Prompt chains, session links, and technical documentation for: Accuracy and reliability of generative artificial intelligence in the statistical analysis of 14 medical datasets.

Abstract

Technical documentation deposited to accompany the manuscript "Accuracy and reliability of generative artificial intelligence in the statistical analysis of 14 medical datasets" (Kabbash IA, Abouzaid MS, Islam T, Mesallam TA, Farahat M, Temsah MH, Geneid A, Malki KH; submitted for publication, 2026). The study compared the GPT-4o model (ChatGPT web interface, May to September 2025) with biostatistician-revised published results for 14 anonymized medical and dental datasets from Tanta University, Egypt, using a 13-item rubric scored by two senior biostatisticians. The deposit contains: (1) the model and interface configuration as far as it can be documented; (2) the standardized five-prompt chain template; (3) the final prompt chain used for each of the 14 datasets, reproduced verbatim; (4) the mapping of datasets to their source publications and read-only shared links to the analysis sessions (available for 12 of the 14 datasets); (5) the 13-item rubric with its scoring anchors, the dataset-level classification rule, and a development note; (6) the consolidated comments of the two biostatisticians for each dataset; and (7) the item-level scores recorded for each dataset. The datasets themselves are not included. Ethics: the parent study was determined exempt by the King Saud University Institutional Review Board as secondary research not involving human subjects (Ref. No. 24/1689/IRB; Project No. E-24-9185; 7 November 2024). Funding: Ongoing Research Funding program, Research Chairs (ORF-RC-2026-4103), King Saud University, Riyadh, Saudi Arabia.