Response to Ventilator Adjustments for Predicting Acute Respiratory Distress Syndrome Mortality. Driving Pressure versus Oxygenation

Abstract

Rationale: Clinicians commonly use short-term physiologic markers to assess the benefit of ventilator adjustments. Improved arterial oxygen tension/pressure (Pa_O2)/fraction of inspired oxygen (Fi_O2) after ventilator adjustment in acute respiratory distress syndrome is associated with lower mortality. However, as driving pressure (ΔP) reflects lung stress and strain, changes in ΔP may more accurately reflect benefits or harms of ventilator adjustments compared with changes in oxygenation.

Objectives: We aimed to compare the association between mortality and the changes in Pa_O2/Fi_O2 and ΔP following protocolized ventilator changes.

Methods: We assessed associations between mortality and changes in Pa_O2/Fi_O2 (ΔPa_O2/Fi_O2) and ΔP (ΔΔP) after postrandomization positive end-expiratory pressure (PEEP) and tidal volume adjustment in reanalyses of the ALVEOLI (Assessment of Low Tidal Volume and Elevated End-Expiratory Volume to Obviate Lung Injury) and ExPress (Expiratory Pressure) trials. We included subjects with available pre- and postintervention Pa_O2/Fi_O2 and ΔP (372 in ALVEOLI and 596 in ExPress). In each separate trial cohort, we performed multivariable Cox regression testing the association between ΔPa_O2/Fi_O2 and ΔΔP with mortality.

Results: In ALVEOLI, when analyzed as separate variables, ΔPa_O2/Fi_O2 was associated with mortality only in subjects in whom PEEP increased, whereas ΔΔP was associated with mortality irrespective of direction of PEEP change. When modeled together, improved ΔPa_O2/Fi_O2 was not associated with mortality, whereas ΔΔP remained associated with mortality (adjusted hazard ratio [aHR], 1.50 per 5 cm H₂O increase; 95% confidence interval [95% CI], 1.21–1.85). When modeled together in ExPress, ΔΔP (aHR, 1.42; 95% CI, 1.14–1.78) was more strongly associated with mortality than ΔPa_O2/Fi_O2 (aHR, 0.95 per 25 mm Hg increase; 95% CI, 0.90–1.00).

Conclusions: Reduced ΔP following protocolized ventilator changes was more strongly and consistently associated with lower mortality than was increased Pa_O2/Fi_O2, making ΔΔP more informative about benefit from ventilator adjustments. Our results reinforce the primacy of ΔP, rather than oxygenation, as the key variable associated with outcome.

Beyond avoidance of large tidal volumes (Vt) and high plateau pressures (1), there is little evidence to guide ventilator settings for acute respiratory distress syndrome (ARDS). Thus, clinicians often use short-term physiologic markers to assess the correctness of ventilator management. Higher positive end-expiratory pressure (PEEP), for example, is often recommended (2), despite multiple large trials testing higher versus lower levels of PEEP failing to demonstrate clinical benefit (3–6). A patient-level meta-analysis of three of these trials—ALVEOLI (Assessment of Low Tidal Volume and Elevated End-Expiratory Volume to Obviate Lung Injury), LOVS (Lung Open Ventilation Study), and ExPress (Expiratory Pressure)—found lower mortality with higher PEEP for subjects with baseline arterial oxygen tension/pressure (Pa_O2)/fraction of inspired oxygen (Fi_O2) ≤200 (7). A separate reanalysis of LOVS and ExPress showed that higher Pa_O2/Fi_O2 after the postrandomization change in PEEP and protocolized reduction in Vt was associated with lower mortality in subjects with baseline Pa_O2/Fi_O2 ≤150 in whom PEEP was increased (8). Because improved oxygenation can reflect recruitment, this study raised the possibility of using a “test dose” of PEEP as a strategy for predictive enrichment in future trials of high versus low PEEP, restricting the trial intervention to the subgroup of PEEP “responders” predicted to benefit from higher PEEP (8, 9). Furthermore, the association with mortality provided justification for using improved gas exchange as a short-term surrogate marker to justify ventilator adjustments.

However, improved oxygenation has not consistently been shown to be an appropriate surrogate for lower mortality in ARDS (1). Rather, lung stress and strain more directly reflect the risk of ventilator-induced lung injury and are consistently associated with outcomes (10–12). Favorable changes in lung stress and strain may be captured by changes in driving pressure (ΔP) following ventilator adjustments. Supporting this concept, lower ΔP mediated the association between higher PEEP and lower Vt with decreased mortality (13). Changes in ΔP (ΔΔP) are potentially more sensitive to overdistension than are changes in Pa_O2/Fi_O2 (ΔPa_O2/Fi_O2), making it a theoretically more useful predictive marker of benefit from changes in PEEP and Vt. However, studies associating ΔΔP with mortality did not adjust for concurrent ΔPa_O2/Fi_O2 (13), and studies associating improved ΔPa_O2/Fi_O2 with mortality did not adjust for ΔΔP (8).

Therefore, to compare the relative strength of the associations between ΔΔP or ΔPa_O2/Fi_O2 and mortality, we performed secondary analyses of the ALVEOLI and ExPress trials. We tested the hypothesis that improved ΔΔP following changes in PEEP and Vt are more informative about the risk of mortality than are improved ΔPa_O2/Fi_O2.

Methods

Study Populations

We performed secondary analyses of ALVEOLI (3) and ExPress (5). In ALVEOLI, subjects were enrolled between 1999 and 2002 from 23 ARDSNetwork hospitals. Briefly, 549 subjects with ARDS (Pa_O2/Fi_O2 ≤300) were randomized to a higher versus lower PEEP strategy using separate PEEP–Fi_O2 tables. ExPress also compared two different PEEP levels in 767 patients with Pa_O2/Fi_O2 ≤300, with the higher PEEP strategy based upon setting the PEEP to result in a plateau pressure of 28–30 cm H₂O. Subjects were enrolled between 2002 and 2005 from 37 French intensive care units. Importantly, in ALVEOLI and ExPress, both PEEP and Vt were adjusted after randomization, with both variables contributing to changes in Pa_O2/Fi_O2 and ΔP. Subjects without a recorded plateau pressure were excluded.

Data Collection

We extracted patient demographics, diagnosis, severity of illness, presence of sepsis, and duration of hospitalization before trial enrollment in both trials. The severity of illness score used in ALVEOLI was the Acute Physiology and Chronic Health Evaluation III, whereas ExPress used the Simplified Acute Physiologic Score II. Mode of ventilation before randomization was recorded in both cohorts. In ALVEOLI, ventilator pressures and Pa_O2/Fi_O2 before randomization and on Day 1 after randomization were recorded, whereas in ExPress we recorded ventilator pressures and Pa_O2/Fi_O2 before and after randomization.

Definitions

ΔPa_O2/Fi_O2 and ΔΔP between Day 1 and prerandomization (postrandomization minus prerandomization) for ALVEOLI, and between post- and prerandomization for ExPress, were the primary exposures. Note that a positive value for ΔPa_O2/Fi_O2 denotes improving oxygenation, whereas a positive value for ΔΔP describes worsening driving pressure. The primary outcome was 90-day mortality for ALVEOLI, and 60-day mortality for ExPress. The primary outcome was available for all subjects in both cohorts. Potential confounders were age, severity of illness score, hospital days before randomization, presence of sepsis, and prerandomization Pa_O2/Fi_O2 and ΔP. Age, severity of illness score, hospital days before randomization, and presence of sepsis were chosen as confounders based on their use in the reanalysis of LOVS and ExPress (8). Prerandomization Pa_O2/Fi_O2 and ΔP were chosen to capture baseline ARDS severity.

Statistical Analysis

Using Cox regression, we assessed the association between ΔPa_O2/Fi_O2 or ΔΔP and mortality. We first tested these variables in separate models, adjusting for all confounders above. We then constructed the model to test our main hypothesis by including both terms in the same model (Figure E1 in the online supplement). All analyses were stratified a priori by the direction of the PEEP change (not the randomization arm) based on the reanalysis of LOVS and ExPress suggesting effect modification by whether PEEP was increased (ΔPEEP >0) or unchanged or decreased (ΔPEEP ≤0) (8). Additionally, we plotted the relationship between ΔPa_O2/Fi_O2 or ΔΔP and the adjusted hazard of mortality. Finally, we combined ALVEOLI and ExPress and repeated our analysis to determine an effect estimate using all available pooled data. Analyses were conducted in Stata 14.2/SE (StataCorp, LP) and R 3.0.1 (www.r-project.org). The proportional hazard assumption was assessed for all Cox regression models and was not violated.

We performed multiple sensitivity analyses. First, we repeated our main analyses after restricting each cohort to subjects with presumed passive ventilation (set respiratory rate = total respiratory rate both before and after randomization), those with moderate or severe ARDS (baseline Pa_O2/Fi_O2 ≤200), and based on randomization arm. We chose to analyze subjects with presumed passive ventilation as we assumed this subgroup had more reliable plateau pressure measurements. Next, because our measures of oxygenation and ΔP were not performed on standardized ventilator settings, we performed an additional analysis adjusting for prerandomization PEEP. Because Vt was also reduced after randomization, we performed an additional analysis including change in Vt (ΔVt) as a confounder (8). Lastly, to specifically examine the ΔP response in the context of PEEP changes, we computed the same model after subtracting changes in ΔP attributable to ΔVt from the overall change in ΔP, which we considered to be a corrected ΔΔP (ΔΔP_corr):

$$\Delta \Delta {\text{P}}_{\text{corr}}=\left(\Delta {\text{P}}_{\text{post}}-\Delta {\text{P}}_{\text{pre}}\right)–\left(\Delta {\text{V}}_{\text{T}}/{\text{V}}_{\text{Tpre}}/\Delta {\text{P}}_{\text{pre}}\right)$$

Results

Association of ΔPa_O2/Fi_O2 or ΔΔP with Mortality in ALVEOLI

Of the 549 subjects in ALVEOLI, 372 (68%) had completely available data before randomization and on Day 1 for calculation of both ΔPa_O2/Fi_O2 and ΔΔP (Table 1). Excluded subjects had lower Acute Physiology and Chronic Health Evaluation scores, lower plateau pressures, and higher Pa_O2/Fi_O2. Excluded subjects were less likely to be on assist control volume control ventilation and more likely to be on pressure support. Mortality did not differ between included and excluded subjects.

Table 1.

Demographics of the ALVEOLI trial cohort stratified by inclusion or exclusion in the present study

Variables	Whole Cohort (N = 549)	Inclusion (n = 372)	Exclusion (n = 177)
Demographics
Age, yr	51 ± 17	51 ± 17	51 ± 17
Sex, F	248 (45)	168 (45)	80 (45)
Severity of illness
APACHE III	25 ± 8	26 ± 8	24 ± 9
Sepsis	208 (38)	145 (39)	63 (35)
Hospital days before randomization	4 ± 5	4 ± 5	4 ± 6
Ventilator after randomization
Assist control (volume)	388 (71)	303 (81)	85 (48)
SIMV	23 (4)	17 (5)	6 (3)
Pressure control	32 (6)	12 (3)	20 (11)
Pressure support	78 (14)	37 (10)	41 (23)
Other	29 (5)	3 (1)	26 (15)
Trial arm
Low PEEP	274 (50)	174 (47)	100 (56)
High PEEP	276 (50)	198 (53)	78 (44)
Actual PEEP direction
ΔPEEP ≤ 0	236 (43)	149 (40)	87 (49)
ΔPEEP > 0	314 (57)	223 (60)	91 (51)
Before randomization
PaO2/FiO2	128 ± 57	120 ± 53	143 ± 63
Plateau pressure	29 ± 7	27 ± 7	25 ± 7
PEEP	10 ± 4	10 ± 4	9 ± 4
ΔP	17 ± 6	17 ± 6	16 ± 5
Day 1 after randomization
PaO2/FiO2	186 ± 81	180 ± 79	207 ± 86
Plateau pressure	26 ± 7	26 ± 7	24 ± 6
PEEP	12 ± 5	12 ± 4	10 ± 5
ΔP	13 ± 5	14 ± 5	13 ± 5
Ancillary therapies by Day 1
Neuromuscular blockade	69 (13)	51 (14)	18 (10)
Prone positioning	9 (2)	8 (2)	1 (<1)
90-d mortality	148 (27)	96 (26)	52 (29)

Definition of abbreviations: ΔP = driving pressure; ALVEOLI = Assessment of Low Tidal Volume and Elevated End-Expiratory Volume to Obviate Lung Injury; APACHE = Acute Physiology and Chronic Health Evaluation; Fi_O2 = fraction of inspired oxygen; Pa_O2 = arterial oxygen tension/pressure; PEEP = positive end-expiratory pressure; SIMV = synchronized intermittent mandatory ventilation.

Data are shown as n (%) or mean ± standard deviation.

ΔPa_O2/Fi_O2 and ΔΔP were not correlated (Figure E2, r = 0.0009, P = 0.986). When analyzed separately (Table 2), there was an association between improved (i.e., increased) ΔPa_O2/Fi_O2 and lower mortality (adjusted hazard ratio [aHR], 0.94 per 25 mm Hg increase; 95% confidence interval [95% CI], 0.84–1.01; P = 0.087) that did not reach the usual threshold for statistical significance. By contrast, higher ΔΔP (i.e., worse) was associated with higher mortality (aHR, 1.52 per 5 cm H₂O increase; 95% CI, 1.24–1.88; P < 0.001). After stratifying the analysis by the direction of change in PEEP, ΔPa_O2/Fi_O2 was associated with mortality only in patients in whom PEEP was increased (aHR, 0.89; 95% CI, 0.79–1.00; P = 0.046; interaction P = 0.27), whereas ΔΔP was associated with mortality irrespective of direction of PEEP change (Table 2). When ΔPa_O2/Fi_O2 and ΔΔP were incorporated in the same model (Table 2 and Figure 1), the association of improved ΔPa_O2/Fi_O2 with lower mortality was attenuated, whereas the association of higher ΔΔP with mortality was unchanged (aHR, 1.50; 95% CI, 1.21–1.85; P < 0.001). There was no evidence of effect modification by direction of PEEP change in the full model (P = 0.463 for interaction of PEEP direction with ΔPa_O2/Fi_O2; P = 0.648 for interaction of PEEP direction with ΔΔP).

Table 2.

Prognostic significance of oxygenation and driving pressure response following postrandomization changes in PEEP in ALVEOLI

Variables	Whole Cohort (N = 372)		Stratified Analysis
			ΔPEEP ≤ 0 (n = 149)		ΔPEEP > 0 (n = 223)
	Hazard Ratio* (95% CI)	P Value	Hazard Ratio* (95% CI)	P Value	Hazard Ratio* (95% CI)	P Value
Model 1†
ΔPaO2/FiO2	0.94 (0.84–1.01)	0.087	1.01 (0.86–1.20)	0.882	0.89 (0.79–1.00)	0.046
Model 2†
ΔΔP	1.52 (1.24–1.88)	<0.001	1.71 (1.12–2.60)	0.013	1.58 (1.22–2.05)	<0.001
Model 3†‡
ΔPaO2/FiO2	0.95 (0.86–1.04)	0.232	1.06 (0.89–1.27)	0.481	0.92 (0.82–1.02)	0.148
ΔΔP	1.50 (1.21–1.85)	<0.001	1.75 (1.15–2.66)	0.009	1.52 (1.17–1.97)	0.002

Definition of abbreviations: ΔP = driving pressure; ΔΔP = change in driving pressure; ALVEOLI = Assessment of Low Tidal Volume and Elevated End-Expiratory Volume to Obviate Lung Injury; APACHE = Acute Physiology and Chronic Health Evaluation; CI = confidence interval; Fi_O2 = fraction of inspired oxygen; Pa_O2 = arterial oxygen tension/pressure; PEEP = positive end-expiratory pressure.

^*Per change in 25 mm Hg in Pa_O2/Fi_O2; per change in 5 cm H₂O in ΔP.

^†All models adjusted for age, APACHE III, days of hospitalization before randomization, presence of sepsis, prerandomization Pa_O2/Fi_O2, and prerandomization ΔP.

^‡Model 3 includes both change in Pa_O2/Fi_O2 and change in ΔP in the same model.

Figure 1.

Left: Relationship between the change in driving pressure (ΔΔP) after ventilator adjustments and the risk of mortality in ALVEOLI and ExPress. ΔΔP was strongly associated with mortality, irrespective of whether positive end-expiratory pressure (PEEP) was increased (red line) or whether it was decreased/unchanged (blue line). There was no evidence of effect modification by direction of PEEP change in either ALVEOLI (interaction P = 0.648) or ExPress (interaction P = 0.407). Right: By contrast, there was no association between ΔPa_O2/Fi_O2 after ventilator adjustments and mortality when stratified by direction of PEEP change. Shaded zones represent 95% confidence intervals. There was no evidence of effect modification by direction of PEEP change in either ALVEOLI (interaction P = 0.463) or ExPress (interaction P = 0.524). ΔP = driving pressure; ALVEOLI = Assessment of Low Tidal Volume and Elevated End-Expiratory Volume to Obviate Lung Injury; ExPress = Expiratory Pressure; Fi_O2 = fraction of inspired oxygen; Pa_O2 = arterial oxygen tension/pressure.

Sensitivity analyses showed that improved ΔΔP, but not ΔPa_O2/Fi_O2, was associated with lower mortality when restricted to patients with presumed passive ventilation (Table E1 in the online supplement), patients with baseline Pa_O2/Fi_O2 ≤200 (Table E2), patients assigned to the high PEEP arm (Table E3), and when adjusting for prerandomization PEEP (Table E4). Neither ΔPa_O2/Fi_O2 nor ΔΔP were associated with mortality in the low PEEP arm (Table E3). ΔΔP, but not ΔPa_O2/Fi_O2, was associated with lower mortality when the analysis adjusted for ΔVt (Table E5) and when modeling ΔΔP_corr attempting to isolate the change in ΔP due to a change in PEEP (Table E6). Results were also unchanged when accounting for use of neuromuscular blockade (Table E7).

In unadjusted analysis, patients with improved ΔP after randomization had lower mortality than subjects with increased ΔP (Fisher’s exact P = 0.016) (Table E8). Patients in whom ΔP decreased were more likely to have received a higher PEEP following randomization compared with patients in whom ΔP increased (64% vs. 49%, P = 0.012). Prerandomization ΔP was the best predictor of improved ΔP after randomization (Tables E8 and E10).

Association of ΔPa_O2/Fi_O2 or ΔΔP with Mortality in ExPress

Of the 767 subjects in ExPress, 596 (78%) had all data available for analysis (Table 3). Pre- and postrandomization Pa_O2/Fi_O2 and ΔP were recorded at a median of 3.7 (interquartile range, 2.5–4.3) hours apart. Mortality did not differ between included and excluded subjects.

Table 3.

Demographics of the ExPress trial cohort stratified by inclusion or exclusion in the present study

Variables	Whole Cohort (N = 767)	Inclusion (n = 596)	Exclusion (n = 171)
Demographics
Age, yr	60 ± 15	60 ± 16	60 ± 15
Sex, F	251 (33)	195 (33)	56 (32)
Severity of illness
SAPS II	51 ± 18	51 ± 18	51 ± 17
Sepsis	663 (86)	515 (86)	148 (87)
Hospital days before randomization	3 ± 4	3 ± 4	2 ± 3
Trial arm
Low PEEP	382 (50)	293 (49)	89 (52)
High PEEP	385 (50)	303 (51)	82 (48)
Actual PEEP direction
ΔPEEP ≤0	269 (35)	258 (43)	11 (6)
ΔPEEP >0	498 (65)	338 (57)	160 (94)
Before randomization
PaO2/FiO2	143 ± 57	144 ± 58	143 ± 55
Plateau pressure	23 ± 5	23 ± 5	23 ± 5
PEEP	8 ± 4	8 ± 4	7 ± 3
ΔP	14 ± 5	14 ± 5	14 ± 5
After randomization
PaO2/FiO2	170 ± 86	170 ± 86	167 ± 89
Plateau pressure	25 ± 5	25 ± 5	25 ± 5
PEEP	12 ± 4	12 ± 4	12 ± 4
ΔP	13 ± 4	13 ± 4	13 ± 4
60-d mortality	289 (38)	219 (37)	70 (41)

Definition of abbreviations: ΔP = driving pressure; ExPress = Expiratory Pressure; Fi_O2 = fraction of inspired oxygen; Pa_O2 = arterial oxygen tension/pressure; PEEP = positive end-expiratory pressure; SAPS = Simplified Acute Physiologic Score. Data are shown as n (%) or mean ± standard deviation.

ΔPa_O2/Fi_O2 and ΔΔP were not correlated (Figure E2, r = −0.0568, P = 0.166). When analyzed separately (Table 4), both ΔPa_O2/Fi_O2 (aHR, 0.93; 95% CI, 0.89–0.99; P = 0.013) and ΔΔP (aHR, 1.47; 95% CI, 1.18–1.83; P = 0.001) were associated with mortality, with stronger associations evident in subjects in whom PEEP was unchanged or decreased after randomization. When ΔPa_O2/Fi_O2 and ΔΔP were incorporated in the same model (Table 4 and Figure 1), ΔΔP (aHR, 1.42; 95% CI, 1.14–1.78; P = 0.002) was more strongly associated with mortality than was ΔPa_O2/Fi_O2 (aHR, 0.95; 95% CI, 0.90–1.00; P = 0.040). As in ALVEOLI, there was no evidence of effect modification by direction of PEEP change in the full model (P = 0.524 for interaction of PEEP direction with ΔPa_O2/Fi_O2; P = 0.407 for interaction of PEEP direction with ΔΔP).

Table 4.

Prognostic significance of oxygenation and driving pressure response following postrandomization changes in PEEP in ExPress

Variables	Whole Cohort (N = 596)		Stratified Analysis
			ΔPEEP ≤ 0 (n = 258)		ΔPEEP > 0 (n = 338)
	Hazard Ratio* (95% CI)	P Value	Hazard Ratio* (95% CI)	P Value	Hazard Ratio* (95% CI)	P Value
Model 1†
ΔPaO2/FiO2	0.93 (0.89–0.99)	0.013	0.81 (0.81–0.99)	0.035	0.96 (0.90–1.03)	0.226
Model 2†
ΔΔP	1.47 (1.18–1.83)	0.001	1.60 (1.16–2.21)	0.004	1.35 (0.98–1.87)	0.071
Model 3†‡
ΔPaO2/FiO2	0.95 (0.90–1.00)	0.040	0.90 (0.81–1.00)	0.051	0.97 (0.90–1.03)	0.300
ΔΔP	1.42 (1.14–1.78)	0.002	1.57 (1.13–2.17)	0.006	1.33 (0.96–1.84)	0.090

Definition of abbreviations: ΔP = driving pressure; ΔΔP = change in driving pressure; CI = confidence interval; ExPress = Expiratory Pressure; Fi_O2 = fraction of inspired oxygen; Pa_O2 = arterial oxygen tension/pressure; PEEP = positive end-expiratory pressure; SAPS = Simplified Acute Physiologic Score.

^*Per change in 25 mm Hg in Pa_O2/Fi_O2; per change in 5 cm H₂O in ΔP.

^†All models adjusted for age, SAPS II, days of hospitalization before randomization, presence of sepsis, prerandomization Pa_O2/Fi_O2, and prerandomization ΔP.

^‡Model 3 includes both change in Pa_O2/Fi_O2 and change in ΔP in the same model.

ΔΔP, but not ΔPa_O2/Fi_O2, was associated with mortality in patients with presumed passive ventilation (Table E1), in patients assigned to the high PEEP arm (Table E3), and when adjusting for prerandomization PEEP (Table E4). Neither ΔPa_O2/Fi_O2 nor ΔΔP were associated with mortality in the low PEEP arm (Table E3). Both improved ΔPa_O2/Fi_O2 and ΔΔP were associated with lower mortality in patients with baseline Pa_O2/Fi_O2 ≤200 (Table E2) and when adjusting for ΔVt (Table E5), with minimal change in effect size when modeling ΔΔP_corr (Table E6).

In unadjusted analysis, there was no difference in mortality between patients with a decrease (improving) versus an increase (worsening) in ΔP (Table E9). As in ALVEOLI, patients in whom ΔP decreased were more likely to have received a higher PEEP following randomization compared with patients in whom ΔP increased (61% vs. 47%, P = 0.003), with prerandomization ΔP again the best predictor of subsequent improved ΔP after randomization (Tables E9 and E10).

Combined Analysis

We performed an analysis using all available data by pooling the ALVEOLI and ExPress cohorts and assessing the association between ΔPa_O2/Fi_O2 and ΔΔP and 60-day mortality (Table E11). When analyzed separately, both ΔPa_O2/Fi_O2 (aHR, 0.94; 95% CI, 0.90–0.98; P = 0.004) and ΔΔP (aHR, 1.52; 95% CI, 1.31–1.78; P < 0.001) were associated with mortality, without evidence of effect modification by direction of PEEP change (both interaction P > 0.8). When ΔPa_O2/Fi_O2 and ΔΔP were incorporated in the same model, ΔΔP (aHR, 1.48; 95% CI, 1.27–1.74; P < 0.001) was more strongly associated with mortality than was ΔPa_O2/Fi_O2 (aHR, 0.95; 95% CI, 0.91–0.99; P = 0.028), without evidence for effect modification by direction of PEEP change (both interaction P > 0.7).

Discussion

In a reanalysis of ALVEOLI and ExPress, ΔΔP following protocolized adjustment of PEEP and Vt was more consistently and strongly associated with mortality than was ΔPa_O2/Fi_O2. ΔΔP after randomization was associated with mortality after adjustment for ΔPa_O2/Fi_O2, whereas ΔPa_O2/Fi_O2 was not associated with mortality after adjusting for ΔΔP. These findings extend previous work showing that a ΔPa_O2/Fi_O2 after ventilator changes is associated with mortality in ARDS when analyzed without adjusting for ΔΔP (8). In both trials, improved ΔP was associated with lower mortality after adjusting for ΔPa_O2/Fi_O2 and ΔVt. These findings, coupled with the rationale that ΔP reflects stress and strain, suggest that ΔΔP after ventilator adjustment is more informative as a surrogate marker for the effectiveness of ventilator adjustments. Our results support the priority of improved ΔP, rather than oxygenation, as the key variable associated with outcome.

The aim of PEEP in ARDS is to maintain lung recruitment and minimize dynamic lung stress and strain (10). However, despite preclinical promise (14, 15) and a strong physiologic rationale (16), multiple trials have failed to demonstrate benefit of higher PEEP (3–6). Meta-analyses and reanalyses have suggested that the response to PEEP is heterogeneous (7, 8), suggesting responders and nonresponders. A strategy to identify responders predicted to benefit from higher PEEP requires a reliable short-term surrogate (9). However, the best method to identify PEEP responders remains uncertain, as recruitment assessed by imaging (10), gas exchange (8), and respiratory mechanics (13) all reflect different aspects of the response to higher PEEP and give different answers regarding “optimal” PEEP. Our analysis of ΔΔP attributable to changes in PEEP (ΔΔP_corr) suggests that improved ΔP, rather than Pa_O2/Fi_O2, could identify PEEP responders.

Interestingly, neither ΔPa_O2/Fi_O2 nor ΔΔP were associated with mortality in subjects assigned to the low PEEP arms of either ALVEOLI or ExPress. One possibility for this lack of association is lower power due to subgroup analysis with fewer nonsurvivors. Alternatively, it may be that changes in oxygenation or ΔP are only informative in the context of the application of high PEEP. We again note that both ALVEOLI and ExPress reduced Vt after randomization and that the association between improved ΔP and lower mortality cannot be solely linked to PEEP changes in this reanalysis. Indeed, the reduction in Vt may be a more important contributor to improved ΔP in both trials, although our analysis of adjusted ΔΔP due to PEEP suggests that there is some contribution by PEEP. Overall, our results support that changes in ΔP following ventilator adjustments are more strongly associated with outcome than changes in oxygenation.

Both ΔPa_O2/Fi_O2 and ΔΔP were associated with mortality in ExPress. This was a larger trial, with greater power to detect an association between ΔPa_O2/Fi_O2 and mortality. Additionally, the time difference between pre- and postrandomization Pa_O2/Fi_O2 and ΔP was considerably shorter in ExPress, increasing plausibility that changes in these variables resulted from protocolized ventilator management, rather than disease resolution. Previous reanalysis of LOVS and ExPress showed ΔPa_O2/Fi_O2 was associated with mortality (8); however, this reanalysis did not account for ΔΔP. When we analyzed ΔPa_O2/Fi_O2 in ALVEOLI without consideration for changes in ΔΔP (Model 1 in Table 2), we reproduced these conclusions, finding an association between ΔPa_O2/Fi_O2 and mortality in subjects with ΔPEEP >0. That this was attenuated when modeled with ΔΔP does not invalidate the utility of ΔPa_O2/Fi_O2. However, it does reinforce that improved oxygenation is unlikely to be causal for reducing mortality. Rather, short-term improvements in oxygenation in response to protocolized ventilator changes should be interpreted as an epiphenomenon of lung recruitment, whereas the effect of recruitment on lung stress and strain is more directly reflected by ΔΔP.

We tested our primary hypothesis in ALVEOLI and ExPress for three reasons. First, reanalysis of trials of PEEP, in contrast to cohort studies, ensured randomization of protocolized changes in PEEP and Vt, which mitigates (but does not eliminate) concern that improved ΔP and Pa_O2/Fi_O2 simply reflect lower ARDS severity. Second, LOVS (4) and ART (Alveolar Recruitment for ARDS Trial) (6) used recruitment maneuvers in addition to PEEP changes throughout the trials, whereas recruitment maneuvers were abandoned after 80 subjects randomized to higher PEEP in ALVEOLI (3). Recruitment maneuvers add yet a third variable (in addition to PEEP and Vt) and could affect mortality in ways unrelated to ΔPa_O2/Fi_O2 and ΔΔP, sometimes dramatically so (6). Third, the recently completed EPVent-2 (Esophageal Pressure-Guided Ventilation) (17) did not achieve substantial separation of PEEP between treatment arms. Thus, ALVEOLI and ExPress were the most appropriate large trials to test the association between ΔPa_O2/Fi_O2 or ΔΔP and mortality, despite having been published more than 10 years ago.

It is notable that ART showed harm with higher PEEP levels despite achieving lower ΔP (6). However, ART was not solely a trial of different PEEP levels, as the high PEEP arm also received maximal recruitment maneuvers and decremental PEEP titration. The recruitment maneuvers achieved plateau pressures up to 60 cm H₂O and may have contributed to severe adverse events prompting modification of the recruitment protocol. Concerns were also raised about the possibility of patient self-inflicted lung injury from absence of paralysis. Overall, it is difficult to disentangle the effects of the recruitment intervention or the higher PEEP on the increased mortality of the intervention arm. Nevertheless, this reinforces that the results of our reanalysis are not definitive, and rather identify ΔΔP as an appropriate surrogate short-term marker of PEEP responsiveness for the purposes of trial enrichment. In light of ART, this also suggests that the procedures by which one achieves lower ΔP also impact eventual outcome and that future trials of PEEP should disentangle recruitment from PEEP.

Limitations

Our study has limitations. We restricted analyses to complete cases without imputation of missing data. Missingness was modest (27% overall in both trials) and nonrandom, precluding imputation. This creates two issues. First, the association between ΔPa_O2/Fi_O2 and ΔΔP with mortality may be biased. Second, the smaller sample size, particularly in ALVEOLI, reduces power to detect an association between ΔPa_O2/Fi_O2 and mortality after adjusting for ΔΔP. Nevertheless, in all models and sensitivity analyses, ΔΔP remained associated with mortality in ALVEOLI, whereas ΔPa_O2/Fi_O2 did not consistently retain an association. This included adjustment for use of neuromuscular blockade, which was used in 13% of ALVEOLI subjects and could plausibly affect both oxygenation and ΔP favorably. In ExPress and combined analyses, the majority of models supported the prognostic primacy of ΔΔP. Additional limitations include no assessment of lung recruitability, either by imaging or by pressure–volume curves. Thus, the physiologic basis for ΔPa_O2/Fi_O2 and ΔΔP in response to PEEP and Vt changes remains speculative. For example, ΔPa_O2/Fi_O2 in response to higher PEEP may be due to reduced pulmonary blood flow and reduced shunt fraction rather than recruitment of atelectatic lung. An additional limitation is the lack of esophageal pressures, transpulmonary pressures, and transpulmonary driving pressures; ΔP in our analysis is reflective of stress and strain over the entire respiratory system, rather than just the lung.

Finally, it is worth emphasizing that we cannot prove causality between response to any given ventilator change and mortality. Although our data support the hypothesis that the ΔP response to PEEP may predict benefit or harm, the observed association does not confirm causality and this hypothesis needs to be tested by an enrichment strategy in future clinical trials. Additionally, the parent trials used protocolized ventilator changes and thresholds for gas exchange, meaning improved ΔP after ventilator changes is not the sole consideration if it comes at significant expense to oxygenation or ventilation. Thus, it should not be assumed that drastic lowering of Vt or ΔP will lower mortality if accompanied by significant hypercarbia. Lastly, the time period between randomization and Day 1 measurements was 12–24 hours in ALVEOLI, raising the possibility that ΔΔP reflects disease resolution, rather than the beneficial effects of ventilator adjustments per se. Reassuringly, ΔΔP was associated with mortality in ExPress, which had a much shorter time period between measurements. The utility of ΔΔP to identify who will benefit from PEEP can best be demonstrated within the context of a trial using prerandomization PEEP responsiveness to stratify treatment (9, 18).

Conclusions

In summary, the change in ΔP after changes in PEEP and Vt following randomization was consistently, independently, and strongly associated with mortality whereas the change in Pa_O2/Fi_O2 showed a less consistent association. We posit that ΔΔP, rather than ΔPa_O2/Fi_O2, may be a more informative surrogate for the effectiveness of ventilator adjustments and reinforces the significance of reduced ΔP as the key variable associated with lower mortality.