Awake Prone Positioning

Background

Patients with coronavirus disease 2019 (COVID–19) may need hospitalization for supplemental oxygen [1, 2]. A substantial proportion of hospitalized COVID–19 patients need admission to the intensive care unit (ICU) for acceleration of respiratory support [1, 2].

Hypoxemia can be treated though the use of low–flow oxygen support systems like a nasal prong or nasal cannula, or non–rebreather or Venturi masks––this type of support is usually provided in a normal ward. Patients with COVID–19 with refractory hypoxemia may benefit from low–flow oxygen support, e.g., high–flow nasal oxygen (HFNO) or continuous positive airway pressure (CPAP)––this type of oxygen support usually mandates admission to an ICU [3]. If oxygenation still does not improve, awake prone positioning, or ‘self–proning’, has the potential to improve oxygenation, as such reducing the need for noninvasive or invasive ventilation [4, 5].

As part of a national multicenter observational study, named the ‘Practice of Adjunctive Treatments in ICU Patients with COVID‒19’ (PRoAcT–COVID) study, we collected detailed information regarding the types of oxygen support used, and rescue therapies for refractory hypoxemia applied in critically ill COVID–19 patients in the second wave of the national outbreak in the Netherlands [6]. In this current analysis, we determined (a) the incidence, timing, frequency and duration of self–proning; (b) the association of self–proning with need for invasive ventilation; and (c) the association of self–proning with mortality. We also determined which types of oxygen supplementation were used during self–proning.

Methods

Design

Preplanned analysis of the PRoAcT–COVID study, a service review performed during the first 3 months of the second wave of the national outbreak.

Patients

Patients were eligible for participation in the PRoAcT–COVID study if: (1) admitted to one of the participating ICUs between October 2020 and January 2021, and (2) diagnosed with COVID–19 confirmed by RT–PCR. Patients aged < 18 years were excluded. For the current analysis, we excluded patients that were transferred from or to another non-participating ICU within the first 2 days of ICU admission.

Exposure

The exposure of interest is use of awake prone positioning.

Endpoints

The primary endpoint of this analysis is a combination of the incidence, timing, frequency and duration of self–proning. Secondary endpoint is a combined endpoint of mortality at day 28 and escalation of respiratory support to invasive ventilation in patients that were initially supported without invasive ventilation. Additionally, length of stay (LOS) in ICU and hospital, and the mortality rates in the ICU and hospital at day 90 will be analyzed.

Statistical analysis plan

Categorical variables are presented as counts (frequencies), and continuous variables are presented as medians (interquartile ranges [IQRs]). In the group of patients that were not intubated immediately or just before ICU admission, we created 2 groups, based on whether or not self–proning was applied during ICU admission. Independent categorical variables were compared with Fisher exact test, and continuous variables with Wilcoxon rank-sum test.

The incidence of self–proning is expressed as the proportion of patients that were placed in a prone position while not receiving invasive ventilation. Timing of self–proning is expressed as the number of days between ICU admission and start of the first session of self–proning. Frequency is expressed by the number of days a patient received one or more self-proning sessions. Duration of self–proning is expressed as the mean number of hours of each self–proning session. Types of oxygen support used at the day of initiation of self–proning are reported as numbers and percentages.

Since the exposure was not randomly assigned, we will use a time-dependent propensity score for adjustment. This approach is designed to account for the fact that the exposure to the intervention of interest might not occur during the study if improvement occurs or termination of efforts (death) occurs first. The propensity score will be calculated based on a time-dependent Cox proportional hazards model with exposure during follow-up as the dependent variable and based on baseline characteristics and daily information. The outcome of this Cox model will be time to intervention during the follow-up, and patients will be censored if the follow-up ended (due to discharge or death) without the intervention. The propensity score for each patient will then be derived from the Cox model as the hazard component (i.e. the linear predictor) at any given moment from the model.

To calculate the propensity score the following baseline variables will be included based on clinical relevance: age, gender, severity of ARDS [7], body mass index (BMI). In addition, the following covariates assessed daily will be included: type of oxygen therapy used, severity of ARDS (based on PaO₂ / FiO₂ ratio) and use of vasopressor.

Then, a 1:1 risk set matching on the propensity score will be performed using a nearest neighbor–matching algorithm with a maximum caliber of 0.01 of the propensity score. Patients receiving the intervention at any given moment will be separately and sequentially propensity score matched with a patient who was at risk of receiving the intervention within the same moment. At-risk patients include those who were still undergoing treatment and did not receive the intervention before or within the same moment. At-risk patients therefore also included patients who received the intervention later, as the matching should not be dependent on future events [8]. As such, the matched group with no intervention includes patients who subsequently received the treatment (although later than their matched counterpart). In all analyses, the time-dependent exposure will be considered a stochastic process (counting process) that equals zero from time 0 until exposure, then it equals to one until the end of observation.

The performance of matching will be assessed through standardized differences between baseline characteristics. Binary outcomes will be compared with mixed-effect generalized linear models with binomial distribution and expressed as odds ratio and 95% confidence interval (CI). Continuous outcomes will be compared with mixed-effect generalized linear models with Gaussian distribution and expressed as mean difference and 95% CI. Time-to-event outcomes will be assessed with (shared-frailty) Cox proportional hazard models. ICU and hospital length of stay will be analyzed with a clustered Fine-Gray competing risk model with death before the event as competing risk. In all models the hospitals will be included as random effect to account for potential clustering. Wherever appropriate, Kaplan-Meier curves will be used to present time-to-event outcomes.

All hypothesis tests will be 2-sided, with a significance level of P < .05. All secondary analyses should be considered exploratory as no adjustments will be made for multiple comparisons. Statistical analyses will be conducted using R v.4.0.3.

References

1. Guan, W.J., et al., Clinical Characteristics of Coronavirus Disease 2019 in China. N Engl J Med, 2020. 382(18): p. 1708-1720.

2. Adhikari, S.P., et al., Epidemiology, causes, clinical manifestation and diagnosis, prevention and control of coronavirus disease (COVID-19) during the early outbreak period: a scoping review. Infect Dis Poverty, 2020. 9(1): p. 29.

3. Azoulay, E., et al., International variation in the management of severe COVID-19 patients. Crit Care, 2020. 24(1): p. 486.

4. Cinesi Gómez, C., et al., Clinical Consensus Recommendations Regarding Non-Invasive Respiratory Support in the Adult Patient with Acute Respiratory Failure Secondary to SARS-CoV-2 infection. Arch Bronconeumol, 2020. 56 Suppl 2: p. 11-18.

5. Rosén, J., et al., Awake prone positioning in patients with hypoxemic respiratory failure due to COVID-19: the PROFLO multicenter randomized clinical trial. Critical Care, 2021. 25(1): p. 209.

6. Valk, C.M.A., et al., Practice of adjunctive treatments in critically ill COVID-19 patients-rational for the multicenter observational PRoAcT-COVID study in The Netherlands. Ann Transl Med, 2021. 9(9): p. 813.

7. Ferguson, N.D., et al., The Berlin definition of ARDS: an expanded rationale, justification, and supplementary material.Intensive Care Med, 2012. 38(10): p. 1573-82.

8. Lu, B., Propensity score matching with time-dependent covariates. Biometrics, 2005. 61(3): p. 721-8.