Evolution of Life-Threatening COVID-19 as the SARS-CoV-2 Pandemic Has Progressed

RationaleAs the SARS-CoV-2 pandemic progressed, hospital mortality among patients admitted with COVID-19 decreased; however, its mechanism remains unclear. ObjectiveTo determine underlying factors that might explain the declining observed mortality of hospitalized patients with COVID-19. MethodsThis observational study used a prospective COVID-19 clinical database compiled by a 15-hospital health system. Participants were persons testing positive for SARS-CoV-2 (n=185,636), among whom 26,872 were hospitalized for COVID-19 from 3/1/2020 to 6/1/2022. MeasurementsEndpoints were hospital and 60-day mortality assessed by randomForests-for-survival machine learning using patient demographics, medical history, symptoms, and admission laboratory test results. Main ResultsMortality of patients hospitalized with COVID-19 fell from 11% in 3/2020 to 3.7% in 3/2022; 60-day mortality was 17% in 5/2020 and 4.7% in 5/2022. Advanced age was the most important predictor of mortality, followed by admission laboratory test results. Risk-adjusted 60-day mortality was 15% had all patients been admitted in 3/2020, minimally unchanged at 12% had they been admitted in 5/2022. Dissociation between observed and predicted mortality was largely explained by change in admission patient profile, particularly admission laboratory test results. Transition to intensive care occurred rapidly for those predicted to do poorly. ConclusionMortality from COVID-19 requiring hospitalization has declined as the pandemic has evolved, but surprisingly, persists for 60 days following admission. Demographics, medical history, and at-admission laboratory results continue to accurately predict mortality despite reduction of observed mortality and in spite of therapeutic advances. Importantly, rapid escalation of care can be predicted at admission using standardly obtained information. There has been a subtle but perceptible change in the at-risk population that explains these findings.