This study investigates the efficiency and thematic completeness of manual versus artificial intelligence (AI)-assisted qualitative analysis of nurses’ insights into the recovery process for patients who have had emergency abdominal surgery (EAS), specifically understanding barriers and facilitators to recovery. EAS is associated with significant mortality and complications, and nurses play a crucial role in patient recovery. Extracting actionable insights from qualitative data is labor-intensive, with traditional methods requiring extensive person-hours and being susceptible to inter-coder drift. Rapid qualitative analysis (RQA) offers a streamlined approach, and commercial large language models (LLMs) may accelerate and augment qualitative research, promising semi-automated data coding and synthesis while maintaining human oversight. This study systematically compared four analytic workflows: manual RQA, ChatGPT-o3 analysis of human-generated summary notes, Microsoft Copilot GPT-4 analysis of cleaned transcripts, and ChatGPT-o3 analysis of cleaned transcripts. Manual RQA took approximately 30 person-hours, while each of the LLM-assisted methods ranged from four to eight person-hours, demonstrating significant time savings. The LLM-generated codebooks captured nearly all concepts and were highly rated by nurse participants for thoroughness, action-orientation, and relevance to their roles. The findings suggest that a pragmatic hybrid approach, using human note-taking, AI analysis, and human review, offers an effective balance of speed, ethical AI use, and rigorous qualitative analysis for implementation science. While LLMs reduced labor and provided comprehensive thematic coverage, human input and analysis remains vital to support qualitative rigor.
Marquard et al. (Mon,) studied this question.