The study team investigated whether explicitly modelling school holidays and public holidays changes short-term respiratory virus forecasts. They reported that holiday periods were represented both as modifiers of transmission and as potential drivers of changes in reporting. The modelling framework allowed the authors to separate changes in the effective reproduction number R(t) from changes in the relative rate of case notification during holidays. Short, discrete holiday effects were therefore treated as distinct inputs rather than as noise absorbed by generic seasonal terms.
Investigators built a Bayesian renewal model in which R(t) is composed of log-additive components. These components included explicit terms for school and public holidays and an annually periodic Gaussian process to capture smoother seasonal variation. The approach integrated transmission and reporting processes in a single probabilistic framework. The authors reported that the model specification allowed holiday-related transmission shifts to be modelled simultaneously with holiday-related reporting changes — when present — rather than conflating the two effects.
The forecasting exercise targeted three respiratory pathogens: RSV, influenza and SARS-CoV-2. Forecasts were produced across a 28-day horizon and applied to data from five Australian jurisdictions. The authors used routinely collected case notification data as model inputs. The preprint does not provide additional methodological details about data cleaning, time windows used for training versus testing, or jurisdiction-specific covariates beyond the holiday and seasonal terms; those specifics were not reported in the abstract.
Applied to the 2025 winter season, the model outperformed a random-walk baseline forecast in every evaluated setting. The authors reported improved predictive performance when holiday and seasonal transmission effects were included. They evaluated forecast skill over the 28-day horizon; the abstract does not list the precise scoring metrics, numerical improvements, or confidence intervals in this summary section, so those statistics are not reproduced here.
Data indicated that school holidays were associated with population-wide reductions in R(t) on the order of 5–15%. The authors noted that these reductions were consistent across the pathogens studied. They also reported that school holidays were generally not associated with substantial changes in reporting rates — that is, reporting appeared relatively stable across holiday periods in most cases. The preprint abstract does not report pathogen- or jurisdiction-specific effect sizes beyond the stated 5–15% range.
The authors made code and forecasting scripts available on GitHub at https://github.com/ruarai/reff_comp_public and archived their code at the DOI listed in the manuscript. Anonymised summary data sufficient to replicate figures was provided, but the underlying case notification datasets are not publicly available. The authors wrote that requests for raw data must be submitted to the respective data custodians and will be assessed independently of the author group. Use of the case notification data was approved by the Health Sciences Human Ethics Sub-Committee, Office for Research Ethics and Integrity, The University of Melbourne (Application ID 2024-26949-50575-3).
The report is a medRxiv preprint and has not been peer reviewed; the authors explicitly state the work should not be used to guide clinical practice without further evaluation. The abstract does not provide details on several methodological choices — for example, exact scoring rules for forecast comparison, the method of cross-validation, or whether jurisdictional heterogeneity in school schedules was modelled beyond binary holiday indicators — and those granular methods were not reported in the abstract. Access to the raw case datasets requires third-party approval, which limits independent replication unless custodians grant access. Remaining questions include how sensitive the reported 5–15% reductions in R(t) are to alternate model specifications, to different seasons beyond 2025 winter, and to jurisdiction-specific reporting practices; those questions would require full-text inspection or subsequent peer review to resolve.