Limitations of Large Language Models in Healthcare Content Provision

Mohsen Khosravi's dataset contains a thematic analysis of 81 studies on the limitations of Large Language Models (LLMs) in healthcare. The analysis was conducted via a systematic review of English articles published between 2018 and 2025. Data was extracted and categorized using Boyatzis's thematic approach and the IPO model.

Use Cases

• Analyze the frequency of cited limitations like 'accuracy gap' or 'data limitations' across the 81 reviewed studies.
• Map identified themes such as 'ethical, safety, and regulatory concerns' or 'model design constraints' to the IPO framework categories.
• Compare the distribution of limitations attributed to input, process, or output stages based on the study's analytical coding.
• Identify proposed mitigation strategies for challenges like 'accessibility issues' or 'response quality' documented in the review.

Strengths

• Systematic review methodology screened studies from PubMed, Scopus, and Cochrane using two independent evaluators.
• Analysis covers 81 studies assessed for quality with the AACODS checklist, showing minimal bias risk.
• Thematic framework identifies 7 key limitation categories, providing a structured taxonomy.

Limitations

• Dataset is small at 53.5 KB, indicating limited raw data or a summary-level analysis.
• Temporal coverage ends in 2025, missing recent LLM developments.
• Focus is exclusively on English-language publications, introducing a language bias.

Provenance

Source

figshare, author Mohsen Khosravi.

Collection Method

Systematic review and qualitative thematic analysis using Boyatzis's approach and the IPO model.

Time Range

Studies published between 2018 and 2025, reviewed in September 2025.

Freshness

Last updated in April 2026, based on a review completed in September 2025.

Geography

null