Table 1_Performance evaluation of large language models in the diagnosis of emergency inte

A retrospective study from the Second Affiliated Hospital of Zhejiang University School of Medicine evaluates the diagnostic performance of five large language models against emergency department junior physicians. The dataset includes 154 anonymized emergency internal medicine patient cases from January to May 2025, covering common acute diseases across multiple systems. It was authored by Jintao Wei and last updated on 2026-05-08.

Use Cases

• Benchmarking LLM diagnostic accuracy based on the reported main diagnostic accuracy scores for 5 models.
• Comparing differential diagnosis comprehensiveness based on the scoring system described in the study results.
• Analyzing response time efficiency for clinical decision support based on the recorded LLM and physician response times.
• Investigating performance variations across disease subgroups, such as respiratory system diseases, as mentioned in the results.

Strengths

• Includes performance metrics for 5 specific LLMs (ChatGPT-4o, Gemini-2.0, Grok3, DeepSeek-V3, Doubao) and 15 emergency department junior physicians.
• Based on 154 real-world, anonymized emergency internal medicine patient cases from a single medical center.
• Reports specific statistical results, including diagnostic accuracy percentages and response times with p-values.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data may reflect geographic and temporal bias inherent to a single-center study from a specific hospital in early 2025.

Provenance

Source

figshare, authored by Jintao Wei.

Collection Method

Single-center retrospective analysis of anonymized patient cases.

Time Range

January to May 2025

Freshness

Last updated 2026-05-08 05:56:42; freshness should be verified.

Geography

Second Affiliated Hospital of Zhejiang University School of Medicine, China.