Using advanced analytics to prevent adverse outcomes from COVID-19
ACT Center paper describes a novel predictive model that identifies patients at high risk of hospitalization or death at Kaiser Permanente Washington
November 12, 2025
Using advanced analytics to prevent adverse outcomes from COVID-19
ACT Center paper describes a novel predictive model that identifies patients at high risk of hospitalization or death at Kaiser Permanente Washington
More than 5 years after the onset of the COVID-19 pandemic, infection with the disease remains an ongoing public health concern and a leading cause of hospitalization and death in the United States. Research has shown that anti-viral medications, such as Paxlovid®, are effective in helping prevent these adverse outcomes in people who are at high risk of developing a serious infection.
This sounds like simple evidence to act on. But it leaves health systems, care teams, and patients with the challenging task of making informed treatment decisions that weigh the patient’s personal risk of hospitalization and death against the downsides of anti-viral medications — such as side effects, high costs, and a short window of time within which treatment is effective.
To meet this challenge, health systems need a quick and accurate way to identify patients at high risk of adverse outcomes from COVID-19, so they can ensure that potentially lifesaving medications are offered to the patients who need them most. Thanks to a novel COVID-19 risk prediction model developed by the ACT Center’s advanced analytics team, Kaiser Permanente Washington has had this capability since 2023. And now, in a new paper published in Prevention Science in October, the team is sharing a closer look at the model and how they developed it in hopes that other health systems can use their results to benefit patients nationwide.
What makes the model unique
Early in the pandemic, experts pointed to accurate prediction of poor outcomes from COVID-19 infection as a key strategy to inform rapid and effective treatment. Since then, several tools to predict hospitalization or death have been developed. But they vary widely in their value across different populations and settings, and most require regular validation and data updates over time, which can make them difficult to implement in routine care. Many risk models also focus specifically on risk of death after hospitalization, which is often past the window when anti-viral treatments are most effective.
The Kaiser Permanente Washington model is unique in part because it focuses on preventing both hospitalization and death in an outpatient population. It was also designed in collaboration with clinicians and health system leaders — to ensure it could be implemented at scale and sustained over time.
“To our knowledge, our COVID risk prediction model is one of few to be developed and validated in a more recent, vaccinated population and to account for how future data on patient outcomes may look different than historical data,” said the paper’s lead author Yates Coley, PhD, an associate biostatistics investigator at Kaiser Permanente Washington Health Research Institute who leads the ACT Center’s predictive analytics work.
“The result is a high-performing, real-time risk prediction tool that’s integrated into our electronic health record,” said coauthor Paul Thottingal, MD, who serves as Kaiser Permanente Washington’s senior medical director for communicable diseases and organizational preparedness. “This is a pivotal new tool to support shared decision-making for clinicians and patients for COVID-19, and it’s being adopted in other care settings to support shared decision-making.”
Senior author Emily Westbrook, MHA, who codirects the ACT Center and is a member of the advanced analytics team, pointed to another of the model’s benefits. “We designed a risk tool that can be easily updated and adapted by other health systems or agencies that have access to clinical records data. Our hope is that this will inspire and inform predictive analytics work that helps prevent COVID-related hospitalizations and deaths in other settings.”
Developing a model that’s both practical and equitable
Racial and ethnic disparities in rates of COVID-19 infection and adverse outcomes emerged early in the pandemic and persist today. More recently, studies have shown that Latino and non-Hispanic Black patients are less likely to be treated with anti-viral medications compared to non-Hispanic white patients.
Coley and the ACT Center team intentionally developed their risk model to ensure that its use in clinical care would help address these disparities rather than exacerbate them. First, they incorporated race and ethnicity into the model to account for the increased risk of hospitalization and death experienced by American Indian or Alaska Native, Asian or Pacific Islander, Black, and Hispanic people. This ensures that more patients from these groups are flagged as high risk — making them more likely to be offered treatment and to receive it. Second, they evaluated how well the model performed across racial and ethnic subgroups to confirm comparable accuracy and strong model performance over time.
Designing analytic tools that help advance health equity is not new to Coley, who published a study in JAMA Psychiatry in 2021 that examined racial inequity in suicide prediction models and urged researchers to test model performance across racial and ethnic groups. In 2023, Coley received an Emerging Leader Award from the Committee of Presidents of Statistical Societies, which highlighted their contributions to the ethical development of clinical prediction models and their commitment to equity and inclusion.
“Many factors contribute to racial inequities in health care,” said Coley. “When we’re intentional about incorporating race and ethnicity into machine learning tools, they have the potential to help health systems deliver equitable treatment that reduces the risk of serious illness or death in populations that have experienced longstanding disparities.”
