Derived data, code and figures for: Research Themes and Evidence Gaps in Pediatric Sepsis Host-Response Studies: A Bibliometric Analysis with Cross-Database and Eligibility Sensitivity Assessments

Version 4.0.0 archives the revised bibliometric and topic-modeling study prepared for Journal of Clinical Medicine. It supersedes the September 29 baseline outputs for the revised study while preserving version 3.0.0 (DOI 10.5281/zenodo.23036618) as the historical 278-publication/106-direct-sepsis baseline. The revised frozen corpora contain 238 Web of Science publications, 243 PubMed publications and a deduplicated union of 267 publications (214 shared; 29 PubMed additions). Sensitivity corpora include 98 direct-sepsis publications, 179 articles, 213 publications before 2026 and 237 publications with abstracts. The publication cutoff is 29 August 2026; 2026 is an incomplete year. The archive supplies permission-compatible aggregate and model-derived results, analysis configurations, frozen-input hashes, data dictionaries, author-created Python and R code, and six main plus three supplementary figures in TIFF, SVG, PDF and PNG formats. It records candidate topic-number and random-seed comparisons, common-vocabulary topic matching, cross-database and eligibility sensitivities, and AI-assisted eligibility and research-intent consistency summaries. Negative and unstable findings are retained. AI consistency is not classification accuracy or clinical expert validation; cross-database comparisons are methodological assessments rather than clinical or biological validation. Reproduction has explicit limits. The supplied R code reproduces seven aggregate-data figures; Figures 3 and S1 require excluded licensed network tables, so only their final numerical plots are public. LDA re-fitting requires separately licensed, exact frozen title/abstract inputs with the documented byte hashes, row order and software environment; the public aggregates cannot reconstruct database retrieval, screening, deduplication or AI evidence judgments. Source records, titles/abstracts, full texts, internal screening/evidence ledgers, individual document tokens/probabilities, raw network edges and native cited-reference strings are excluded. The 47-entry cited-reference display key contains attribution for plotted references only. Derived data, documentation and figures are licensed CC BY 4.0; author-created code is MIT. No licence is granted to excluded third-party source material. SHA-256 checksums identify each public file and the uploaded release package. Data publication is separate from final all-author approval and submission of the manuscript.