Improving Large Language Models for Clinical Named Entity Recognition via Prompt Engineering

TL;DR

This study evaluates GPT-3.5 and GPT-4 for clinical named entity recognition and introduces a prompt engineering framework that significantly improves their performance, making them more feasible for clinical applications despite not surpassing specialized models.

Contribution

The paper presents a novel prompt engineering framework tailored for clinical NER tasks that enhances GPT models' performance with minimal training data.

Findings

Prompt components improve GPT performance across tasks

GPT models outperform baseline prompts with all components

BioClinicalBERT still outperforms GPT models in accuracy

Abstract

Objective: This study quantifies the capabilities of GPT-3.5 and GPT-4 for clinical named entity recognition (NER) tasks and proposes task-specific prompts to improve their performance. Materials and Methods: We evaluated these models on two clinical NER tasks: (1) to extract medical problems, treatments, and tests from clinical notes in the MTSamples corpus, following the 2010 i2b2 concept extraction shared task, and (2) identifying nervous system disorder-related adverse events from safety reports in the vaccine adverse event reporting system (VAERS). To improve the GPT models' performance, we developed a clinical task-specific prompt framework that includes (1) baseline prompts with task description and format specification, (2) annotation guideline-based prompts, (3) error analysis-based instructions, and (4) annotated samples for few-shot learning. We assessed each prompt's effectiveness and compared the models to BioClinicalBERT. Results: Using baseline prompts, GPT-3.5 and GPT-4 achieved relaxed F1 scores of 0.634, 0.804 for MTSamples, and 0.301, 0.593 for VAERS. Additional prompt components consistently improved model performance. When all four components were used, GPT-3.5 and GPT-4 achieved relaxed F1 socres of 0.794, 0.861 for MTSamples and 0.676, 0.736 for VAERS, demonstrating the effectiveness of our prompt framework. Although these results trail BioClinicalBERT (F1 of 0.901 for the MTSamples dataset and 0.802 for the VAERS), it is very promising considering few training samples are needed. Conclusion: While direct application of GPT models to clinical NER tasks falls short of optimal performance, our task-specific prompt framework, incorporating medical knowledge and training samples, significantly enhances GPT models' feasibility for potential clinical applications.