Dear Editor,

Continuous positive airways pressure (CPAP) treatment was used increasingly in the UK and elsewhere for coronavirus disease 2019 (COVID-19) to avoid the need for mechanical ventilation [1,2,3]. Early identification of patients benefitting from CPAP would assist decision making however we found that intensive care unit (ICU) admission respiratory parameters (rate, PaO2/FiO2 ratio) were poorly discriminatory. Using electronic healthcare records we retrospectively studied COVID-19 patients admitted to the University College London Hospital ICU between 8 March and 5 June 2020 in moderate-to-severe respiratory failure who received initial management with CPAP to see whether better prognosticators could be identified.

Comparison of demographic, clinical and biochemical parameters measured on ICU admission was made between patients adjudged CPAP success (hospital survival with CPAP alone) or failure (either death where CPAP was a ceiling of treatment, or need for mechanical ventilation regardless of hospital outcome). Requirement for other organ support was also compared. Patients managed on CPAP in non-ICU wards were excluded as detailed clinical and lab data were often lacking.

Of 108 within-hospital admissions, 93 (86%) received CPAP as initial respiratory failure management (median[IQR] PaO2/FiO2 ratio 13 [10–18] kPa) (Supplementary Fig. 1). Thirty-two (34%) were adjudged CPAP successes and 61 (66%) failures (14 deaths (23%) with CPAP as the ceiling of treatment, 47 (77%) requiring invasive ventilation of whom 26 died). Demographics and ICU admission values of respiratory rate, inspired oxygen concentration (FiO2) and PaO2/FiO2 ratio were similar between groups (Supplementary data). Admission values of C-reactive protein (CRP) (p < 0.0001), N Terminal-pro-B-type natriuretic peptide (NT-proBNP) (p < 0.001), troponin-T (p < 0.001) and d-dimers (p < 0.05) were significantly higher in CPAP failure patients (Fig. 1). Other organ support was only required for CPAP failure patients receiving invasive ventilation (46 vasopressors, 28 renal replacement therapy), but none for CPAP successes. At 6 h post-CPAP the PaO2/FiO2 ratio rose by 76.7% (37.9 to 99.8%) in CPAP success patients but only by 38.1% (−24.4 to 100.5) in the failure group (p = 0.015).

Fig. 1
figure 1

Biomarkers and variables recorded on ICU admission in patients receiving CPAP. Shaded area denotes normal range. The normal range for C-Reactive Protein (≤ 5 mg/L) is not indicated, FiO2 fraction of inspired oxygen concentration; PaO2/FiO2 ratio of arterial partial pressure to the fraction of inspired oxygen concentration. Shaded area denotes normal range except for PaO2/FiO2 where the increasing shades of grey denotes mild, moderate and severe respiratory failure according to the Berlin definition, NT-proBNP N terminal-pro B-type natriuretic peptide

Full size image

For outcome prediction (CPAP failure), sixteen potential predictor physiological or biochemical variables were assessed. Details are provided in the Supplementary Index. In brief, separate univariable logistic regression models were fitted for each predictor, with p values < 0.05 determining variables entered into a multivariable logistic regression model. A model combining elevated CRP and NT-proBNP predicted CPAP failure with a sensitivity of 0.75 (95% CI 0.62–0.86), specificity of 0.83 (0.61–0.95), positive predictive value of 0.91 (0.78–0.97) and negative predictive value of 0.59 (0.41–0.76).

Failure of CPAP to avert death or invasive mechanical ventilation was associated with increased blood levels of thrombo-inflammatory and cardiac injury/dysfunction biomarkers on ICU admission. This may reflect an increased incidence of pulmonary thrombi either identifiable on imaging or multiple microthrombi, with subsequent right heart strain. The much-increased requirement for vasopressor and renal support in invasively ventilated patients reflects disease severity, but a significant iatrogenic contribution from high airway pressures, heavy sedation and hypovolaemia cannot be excluded. The high positive predictive value using elevated CRP and NT-proBNP may identify patients likely to fail CPAP. If validated, this could be gainfully used in management pathways and to stratify patients for intervention studies.