A leakage-aware comparison of static and online LSTM forecasting of influenza positivity under distribution shift in Tianjin, China

Weekly influenza sentinel surveillance data and the complete analysis code supporting the study “A leakage-aware comparison of static and online LSTM forecasting of influenza positivity under distribution shift in Tianjin, China” (Scientific Reports, under revision). Contents data/tianjin_influenza_weekly_aggregate_2015_2025.csv — the aggregated 574-week series (ISO week 1 of 2015 to week 52 of 2025): ISO year/week, week start date, weekly specimen counts (total 81,421), subtype-specific positive counts A(H3N2), A(H1N1), B/Yamagata, B/Victoria, B-unsubtyped positives, positives for any subtype, and the combined weekly positivity rate (the analysis target). data/tianjin_influenza_specimen_level_deidentified.csv — the de-identified specimen-level dataset (81,421 records: ISO year/week, collection date, sex, age, RT-PCR result; all direct identifiers removed; no free-text fields). This is the exact analysis population defined in Methods §2.1; every number in the manuscript can be regenerated from it. code/ — the complete pipeline (data construction, sliding-window construction with the leakage-free boundary rule, training of all 14 model configurations × 5 seeds, evaluation including Diebold–Mariano and Holm-corrected comparisons, and figure generation). Mirror of https://github.com/JonasTaube/influenza-lstm-leakage-evaluation results/ — per-seed evaluation outputs underlying Tables 1–5 and Supplementary Tables S3–S11. README.md — file inventory, data dictionary, software environment, reproduction instructions. Ethics. The study used only aggregated, de-identified routine surveillance data involving no individually identifiable information; secondary use was approved by the institutional ethics committee of Tianjin Nankai District Center for Disease Control and Prevention (Approval No. 2026-002-001), which waived individual consent. Licenses. Data: CC-BY-4.0. Code: MIT (see code/LICENSE).