Collaborative modeling attempts could improve response to the COVID-19 pandemic

By Pooja Toshniwal PahariaReviewed by Aimee MolineuxApr 8 2022

In a recent commentary published in PNAS*, researchers retrospectively analyzed the collaborative modeling attempt developed by the University of Texas (UT) at Austin, and described by Fox et al, for guiding public health decisions during the coronavirus disease 2019 (COVID-19) pandemic.

Modeling and prediction attempts have aided public health decision-making during the COVID-19 pandemic by the local, state, and national authorities by increasing awareness of the situation at hand, providing information on important severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) characteristics, and optimizing strategies for mitigating COVID-19.

Forecasting attempts are the most evident modeling outputs for the public since predictions are frequently pointed out by the media. However, several other modeling attempts have also played an essential role in mitigating COVID-19.

Model methods

The UT model predicted the COVID-19 burden on healthcare by focusing on hospital metrics (hospital admissions, usage of intensive care unit [ICU] beds, and hospital beds) which are important indicators of the COVID-19 burden on healthcare.

For external validation, the team compared their model predictions for the cumulative fraction of infected people with an independent source of data, the Centers for Disease Control and Prevention’s (CDC) infection rate estimates.

Additionally, the authors carefully incorporated mobility data to improve model accuracy. However, the extent to which mobility data accounts for uncertainty depends on the situation, for example, with high vaccine uptake and high mask-wearing sites, mobility data would have a small role.

Model results and interpretation

Hospital admissions data could precisely and timely indicate recent viral transmission and forthcoming usage of hospital beds and ICU beds in the short term (one to two weeks). On the other hand, case data were poor indicators of prospective COVID-19 burden on health care since they showed significantly low correlation, probably due to changing trends in care-seeking and case reporting.

The breadth of the UT model was commendable. The model not only predicted the healthcare burden but also provided a real-time estimate of the reproduction number, which is an estimate of the rate of pathogen transmission. Thus, it could enable the provision of instant and time-sensitive feedback to policymakers that would aid the planning of resources by local hospitals, provide requests to federal and state authorities for additional resources in case of future pandemic surges, and increase care sites to improve healthcare capacities.

In addition, the availability of experts in the field of modeling and data processing to monitor the COVID-19 pandemic data enabled more confident government operations. The experts could make data adjustments and appropriately interpret information considering unforeseen disruptions and situations such as Texas’ uncommon winter freeze in the initial 2021 period.

Conclusion

The authors believed that the mobility-driven mechanistic UT model was an honest, careful, precise, externally valid, and judicious attempt at real-time and extensive collaborative modeling during the COVID-19 pandemic, that was tailored to the particular healthcare needs put forth by city officials.

Additionally, according to the authors, the model had a vast breadth as the model could not only predict the prospective short-term COVID-19 burden on healthcare but also provide instant policy feedback. The data adjustments by the model experts and the incorporation of robust evidence could also instill more confidence in the government for public decision-making in Austin.

The authors of the present study believe that the collaborative modeling attempt, described by Fox et al, is noteworthy due to its high accuracy and extensive collaboration, built on trust, with the Austin city officials. They believe that while this attempt would serve as an exemplary model for future collaborations of a similar kind, the scalability of the model is questionable.

The identification of the most valuable and real-time data streams for model estimations, incorporating methods other than mobility data to account for uncertainty would improve epidemic modeling. Furthermore, the use of ‘modeling hubs’ such as the United States (US) COVID-19 Forecast Hub and the Scenario Modeling Hub that would aggregate model outputs from several academic and industry modeling teams and generate results for several locations instantly would improve the availability of the model.

Continued investments in data modernizing, modeling technology, and the development of the workforce would improve model scalability and modeling capacity to enable local jurisdictions to benefit from the models.