Prediction of the need for intensive oxygen supplementation during hospitalisation among subjects with COVID-19 admitted to an academic health system in Texas: a retrospective cohort study and multivariable regression model

ObjectiveSARS-CoV-2 has caused a pandemic claiming more than 4 million lives worldwide. Overwhelming COVID-19 respiratory failure placed tremendous demands on healthcare systems increasing the death toll. Cost-effective prognostic tools to characterise the likelihood of patients with COVID-19 to progress to severe hypoxemic respiratory failure are still needed.DesignWe conducted a retrospective cohort study to develop a model using demographic and clinical data collected in the first 12 hours of admission to explore associations with severe hypoxemic respiratory failure in unvaccinated and hospitalised patients with COVID-19.SettingUniversity-based healthcare system including six hospitals located in the Galveston, Brazoria and Harris counties of Texas.ParticipantsAdult patients diagnosed with COVID-19 and admitted to one of six hospitals between 19 March and 30 June 2020.Primary outcomeThe primary outcome was defined as reaching a WHO ordinal scale between 6 and 9 at any time during admission, which corresponded to severe hypoxemic respiratory failure requiring high-flow oxygen supplementation or mechanical ventilation.ResultsWe included 329 participants in the model cohort and 62 (18.8%) met the primary outcome. Our multivariable regression model found that lactate dehydrogenase (OR 2.36), Quick Sequential Organ Failure Assessment score (OR 2.26) and neutrophil to lymphocyte ratio (OR 1.15) were significant predictors of severe disease. The final model showed an area under the curve of 0.84. The sensitivity analysis and point of influence analysis did not reveal inconsistencies.ConclusionsOur study suggests that a combination of accessible demographic and clinical information collected on admission may predict the progression to severe COVID-19 among adult patients with mild and moderate disease. This model requires external validation prior to its use.