Data from Associations of an inflammation–nutrition composite score across strata of gastric mucosal lesions: a retrospective cross-sectional study

This deposit contains the de-identified dataset and the complete reproducible analysis scripts for the manuscript "An inflammation–nutrition composite score for stratification of gastric mucosal lesions in 445 hospitalized patients" (submitted to PeerJ). Study summary. Retrospective single-center study of 445 hospitalized patients (March 2021 – June 2026) stratified into four gastric mucosal lesion groups: chronic non-atrophic gastritis (n = 166), chronic atrophic gastritis with intestinal metaplasia (n = 117), early gastric cancer (n = 49), and advanced gastric cancer (n = 113). A composite score (0–4) built from four routine admission laboratory markers — high-sensitivity C-reactive protein, neutrophil-to-lymphocyte ratio, albumin, and prealbumin — discriminated gastric cancer from non-gastric cancer with an AUC of 0.823 (Youden-optimized cut-offs) and 0.789 (pre-specified rounded cut-offs). hs-CRP values below the detection limit (< 0.20 mg/L) were imputed as LOD/2 = 0.10 mg/L in the primary analysis. Contents. data_deidentified_445.csv — de-identified analysis dataset, 445 patients × 80 columns (direct identifiers removed; admission year retained to the calendar year only). scripts/ — six scripts (Python and R) that reproduce every statistical result in the manuscript: core statistics and Tables 1–3, figures and the summary Excel workbook, 1000× bootstrap optimism correction and internal cross-validation, DeLong tests (R pROC), reviewer-requested supplementary analyses, and ordinal regression models. outputs/ — regenerable figures (Figure 1 boxplots; Figure 4 forest plot of the 10 pre-specified markers plus age) and the summary workbook (Tables 1–6 + dataset sheet). README.md — full documentation, data dictionary, and a verification table comparing every reproduced value with the manuscript value. De-identification. All fields that could directly identify patients (name, medical record number, exact admission and laboratory dates, department) were removed before sharing; attempts to re-identify individuals are prohibited by the data use terms. Please cite this record as: [manuscript title, authors, journal, DOI when published] and this dataset via its DOI: https://doi.org/10.5281/zenodo.21859066.