Effect of imaging timepoint on AI predictions of ICU need and mortality

Ascertain whether the predictive values of artificial intelligence models for forecasting the need for intensive care or mortality in COVID-19 patients remain consistent when imaging is performed at admission versus at other time points during hospitalization.

Background

The authors highlight that most included studies acquired imaging at admission and explicitly state that it has not been evaluated whether AI model performance holds when imaging is obtained at other times.

Understanding temporal robustness is crucial for applying AI prognostic tools throughout a patient's clinical course and for defining optimal imaging timing for prediction.