Study design and population
This was a retrospective, single center cohort study in the Maasstad Medical Centre, which is the largest (600-bed) non-academic teaching hospital in the Rotterdam area, the Netherlands. The study was approved by the Hospitals Medical Ethical Committee, application number W20.081.
The study included patients older than 18 years who were admitted to the hospital with a COVID-19 infection and severe respiratory insufficiency between the 9th of March 2020 and 1st of May 2020. SARS-CoV-2 diagnoses were confirmed using PCR techniques on a nasopharyngeal swab with a Roche analyzer upon admission.
Mechanical ventilation in the ICU was declined in a portion of the admitted patients because of clinical frailty status, performance status or history of disease (e.g. severe cardiopulmonary conditions). These decisions were made in accordance with the patient and his or her relatives and were discussed on a daily basis in a multidisciplinary team together with Critical Care, Internal Medicine, Pulmonology and Palliative Care Specialists. During these meetings, performance scores, comorbidities, cognitive functioning, and frailty informed decision-making. When there was unanimous agreement that a patient was too frail for mechanical ventilation, HFNC was the only alternative treatment. All of these patients were included in the study. When starting HFNC, a flow of 60 liter per minute with a temperature of 37 degrees Celsius was used and further oxygen fractions were titrated based on oxygen saturation (SpO2 >92%) and respiratory rate (<25 per min).
Data were collected using SQL Server Management Studio (version 18.3) from the electronic patient record (Chipsoft: Healthcare Information X-change). The hospital’s data manager (GW) performed a first check after automatic extraction. A final check on the database was then performed by the two principal investigators (JvS and MvH).
Demographics, Medical history and Drug use
Demographic data and medical history were extracted from the medical records. The following demographic data were extracted: age, gender, date of admission, days of hospital stay, body mass index (BMI) and survival. Comorbidities were scored, including hypertension (defined as the use of antihypertensive drugs), diabetes mellitus type 2 (defined as fasting plasma glucose level ≥ 6.1 mmol/L), asthma (bronchodilator use and spirometry with reversibility), chronic obstructive pulmonary disease (COPD Gold classification), smoking (current or former), chronic kidney disease (eGFR <60ml/min/1,73 m2), malignancy (history of malignant neoplasm), occlusive peripheral arterial disease, ischemic heart disease (defined as obstructive coronary artery disease), non-ischemic heart disease, liver disease (radiological or pathological steatosis or cirrhosis) and the total number of comorbidities. Furthermore, out of hospital drug use (immunosuppressive drugs, ACE inhibitors and non-steroidal drugs) and in hospital drugs were scored (cefuroxime and azithromycin).
Frailty assessment and clinical prediction scores
The Clinical Frailty Score  ranging from 1 (very fit) to 9 (terminally ill) was used for clinical frailty assessment. In addition, WHO performance status score ranging from 1 (fully active) to 4 (completely disabled)  was extracted from the medical record. Sequential organ failure assessment scores (SOFA) were collected within 24 hours of admission for the prediction of clinical outcomes. Points for SOFA item ‘mechanical ventilation’ were assigned for patients on HFNC within the first 24 hours of admission.
Signs and clinical parameters
Clinical parameters were extracted from the medical record for every patient on admission, these include pulse rate, blood pressure, temperature, respiratory rate and oxygen saturation. Patients were asked about the presence and onset of the following symptoms: cough, dyspnea, weight loss, diarrhea and nausea.
Laboratory values and Radiological findings
For each patient blood examinations including hemoglobin, leukocyte count with differential count, platelet count, d-dimer, alanine transaminase (ALAT), aspartate transaminase (ASAT), creatine kinase (CK), lactate dehydrogenase (LDH), troponin T and ferritin levels were collected. These specific laboratory measures were chosen because most of them have already proved to be discriminating between mild and severe clinical courses . A chest X-ray was performed upon admission in every patient. If there was a high clinical suspicion of pulmonary embolism with elevated d-dimer levels in absence of pulmonary infiltrates on chest x-ray, an additional computed tomography (CT) pulmonary angiogram was performed.
High Flow Nasal Cannula therapy
The number of days since admission was registered before the start of HFNC. Furthermore, flow (liters per minute), fraction of inspired oxygen (FiO2) and reasons for failure of HFNC therapy were extracted from the medical record. FiO2 increase of ten percent in the first 24 hours were scored separately.
Fever was defined as a tympanic temperature lower than 36 or greater than 38 degrees Celsius. Patients were considered infected if they were proven positive for SARS-CoV-2 using a nasopharyngeal PCR test.
Respiratory failure was defined as persisting hypoxemia despite maximum conventional oxygen administration e.g. saturation (SpO2) lower than 92% despite maximum oxygen administration with a non-rebreathing mask (15 liters per minute). To assess the severity of Acute Respiratory Distress Syndrome (ARDS) SpO2/FiO2 ratios were calculated at start of HFNC therapy, as it was impossible to collect pO2/FiO2 ratios for all patients and because SpO2/FiO2 can be considered equally sensitive and specific as compared to PO2/FiO2 ratios.
Baseline data is presented as median (IQR) or n (%). Differences between survivors and non-survivors were compared using Mann Whitney U test for continuous data and Fisher’s exact test for categorical data. P-values <0.05 were considered to be statistically significant. All statistical analysis were performed using The R Project for Statistical Computing (version 4.0.3) . Variables with clinical relevance and in between group differences were used in a univariate logistic regression model to calculate odds ratios in order to assess factors associated with in hospital mortality. No multivariate analysis was performed because of high risk of overfitting the model due to the size of the cohort.