The unmet need for accurate early predictors of treatment failure and death during noninvasive ventilation (NIV)

NIV is being increasingly used as first-line ventilatory support in de novo (i.e., not due to exacerbation of chronic lung disease or cardiac failure) acute hypoxemic respiratory failure (AHRF). However, concern has been raised that mechanical ventilation may exacerbate lung injury (i.e., ventilation-induced lung injury, VILI) and worsen outcome in spontaneously breathing patients with AHRF [1]. Different ventilatory variables have been proposed to contribute to VILI progression, including high tidal volume [2], minute ventilation [3] and inspiratory effort [4]: these parameters have been separately addressed in clinical studies, but data in noninvasively ventilated patients are scarce, inconsistent and inconclusive, at least in part due to the difficulty in controlling and monitoring time-varying individual ventilatory parameters in spontaneously breathing patients.

Potential contribution of Mechanical Power (MP) to clinical prediction in the framework of PPPM

Predictive, Preventive, and Personalized Medicine (PPPM) is an effective integrative approach, which has been promoted by the European Association for Predictive, Preventive and.

Personalized Medicine (EPMA, http://www.epmanet.eu/) [5]. It contains three aspects: individual predisposition prediction, targeted preventive measures and personalized treatment algorithms [6].

In recent years, a unifying patho-physiological theory, based on thermodynamic principles, has been proposed to underlie VILI: this theory attributes lung injury to the energy transfer from the ventilator to the pulmonary parenchyma, with energy dissipation within the lungs leading to heat production, cell integrity and extracellular matrix disruption, and inflammatory cell recruitment [7-9]. Consistently, mechanical power (MP), a measure of the energy transfer rate from the ventilator to the respiratory system, predicted mortality in invasively ventilated ARDS patients, irrespective of the combination of each ventilatory component [10-13].

Whether the energy delivered to the respiratory system during noninvasive ventilation (NIV) affects clinical outcomes in AHRF is unexplored. During COVID-19 pandemic, NIV intensity and duration were linked to an increased mortality in COVD-19-related AHRF [14-16], but no individual ventilatory parameter, including respiratory drive and inspiratory effort, was able to predict clinical outcomes [17-20]. Furthermore, clinical benefits from awake prone position (PP) during NIV have been reported by several, but not all studies, through still unclear mechanisms [21, 22].

Working hypothesis

We hypothesized the energy delivered by noninvasive ventilatory assistance to the respiratory system could contribute to physio-anatomical and clinical responses to NIV in Severe Acute Respiratory Syndrome Corona Virus 2 (SARS‐CoV‐2) pneumonia-related AHRF, and that the clinical benefits of awake PP in these patients could be at least in part mediated by a reduced MP delivery during prone NIV. We therefore investigated:

1)

the effect of prone position (PP) on different measures of MP during NIV.

2)

the contribution of MP measures delivered early during either supine or prone NIV to physio-anatomical and clinical outcomes in COVID-19 pneumonia

Methods

In this secondary analysis of the non-randomized, controlled Prone position in NonInvasive Ventilation (PRO-NIV) study(study ID: ISRCTN23016116) [23], we studied 216 SARS‐CoV‐2 pneumonia patients with acute (i.e. symptom onset < 14 days of hospital admission) moderate-to-severe AHRF(paO2/FiO2 ratio < 200 mmHg while on a FiO2 50% Venturi mask or a non-rebreather reservoir bag mask): 108 patients treated with NIV (CPAP or PSV) plus early PP and 108 matched controls treated with supine NIV) at HUMANITAS Gradenigo COVID Subintensive Care Unit between June 1st 2020 and June 30th 2021.

In both groups, NIV was initiated within 24 h of admission to Subintensive Care Unit and delivered continuously for ≥ 48 h or until discharge or death; full-face mask was the initial interface of choice. PP was initiated within 24 h of admission in the PP group. NIV and PP duration, equipment, settings, standard care, monitoring, treatment failure criteria were protocolized a priori before patients enrollment (see supplementary text, Supplementary Fig. 1, Supplementary Table 1) [23].

Measurements Respiratory parameters

NIV duration, posture, ventilatory settings and parameters (spO2, RR, VTe, MV) were continuously monitored and recorded on an hourly basis on a predefined form.

ABGs were performed during NIV, every 24 h and ≥ 1 h after each postural change, after achievement of ventilatory stability (defined by a ≤ 10% variation in RR and VTe and air leaks < 10% for ≥ 30 min).

The following gas exchange parameters were calculated from ABG: paO2/FiO2 ratio and dead space indices (DSIs) [Ventilatory Ratio (VR) and corrected Minute Ventilation (MVcorr)]

The energy delivered by ventilatory assistance to respiratory system per time unit (mechanical power, MP) was estimated via Becher formula [24], which was already evaluated in noninvasively ventilated patients and was recommended to improve noninvasive ventilatory support monitoring in COVID-19 pneumonia [14, 25]. In this formula, ∆Pinsp was replaced by ∆Paw(airway pressure over PEEP) to reflect the energy imparted by ventilator during inspiration [1, 26].

Since NIV isa dynamic process, to reflect the power delivered during the entire ventilatory sessions, time weighted–average of hourly MP values during each NIV session (supine/prone) was calculated as the area under the parameter–versus–time plot (detailed in supplementary text) [27]. The same procedure was followed to calculate time weighted–average values of RR, VTe, and MV for each ventilator session and for each day.

For MV, we also planned to verify consistency between time-averaged values and those obtained at the time of ABGs, which were used to calculate DSIs.

The primary exposure variable of interest was MP normalized by the volume of well-aerated lung (MPWAL), to reflect the “intensity” of the power, i.e., the volume of well-aerated lung exposed to energy load during mechanical ventilation, during the 1st 24 h of NIV [MPWAL(day 1)].

We also normalized MP by predicted body weight (MPPBW) to account for individual lung size variation.

Lung imaging

All patients underwent a lung CT scan on admission: the nature and extent of parenchymal involvement were scored using a validated index [28] and the volume of well-aerated lung (WAL), of poorly aerated lung (PAL) and of non-aerated lung (NAL) were quantified on CT scans via a validated open-source software (3D Slicer ver.4.13.2) (see supplementary text) [29].

Lung ultrasound was performed daily from admission (day 0) to day 7 by three intensivists with expertise in lung and cardiac recording and interpretation (each operator having performed at least 50 supervised procedures and at least 200 non-supervised procedures) [30] using the same equipment (HM70A Samsung, Seoul, Korea), the same convex-array probe and the same setting.

The accuracy of ultrasound examinations in staging lung disease severity was preliminarily evaluated at baseline against the CT scan (double-blinded operators, LUS performed within 24 h of CT examination).

The severity and extent of parenchymal involvement of each of 6 lung regions (2 anterior, 2 lateral, 2 dorsal) were scored (range 0–3) [31] and recorded on a predefined form and the following indices were calculated (supplementary text):

regional and global lung ultrasound score (LUS);

regional and global LUS reaeration score, a validated index of lung recruitment (i.e., change from consolidated, non-aerated tissue to aerated tissue) [32, 33];

additionally, using software-based lung parenchyma segmentation and analysis function [29], each lung was divided into six areas to mirror as much as possible the regions explored by ultrasound, and regional and global LUS-derived WAL volume (i.e. lung volume with LUS score 0–1), PAL (i.e. lung volume with LUS score 2) and NAL (i.e. lung volume with LUS score 3) were derived from admitting LUS scan examinations as previously described in ARDS [34]. Global LUS-derived WAL volume, a predictor of COVID-19 pneumonia outcome [24], was then calculated from daily LUS scans through day 0–7(detailed in supplementary text).

The agreement between regional LUS score and regional CT classification was assessed with Cohen’s kappa coefficient, and the association between regional CT-derived gas/tissue content and regional LUS categories was assessed by simple linear regression and Spearman’s rank correlation (rs).

Details on full derivations are provided in supplementary text.

The PEEP at which each LUS examination was made was recorded.

Definitions and timepoints

We adopted the following definitions:

day 0 (baseline): the time of NIV initiation;

day 1: first 24 h after NIV initiation;

day 7: day 7 after NIV initiation;

timepoint sp0: supine position. In PP group, the session preceded the first PP session. ABG was performed ≥ 1 h after supine NIV initiation, after achieving ventilatory stability;

timepoint pp1: first PP session in PP group. ABG was performed ≥ 1 h after prone NIV initiation, after achieving ventilatory stability;

timepoint sp1: supine position; ABG was performed ≥ 1 h after resupination following the first PP session in the PP group and 24 h after NIV initiation in the supine group, after achieving ventilatory stability;

This schedule allowed comparing all groups in the supine position, after 24 h of NIV, while taking into account the effect of the first PP session (pp1) in the PP group.

Further definitions are provided in legends to Table 1 and in a study scheme (Supplementary Fig. 1).

Table 1 Baseline, treatment-related parameters and clinical outcomes of supine and PP group; within each treatment group, patients were divided according to median values of Mechanical PowerWAL during the 1st 24 h of NIV (MPWAL at day 1) into High MPwal (MPwal ≥ 9.1 J/min/L) or Low MPwal (MPwal < 9.1 J/min/L)(n = 216) Outcomes

Primary outcomes were the occurrence of.

NIV failure within 28 days of enrolment, defined as intubation OR death;

death, censored at 28 days after enrollment.

Secondary outcomes were:

endotracheal intubation (ETI) at 28 days (after excluding patients with a do-not-intubate, DNI, order);

60-day death

O2-response: paO2/FiO2sp1–paO2/FiO2sp0 (or ∆paO2/FiO2sp0-1).

CO2-response: ∆VRsp0-1;

C-reactive protein (CRP) response: ∆CRP0-1;

global LUS response at day 1: ∆global LUS0-1;

global reaeration score at day 1: global reaeration score1;

change in LUS- assessed WAL (%) at day 1 (∆WAL0-1)

Statistical analysesSample size calculation and propensity score (PS)-matching of PP and supine group.

Sample size calculation and PS-matching of PP and supine groups for relevant baseline covariates are detailed in supplementary text.

Descriptive statistics

Data are given as median (IQR) or n (%) as appropriate.

We used chi-square test or Fisher´s exact test for categorical variables, T-Test for normally distributed and Kruskal–Wallis test for non-normal continuous variables.

Time change in continuous variables was assessed by computing the AUC with the trapezoid method [36].

Repeated measures two-factor (within subject and between group) ANOVA was used to compare continuous variables assessed at multiple timepoints (i.e., respiratory and biochemical parameters), after log-transformation of non-normal variables.

To explore the effect of early MPWAL delivery on clinical outcomes, the whole cohort was split into 2 groups according to median MPWAL values at day 1. The probability of 28-day NIV failure, death and ETI in low vs high MPWAL group at day 1 was compared using Kaplan–Meier procedure and log-rank test. Data were analyzed on an intention-to-treat basis.

Beside categorizing patients into low/high MPWAL, we explored dose–response relationship between early power delivery and clinical outcomes by dividing the entire cohort into quartiles of power measures at day 1. Comparison between quartiles was made by ANOVA with post-hoc comparison from the first quartile performed using the Tukey test. We planned to assess also reciprocal relationship between MP, MPPBW and MPWAL at day 1 by univariate analysis and Spearman correlation coefficient (rs).

Multivariable Cox proportional regression analysis adjusting for imbalanced covariates between groups was used to assess the effect of confounders on 28-d NIV failure, death and ETI in the whole cohort, with the maximum number of covariates allowed in each model set at (event rate x N)/10, where N is the sample size [37]. The allocation assignment (PP or supine) was entered as a predefined covariate into the models. Calendar month of admission was forced into all models to account for unmeasured temporal disease trends during the pandemic.

Anticipating high collinearity between ventilatory variables, in all (Cox and linear) multivariable models we used a combination of backward procedure and exclusion of highly collinear variables through model-dependent Variance Inflation Factor(VIF) cut-off values to select covariates [38].

MP, MPPBW and MPWAL at day 1 were included into all backward multivariate models to assess relative robustness of the association of each power measure (MP, MPPBW, MPWAL) with outcomes.

We also explored the predictive performance of each power measure and of other ventilatory variables at day 1 for 28-day NIV failure and death and the optimal cut-offs using the area under receiver operating characteristic curve (AUROC) analysis and Youden index.

Comparison between AUROC curves was made by DeLong’s method.

In the physio-anatomical analysis, we explored dose–response relationship between the power delivered at day 1 by the ventilator to the respiratory system and gas exchange, ultrasonographic and circulating biomarker changes after the 1st 24 h of NIV by univariable and multivariable regression analysis, after log transformation of skewed parameters; the best fit among four predictive models (linear, exponential, logarithmic, binomial) was searched using R2 values.

Two-tailed p values < 0.05 were considered statistically significant (MedCalc 19.7, Ostend, Belgium).