Ventilatory Ratio is a Valuable Prognostic Indicator in a Representative Observational Cohort of Patients with Acute Respiratory Distress Syndrome

Estimating mortality risk is essential for prognostic enrichment. How various indices specic to respiratory compromise contribute to prognostication in patients with acute respiratory distress syndrome (ARDS) is not well-characterized in general clinical populations. The primary objective of this study was to identify variables specic to respiratory failure that add prognostic value to indicators of systemic illness severity. We tested the added benet of respiratory variables in a representative observational cohort of patients with ARDS. 50 patients with ARDS were enrolled in a single-center, prospective, observational cohort. We tested the contribution of respiratory variables (oxygenation index, ventilatory ratio [VR], and the radiographic assessment of lung edema score) to logistic regression models of 28-day mortality adjusted for indicators of systemic illness severity (the Acute Physiology and Chronic Health Evaluation [APACHE] III score or severity of shock as measured by the number of vasopressors required at baseline). We also compared a model utilizing APACHE III with one including baseline number of vasopressors using the areas under their receiver operating curves. is easily obtained at the bedside and offers promise for both clinical prognostication and enriching clinical trial populations.


Introduction
Estimating mortality risk in the acute respiratory distress syndrome (ARDS) is a crucial component of clinical trial design because prognostic enrichment increases statistical power to detect an effect, a principle especially important in such a heterogeneous syndrome [1,2]. Mortality in ARDS can be driven predominantly by a patient's underlying ARDS risk factor (such as sepsis), underlying comorbidities, or the severity of lung injury, in part measured by the degree of hypoxemia [3]. Respiratory variables alone do not fully capture mortality risk in ARDS given the heterogeneity of clinical risk factors and frequent concomitant multiorgan failure. Additionally, since respiratory failure is a substantial contributor to mortality in patients with ARDS [3], general severity of illness scores do not fully explain ARDS mortality risk.
Thus, metrics that are speci c to both severity of critical illness and respiratory failure are necessary to best understand mortality in ARDS. For severity of illness, the Acute Physiology and Chronic Health Evaluation (APACHE) III score is highly associated with hospital mortality risk for critically ill adults [4]. APACHE III is not readily available in most clinical settings, however, and it is burdensome to calculate.
The ventilatory ratio (VR), representing the ratio of measured minute ventilation x measured arterial carbon dioxide tension (PaCO 2 ) / (ideal minute ventilation x ideal PaCO 2 ), is an index of impaired ventilation that can be easily obtained at the bedside, correlates with pulmonary dead space, and is independently associated with an increased odds of hospital mortality even after adjusting for positive end-expiratory pressure, hypoxemia, and severity of illness [5]. A physiologic pulmonary dead space fraction of <0.3 and VR of 1 are considered normal, and higher values represent worse physiological derangements. A ventilatory ratio of at least 2 in ARDS ("high VR") has been shown to be associated with a higher risk of mortality [5]. Plasma biomarkers may also enhance prognostication in ARDS [6-8], and they have been integrated into parsimonious models that identify hypoin ammatory and hyperin ammatory ARDS phenotypes, which are associated with differential outcomes and treatment responses [9][10][11][12][13].
Many previous studies of mortality predictors in ARDS have utilized large clinical trial populations. Importantly, however, mortality from ARDS is higher in observational studies compared to randomized controlled trials (RCTs) [14], partially because observational studies often include patients who would be excluded from RCTs, such as those with severe comorbidities [14]. The aim of this research was to study the following questions in a cohort of critically ill patients with comorbidities and illness severity representative of the ARDS population outside of clinical trials: Which candidate variables speci c to respiratory compromise will improve the performance of a logistic regression model describing 28-day mortality? Will the number of vasopressors required at baseline have comparable discriminatory power for mortality to APACHE III? Finally, will the hyperin ammatory and hypoin ammatory phenotypes with differential outcomes, previously identi ed in parsimonious models from large, randomized controlled trials, be present in a small, observational cohort?

Study design and oversight
This was a secondary analysis of a single-center prospective observational cohort study of subjects with ARDS who required endotracheal intubation and mechanical ventilation and were admitted to an intensive care unit in the Mo tt-Long Hospital (Parnassus Campus) at the University of California, San Francisco Medical Center. This study was approved by the institutional review board (#17-21982).

Human subjects: patient selection and informed consent
Daily screening of the electronic medical record (EMR) from July 1, 2017 to March 12, 2019 was used to identify mechanically ventilated subjects who were endotracheally intubated and met the Berlin criteria for ARDS (including PaO 2 /FiO 2 < 300 mmHg, PEEP ≥ 5 cm H 2 O and bilateral in ltrates on chest radiography not fully explained by cardiogenic pulmonary edema) [15]. Patients were excluded if ARDS was present for 72 hours or longer prior to enrollment; they were younger than 18; had an expected survival of <96 hours; had a previous diagnosis of pulmonary hypertension; or were receiving mechanical circulatory support (extracorporeal membrane oxygenation, ventricular assist device, or intra-aortic balloon pump), on pulmonary vasodilators, or had been subject to externally-induced cardioplegia (such as in the setting of cardiopulmonary bypass) within 24 hours. Informed consent was obtained for all participants. If the patient was unable to provide consent, consent was obtained from a surrogate.

Data Collection
Clinical data were extracted from the EMR. Data collected at enrollment included patient demographics, primary and secondary risk factors for ARDS, comorbid conditions, and components of the APACHE III score. PaO 2 /FiO 2 , oxygenation index (OI, [FiO 2 x mean airway pressure x 100]/PaO 2 ), ventilator parameters (plateau airway pressure, mean airway pressure, positive end-expiratory pressure [PEEP], tidal volume, respiratory system compliance, driving pressure), mode of ventilation, and pulmonary dead space fraction (V D /V T ) were also recorded at enrollment. V D /V T was calculated from the Enghoff modi cation of the Bohr equation [16] utilizing simultaneous mixed-expired CO 2 measured utilizing volumetric capnography (NICO Cardiopulmonary Management System, Novametrix, Wallingford, Connecticut) and arterial blood gas measurements. Two sequential calculations of V D /V T were made using measurements with a ve-minute interval, and these two values were then averaged to represent the patient's baseline pulmonary dead space fraction. Ventilatory ratio was calculated as (minute ventilation [ml/min] x PaCO 2 [mmHg]/predicted body weight [kg] x 100 x 37.5). Ventilator-free days (VFD) was calculated as days alive and free of mechanical ventilation to 28 days. ICU-free days was calculated as days alive and outside of the ICU to 28 days. For both outcomes, patients who die before 28 days were assigned zero.
Biologic specimen collection, processing, and storage Blood samples were collected on the day of enrollment. Plasma obtained from two 10 mL EDTA anticoagulated blood samples was divided immediately after centrifugation into 0.5 mL and 1 mL aliquots and frozen at -70°C.

Biomarker Measurements and Phenotype Assignment
Angiopoietin-2 (Ang-2), receptor for advanced glycation end-products (RAGE), interleukin-6 (IL-6), interleukin-8 (IL-8), protein C, and soluble tumor necrosis factor receptor 1 (sTNFR-1) were measured in plasma samples obtained at study enrollment. Biomarkers were measured by ELISA (R&D Systems, Minneapolis, MN, USA and Helena Laboratories, Beaumont, TX, USA). Biomarker measurements were made by a laboratory manager who was blinded to patient characteristics outside of inclusion/exclusion criteria. Hyperin ammatory and hypoin ammatory phenotype assignments were performed using previously validated model coe cients [9]. The probability cut-off was set at 0.5, and the 3-variable model was used given its higher speci city.

Statistical Analyses
Continuous variables are expressed as mean ± SD or median (interquartile range [IQR]). Categorical variables are presented as count (percentage). Biomarker concentrations were log 10 transformed to improve interpretability and better meet model assumptions.
For between-group comparisons, Fisher's exact test was used to compare categorical variables, two-sided unpaired t-test was used for normally distributed continuous variables, and the Mann-Whitney U test was used for non-normally distributed continuous variables. Spearman correlation coe cients were used to describe all correlations and investigate collinearity between variables. Multivariable logistic regression utilizing manual step-wise backward selection and likelihood ratio testing for nested models was used for analyses of the association between candidate variables and 28-day mortality with appropriate model checking. Areas under receiver operating curves (ROC) were compared by the Delong test for non-nested models [19]. A two-sided P value of < 0.05 was considered signi cant. All analyses were performed using Stata SE 16.1 (StataCorp, College Station, TX).

Results
Of 322 patients screened for inclusion, 272 were excluded, and 50 patients were enrolled ( Figure 1). Pneumonia was the most common primary ARDS etiology (46%) ( Table 1). The baseline APACHE III score was 107 ± 30, and 82% of the subjects had vasopressor-dependent shock. Many patients had major comorbid conditions such as cirrhosis, chronic lung disease, end stage renal disease, and heart failure. Most patients had at least one major comorbid condition. Baseline respiratory parameters are presented in Table 2. Most patients had a PaO 2 /FiO 2 ratio less than 200 mmHg. Both V D /V T (0.59 ± 0.13) and VR (1.9 [1.6 -2.3]) were elevated in this cohort. 28-day mortality was 56% (Table 3). High mortality accounted for a median of zero ventilator-free days (VFD) and ICU-free days. Median duration of mechanical ventilation among survivors was 9 days (IQR 4 -20 days). Median ICU duration among survivors was 11 days (IQR 7 -21 days), and median length of hospitalization was 27 days (IQR 13 -28 days).   The APACHE III score and the non-pulmonary SOFA score were signi cantly higher among non-survivors as compared to survivors (Table 4). Non-survivors required a signi cantly greater number of vasopressors at baseline. VR was signi cantly higher among non-survivors, p = 0.01 ( Figure 2). Mortality among patients with high VR (VR ≥ 2) was 78% vs. 37% in the low VR group (p=0.005). Among plasma biomarkers, IL-8 and sTNFR-1 concentrations were signi cantly higher and protein C was signi cantly lower among non-survivors (Table 4).

RAGE = Receptor for Advanced Glycation End-products
We rst tested correlations among candidate variables to minimize collinearity within candidate logistic regression models. There was a strong negative correlation between PaO 2 /FiO 2 ratio and OI ( = -0.91, p < 0.0001). Moving forward, we included OI as a prognostic indicator in candidate models instead of PaO 2 /FiO 2 given the existing body of evidence that OI is an independent risk factor for mortality in adults with ARDS [20][21][22] and that PaO 2 /FiO 2 ratio is not consistently associated with mortality [22][23][24].
Univariable logistic regression for each candidate variable, the full candidate logistic regression model, and the nal model for 28-day mortality using physiological variables is depicted in  Lastly, we tested whether ARDS phenotypes that have been identi ed in secondary analyses of randomized clinical trials and large, well-phenotyped observational cohorts are also present in a small observational cohort. These phenotypes can be accurately identi ed with a parsimonious classi er model consisting of three plasma biomarkers: IL-8, protein C, and bicarbonate [9]. We found similar proportions of patients were classi ed as hyperin ammatory (38%) and hypoin ammatory (62%) in our cohort as in prior work [9]. Mortality in the hyperin ammatory group was higher (71%) than in the hypoin ammatory group (46%), although this difference did not meet statistical signi cance (p=0.11) (Table 6). Thus, even in this small observational cohort with high baseline mortality, ARDS sub-phenotypes with distinct clinical trajectories were readily identi ed. Dead space fraction, mean ± SD 0.61 ± 0.14 0.57 ± 0.14 0.40 Oxygenation index (IQR) 12.6 (7.5 -18.

Discussion
In this observational cohort of 50 patients that is representative of ARDS populations outside of clinical trials, the ventilatory ratio and the number of vasopressors required at baseline were strongly associated with mortality. Both of these clinical variables are easily obtained and could enhance bedside assessment of patient prognosis. In contrast to the oxygenation index, which did not improve model performance, the ventilatory ratio was the most discriminant respiratory variable for mortality. VR has previously been demonstrated to be associated with mortality in large cohorts of patients [5,25], but this is the rst study to our knowledge to test the association between VR and mortality in a small observational cohort. These ndings support the generalizability of VR as a prognostic variable in general populations of ARDS patients. Additionally, a three-variable parsimonious classi er model that included IL-8, protein C, and the lowest bicarbonate identi ed two sub-phenotypes of ARDS that have been previously identi ed predominantly in large RCTs, and those with a hyperin ammatory phenotype had substantially greater mortality.
Several measures of the severity of respiratory failure have been studied for their prognostic validity in ARDS. We tested PaO 2 /FiO 2 , oxygenation index, VR, and the RALE score because of their frequent use in clinical and research settings and their established associations with ARDS outcomes [5,18,20,22,26].
In previous work, a novel composite score that includes PaO 2 /FiO 2 , the RALE score, and VR had high discrimination for the need for ECMO or death from severe pulmonary dysfunction [27]. However, this score was derived from a single cohort without external validation and had poor discrimination for overall hospital mortality, so its generalizability remains unclear. In our study, the ventilatory ratio improved logistic regression models of 28-day mortality when combined with APACHE III, the most commonly used severity of illness score in ARDS, or with the number of vasopressors required at baseline. Additionally, a model that replaced the multi-component APACHE III with the number of vasopressors required at baseline performed similarly to the model with APACHE III. These ndings are especially important given that this cohort was more representative of the general ARDS population than the subjects enrolled in clinical trials, as our subjects were critically ill with comorbidities and poor prognoses that often would lead to their exclusion from RCTs.
ARDS is a complex and heterogeneous syndrome that often involves both respiratory failure and multisystem organ dysfunction, leading to severe critical illness. Although the APACHE III score is a reliable measure of severity of illness, it is not designed to speci cally measure the severity of lung injury or respiratory failure. Thus, we tested whether it would be valuable to supplement a measure of severity of illness with variables speci c to respiratory compromise when modeling mortality in ARDS. Several indices of respiratory failure have been shown to have stronger independent associations with ARDS outcomes than the PaO 2 /FiO 2 ratio [24,28,29]. Pulmonary dead space fraction (V D /V T ) is an independent predictor of mortality in ARDS [29]; however, estimation of dead space requires specialized equipment to measure the partial pressure of carbon dioxide (PCO 2 ) in mixed expired air. In this study, VR improved a model of 28-day mortality including APACHE III, which suggests that VR captures a domain of ARDS severity and lung injury that is not re ected by APACHE III alone. These ndings are also consistent with the existing literature demonstrating that VR is independently associated with mortality even after adjusting for oxygenation, PEEP, and severity of illness with APACHE II [5]. Previous work has demonstrated the importance of identifying patients at risk of ARDS-attributable mortality [3], and our ndings imply that VR can play a valuable role in identifying these high-risk patients. In addition, we found that a model including vasopressor number performed similarly to a model including the APACHE III score. Given its complexity and limited availability outside the research setting, the APACHE III score is rarely used in clinical practice, and these results indicate that vasopressor number may represent similar prognostic information. Additionally, likelihood ratio testing indicates that the performance of a model with vasopressor number, like the APACHE III score, is augmented when VR is added.
VR alone was also associated with mortality, while pulmonary dead space fraction was not. This can likely be attributed to the fact that VR provides information about both pulmonary dead space and metabolic derangements, including carbon dioxide excretion and production [17]. Additionally, when VR was dichotomized into "high VR" (≥2) and "low VR," (<2) mortality was signi cantly greater among patients with high VR. VR had high mortality discrimination even in a modest cohort of patients and is easily obtained without the use of specialized equipment. Thus, VR may be a reliable and underappreciated tool for prognostication even outside of large RCTs.
Two ARDS phenotypes, termed hyperin ammatory and hypoin ammatory, have been accurately and consistently identi ed in numerous cohorts from randomized controlled trials of ARDS. Pending rapid biomarker quanti cation, parsimonious models offer a simple and unique method for prognostic enrichment. Interpreting parsimonious classi er models has largely been limited to the clinical trial populations in which they were derived and validated, although there is a growing body of evidence that these subphenotypes may be generalizable to unselected populations of ARDS [30]. The present study demonstrates a substantial trend (although not statistically signi cant) towards greater mortality in the hyperin ammatory phenotype, suggesting that these classi er models may be more widely applicable.
Although the sample size of this study potentially limits its generalizability, we deliberately tested our hypotheses in a modest-sized cohort in order to elucidate which physiologic and biologic factors were most relevant to a cohort of ARDS patients representative of clinical practice. This study had very few exclusion criteria. As a result and in contrast to many randomized controlled trials, our cohort included severely ill patients with multiple comorbidities. Chronic respiratory failure, chronic liver disease, previous bone marrow transplantation, and prior lung transplantation are all medical problems that were represented in our cohort but have been exclusion criteria in many of the landmark ARDS studies [31][32][33].
As was observed in the LUNG-SAFE study [34], patients in observational cohorts are often more systemically ill and have higher mortality rates. Observational studies such as this one are representative of clinical practice and have a crucial role in the generalizability of ARDS research.

Conclusions
The results of this study underscore that ventilatory ratio is a valuable tool for mortality risk assessment and captures a domain of ARDS severity that is not re ected by general severity of illness indicators. Although ventilatory ratio improved a model of 28-day mortality when added to a severity of illness score, it also performed well on its own at discriminating between those who lived and those who died. This severely ill cohort is representative of ARDS in clinical practice, more so than the carefully selected populations that generally meet all criteria for enrollment in clinical trials. Thus, ventilatory ratio, a respiratory variable that is convenient to calculate using information already collected in critically ill patients, may be valuable for both clinical trials that aim to enrich for the patients with the greatest chance of mortality and for risk assessment and shared decision making in the clinical setting.  Flowchart of patient screening and enrollment.

Figure 2
Box plot of ventilatory ratio dichotomized by Alive (n=22) vs. Dead (n=28) at Day 28. Ventilatory ratio was signi cantly higher among non-survivors, p = 0.01 by Wilcoxon Rank Sum.