New LLM prompting method improves fairness in ICU mortality prediction

Researchers have developed a new framework called Case-Based Prompting (CAP) to improve the fairness of large language models (LLMs) used for predicting mortality risk in intensive care units. This training-free method incorporates similar historical cases, including those with demographic disparities, as contextual evidence during inference. Experiments on the MIMIC-IV dataset showed CAP significantly boosted predictive performance, increasing AUROC from 0.806 to 0.873 and AUPRC from 0.497 to 0.694 compared to a baseline prompt. While CAP reduced disparities related to sex and race, age-related disparities persisted, indicating that simply suppressing demographic information does not uniformly improve both performance and fairness. AI

IMPACT Enhances LLM fairness in critical healthcare applications, potentially improving patient outcomes and reducing bias in clinical decision-making.

RANK_REASON Academic paper detailing a new method for LLM fairness in a specific domain. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• Baseline Prompt
• Case-Based Prompting
• Gangxiong Zhang
• intensive care unit
• large-language models
• MIMIC-IV

• paper
• safety
• product

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →