Exploratory multimodal artificial intelligence biomarker analysis for in-hospital mortality in adult patients with pneumonia: a proof-of-concept study – Anonymized Research Dataset

Anonymized multimodal research dataset from a single-centre retrospective cohort study (Clínica CES / Hospital Alma Máter de Antioquia, Medellín, Colombia, 2024). Ethics approval IN82-2021. n = 27 patients (3 non-survivors, 24 survivors) with complete chest X-ray (CXR), ECG, and clinical note from admission for community-acquired pneumonia. Three AI modules applied: (1) ResNet-18 CXR consolidation quantification (Class Activation Mapping); (2) Spanish NLP pipeline extracting IDSA/ATS severity scores; (3) NeuroKit2 HRV analysis of digitised ECG signals. Dataset includes: ECG HRV features (94 columns), merged multimodal features (111 columns), radiologist chest X-ray annotations (JSON, 27 patients), anonymised ECG images, Table 1 patient characteristics, kappa analysis results, and full analysis code/notebooks. Patient MRNs replaced with sequential codes P001–P027. No names, admission dates, or other direct identifiers present. Submitted to PLOS Digital Health (manuscript PDIG-D-25-00464).