把流行病建模定义为公共服务

The world invests heavily in pandemic firewalls such as vaccines and therapeutics, but far less in the analytic intelligence needed to guide their use. Like weather forecasting, epidemic modeling and analytics should be treated as a standing component of the public-health infrastructure.

Revisiting the SARS-CoV-2 pandemic often divides people into two groups. One side argues that science and technology helped us manage the crisis: although medical knowledge, vaccines, and public health policies were not perfect, they prevented greater harm. The other side sees mainly failure: mismanagement, inequalities, confusing communication, and only partly effective vaccines and treatments are blamed for everything that went wrong. Regardless of a person’s perspective, one thing is clear: next time, we should do better. And in many important ways, the international community is working toward that goal. Governments and agencies now recognize that pandemics are not once-in-a-century events and are planning investments in emergency logistics, new pathogen-surveillance systems, and ultra-fast vaccine platforms^1,2.

In several high-level meetings on “Pandemic X” the authors have attended over the past few years, this agenda has been remarkably consistent. The focus is on stockpiles, manufacturing capacity, cold chains, and rapid clinical trials. By contrast, modeling and outbreak analytics (including forecasting, scenario projection, and situational awareness) often appear as a sidebar in a few slide decks: interesting add-ons, viewed with a pinch of skepticism, rarely considered core infrastructure. The issue is mirrored in funding discussions as well: billions of dollars are allocated to therapeutics, supplies, and vaccines, yet support for forecasting and analytic capacity is only a tiny fraction, if mentioned at all. This neglect is misguided, as many of the worst side effects of the pandemic, from broken supply chains and inflation to prolonged school closures, can be traced to poor assumptions and uncertainty about its duration and severity.

Commendable efforts are underway. The CDC’s Center for Forecasting and Outbreak Analytics, launched in 2023, aims to create internal modeling capabilities at CDC and supports Insight Net, a network that brings together more than 100 academic, private, and public health organizations to work on training, tool development, and improved analytics for infectious disease outbreaks³. The ECDC has initiated collaborative programs in forecasting and scenario modeling, and the WHO has established a hub for pandemic and epidemic intelligence⁴. However, these efforts remain largely concentrated in high-income settings and still lack the sustained funding, staffing, and institutional commitment needed to build the lasting infrastructure that can continuously translate new modeling advances into operational practice, and scale capabilities globally for pandemic response⁵. Even in the United States, one of the better-resourced examples, dedicated support for outbreak analytics remains only a small fraction of overall preparedness spending. Further, the scarcity of epidemic modeling and analytics capabilities in low- and middle-income countries remains stark⁶.

Once more, we risk overlooking one of the most important lessons COVID-19 taught us. During a crisis, the capacity to anticipate the coming weeks and months can be just as important as the medical tools we are ready to deploy. It is unlikely that an outbreak of a novel pandemic pathogen will be contained at the source, even with the best available warning systems and border controls. In most realistic scenarios for “Pandemic X,” our optimistic timelines for detection, vaccine deployment, or targeted containment will not hold. The real challenge, as we experienced in 2020, is mitigating the epidemic that is already underway⁷. Policymakers, business leaders, and politicians will, as in 2020, face difficult questions in the first few weeks or months of an outbreak. Will this lead to a full-blown epidemic? What can we do to control it? Is transmission happening undetected, and where? Which areas are likely to see surges in hospital admissions over the next two or three weeks? Who gets priority for limited vaccines, antivirals, and protective gear?

These are the questions that public health authorities, hospital networks, and governments struggled with during the COVID-19 pandemic. And they are, at their core, modeling problems^5,8 that require integrating clinical and laboratory surveillance, mobility patterns, demographic information, serology, wastewater, and genomic data, each of which is noisy, incomplete, or delayed.  Epidemic modeling and outbreak analytics, including forecasting, scenario projection, and situational awareness, are public health infrastructure. They should be funded as such, not assembled as emergency research once a crisis begins.

Lessons from the modeling frontline

In recent years, modeling has demonstrated tangible value to public health practice, such as forecasting and scenario modeling during COVID-19, and has gained significant insights into how to improve reliability and impact^{8,9,10,11,12,13,14} (see Fig. 1). One of the most important changes in infectious-disease modeling over the past decade, accelerated by COVID-19, is a fundamental shift in what we expect from models. For a long time, simple models have helped us to understand fundamental mechanisms, ranging from how disease transmission depends on individuals’ contact networks to how immunity, interventions, and pathogen characteristics affect epidemic dynamics. These “textbook” models are invaluable for teaching and for building intuition about how diseases spread. But a key lesson from recent crises is to embrace the real-world complexity these models are meant to represent. Populations are heterogeneous with respect to age, household structure, mobility, health status, and behavior. Contacts among individuals happen in specific settings, such as homes, schools, and workplaces. Policies do not target an abstract “average person”; instead, they close schools in particular districts or limit the capacity of specific commercial venues. In response, the modeling community has developed approaches that incorporate more of these structures and novel data, increasing the detail and flexibility in how critical transmission behaviors are integrated into models and their utility^11,12.

At the same time, no single model fits all situations; different questions require multiple tools. To understand how pathogens spread globally, we need models that explicitly consider travel networks and regional connectivity. To explore counterfactuals and scenario analysis, we must rely on more detailed compartmental or agent-based models that account for specific social behaviors. In practice, no single approach or modeling team can cover this entire spectrum: we need a toolbox and the corresponding mix of expertise to navigate the different stages of a pandemic^9,12.

During the COVID-19 and seasonal influenza forecasting challenges, we have also learned that models disagree in valuable ways^13,14. For example, one model may be more sensitive to changes in mobility or policy, while another one may be more sensitive to signals in hospital admissions. However, the characteristics of each modeling approach are not problems to be eliminated; they are features to leverage. Indeed, recognizing that no model is an oracle requires each modeling team to be explicit about assumptions and to avoid overstating the model’s conclusions. More importantly, we should favor multi-model ensembles that aggregate results from different approaches, each with its own structure and assumptions, which generally deliver more stable predictions than any single model^15,16. This approach is alike modern weather prediction, and depends also on continuous observation, data assimilation, and communication standards, and the public trust built over decades. It is not just a matter of ensemble mathematics but a means of communicating results that reflect the consensus view of a community.

Finally, the past decade has demonstrated clear best practices for epidemic modeling and forecasting. They are not perfect, but they exist. We know that clear documentation of methods and standardized targets (for example, clearly defined horizons and targets such as weekly hospital admissions) enables model comparison and faster operational progress¹⁷. We have a better sense of what can and what cannot be forecasted, and distinguishing forecasts from what-if scenarios reduces communication problems and builds trust. These best practices were developed through experience with forecast and scenario modeling hubs, collaborative challenges, and ongoing interactions with public health partners. They represent the “operational wisdom” of the modeling community¹⁸. Most importantly, they are not just methodological preferences; they generate operational requirements. To maintain these requirements, the community needs consistent data flows, shared infrastructure for model development and evaluation, and skilled professionals to keep these systems running.

These lessons point to the need for a true ecosystem. No single model or single team can cover what’s required. Supporting this ecosystem means investing in data infrastructure, coordination mechanisms, and the long-term stability that allows these capabilities to persist between crises. Pre-pandemic analyses can, for example, build consensus on when travel restrictions meaningfully slow introduction and what metrics should trigger their relaxation. Scenario modeling can develop allocation frameworks for scarce vaccines and antivirals. This peacetime work also limits the harms of reactive modeling. Models built under crisis pressure are more prone to errors and rushed assumptions. And yes, modeling and analytics remain ongoing efforts with significant room for improvement. But without support and the necessary infrastructure, we will continue to demand more from models while keeping them stuck at the same frontier, leaving us unprepared for real-world threats.

Shared responsibilities

Establishing epidemic modeling and outbreak analytics as a public service, however, requires changes in practice among modelers, policymakers, and funders (Box 1). For the modeling community, the primary responsibility is to adopt a service mindset. This service mindset contrasts with the traditional academic incentives, which include publications in high-impact journals, grant competitiveness measured by novelty rather than utility, and success tied to individual attribution rather than collaborative impact. Too often, the messy constraints of real-world application and incomplete data are seen as compromises that dilute scientific rigor rather than as the essential challenges of public health work. The modeling community should move beyond a purely academic mindset, focusing on the questions that public health stakeholders truly need answered, prioritizing timeliness and robustness over elegance, and valuing reproducible pipelines and shared standards over individual methodological innovations.

On the other hand, governments, public health agencies, and funders must create the conditions that enable such a service-oriented modeling approach to exist and thrive. This begins by recognizing that epidemic modeling and outbreak analytics are strategic capabilities^5,8. This commitment goes beyond the core modeling capacity. It includes the training programs that build the talent pipeline, the data systems that support modeling, the coordination hubs that align efforts across institutions, and the evaluation frameworks that keep improving quality standards. And it should cover the partnerships between academic modelers, government analysts, and public health practitioners that allow capabilities to mature over decades. Team science and collaborative infrastructure should be incentivized while also preserving support for smaller innovative efforts that feed novel methods and ideas into the broader ecosystem. Just as we don’t build vaccine manufacturing capacity from scratch when a pandemic arrives, we shouldn’t assemble modeling capacity on the fly during a crisis.

Fighting the fire, knowing the wind

Unfortunately, investment in epidemic modeling as a public service remains the exception rather than the rule in the international community. As noted, initiatives such as the CDC’s Center for Forecasting and Outbreak Analytics and a handful of other efforts represent important steps. Still, there is a large gap between current investments and what’s needed. Most of these efforts are still treated as time-limited projects rather than long-term public services. Recently, they have also been driven mostly by the belief that a small investment in “cool” artificial intelligence will solve the analytics problem, not by a coherent long-term strategy. AI tools are potentially valuable components of outbreak analytics, but they cannot substitute for domain expertise and the public-health decision processes that turn analyses into action. If we accept that every country deserves weather and economic forecasting services that inform decisions by hospitals, schools, businesses, and governments alike, every country deserves access to reliable epidemic modeling and outbreak analytics as well.

The next pandemic will test our biomedical tools as well as our ability to anticipate weeks ahead and act preemptively. Forecasts, scenarios, and analytics are not optional extras; they are the backbone of a modern response. Without a global outbreak analytics infrastructure as a public service, we will enter the next crisis with better biomedical tools but remain effectively blind in using them. Building it, will transform the lessons from COVID-19 into something enduring: a world that doesn’t fight fires without understanding which way the wind is blowing.

References

1. Boyce, M. R., Sorrell, E. M. & Standley, C. J. An early analysis of the World Bank’s Pandemic Fund: a new fund for pandemic prevention, preparedness and response. BMJ Glob. Health 8, e011172 (2023).

   Article  PubMed  PubMed Central  Google Scholar 

2. The Pandemic Fund. Medium-term strategic plan: 2024-2029. https://www.thepandemicfund.org/medium-term-strategic-plan (2024).

3. Insight Net, established in 2023 by the Centers for Disease Control and Prevention’s (CDC) Center for Forecasting and Outbreak Analytics (CFA): https://www.cdc.gov/insight-net/php/about/index.html.

4. European Respiratory Diseases Forecasting Hub: https://respicast.ecdc.europa.eu/; The World Health Organization (WHO) Hub for Pandemic and Epidemic Intelligence, https://pandemichub.who.int/.

5. Kretzschmar, M. Disease modeling for public health: added value, challenges, and institutional constraints. J. Public Health Policy 41, 39–51 (2020).

   Article  PubMed  PubMed Central  Google Scholar 

6. Teerawattananon, Y. et al. Recalibrating the notion of modelling for policymaking during pandemics. Epidemics 38, 100552 (2022).

   Article  CAS  PubMed  PubMed Central  Google Scholar 

7. Marani, M., Katul, G. G., Pan, W. K. & Parolari, A. J. Intensity and frequency of extreme novel epidemics. Proc. Natl. Acad. Sci. USA. 118, e2105482118 (2021).

   Article  CAS  PubMed  PubMed Central  Google Scholar 

8. Lofgren, E. T. et al. Mathematical models: a key tool for outbreak response. Proc. Natl. Acad. Sci. USA. 111, 18095–18096 (2014).

   Article  CAS  PubMed  PubMed Central  Google Scholar 

9. Poletto, C., Scarpino, S. V. & Volz E. M. Applications of predictive modelling early in the COVID-19 epidemic. Lancet Digit. Health, 2, e498–e499 (2020).

10. Imai, N. et al. (eds) Principles and Practice of Emergency Research Response Ch. 25 (Springer, 2024).

11. Moran, K. R., Lopez, T. & Del Valle, S. Y. The future of pandemic modeling in support of decision making: lessons learned from COVID-19. BMC Glob. Public Health 3, 24 (2025).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

12. Holmdahl, I. & Buckee, C. Wrong but useful—what Covid-19 epidemiologic models can and cannot tell us. N. Engl. J. Med. 383, 303–305 (2020).

    Article  CAS  PubMed  Google Scholar 

13. Mathis, S. M. et al. Evaluation of FluSight influenza forecasting in the 2021–22 and 2022–23 seasons with a new target laboratory-confirmed influenza hospitalizations. Nat. Commun. 15, 6289 (2024).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

14. Cramer, E. Y. et al. Evaluation of individual and ensemble probabilistic forecasts of COVID-19 mortality in the United States. Proc. Natl. Acad. Sci. USA. 119, e2113561119 (2022).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

15. Reich, N. G. et al. Collaborative hubs: making the most of predictive epidemic modeling. Am. J. Public Health 112, 839–842 (2022).

    Article  PubMed  PubMed Central  Google Scholar 

16. Howerton, E. et al. Evaluation of the US COVID-19 Scenario Modeling Hub for informing pandemic response under uncertainty. Nat. Commun. 14, 7260 (2023).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

17. Consortium of Infectious Disease Modeling Hubs. Coordinating collaborative infectious disease modeling projects with the hubverse. medRxiv https://doi.org/10.1101/2025.10.03.25337284 (2025).

18. Borchering, R. K. et al. Public health impact of the U.S. Scenario Modeling Hub. Epidemics 44, 100705 (2023).

    Article  CAS  PubMed  Google Scholar 

Download references

Funding

A.V., M.A.J. and M.M. acknowledge support from the CDC-RFA-FT-23-0069 cooperative agreement from the CDC’s Center for Forecasting and Outbreak Analytics. A.V. acknowledges support from the Gates Foundation INV-094330. M.M. acknowledges support from the Centers for Disease Control and Prevention (CDC) through the Pathogen Genomics Centers of Excellence network (PGCoE) grant 6NU50CK000555-03-01, the Defense Threat Reduction Agency (DTRA) contract HDTRA120F0017, and the National Science Foundation (NSF) through the Expeditions in Computing Grant CCF-1918656. The findings and conclusions in this study are those of the authors and do not necessarily represent the official position of the funding agencies, the National Institutes of Health, or the Department of Health and Human Services. Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the U.S. Government. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Author information

Authors and Affiliations

1. Bouvé College of Health Sciences and Network Science Institute, Northeastern University, Boston, MA, USA

   Alessandro Vespignani & Michael A. Johansson

2. Institute for Scientific Interchange Foundation, Turin, Italy

   Alessandro Vespignani

3. Department of Epidemiology, University of North Carolina Gillings School of Global Public Health, Chapel Hill, NC, USA

   Justin Lessler

4. Carolina Population Center, University of North Carolina at Chapel Hill, Chapel Hill, NC, USA

   Justin Lessler

5. Department of Epidemiology, Johns Hopkins Bloomberg School of Public Health, Baltimore, MD, USA

   Justin Lessler

6. Biocomplexity Institute and Department of Computer Science, University of Virginia, Charlottesville, VA, USA

   Madhav Marathe

7. Fogarty International Center, National Institutes of Health, Bethesda, MD, USA

   Cécile Viboud

Authors

1. Alessandro Vespignani
2. Michael A. Johansson
3. Justin Lessler
4. Madhav Marathe
5. Cécile Viboud

Contributions

A.V., M.A.J., J.L., M.M. and C.V. contributed to the conceptualization and writing of this article. All authors reviewed and approved the final manuscript.

Corresponding author

Correspondence to Alessandro Vespignani.

Ethics declarations

Competing interests

The authors declare no competing interests.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

Reprints and permissions

About this article

Cite this article

Vespignani, A., Johansson, M.A., Lessler, J. et al. Defining epidemic modeling as a public service. Commun. Health 1, 35 (2026). https://doi.org/10.1038/s44528-026-00019-0

Download citation

• Received: 28 March 2026

• Accepted: 26 June 2026

• Published: 07 October 2026

• Version of record: 07 October 2026

• DOI: https://doi.org/10.1038/s44528-026-00019-0