Risk stratification using SpO₂/FiO₂ and PEEP at initial ARDS diagnosis and after 24 h in patients with moderate or severe ARDS

Abstract

Background

We assessed the potential of risk stratification of ARDS patients using SpO₂/FiO₂ and positive end-expiratory pressure (PEEP) at ARDS onset and after 24 h.

Methods

We used data from a prospective observational study in patients admitted to a mixed medical–surgical intensive care unit of a university hospital in the Netherlands. Risk stratification was by cutoffs for SpO₂/FiO₂ and PEEP. The primary outcome was in-hospital mortality. Patients with moderate or severe ARDS with a length of stay of > 24 h were included in this study. Patients were assigned to four predefined risk groups: group I (SpO₂/FiO₂ ≥ 190 and PEEP < 10 cm H₂O), group II (SpO₂/FiO₂ ≥ 190 and PEEP ≥ 10 cm), group III (SpO₂/FiO₂ < 190 and PEEP < 10 cm H₂O) and group IV (SpO₂/FiO₂ < 190 and PEEP ≥ 10 cm H₂O).

Results

The analysis included 456 patients. SpO₂/FiO₂ and PaO₂/FiO₂ had a strong relationship (P < 0.001, R ₂ = 0.676) that could be described in a linear regression equation (SpO₂/FiO₂ = 42.6 + 1.0 * PaO₂/FiO₂). Risk stratification at initial ARDS diagnosis resulted in groups that had no differences in in-hospital mortality. Risk stratification at 24 h resulted in groups with increasing mortality rates. The association between group assignment at 24 h and outcome was confounded by several factors, including APACHE IV scores, arterial pH and plasma lactate levels, and vasopressor therapy.

Conclusions

In this cohort of patients with moderate or severe ARDS, SpO₂/FiO₂ and PaO₂/FiO₂ have a strong linear relationship. In contrast to risk stratification at initial ARDS diagnosis, risk stratification using SpO₂/FiO₂ and PEEP after 24 h resulted in groups with worsening outcomes. Risk stratification using SpO₂/FiO₂ and PEEP could be practical, especially in resource-limited settings.

Electronic supplementary material

The online version of this article (doi:10.1186/s13613-017-0327-9) contains supplementary material, which is available to authorized users.

Introduction

In the Berlin definition for acute respiratory distress syndrome (ARDS), risk stratification was suggested by categorizing patients as having ‘mild,’ ‘moderate’ or ‘severe’ ARDS based on the ratio of the arterial oxygen tension (PaO₂) to the fraction of inspired oxygen (FiO₂) at initial diagnosis [1]. This approach knows several challenges. First, it requires invasive and expensive arterial blood sampling and analysis that is frequently not available, in particular in resource-limited settings. Second, the robustness of this way of classifying patients was not confirmed by external validation [2]. Most important, though, is that the level of positive end-expiratory pressure (PEEP) may affect the PaO₂/FiO₂ [3] and that PaO₂/FiO₂ and PEEP collected at later time points, for instance 24 h after the initial ARDS diagnosis, largely improve risk stratification [4, 5].

Due to the sigmoidal nature of the oxyhemoglobin dissociation curve, the pulse oximetry saturation (SpO₂) may serve as a reliable alternative for the PaO₂ in patients with SpO₂ levels lower than 97% [6]. Indeed, patients with ARDS diagnosed by means of the SpO₂/FiO₂ have similar characteristics and outcomes as patients in whom ARDS is diagnosed using the PaO₂/FiO₂ [7], and persistence of a high oxygen saturation index at 24 h has been found to be associated with worse outcomes in cases of pediatric ARDS [8]. As pulse oximetry is noninvasive, inexpensive and widely available, SpO₂/FiO₂ could serve as an attractive alternative for PaO₂/FiO₂ in risk stratification of ARDS patients.

The overarching aim of the present investigation was to determine whether classification of patients using SpO₂/FiO₂ and PEEP at initial ARDS diagnosis and after 24 h could be used to stratify for risk of mortality. Specifically, we hypothesized that the SpO₂/FiO₂ could serve as an alternative for the PaO₂/FiO₂ in risk classification and that re-classification using SpO₂/FiO₂ and PEEP after 24 h improves risk stratification in a cohort of consecutive adult patients with moderate or severe ARDS in an intensive care unit (ICU) in the Netherlands.

Methods

Study design

We used data from a prospective observational conveniently sized cohort of well-defined critically ill patients admitted to the mixed surgical–medical intensive care unit of one university hospital in the Netherlands [9]. These data were matched with nurse-validated oxygenation data at baseline and after 24 h that was retrospectively collected from the patient data management system (Metavision^®, iMDSoft, Tel Aviv, Israel). The Institutional Review Board of the Academic Medical Center approved the parent study protocol and use of an ‘opt-out’ consent method (IRB no. 10-056C).

Inclusion and exclusion criteria

A team of trained ICU researchers scored patients for the absence or presence of ARDS according to the criteria stated by the American–European Consensus Conference on ARDS [10]. Although this study started in 2011 (i.e., before publication of the Berlin definition for ARDS [1]), all patients diagnosed with ARDS fulfilled the criteria of the latest definition for ARDS. Patients with mild ARDS were excluded as were patients discharged or transferred to another ICU before 24 h after the initial ARDS diagnosis.

Outcomes

The primary outcome was all-cause in-hospital mortality. Secondary outcomes included ICU mortality, 30- and 90-day, and 1-year mortality, and the number of ventilator-free days and alive at day 28 (VFD-28), defined as the number of days from days 1 to 28 that the patient is alive and breathing without invasive assistance of the mechanical ventilator for at least 24 consecutive hours.

Data collection

At initial ARDS diagnosis and after 24 h, pulse oximetry results were retrospectively collected, and the corresponding SpO₂/FiO₂ was calculated. For this, we first collected ten successive nurse-validated SpO₂ and FiO₂ values over 10 min directly preceding, and at the time, an arterial blood sample was drawn for blood gas analysis. The median of these ten SpO₂ and FiO₂ values was used to calculate the SpO₂/FiO₂ to alleviate potential artifactual SpO₂ values. In all patients, pulse oximetry was measured by conventional two-wavelength finger pulse oximeters. The local protocol dictated a change in finger position every 3–4 h to avoid decubitus lesions and to use alternative probes, such as nose and forehead probes, only in patients in whom no adequate signal could be obtained with a conventional pulse oximeter on a finger. In addition to this, PaO₂ and PaCO₂ levels, pH levels and body temperature were collected, and it was determined whether or not patients received treatment with vasopressors at both time points.

Baseline data collected from the original database included demographic and ventilator data, comorbidities, admission type, patient category and cause of ARDS. The Acute Physiology and Chronic Health Evaluation (APACHE) IV score [11], the Lung Injury Prediction Score [12] and Charlson comorbidity score [13] were calculated.

Analysis plan

The correlation between SpO₂/FiO₂ and PaO₂/FiO₂ was studied only using SpO₂/FiO₂ and PaO₂/FiO₂ at initial ARDS diagnosis. SpO₂ values of > 97% were excluded from the correlation analysis as the oxyhemoglobin dissociation curve flattens above this level and large changes in the PaO₂ may result in little or no change in the SpO₂, in line with previous investigations that analyzed this relationship [6, 14, 15]. A nonlinear equation (the ‘Ellis formula’) was also used to estimate PaO₂/FiO₂ from SpO₂ and FiO₂ values [16, 17].

Next, we classified all patients at initial ARDS diagnosis and re-classified patients after 24 h, into four risk groups based on predefined SpO₂/FiO₂ and PEEP cutoffs. As two recent studies of risk stratification of ARDS patients used a cutoff for the PaO₂/FiO₂ of 150 mmHg [4, 5], which corresponds to a SpO₂/FiO₂ of 190 [6], we used this value as a cutoff for SpO₂/FiO₂ in the present analysis. We used the same cutoff for PEEP as in two previous studies [4, 5]. Accordingly, we created four groups: SpO₂/FiO₂ ≥ 190 and PEEP < 10 cm H₂O (group I); SpO₂/FiO₂ ≥ 190 and PEEP ≥ 10 cm (group II); SpO₂/FiO₂ < 190 and PEEP < 10 cm H₂O (group III); and SpO₂/FiO₂ < 190 and PEEP ≥ 10 cm H₂O (group IV). The comparison among groups included all pairwise comparisons across the four risk groups.

Statistical analysis

Data were expressed as mean ± standard deviation (SD), median with interquartile range (IQR) or number with percentage, where appropriate.

A two-way scatterplot and Pearson correlation analysis were used to characterize the relationship between SpO₂/FiO₂ (linear and nonlinear estimations) and PaO₂/FiO₂. Linear regression allowed quantification of the best regression line and derive a predictive equation for the relationship. Hence, based on the derived regression equation, we obtained the SpO₂/FiO₂ values that correspond to PaO₂/FiO₂ ratio values of 100, 150 and 200 mmHg. PaCO₂, arterial pH, body temperature, PEEP and use of vasopressors were tested as interaction terms in the model to evaluate moderation of the association between SpO₂/FiO₂ and PaO₂/FiO₂. The area under the receiver operating characteristic (ROC) curve evaluated the applicability of SpO₂/FiO₂ in discriminating moderate from severe ARDS.

Differences between risk groups were tested with the Pearson Chi-square or Fisher exact test for categorical variables and with one-way ANOVA or Kruskal–Wallis test for continuous variables. In-hospital mortality and other mortality endpoints were calculated for each of the four groups, and a P value for trend was calculated from a chi-squared test for trend in proportions (i.e., the Cochrane–Armitage test), testing the null hypothesis that the proportions in several groups are the same. A pairwise comparison at ARDS diagnosis and after 24 h was performed using a contrast matrix predictor approach in which odd ratios (ORs) and 95% confidence intervals are generated for each mutual comparison between groups.

Post hoc analyses

Several post hoc analyses were performed. First, we evaluated whether transforming the data to fractions (1/PaO₂/FiO₂ and 1/SpO₂/FiO₂) improved the fit between SpO₂/FiO₂ and PaO₂/FiO₂, as previously reported [18].

Second, risk classification using the Berlin definition threshold for severe ARDS (i.e., 100 mm Hg) [6] was performed.

Third, in the last post hoc analysis we analyzed whether the association between risk stratification and in-hospital mortality was confounded by factors such as disease severity, and other readily available parameters in the database, such as arterial pH and plasma lactate levels, blood pressure levels and vasopressor therapy. For this analysis, a multivariable logistic regression model was built to assess the confounding effect of these variables on the association between risk groups and the primary outcome.

All analyses were performed in R via the R-studio interface (R version 3.0, www.r-project.org). A P value below 0.05 was considered significant.

Results

Patients

Of 554 patients with ARDS, 456 were classified as having moderate or severe ARDS according to the Berlin definition and had complete datasets. Of them, 382 could be used for the correlation analyses, and all patients for the classification and re-classification analyses (Fig. 1). Pneumonia, sepsis, major surgery and trauma were the most common causes of ARDS (Table 1). At baseline, there were no differences between survivors and non-survivors with regard to oxygenation parameters, tidal volume size, the Lung Injury Prediction Score and the etiology of ARDS. Also, there were no differences with regard to the comorbidity score. Overall all-cause in-hospital mortality rate was 39.7%. Non-survivors were older, had a lower arterial pH levels at ARDS diagnosis, were ventilated at higher respiratory rates and had higher disease severity scores. Berlin definition class distribution was not different between survivors and non-survivors.

Fig. 1.

Study flowchart. SpO₂, pulse oximetry oxygen saturation; ARDS, acute respiratory distress syndrome

Table 1.

Baseline characteristics of 456 survivors and non-survivors with moderate and severe ARDS

Variables	All patients	Survivors N = 275	Non-survivors N = 181	P value
Age, years	62 (51–72)	60 (48–70)	63 (55–72)	0.008
Female		113 (41.1)	59 (32.6)	0.075
Weight, kg	80 (70–90)	80 (70–90)	80 (68–87)	0.347
APACHE IV	68 (51–89)	72 (58–95)	90 (74–113)	< 0.001
Charlson comorbidity index 2	1 (0–3)	1 (0–2)	1 (0–3)	0.103
Lung Injury Prediction Score	8.5 (7–10)	8 (6.5–10)	9 (7–10)	0.074
Tidal volume, ml/kg	6.3 (5.3–7.6)	6.3 (5.3–7.8)	6.3 (5.4–7.3)	0.412
Respiratory rate	23 (17–29)	20 (16–27)	25 (19–30)	0.001
PEEP, cmH2O	10 (8–12)	10 (8–12)	10 (8–13)	0.436
FiO2, %	60 (50–75)	60 (50–70)	60 (50–80)	0.462
Arterial pH	7.34 (7.26–7.41)	7.35 (7.28–7.41)	7.31 (7.22–7.39)	0.001
PaCO2, kPa	5.8 (5.1–6.9)	5.7 (5–6.9)	6 (5.2–7.3)	0.072
PaO2/FiO2	120 (89–149)	123 (91–149)	115 (89–149)	0.353
SpO2/FiO2	160 (126–190)	163 (127–191)	157 (124–190)	0.344
Vasopressor use in first 24 h	293 (64.3)	164 (59.6)	129 (71.3)	0.015
ARDS class
Moderate ARDS	300 (65.8)	181 (65.8)	119 (65.7)	1
Severe ARDS	156 (34.2)	94 (34.2)	62 (34.3)	1
Cause of ARDS
Pulmonary origin	325 (71.3)	200 (72.7)	125 (69.1)	0.732
Non-pulmonary origin	185 (40.6)	104 (37.8)	81 (44.8)	0.382

Categorical variables: number (percentage); continuous variables: median (25–75 percentile)

PEEP positive end-expiratory pressure, FiO2 fraction of inspired oxygen, PaCO ₂ arterial carbon dioxide tension, PaO ₂ arterial oxygen tension, SpO ₂ pulse oximetry saturation, ARDS acute respiratory distress syndrome, APACHE IV acute physiology and chronic evaluation IV

Relationship between SpO₂/FiO₂ and PaO₂/FiO₂

The correlation between SpO₂/FiO₂ and PaO₂/FiO₂ was strong (P < 0.001, R ² = 0.676; Fig. 2) and could be described in a linear regression equation:

$${\text{SpO}}_2/{\text{FiO}}_2=42.6+1.00\ast {\text{PaO}}_2/{\text{FiO}}_2$$

Fig. 2.

Scatterplot of SPO₂/FiO₂ versus PaO₂/FiO₂ at initial ARDS diagnosis. The line represents the best-fit linear relationship: SpO₂/FiO₂ = 42.6 + 1.0 * PaO₂/FiO₂ [P < 0.001, R ² = 0.676] at initial ARDS diagnosis

This relationship was neither moderated by arterial PaCO₂ and PEEP, nor by body temperature and vasopressor therapy. There was a moderation effect by arterial pH and FiO₂, as the steepness of the slope between PaO₂/FiO₂ and SpO₂/FiO₂ was inclined by arterial pH (1.4, P < 0.001) and declined by FiO₂ (− 0.01, P < 0.001).

Based on the found equation, a PaO₂/FiO₂ of 200 mm Hg corresponded to a SpO₂/FiO₂ of 243 [95% CI 220–265], a PaO₂/FiO₂ of 150 mm Hg to a SpO₂/FiO₂ of 193 [95% CI 174–211] and a PaO₂/FiO₂ of 100 mm Hg to a SpO₂/FiO₂ of 143 [95% CI 127–158].

The SpO₂/FiO₂ had an excellent ability to discriminate moderate from severe ARDS, with a ROC area under the curve of 0.928 (Fig. 3). The nonlinear formula (Ellis formula) to estimate the PaO₂/FiO₂ from the SpO₂ and FiO₂ was not superior to the linear model when SpO₂ ≤ 97% (N = 382, R ² = 0.656).

Fig. 3.

ROC curve for SpO₂/FiO₂ versus PaO₂/FiO₂ < 100. Dotted lines represent 95% confident intervals, AUC = 0.928. AUC, area under the curve

The use of fractions (1/SpO₂/FiO₂ and 1/PaO₂/FiO₂) did not result in a stronger linear relationship (R ² = 0.629). The regression equation, though, was very similar to one previously reported in pediatric patients (1/SpO₂/FiO₂ = 0.0024 + 0.46/PaO₂/FiO₂) [18].

Risk stratification

The distribution and outcome data for the four risk groups are presented in Table 2. There was no trend for increasing in-hospital mortality rate among the four risk groups at baseline (P value for trend = 0.90). A pairwise comparison also showed no significant differences in hospital mortality between groups at ARDS diagnosis. In contrast, the same categorization after 24 h resulted in risk groups with increasing rates of in-hospital mortality (P value for trend < 0.001). A pairwise comparison showed that the differences in in-hospital mortality between group IV and group II (OR 2.40 [1.15–4.82]; P value = 0.012), and between group IV and group I were significant (OR 2.47 [1.26–4.85]; P value = 0.003) (Table 3).

Table 2.

Distribution and outcomes of each subset of patients with ARDS at initial diagnosis and after 24 h

Outcome	Group I	Group II	Group III	Group IV	P value
	SpO2/FiO2 ≥ 190 and PEEP < 10	SpO2/FiO2 ≥ 190 and PEEP ≥ 10	SpO2/FiO2 < 190 and PEEP < 10	SpO2/FiO2 < 190 and PEEP ≥ 10
At onset of ARDS (N)	87	43	100	226
ICU mortality (%)	19.5	23.3	18.0	30.1	0.042
In-hospital mortality (%)	44.8	30.2	36.0	41.2	0.897
30-day mortality (%)	27.6	27.9	24.0	33.2	0.276
90-day mortality (%)	47.1	34.9	41.0	42.0	0.636
1-year mortality (%)	60.9	41.9	47.0	49.6	0.182
VFD (days) (IQR)	19 (7–24)	22 (6.5–25)	20.5 (6–25)	18 (0–23)	0.070
After 24 h (N)	213	145	17	81
ICU mortality (%)	15.5	24.1	35.3	48.2	< 0.001
In-hospital mortality (%)	34.7	35.9	52.9	56.8	< 0.001
30-day mortality (%)	23.9	26.9	47.1	45.7	< 0.001
90-day mortality (%)	38.0	38.6	52.9	56.8	0.003
1-year mortality (%)	47.4	45.5	58.8	65.4	0.006
VFD 28 (days) (IQR)	23 (15–26)	18 (0–23)	6 (0–17)	0 (0–17)	< 0.001

VFD-28 ventilator-free days and alive at day 28, IQR interquartile range, CI confidence interval

Table 3.

Intergroup comparisons at ARDS diagnosis and after 24 h for in-hospital mortality

Comparison	At ARDS diagnosis		After 24 h
	OR (95% CI)	P value	OR (95% CI)	P value
Group II versus I	0.53 (0.19–1.46)	0.376	1.05 (0.59–1.86)	0.996
Group III versus I	0.69 (0.32–1.49)	0.600	2.11 (0.59–7.62)	0.433
Group IV versus I	0.86 (0.45–1.65)	0.933	2.47 (1.26–4.85)	0.003
Group III versus II	1.30 (0.48–3.53)	0.907	2.01 (0.55–7.43)	0.509
Group IV versus II	1.61 (0.65–4.03)	0.531	2.40 (1.15–4.82)	0.012
Group IV versus III	1.24 (0.66–2.34)	0.811	1.17 (0.30–4.53)	0.991

ARDS acute respiratory distress syndrome, OR odd ratio, CI confidence interval

Findings were comparable for the secondary outcomes. While classification using cutoffs for SpO₂/FiO₂ and PEEP at initial ARDS diagnosis were not related to any of these outcomes, risk stratification at 24 h resulted in significant positive trends across groups, and significantly worse secondary outcomes in group IV compared to group I and to group II (Additional file 1: Table E1).

Distribution of patients in each subset noticeably changed at 24 h. Less than a quarter of patients maintained to have SpO₂/FiO₂ < 190; their in-hospital mortality was much higher than that of patients with a SpO₂/FiO₂ ≥ 190 (56.1 vs. 35.2%, P = 0.001). Half of the patients were ventilated at PEEP < 10 cm H₂O; their in-hospital mortality was similar to that of patients who were kept at PEEP ≥ 10 cm H₂O (36.1 vs. 43.4 P = 0.107). When considering the characteristics of the four risk groups at 24 h, we found statistical differences in the APACHE IV score, arterial pH and FiO₂ at 24 h, maximum plasma lactate levels and vasopressor use, which could explain the significant differences in outcome (Table 3).

Post hoc analyses

The post hoc analysis with the adapted SpO₂/FiO₂ cutoff of 150, in line with the Berlin definition cutoff of PaO₂/FiO₂ of 100 mm Hg, showed similar results to the primary analysis, albeit group sizes changed. No trends in risk were observed other than for ICU mortality when using this cutoff at onset of ARDS, while a significant trend across the four risk groups was noted using this cutoff after 24 h (Additional file 1: Table E2). The intergroup comparisons also mirrored the primary analysis, with no significant contrasts at ARDS diagnosis, but clear contrast after 24 h (group IV vs. group I and II) (Additional file 1: Table E3).

The multivariable model showed that APACHE IV scores, plasma lactate levels and arterial pH significantly confounded the relationship between the group stratification and in-hospital mortality (Additional file 1: Table E4). The same applied when restricting this analysis to the extreme risk groups (i.e., groups I and IV), in which even vasopressor therapy was a significant confounder (Additional file 1: Table E5).

Discussion

To our best knowledge, this is the first study in which ARDS patients were classified using predefined cutoff values for SpO₂/FiO₂ and PEEP. The most relevant findings of this study are: (1) SpO₂/FiO₂ confirms to be a valid surrogate for PaO₂/FiO₂ in patients with moderate or severe ARDS; (2) re-classification after 24 h using predefined cutoffs for SpO₂/FiO₂ and PEEP improves risk stratification compared to the same cutoffs at baseline; and (3) the association between risk group and outcome, however, is confounded by disease severity, arterial pH and lactate levels, and vasopressor therapy. We believe that the approach of using SpO₂/FiO₂ and PEEP to classify patients could be useful for the implementation of an individualized approach for appropriate diagnosis and therapy in ARDS patients, in particular in settings where it is difficult, if not impossible, to perform (repeated) blood gas analyses (Table 4).

Table 4.

Main characteristics of 456 patients with ARDS—classification 24 h after initial ARDS diagnosis based on SpO₂/FiO₂ cutoff of 190 and PEEP cutoff of 10 cm H₂O

Variables	Group I (N = 213)	Group II (N = 145)	Group III (N = 17)	Group IV (N = 81)	P value
	SpO2/FiO2 ≥ 190 and PEEP < 10	SpO2/FiO2 ≥ 190 and PEEP ≥ 10	SpO2/FiO2 < 190 and PEEP < 10	SpO2/FiO2 < 190 and PEEP ≥ 10
APACHE IV score	75 (60–93)	84 (65–108)	78 (65–112)	86 (68–110)	0.001
Age, years	63 (53–73)	60 (49–69)	57 (42–70)	59 (49–69)	0.018
Female	80 (37.6)	61 (42.1)	5 (29.4)	26 (32.1)	0.436
CCI 2	1 (0–3)	1 (0–2)	2 (0–3)	1 (0–2)	0.423
LIPS	8 (6.5–9.5)	9 (7.5–10.5)	7 (6.5–8)	9.5 (7.5–11.5)	< 0.001
Tidal volume, ml/kg	6.4 (5.3–7.7)	6.2 (5.2–7.5)	7.2 (6.5–8.2)	6.4 (5.6–7.4)	0.166
FiO2, %	40 (36–40)	40 (40–50)	60 (55–60)	60 (55–65)	< 0.001
PEEP, cm H2O	5 (5–8)	12 (10–13)	7 (5–8)	14 (12–15)	< 0.001
Arterial pH	7.4 (7.4–7.5)	7.4 (7.3–7.4)	7.4 (7.3–7.5)	7.3 (7.3–7.4)	< 0.001
PaCO2, kPa	5.3 (4.7–6.2)	5.6 (4.9–6.2)	5.6 (5–7)	5.8 (5.1–6.6)	0.018
Plasma lactate level, mmol/l	2.4 (1.4–4.1)	3.1 (1.7–6.2)	1.6 (1.4–3.9)	5.4 (2.6–10.9)	< 0.001
Respiratory rate	22 (17–27)	24 (17.2–30)	22 (18–28)	24 (16–34)	0.183
PaO2/FiO2	207 (175–252)	193 (167–226)	128 (120–149)	133 (105–153)	< 0.001
SpO2/FiO2	243 (231–265)	232 (200–245)	165 (154–170)	158 (136–176)	< 0.001
Vasopressor use in first 24 h	100 (47.0)	102 (70.3)	6 (35.3)	67 (82.72)	< 0.001
ARDS class after 24 h
Mild ARDS	122 (57.3)	61 (42.1)	0 0	5 (6.2)	< 0.001
Moderate ARDS	91 (42.7)	84 (57.9)	15 (88.2)	59 (72.8)	< 0.001
Severe ARDS	0 0	0 0	2 (11.8)	17 (21)	< 0.001
Cause of ARDS
Pulmonary	156 (73.2)	99 (68.3)	12 (70.6)	58 (71.6)	0.509
Non-pulmonary	67 (31.5)	78 (53.8)	5 (29.4)	40 (49.4)	0.426

Categorical variables: number (percentage); continuous variables: median (25–75 percentile)

CCI Charlson Comorbidity Index, LIPS Lung Injury Prediction Score, APACHE acute physiology and chronic health evaluation, PEEP positive end-expiratory pressure, FiO ₂ fraction of inspired oxygen, PaCO ₂ arterial carbon dioxide tension, PaO ₂ arterial oxygen tension, SpO ₂ pulse oximetry oxygen saturation, ARDS acute respiratory distress syndrome

Our study has several strengths. The prospective design of the collection of data, the completeness of follow-up and the fact that the ARDS diagnosis was scored by a team of trained ICU researchers prevented against bias. In addition, we could re-categorize patients from the criteria stated by the American–European Consensus Conference on ARDS [10] to the newer Berlin definition for ARDS [1], after we could select patients with moderate or severe ARDS. Finally, the number of participants was large, patients were homogeneous regarding their clinical characteristics as well as type of ARDS and the overall in-hospital mortality rate of our cohort was comparable to the pooled mortality for moderate and severe ARDS patients in the recently published LUNG SAFE study [19].

SpO₂/FiO₂ is an increasingly appreciated parameter in the diagnosis and management of patients with ARDS [6, 15, 17, 20, 21]. The current pediatric ARDS definition includes oximetry-based measures in preference of the PaO₂/FiO₂, captured in the oxygenation saturation index [22]. Unfortunately, mean airway pressures, necessary for calculation of the oxygenation saturation index, were not reliably captured in our cohort preventing us from comparing our results to those from previous studies in children [8, 18, 23]. We used a previously reported equation (SpO₂/FiO₂ = 64 + (0.84 * PaO₂/FiO₂) [6] to convert a previously used cutoff for PaO₂/FiO₂ (i.e., 150 mm Hg) to a cutoff for SpO₂/FiO₂ of 190. Using the equation that came from our own data a very similar cutoff for SpO₂/FiO₂ would have been chosen. Due to its increased feasibility, the SpO₂/FiO₂ was actually considered for inclusion during the Berlin criteria definition process [24], but was successively excluded on the possibility that it may misclassify mild into severe ARDS patients. A cutoff for PaO₂/FiO₂ of 150 mm Hg has frequently been shown to accurately stratify for in-hospital mortality [4, 5, 25] and was also used in two of the largest, and ultimately positive, randomized controlled trials in ARDS patients [26, 27].

The present findings on the relationship between SpO₂/FiO₂ and PaO₂/FiO₂ mirror results from previous investigations in adult [6, 15] and pediatric patients [14, 18, 23, 28], albeit that a marginally different regression equation was found. One investigation did not find a strong relationship between SpO₂/FiO₂ and PaO₂/FiO₂, but in that study the data were collected from automated anesthesia information system with a very high portion of the data excluded due to high SpO₂ values [29]. Recently, the linear relation between SpO₂/FiO₂ and PaO₂/FiO₂ was re-challenged and both a fractional transformation [14] and a nonlinear imputation method [17, 30] were proposed to improve the model fit or better represent the sigmoidal shape of the oxyhemoglobin dissociation curve. Despite the clear physiological rationale, these approaches did not improve the relationship in the present dataset.

It has been described before that FiO₂ at levels > 70% alters the PaO₂/FiO₂: an increase in FiO₂ > 70% gradually increases the PaO₂/FiO₂ [31], particularly at low levels of shunt fraction [32]. The effect of FiO₂ on SpO₂/FiO₂, however, could be opposite as pulse oximetry has an intrinsic upper limit much tighter than PaO₂. An increase in FiO₂ will gradually decrease the SpO₂/FiO₂. This intrinsic mathematical limitation of the SpO₂/FiO₂ can possibly lead to misclassification of individual ARDS cases. Of note, FiO₂ > 70% was used in 25% of patients at ARDS diagnosis, and only in 3% after 24 h.

The level of PEEP is known to be particularly relevant as the evolution and prognosis of ARDS are related to changes in PaO₂/FiO₂ in response to changes in PEEP greater than or equal to 10 cm H₂O [3, 33, 34]. While PEEP does not affect the oxyhemoglobin curve and may not significantly alter the relationship between SpO₂ and PaO₂ [15], it may impact the SpO₂/FiO₂ ratio by improving ventilation–perfusion matching and thus was considered in our analysis. Of note, the approach of using PaO₂/FiO₂, or SpO₂/FiO₂, and PEEP is not new. Indeed, risk classification in the Berlin definition uses PaO₂/FiO₂ and PEEP levels [1], though only at onset of ARDS and not after 24 h of standard care. The empathic aim of the present investigation was to see whether two easy to collect and almost always-available ventilator parameters, even in resource-limited settings, i.e., SpO₂/FiO₂ and PEEP, would alter risk groups.

Despite several classification and prediction systems for ARDS patients [1, 3, 25, 33, 35–37], we still lack a proper classification system for clinical management and research purposes that can serve as a practical model for setting individual therapeutic targets. Several investigations have shown that the PaO₂/FiO₂ measured at onset of ARDS cannot be used for risk stratification in patients with moderate or severe ARDS [35, 38–41]. Recently, it was shown that standardization of ventilator settings at the moment of collecting PaO₂/FiO₂ data [33] as well as re-categorization based on both PaO₂/FiO₂ and PEEP level cutoffs, measured 24 h after the initial ARDS diagnosis, largely improves risk stratification for hospital mortality [4, 5]. The findings of the present study are in agreement with and extend these findings showing for the first time that pulse oximetry after 24 h perform better than at the initial ARDS diagnosis in short- and long-term outcome stratification in adult patients with moderate or severe ARDS. Similar findings derive from the pediatric population, where oximetry derived indexes after 24 h reliably stratified outcome while initial values were not helpful in prognostication [8].

In the studied cohort of patients with moderate or severe ARDS, improvement or worsening of the SpO₂/FiO₂ in the first 24 h was strongly associated with outcome. Group I represents the less complicated patients with ARDS and a great portion of our patients fit in this group after 24 h (46.7 vs. 19.1% at baseline), underlining the ample clinical change that characterizes the initial hours of care. Few patients (17, 3.7%) were classified after 24 h as having a SpO₂/FiO₂ < 190 and PEEP < 10 cm H₂O (group III), in accordance with previous observations in which only few patients with a worsening oxygenation are managed with low levels of PEEP [4]. The use of a different SpO₂/FiO₂ cutoff (150, corresponding with 100 mm Hg as used in the Berlin definition for ARDS to classify severe ARDS) did not essentially change the results. This underlines the rationale of the 24 h re-classification.

The comparisons between risk groups emphasize the improvement in risk stratification after 24 h but also suggest that the proposed classification by SpO₂/FiO₂ and PEEP in four risk groups is dependent on other factors. The fact that differences between group III and the other groups did not reach significance could be explained in part by the small number of patients in this group after 24 h. Also interestingly, differences in outcomes between groups separated by PEEP were not significant (group I vs. II and group III vs. IV). This does not weaken the improvement in risk stratification after 24 h, but probably suggests that our findings are driven more by SpO₂/FiO₂ than by PEEP. Hence, although PEEP is an attractive stratification tool due to its ubiquity, alternative variables might result in a better classification when combined to the SpO₂/FiO₂ ratio.

The high mortality rates, especially in the lower risk groups, are high, but confirm those from previous investigations [4, 5]. Re-classification after 24 h lead to lower mortality rates in the lowest risk group, but still the proportion of patients that died was high. This too is in line with findings from the earlier studies [4, 5]. Noticeable differences were seen in APACHE IV, arterial pH, plasma lactate levels in the first 24 h, and vasopressor use between the risk groups, and all could, at least in part be associated with the significant differences in mortality rates. In fact, the post hoc multivariate analysis showed that stratification based on SpO₂/FiO₂ and PEEP looses it predictive capability when controlled for APACHE IV scores, plasma lactate levels, arterial pH, blood pressure levels and vasopressor therapy. This at least suggests that persistence of hypoxemia after 24 h could be a surrogate, e.g., overall disease severity but also underlines the weakness of a simplified classification system. Nevertheless, we think that the proposed risk stratification may still be useful in settings where only pulse oximetry is available for monitoring and where laboratory examinations to compute disease severity scores are lacking.

The results of this study suggest that the SpO₂/FiO₂ can accurately surrogate the PaO₂/FiO₂ in stratifying for mortality of adult patients with established ARDS in resource-rich settings. This simple approach may in particular be useful in settings where repeated blood gas analyses are difficult to obtain or unavailable, such as in pediatric patients [23] and resource-limited settings [42]. However, it must be acknowledged that oximetry may never fully replace arterial blood gas analyses completely. Indeed, acid–base status and PaCO₂ levels are clinically important. However, we stress that pulse oximetry is consistently present in low-resource settings while blood gas analyzers are scarce, if not completely absent. Moreover, several point-of-care devices allow to measure capillary or venous acid–base, potentially alleviating the need for an arterial blood sample and expensive blood gas analysis devices.

The SpO₂/FiO₂ may also turn useful in identifying patients at risk for ARDS. It has already been shown to be a useful independent indicator of ARDS in the Kigali criteria for ARDS, in which SpO₂/FiO₂ and lung ultrasound were used instead of PaO₂/FiO₂ and chest radiography [42]. Further studies in resource-rich and resource-poor settings are still needed to assess the utility of these adapted criteria in patients at risk for ARDS.

Limitations of our analysis include its single-center design and the lack of standardization of the SpO₂ measurements, due to their retrospective collection. Indeed, standardization of ventilatory settings at baseline and after 24 h, and thus calculation of the PaO₂/FiO₂, may be crucial for appropriate stratification and patient selection bias minimization [43], and the same may apply for SpO₂-based markers. While we tried to minimize potential errors by increasing the number of SpO₂ values collected, we are aware that potential artifactual SpO₂ values or ones resulting from acute events unrelated to the disease process may have been included (such as patient–ventilator asynchrony, obstruction of endotracheal tubes, suctioning and hemodynamic instability). We also did not collect information on skin pigmentation, motion artifacts, skin perfusion, ambient light, methemoglobinemia and carboxyhemoglobinemia levels—factors that all may be associated with the accuracy of SpO₂ measurements. Its retrospective design allowed us neither to determine the type of pulse oximeter used nor to capture the positions of the probes used in individual patients. The minor but not quantifiable amount of measurements from a different probe or location other than the conventional finger oximeter represents a methodological weakness, but we do not believe affects the outcome or correlation analysis in an important way. We focused on only two time points, i.e., the time of initial ARDS diagnosis and after 24 h. It may be possible that SpO₂/FiO₂ and PEEP at other time points add to risk stratification. Finally, re-classification after 24 h means that patients who died in the first 24 h are missed. The same happens with the small proportion of patients who are extubated or transferred before 24 h. We also excluded patients with a PaO₂/FiO₂ > 200 mm Hg; however, we do not believe this weakens our results. We did this for the two following reasons. Patients classified as having mild ARDS [1] are very ‘heterogeneous’ with respect to outcomes. Moreover, the milder degree of oxygenation impairments in these patients increases the number of patients with SpO₂ values > 97%, above which the relationship between PaO₂ and SpO₂ becomes weak.

Conclusions

SpO₂/FiO₂ and PaO₂/FiO₂ have a strong linear relationship. In contrast to risk stratification at initial ARDS diagnosis, risk stratification using SpO₂/FiO₂ and PEEP at 24 h leads to risk groups with worsening outcomes. Despite the fact that the association between risk group assignment and outcome is confounded by several factors, risk stratification using SpO₂/FiO₂ and PEEP could be useful and practical, especially in settings where repeated blood analyses are difficult or impossible.

Abbreviations

SpO₂

peripheral capillary oxygen saturation

FiO₂

fraction of inspired oxygen

PaO₂

arterial oxygen partial pressure

PEEP

positive end-expiratory pressure

ARDS

acute respiratory distress syndrome

ICU

intensive care unit

VFD

ventilator-free days

APACHE

acute physiology and chronic health evaluation

LIPS

Lung Injury Prediction Score

ROC

receiver operating characteristic

AUC

area under the curve

AECC

American–European Consensus Conference on ARDS