Factors associated with high-flow therapy failure in COVID-19-related acute hypoxemic respiratory failure: A prospective single-centre cohort study.

INTRODUCTION The optimal selection of non-invasive respiratory support for patients with COVID-19-related acute hypoxemic respiratory failure remains uncertain. High-flow therapy (HFT) is frequently used as initial support, but early identification of patients at risk of escalation is clinically relevant. We evaluated clinical and laboratory variables associated with escalation from HFT to non-invasive positive-pressure ventilation (NPPV).

Methods

We conducted a prospective single-centre cohort study including 139 consecutive patients initiated on HFT in a respiratory intermediate care unit. The primary outcome was HFT failure, defined as escalation from HFT to NPPV. Time to HFT failure was analysed from HFT initiation using cause-specific Cox regression; successful HFT withdrawal without escalation was treated as censoring. A parsimonious adjusted model included age, time from symptom onset to HFT initiation, baseline PaO2/FiO2 ratio, and lactate dehydrogenase (LDH). Logistic regression and a supplementary bicarbonate model were used as sensitivity analyses.

Results

Seventy-two of 139 patients (51.8%) required escalation to NPPV. Seventeen escalations were recorded on day 0 and 44 of 72 failures (61.1%) occurred by day 1. In the adjusted Cox model, higher LDH levels were associated with an increased hazard of escalation (adjusted hazard ratio [aHR] per 100U/L, 1.33; 95% confidence interval [CI], 1.06-1.66; p=0.014), whereas higher baseline PaO2/FiO2 ratios were associated with a lower hazard (aHR per 25mmHg, 0.78; 95% CI, 0.67-0.90; p=0.001). A longer interval from symptom onset to HFT initiation was also associated with a lower hazard of escalation (aHR per day, 0.93; 95% CI, 0.86-0.99; p=0.033). In a supplementary sensitivity model, higher arterial bicarbonate levels were associated with a lower hazard of escalation (aHR per 3mmol/L, 0.81; 95% CI, 0.66-0.99; p=0.039).

Conclusion

In this prospective single-centre cohort, LDH and baseline PaO2/FiO2 were associated with escalation from HFT to NPPV after adjustment. Arterial bicarbonate showed a supportive exploratory signal in sensitivity analysis. These findings require external validation before incorporation into treatment-selection algorithms.