Predicting Failure at Initiation of High‐Flow Nasal Oxygen in Patients With COVID ‐19: Literature Review, Development and Internal Validation of a Prediction Model

ABSTRACT Background and Objective High‐Flow Nasal Oxygen (HFNO) can reduce the need for invasive mechanical ventilation in patients with acute hypoxemic respiratory failure (AHRF) from viral pneumonias, like COVID‐19. Early prediction of HFNO failure is useful for timely decision‐making at HFNO initiation. This study aimed to develop a prediction model for HFNO failure using predictors available just prior to HFNO initiation in patients with COVID‐19 AHRF and compare its performance to existing models. Methods This multicenter, prospective observational cohort study included hospitalized patients from 10 centers in the Netherlands between December 2020 and July 2021. Adults who tested positive for SARS‐CoV‐2, had no treatment limitations, and initiated HFNO for hypoxemia were included. The primary outcome was HFNO failure, defined as the event of endotracheal intubation. Pre‐defined candidate predictors were selected by multivariable logistic regression for prediction model development. Internal validation was conducted using bootstrapping. Results Out of 608 patients, 277 (46%) experienced HFNO failure. Independent predictors of HFNO failure included (odds ratio 95% CI): age (1.02 1.00–1.03), urea (1.04 1.00–1.08), platelet count (0.94 0.92–0.97), respiratory rate (1.05 1.02–1.08), oxygen saturation (0.89 0.84–0.94), and FiO 2 (conventional oxygen 10–15 L/min vs. < 10 L/min: 3.00 1.71–5.29, 15 L/min vs. < 10 L/min: 4.95 3.19–7.70) prior to HFNO initiation. The model C‐statistic was 0.767; 95% CI 0.727–0.803, with excellent calibration (intercept: −0.005, slope: 1.001), and stable performance after internal validation. Conclusions This newly developed model, using variables available at HFNO initiation, effectively predicted HFNO failure in hospitalized hypoxemic patients due to COVID‐19 pneumonia with good performance. Registration Number Clinical Trial Dutch Trial Registry: DTR, NL9067.