Search research articles

ABOUT JoVE

Overview Leadership Blog JoVE Help Center

AUTHORS

Publishing Process Editorial Board Scope & Policies Peer Review FAQ Submit

LIBRARIANS

Testimonials Subscriptions Access Resources Library Advisory Board FAQ

RESEARCH

JoVE Journal Methods Collections JoVE Encyclopedia of Experiments Archive

EDUCATION

JoVE Core JoVE Business JoVE Science Education JoVE Lab Manual Faculty Resource Center Faculty Site

Terms & Conditions of Use

Search research articles

Related Experiment Video

Updated: Apr 23, 2026

Augmenting Large Language Models via Vector Embeddings to Improve Domain-Specific Responsiveness

Augmenting Large Language Models via Vector Embeddings to Improve Domain-Specific Responsiveness

Published on: December 6, 2024

Enhancing large language model clinical support information with machine learning risk and explainability: a

Yu-Chang Yeh¹, Hsin-Yu Yang², Ching-Tang Chiu²

¹Departments of Anesthesiology, Information Technology Office, National Taiwan University Hospital, No 7, Chung Shan South Road, Taipei, Taiwan. tonyyeh@ntuh.gov.tw.

Intensive Care Medicine Experimental

|April 21, 2026

Summary

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by

Same author

Constructing a lower-bound estimate of the global number of insect species on a hyperdiverse empirical foundation.

Proceedings of the National Academy of Sciences of the United States of America·2026

Same author

Temperate forest heterogeneity decreases local and landscape-scale spider diversity through habitat filtering despite increasing species turnover.

The Journal of animal ecology·2026

Same author

Implementing a Safety Protocol for Thirst Management to Improve Postoperative Thirst Using the Iowa Model of Evidence-Based Practice: A Propensity Score-Matched Evaluation.

Journal of clinical nursing·2026

Same author

The IMPACT framework for evaluating generative AI in critical care: development and multinational consensus validation.

Annals of intensive care·2026

Same author

Restoring structural complexity in temperate forests increases bat and bird diversity.

Current biology : CB·2026

Same author

Inflammation and sarcopenia exhibit distinct associations with psychopathology and executive function in schizophrenia.

Journal of the Chinese Medical Association : JCMA·2026

Same journal

Decoy receptor 3 as a diagnostic marker for sepsis: a meta-analysis.

Intensive care medicine experimental·2026

Same journal

Hidden effects of dobutamine on cardiac output.

Intensive care medicine experimental·2026

Same journal

Behaviour artificial intelligence technology to support the decision-making process of continuation or withdrawal of life-sustaining therapy.

Intensive care medicine experimental·2026

Same journal

Continuous intratracheal gas suctioning combined with ventilator leak-compensation may enhance carbon dioxide removal efficiency: a proof-of-concept study using a porcine model.

Intensive care medicine experimental·2026

Same journal

Timing-dependent renal protection of dapagliflozin in endotoxemic diabetic mice by real-time GFR and biomarkers.

Intensive care medicine experimental·2026

Same journal

Improving mortality prediction in critically ill cancer patients with a multidimensional machine learning model.

Intensive care medicine experimental·2026

See all related articles

This summary is machine-generated.

Integrating machine learning (ML) risk predictions and Shapley Additive Explanations (SHAP) with large language models (LLMs) can enhance clinical decision support. LLM-based evaluation shows promise for assessing clinical content quality.

Area of Science:

Artificial Intelligence in Medicine
Clinical Decision Support Systems
Machine Learning Interpretability

Background:

Current machine learning (ML) models offer limited actionable insights for individualized patient management.
Large language models (LLMs) can translate ML risk estimates (using Shapley Additive Explanations - SHAP) into clinically useful information.
The added value of ML data in LLMs and the comparative performance of different LLMs require further investigation.

Purpose of the Study:

To evaluate the quality of LLM-generated clinical support information using the IMPACT framework.
To determine if augmenting LLM inputs with ML-derived risk and SHAP values improves clinical response quality.
To compare the performance of GPT-4o with seven other contemporary LLMs.

Main Methods:

Keywords:

Clinical support system Critical care Generative artificial intelligence Large language model Machine learning

More Related Videos

Evidence-based Knowledge Synthesis and Hypothesis Validation: Navigating Biomedical Knowledge Bases via Explainable AI and Agentic Systems

Evidence-based Knowledge Synthesis and Hypothesis Validation: Navigating Biomedical Knowledge Bases via Explainable AI and Agentic Systems

Published on: June 13, 2025

Related Experiment Videos

Last Updated: Apr 23, 2026

Augmenting Large Language Models via Vector Embeddings to Improve Domain-Specific Responsiveness

Augmenting Large Language Models via Vector Embeddings to Improve Domain-Specific Responsiveness

Published on: December 6, 2024

Evidence-based Knowledge Synthesis and Hypothesis Validation: Navigating Biomedical Knowledge Bases via Explainable AI and Agentic Systems

Evidence-based Knowledge Synthesis and Hypothesis Validation: Navigating Biomedical Knowledge Bases via Explainable AI and Agentic Systems

Published on: June 13, 2025

A retrospective analysis of MIMIC-IV ICU admissions was conducted.
An XGBoost model predicted ICU mortality risk and derived SHAP values.
GPT-4o and seven other LLMs generated clinical support responses, which were evaluated using the IMPACT framework and scored by Claude 3.7 Sonnet.

Main Results:

Augmenting LLM inputs with predicted ICU mortality risk and SHAP values significantly improved GPT-4o's IMPACT scores.
GPT-5 mini and gpt-oss-120B demonstrated superior interpretability and quality compared to GPT-4o.
Claude 3.7 Sonnet showed high agreement with human ratings for evaluating LLM outputs.

Conclusions:

Combining ML-derived risk, SHAP explanations, and LLMs may offer modest improvements in ICU clinical support.
LLM-based evaluators are feasible for scalable assessment of generated clinical content.
Further research is needed to optimize LLM integration for clinical decision support.