Trial design and oversight
We conducted a parallel randomized controlled trial at two university hospitals in western Mexico, in which three high-acuity units dedicated to treat COVID-19 patients were included, with patient-to-nurse ratio of 4:1 in two units and 2:1 in one unit. The protocol was approved by the ethic committees at both hospitals. This trial was registered at clinicaltrials.gov with the identification number NCT04477655.
Patients aged ≥ 18 years with reverse-transcriptase polymerase chain reaction (RT-PCR) confirmed COVID-19, and pulse oximetry (SpO2) < 90% despite receiving oxygen at 15 L/min through a non-rebreather mask, were initiated on high-flow nasal cannula (HFNC) and assessed for eligibility. Inclusion criteria were the requirement of a fraction of inspired oxygen (FIO2) ≥ 0.3 through HFNC at the maximum tolerated flow to maintain a SpO2 ≥ 90%. Exclusion criteria included severe respiratory failure requiring immediate intubation, do-not-intubate/resuscitate orders, laparotomy within 2 weeks, pregnancy, vasopressor requirement to maintain median arterial pressure > 65 mmHg, and refusal to participate.
For all COVID-19 patients admitted to our hospitals, silent hypoxemia was assessed before initiating oxygen therapy and documented by clinicians at hospital admission using a prespecified definition of SpO2 < 90% at ambient air without perception of dyspnea or shortness of breath [13, 14].
Randomization and intervention
After signing informed consent, patients were randomly assigned to receive standard care or APP using a predetermined randomization sequence prepared in sealed opaque envelopes. The sequence was generated by computer and stratified by center, with a 1:1 allocation ratio, using permuted blocks with a size of 4. Due to the nature of the intervention, blinding was impossible.
HFNC was initiated at maximal flow of 40 L/min (Vapotherm, Precision flow, Exeter, NH, USA), with FIO2 titrated to maintain SpO2 between 92% and 95%. Flow and temperature were then titrated based on patient comfort. HFNC was maintained until patients met weaning criteria, defined as a requirement of a FIO2 ≤ 0.4 and flow ≤ 20 L/min to maintain SpO2 ≥ 90% for ≥ 2 h; death; need for noninvasive ventilation (NIV) or intubation.
Study investigators maintained daily communication with on-site clinicians minimally three times/day to emphasize the protocol adherence. Labels indicating the allocation arm were placed inside patient rooms to remind the patients to stay in the assigned group. Patients in the APP group were consistently encouraged by the bedside clinicians to remain in APP as long as possible, and assistance to reposition was offered as needed, with pillows placed at chest, pelvis, and/or knees to improve comfort. Personal cell phone use with Internet connection was encouraged, in order to distract patient attention and maximize their tolerance for APP.
During the first APP session, patient’s physiological variables pre-APP, 1 h during APP, and 1 h after returning to supine position were recorded, including respiratory rate, SpO2/FIO2, and the ratio of SpO2/FIO2 to respiratory rate, known as the ROX index . The APP duration in each session was recorded by nurses. Patients who received APP < 1 h/day during HFNC were excluded from the per-protocol analysis.
In the standard care group, APP was discouraged. If APP was performed for ≥ 1 h, patients were excluded from the per-protocol analysis.
Based on early evidence of prognostic role of D-dimer levels, they were prospectively recorded for all patients at enrollment .
Attending physicians, who were all certified trainers of WINFOCUS (World Interactive Network Focused on Critical Ultrasound, https://www.winfocus.org/), recorded daily lung ultrasound (LUS) score in all enrolled patients under supine position from study enrollment and until the third day. The LUS score was interpreted and recorded by the same clinicians, based on the lung ultrasound pattern in the ventral, lateral, and dorsal sections of both lungs (12 sections), with maximal score of 3 in each section and 36 in total, higher score indicating lower aeration [17, 18].
All patients were followed up until 28 days of enrollment, with intubation as the primary outcome. Intubation criteria were predetermined: respiratory rate ≥ 40 breaths/min, respiratory muscle fatigue, respiratory acidosis with pH ≤ 7.25, copious tracheal secretions, SpO2 < 80% despite an FIO2 of 1.0, hemodynamic instability, anxiety due to dyspnea, or deteriorating mental status.
Secondary outcomes included treatment success, which was defined as being alive without intubation at day 28, mortality at 28 days, HFNC duration, use of NIV, time to intubation, days of invasive ventilation, hospital length of stay, physiological response to the first APP session, and adverse events.
Exploratory outcomes included the predictors for treatment success in the overall population and in the APP group, the association between treatment success and mean daily duration of APP, and the difference in outcomes among patients with silent hypoxemia.
Sample size and statistical analysis
Assuming an intubation rate of 38%, the calculated sample size was 468 patients (234 per group) in order to detect an absolute difference of 12% to provide a statistical power of 80% and α = 0.05.
The predictors for treatment success in the overall population and in the APP group were assessed with multivariate logistic regression analysis, using the enter method. A linear relationship was investigated between probability of success and the duration of APP within the first 3 days, and it was plotted in a scatter diagram. Receiver operating characteristic (ROC) curve analysis was performed for duration of APP and other continuous predictive variables, using treatment success as the end point; Kaplan–Meier curves were plotted for treatment success and death between patients above and under the best cutoff according to the Youden index (8 h) and were compared with the log-rank test and adjusted with a Cox proportional hazards regression. We performed a subgroup analysis of the outcomes in patients with silent hypoxemia at hospital admission. A two-tailed P value of less than 0.05 was considered as statistically significant. We used SPSS software, version 27 (IBM), and GraphPad Prism software (version 9) for all the analyses and graphics.