Chile-ED-Resp: A Curated and Reproducible Weekly Hospital Dataset for Respiratory Emergency-Demand Forecasting in Chile

Abstract

Respiratory emergencies create recurrent seasonal pressure on hospital services, yet reproducible forecasting research is often hindered by the effort required to retrieve, interpret, harmonize, and document public administrative data. Chile-ED-Resp is a curated secondary dataset and reproducible processing pipeline derived from the official Chilean Ministry of Health/DEIS dataset “Atenciones de urgencias de causas respiratorias por semana epidemiológica”, produced by the Sistema de Atención Diaria de Urgencias (SADU). Version 1.1.0 uses the closed 2022–2025 period and was built from the official Parquet snapshot retrieved on 30 August 2026. After restriction to hospitals and hospital emergency units (UEH), the curated long artifact contains 449,844 rows and 32 variables, representing 180 hospitals and 12 official respiratory cause labels. A complementary forecasting artifact contains 37,620 hospital–week rows and 38 variables on a fixed 209-position statistical-week calendar (52 weeks in 2022–2024 and 53 in 2025); 178 of 180 hospitals (98.9%) satisfy the predefined ≥95% target-coverage criterion. Acute upper respiratory infection is retained under its official Spanish label and acronym, IRA Alta. Missing source records remain missing and are distinguished from explicitly reported zeros. Precomputed target-derived predictors use outcomes available through the preceding week and are intended for rolling-origin one-week-ahead evaluation, not an unqualified fixed-origin forecast of the entire test year. Reproducibility is supported through source metadata, SHA-256 checksums, machine-readable descriptive audits, 15 validation checks, unit tests, and an example that recomputes predictors at each forecast origin. The resource provides a transparent foundation for forecasting, benchmarking, health-informatics teaching, and secondary analyses while preserving source provenance and licensing conditions.