Computational forecasting using Swedish data in the vaccination phase of the COVID-19 pandemic: a systematic literature review deliberating modelling relevance for public health and healthcare.

Clinical Snapshot

Key Findings

• P Value: Not reported

• Effect Size: Not applicable — no pooled quantitative effect size was calculable. Qualitative synthesis identified Bayesian calibration as the predominant modelling methodology across included studies.

• Primary Outcome: Of 2034 unique publications identified, only 6 articles satisfied selection and quality criteria. Quantitative meta-analysis of forecasting performance was infeasible due to variability in performance reporting across studies.

• Nnt Or Sensitivity: Not applicable to this review type. Key descriptive finding: 5 of 6 included articles reported internal model corroboration data; 0 of 6 reported external model validation. 0 of 6 articles identified policymaking as the primary research context.

• Confidence Interval: Not reported — qualitative synthesis only

Clinical Application

The review does not evaluate a clinical intervention per se, but rather the research ecosystem surrounding pandemic forecasting tools. Its findings are most actionable for research funders, public health agencies, and academic modelling groups seeking to align future forecasting research with real-world policymaking needs. The conclusion that model design dominated over policy application suggests a structural gap in research commissioning that could be addressed through targeted funding mechanisms. While the review is scoped to Swedish data, its core finding — that computational forecasting research during the vaccination phase was predominantly model-design oriented rather than policymaking-driven — has direct relevance to the Australian context. During Australia's COVID-19 vaccination rollout (2021–2023), the Australian Health Protection Principal Committee (AHPPC) and the Doherty Institute produced influential modelling outputs that informed National Cabinet decisions on vaccination thresholds and reopening strategies. The degree to which these modelling efforts were prospectively designed to answer specific policy questions, versus retrospectively applied, warrants similar systematic evaluation in the Australian context. The TGA's regulatory framework for software as a medical device (SaMD) and the RACGP's pandemic preparedness guidelines do not currently address computational forecasting model standards. This review's findings support the case for establishing clearer commissioning frameworks for pandemic modelling in Australia, potentially through the Australian Centre for Disease Control (established 2024). Public health authorities, pandemic preparedness planners, health service researchers, and infectious disease modellers seeking to understand the evidence base for computational forecasting tools during vaccine-era pandemic phases. Directly applicable to Swedish healthcare context; indirectly relevant to comparable universal healthcare systems.

Abstract

BACKGROUND: The benefits of computational forecasting in the later phase of the COVID-19 pandemic when vaccines and clinical pharmaceutical interventions were available have seldom been assessed. We aimed to evaluate computational forecasting research applied to Swedish populations in the vaccination phase of the pandemic. METHODS: A systematic search was performed on March 8, 2024 in the electronic databases PubMed, Scopus, Cochrane library, Embase, Love platform and Epistemikos. An updated version of the Risk of bias Opinion Tool (ROBOT) was used to assess the quality of evidence reported in the papers identified in the search. The articles fulfilling the quality criteria were assessed for suitability for meta-analysis. Data were extracted from the selected articles for synthesis of characteristics, and a thematic analysis was used for a qualitative synthesis of the contents. RESULTS: Of 2034 unique publications identified in the database search, 6 articles satisfied the selection and quality criteria. Variability in the reporting of forecasting performance results was found to make a quantitative meta-analysis of forecast performance infeasible. The data synthesis showed that statistical modeling using Bayesian calibration was the most common methodological approach. No external model validation was reported, but 5/6 articles included internal model corroboration data. The primary theme resulting from the qualitative synthesis of article content was design or refinement of computational models with demonstration of model use in health service practice as a secondary theme. None of the articles referred to health service policymaking as the primary research context. CONCLUSION: Computational forecasting research using Swedish population data from the vaccination phase of the COVID-19 pandemic was deployed in a model design context. While methodological knowledge was developed, most of the research was not initiated to solve the public health and healthcare problems at hand. Our results indicate that the alignment between computational forecasting research and policymaking needs in the vaccination phase of pandemics can be enhanced.

References

1. 1.Jöud, A. S., Thorén, H., Spreco, A., Lundh, T., Timpka, T., & Gerlee, P. (2026). Computational forecasting using Swedish data in the vaccination phase of the COVID-19 pandemic: a systematic literature review deliberating modelling relevance for public health and healthcare. PLOS Global Public Health. https://doi.org/10.1371/journal.pgph.0004675