discoveries magazine

Model Predicts How Many COVID-19 Patients Will Need Care

An illustration of scientists looking for predictive models for COVID-19.

To help uncloud the future impact of the pandemic, forecasting models developed at Cedars-Sinai are predicting, with sharp accuracy, how many patients are likely to need care at the medical center week by week.

Sophisticated algorithms make deductions based on rates of infection and hospitalizations across Los Angeles County, which are then combined with nuanced knowledge of Cedars-Sinai’s current patient population. The predictions ensure that the medical center is prepared with enough ventilators, personal protective equipment and intensive care beds.

The enterprise data intelligence team has been training machine-learning software for years to help best prepare Cedars-Sinai for patients with any number of conditions. For COVID-19 predictions, the team runs about 150 models every day, then compares them to reality to better inform the next day’s models.

One big variable: shifting physical-distancing guidelines, and whether people adhere to them. Even a 10% change in social interaction is likely to affect how many people become sick enough to require hospitalization.