviews.

In practice

• Implement LLM feature extraction for dementia assessment.
• Utilize pause-enriched transcripts for automated screening.
• Apply zero-shot LLMs for initial depression screening.

Topics

• Large Language Models
• Dementia Assessment
• Depression Assessment
• Neuropsychiatric Disorders
• Speech Analysis
• Zero-shot Learning

Best for: NLP Engineer, AI Scientist, Research Scientist, Machine Learning Engineer

Related on AIssential

• Predicting Post-Traumatic Epilepsy from Clinical Records using Large Language Model Embeddings — LLM embeddings from clinical records can predict post-traumatic epilepsy with high accuracy.
• An Agentic LLM-Based Framework for Population-Scale Mental Health Screening — An agentic LLM framework enhances mental health screening by ensuring robust, adaptable, and trustworthy AI pipelines.
• Beyond Augmentation: Score-Guided Pathological Prior for EEG-based Depression Detection — A score-guided classification framework improves EEG-based depression detection by modeling pathological priors without data augmentation.
• Benchmarking large language models for cell-free RNA diagnostic biomarker discovery — Large language models show promise for diagnostic biomarker discovery, particularly in gene panel nomination.
• Multilingual Detection of Alzheimer's Disease from Speech: A Cross-Linguistic Transfer Learning Approach — Cross-language transfer learning effectively detects Alzheimer's Disease from speech across multiple languages, achieving strong generalization and rapid inference.
• A domain-adapted large language model to support clinicians in psychiatric clinical practice — Domain-adapted LLMs can significantly enhance psychiatric clinical workflows, improving accuracy and efficiency.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computation and Language.