Designing COVID-19 staged alert systems

Like Comment
Read the paper

Throughout the COVID-19 pandemic, governments worldwide have implemented social distancing measures in an effort to curb transmission. Some of the more stringent interventions can exact substantial socioeconomic costs, including lost wages, educational setbacks, and harms to physical and mental health. To balance competing public health and societal demands, policymakers have adopted adaptive guidelines for enacting and lifting restrictions. In particular, there are numerous COVID-19 alert systems that monitor case counts, test positivity, hospital usage, or other key indicators to trigger lockdowns or more moderate combinations of non-pharmaceutical interventions. Although alert systems vary greatly in their specifics, they are generally designed to promote public awareness and compliance with policy changes. 

As the world continues to combat COVID-19 and faces the specter of future pandemics, carefully designed alert systems may provide a path forward. We should look back and ask how existing systems have fared over the last 18 months and develop a roadmap for building robust systems---for rapidly identifying reliable combinations of data, triggers, and measures---in advance of or during future threats.

In our paper, “Design of COVID-19 staged alert systems to ensure healthcare capacity with minimal closures” we describe the optimized alert system that has anchored public awareness and governed COVID-19 responses in Austin, Texas since May of 2020. The system’s design rests on two pillars. First, the system tracks daily new COVID-19 hospital admissions as a reliable real-time indicator of recent transmission and an imminent healthcare surge. Second, using this indicator, thresholds were derived to trigger changes in alert level using a data-driven optimization model, which was pressure tested as the pandemic evolved. Specifically, we coupled stochastic optimization with a high-fidelity simulation of COVID-19 transmission and solved for policy thresholds that minimize the expected number of days spent in costly stages of lockdown while guaranteeing sufficient healthcare capacity with high probability.

In Austin, a task force including elected leadership, public health authorities, executives from all area hospitals, and scientists has guided the city’s COVID-19 reponse policies. Since May of 2020, the task force has tracked daily COVID-19 hospital admissions and enacted staged restrictions when the seven-day moving average crossed our optimized thresholds, as publicly displayed on a City of Austin’s COVID-19 Dashboard. On December 23, 2020, Austin transitioned to the most stringent alert stage (red). In February, March, and May 2021, as hospital admissions dropped below the corresponding thresholds, Austin relaxed to the orange, yellow, and then blue stages. As of August 2021, a resurgence fueled by the Delta variant has propelled Austin back into the orange stage. Throughout the pandemic, COVID-19 hospital admissions have not been routinely available. Austin’s efforts to track these counts, socialize the alert system, and provide timely risk communications have been an exemplar of collaborative data-driven local leadership. To date, the policy has achieved its public health and socioeconomic objectives. The COVID-19 mortality rate in Austin remains under 50% the overall rate in Texas. During the major COVID-19 surge in the winter of 2020-2021, area hospitals remained under capacity, even at the peak, and Austin spent the fewest days under a restrictive statewide order among all major Texas cities. 

How does Austin’s alert system compare to others around the globe? Our manuscript takes a first step towards answering this question. We simulated two other systems, ICU-based closure thresholds used by France and incidence-based triggers recommended by the Harvard Global Health Institute (HGHI). Our analysis suggests that France’s policy is not stringent enough to prevent overwhelming hospital surges and the HGHI policy forces longer lockdowns than needed to preserve healthcare capacity. 

Looking ahead, SARS-CoV-2 may become a seasonal threat, like influenza.  If COVID-influenza twindemics become reality and variants of concern continue to emerge, staged-alert systems may become permanent fixtures on multiple scales, from individual schools to countries. Having a validated model under the hood, like the one described in our paper, will be key to simultaneously protecting our healthcare systems and allowing our communities to thrive.

Bismark Singh

Dr., Friedrich-Alexander-Universität Erlangen-Nürnberg

Researcher in mathematical optimization