Physiological predictors of survival during high-frequency oscillatory ventilation in adults with acute respiratory distress syndrome

Introduction Data that provide clinical criteria for the identification of patients likely to respond to high-frequency oscillatory ventilation (HFOV) are scarce. Our aim was to describe physiological predictors of survival during HFOV in adults with severe acute respiratory distress syndrome (ARDS) admitted to a respiratory failure center in the United Kingdom. Methods Electronic records of 102 adults treated with HFOV were reviewed retrospectively. We used logistic regression and receiving-operator characteristics curve to test associations with oxygenation and mortality. Results Patients had severe ARDS with a mean (SD) Murray's score of 2.98 (0.7). Partial pressure of oxygen in arterial blood to fraction of inspired oxygen (PaO2/FiO2) ratio and oxygenation index improved only in survivors. The earliest time point at which the two groups differed was at three hours after commencing HFOV. An improvement of >38% in PaO2/FiO2 occurring at any time within the first 72 hours, was the best predictor of survival at 30 days (area under the curve (AUC) of 0.83, sensitivity 93%, specificity 78% and a positive likelihood ratio (LR) of 4.3). These patients also had a 3.5 fold greater reduction in partial pressure of carbon dioxide in arterial blood (PaCO2). Multivariate analysis showed that HFOV was more effective in younger patients, when instituted early, and in patients with milder respiratory acidosis. Conclusions HFOV is effective in improving oxygenation in adults with ARDS, particularly when instituted early. Changes in PaO2/FiO2 during the first three hours of HFOV can identify those patients more likely to survive.


Introduction
Patients with acute respiratory distress syndrome (ARDS) exhibit a highly inhomogeneous, compliancedependent distribution of regional ventilation during conventional mechanical ventilation (CMV) [1]. Consequently, CMV can lead to further lung injury through tidal hyperinflation and shear stress injury, even when it is administered according to a 'lung protective strategy' that limits tidal volumes and plateau pressure [2], and employs recruitment maneuvers to maximize the proportion of aerated alveolar tissue [1]. High-frequency oscillatory ventilation (HFOV) can theoretically offer effective lung protective ventilation by delivering very low tidal volumes (1 to 3 mL Kg -1 ) around a fixed mean airway pressure at frequencies of 5 to 12 Hz (lower frequencies are used in adults). At high respiratory frequencies, the short inspiratory time results in a distribution of ventilation which, compared to mechanical ventilation at conventional breathing rates, is more homogeneous and less dependent on the distribution of regional lung compliance [3,4]. This results in the protection of the recruited lung (with greater compliance) from excessive cyclic variations in alveolar pressure. In addition, if the continuous distending pressure (CDP) is optimized following a stepwise recruitment maneuver, the more compliant lung regions are less susceptible to static hyperinflation [3,5,6], thereby reducing lung strain and ventilation-induced inflammation [7,8].

Patients
This was a single center observational study of patients with ARDS admitted to the Adult Intensive Care Service at Guy's and St Thomas' Hospital in London between 1998 and 2002. We included patients who were treated with HFOV because of severe gas exchange impairments while on CMV. Medical records and physiological data before, during and after HFOV were retrieved from our ICU electronic patient record (Intellivue Clinical Information Portfolio, Philips Medical Systems UK Limited). Patients' demographic data, hemodynamic variables, oxygenation and ventilator settings were recorded while on CMV prior to HFOV, during HFOV, prior to discontinuation of HFOV and on recommencement of CMV. Oxygenation index (OI = (FiO 2 CDP 100)/PaO 2 ) and partial pressure of oxygen in arterial blood to fraction of inspired oxygen (PaO 2 /FiO 2 ) ratio were calculated at the same intervals. Acute Physiology and Chronic Health Evaluation (APACHE II) and Murray lung injury severity score (which combines degree of lung infiltration on chest × ray, lung compliance, PaO 2 /FiO 2 and positive end expiratory pressure (PEEP)) [29] at admission and on commencement of HFOV were determined (with a blinded and independent radiologist scoring the chest xray appearance). Hemodynamic data were obtained using cardiac output monitors PiCCOplus (Pulsion, Munich, Germany, Software version 7.0 non USA), or a LiDCO (LiDCO Ltd, London, UK). This study was considered by the National Research Ethics Service as 'service evaluation' and therefore did not require Research Ethics Committee review [30].
HFOV was delivered using an adult high-frequency oscillatory ventilator (3100B, Viasys (CareFusion), Yorba Linda, CA, USA). All patients were initiated onto HFOV using the following settings: a FiO 2 of 1.0, a frequency of 4 to 6 Hz, inspiratory time of 33%, a bias flow of 30 to 40 Lmin -1 , a CDP set 3 to 5 cmH 2 O above the CDP during CMV and a Power to obtain transmitted oscillation ('wiggles') up to the level of mid-thigh. The power dial determines the amount of power that drives the oscillator piston to and fro. The Power control is a 10turn locking dial, electrical potentiometer covering the power range of 0 to 100%. The effect of this control is to change the displacement of the oscillator piston and hence to determine the oscillatory pressure ΔP. The Power setting interacts with the pulmonary artery wedge (Paw) and the conditions existing within the circuit to produce the resultant ΔP [33].
On starting HFOV, patients underwent a standardized slow recruitment maneuver (SRM), which represents the standard of care for patients receiving HFOV in this Institution. The SRM is derived from the maneuver included in the original MOAT Study protocol [34]. The SRM was performed by a stepwise increase in CDP by increments of 3 cmH 2 O every 10 minutes, starting from the CDP on CMV + 3 to 5 cmH 2 O, up to 50 cmH 2 O. SRM was interrupted if the mean arterial pressure fell below 55 mmHg or if desaturations (SpO2 <85%) or arrhythmias occurred. Subsequently, CDP was reduced by 2 cmH 2 O every 5 minutes. Arterial blood gases were taken every 10 minutes (every step during the incremental CDP, phase, and every two steps of the decremental CDP, phase). The 'optimal' CDP was established as the lowest CDP that achieved the most favorable combination between the highest PaO 2 and/or lowest PaCO 2 , while maintaining the FiO 2 constant at 1.0.
The protocol for the adjustment of HFOV was published previously [25]. A reduction of CDP was initiated when the FiO 2 was ≤0.5. Once CDP ≤20 to 22 cmH 2 O was achieved on a FiO 2 of 0.4, the patients returned to CMV. CMV was restarted in the pressure-control mode with CDP close to the CDP on HFOV, plateau pressures <28 cmH 2 O and PEEP adjusted to a tidal volume of 6 mLKg -1 of predicted body weight.

Outcome measures
The primary outcome measures were: improvement in PaO 2 /FiO 2 and OI and identification of physiological variables associated with 30-day survival.
To stratify patients we used an empirical score generated from the available data that was solely designed to give a pragmatic quantification of disease severity and not intended to have diagnostic or prognostic value. The score included PaO 2 /FiO 2 , basal PaCO 2 , respiratory system compliance, minute ventilation and mean airway pressure. This score was used as a dichotomous variable to separate patients with a more severe index of disease (score >50th percentile) from those with a less severe index of disease (score <50th percentile).

Statistical analysis
Distribution of baseline variables was assessed using the Kolmogorov-Smirnov test. Differences in baseline variables between survivors and non-survivors were compared using the two-tailed t-test or Mann-Withney U test for continuous data, and χ 2 or the Fisher test for qualitative data. Differences in physiological variables over time between the two outcome groups were evaluated using repeated-measure analysis of variance (ANOVA). The Friedman test and Dunn's post-hoc analysis was performed for non-normally distributed data. Multiple regression analysis and analysis of co-variance (ANCOVA) were used to test the effect of various physiological variables on oxygenation indices. Continuous outcome variables were corrected for confounding variables at baseline. Post hoc analyses were performed using Bonferrroni's correction. Variables associated with mortality in an analysis of covariance were entered in a multivariate logistic backward-likelihood ratio regression analysis, to identify predictors of mortality at different end-points. The Hosmer-Lemeshow goodness-of-fit test was used to test the validity of the model. Receiveroperating characteristic (ROC) curves were plotted to determine the best predictor of survival. The value with the best sensitivity, specificity and positive predictive value was selected as the cut-off point to predict survival.
Analyses were performed using SPSS software (version 12; SPSS, Chicago, IL, USA) and MediCalc (Mariakerke, Belgium) for ROC curve analysis. Two-tailed tests for significance were used, and a P value less than 0.05 was considered statistically significant.

Results
We report the results on 102 consecutive ARDS patients who received HFOV. The median (IQR) duration of ARDS prior to HFOV was 48 hours (24 to 120 hours). The median (IQR) duration of CMV prior to HFOV was 45 hours (9 to 138 hours). Table 1 presents the baseline patient demographics, physiological variables and severity scores at study entry. Table 2 summarizes patients' outcome and complications from HFOV.

Effects of HFOV on gas exchange
During the first 72 hours of HFOV, PaO 2 /FiO 2 improved significantly from baseline only in survivors ( Figure 1a). The earliest time-point at which PaO 2 /FiO 2 was statistically different from baseline in the survivor group was at three hours of HFOV (P <0.05) ( Figure  1a). The improvement in PaO 2 /FiO 2 was not determined by the level of CDP (Figure 1b) with mean CDP ± SD of 33.9 ± 5.4 cmH 2 O versus 32.0 ± 7.05 cmH 2 O (P = 0.08), respectively, for survivors and non-survivors. The Multivariate logistic regression analysis identified the following four independent predictive factors of mortality at 30 days: 1) days with ARDS prior to HFOV (OR 1.5, 95%CI 1.08 to 1.92; P = 0.01); 2) improvement in PaO 2 /FiO 2 in the first 72 hours (OR 0.8, 95 % CI 0.77 to 0.9; P <0.001), 3) age (OR 1.1, 95 % CI 1.02 to 1.14; P = 0.03); and 4) pre-HFOV pH (OR 0.8, 95 % CI 0.7 to   There was an interaction between change in PaO 2 / FiO 2 and the etiology of ARDS, with a greater change in PaO 2 /FiO 2 for patients with extra-pulmonary ARDS, independent of baseline PaO 2 /FiO 2 , which may reflect the different degree of lung recruitability.
There was a trend towards greater reduction in PaCO 2 in 'responders' as defined on the ROC curve by an increase in PaO 2 /FiO 2 ratio of >38 % compared to 'nonresponders' (increase in PaO 2 /FiO 2 ratio of <38 %), with a median per cent change (IQR) of -17.4 % (-33.4 to 5.48) versus -4.9 % (-19.8 to 11.9) (P = 0.07) (Figure 2b). Furthermore, patients with a worse empirical disease severity score showed a more rapid clearance of PaCO 2 during the first 12 hours of HFOV (Figure 2c) despite similar settings of frequency, amplitude and power (Figure 3). The absolute PaCO 2 remained higher in the more severe group. This result may suggest that patients with more severe disease have a greater proportion of recruitable lung and an increase in alveolar ventilation following HFOV. Overall, patients with lower respiratory system compliance had a trend towards a greater change in PaCO 2 post-SRM (-20.5% versus -2.4 %; P = 0.08), and there was a correlation between change in PaCO 2 post-SRM and change in compliance post-HFOV (r 2 = 0.6; P = 0.04).

Discussion
This study aimed to identify potential predictors of survival in patients with severe ARDS who received HFOV after failing lung-protective CMV. The key results of our study are that: 1) an early improvement in PaO 2 / FiO 2 ratio is a predictor of survival at 30 days; 2) patients with more severe disease and lower respiratory system compliance pre-HFOV show greater CO 2 clearance; and 3) if there is no improvement in gas exchange within three hours, patients can be considered to have failed HFOV and perhaps should be considered for alternative treatment (for example, extracorporeal support). Despite theoretical beneficial effects on minimizing lung injury and improving gas exchange, HFOV is not widely utilized because of the lack of evidence supporting a clear survival benefit over CMV [25,26,34]. Furthermore, available clinical trials do not help the clinician decide when to consider HFOV and, importantly, how long HFOV should be continued to enhance patient survival. Our study has a similar scope to the series reported by Adhikari et al. [23]; however, there are important methodological differences in the HFOV protocols and the type of recruitment maneuver (that is, a slow stepwise maneuver in our study versus a sustained inflation in Adhikari et al.) used in the two studies. Furthermore, recruitment maneuvers were performed in all patients in our case series, whereas in the study by Adhikari et al only 49.5% of the patients received a recruitment maneuver. The rationale of the stepwise recruitment we used in this study was similar to the stepwise recruitment used in neonates [35], in that it allowed for setting of the optimal CDP but it differed in two aspects. First, we used a fixed FiO 2 of 1.0 and response to the recruitment was assessed as changes in PaO 2 . Second, in order to allow time for equilibration of PaO 2 [36] and to minimize hemodynamic instability, our protocol required longer times between changes in CDP (ten minutes during the incremental phase and five minutes during the decremental phase) compared to the recruitment used in neonates [35]. It is possible that the slower and early recruitment, as performed in this study, can explain the early identification of responders to HFOV.
Although our study is not a randomized comparative study, we believe it identifies clinically important predictors of clinical outcome within the first few hours of initiation of HFOV, possibly in response to the initial SRM. Our study population included, as might be expected for a rescue study, patients with more severe ARDS than in the MOAT trial where patients received HFOV as a primary ventilation mode [26] and similar to that of the EMOAT trial [25] and the recent case series [23].
In contrast to other published reports [17,18], in our study gas exchange variables (PaO 2 /FiO 2 , and OI) improved significantly only in survivors, and change in PaO 2 /FiO 2 remained significantly different between the two outcome groups after adjusting for baseline confounding factors. Although the CDP on HFOV was higher than the CDP on CMV, there was no difference in CDP between survivors and non-survivors. Despite similar levels of CDP, the response to HFOV within the first three hours could identify patients with a favorable outcome based on PaO 2 /FiO 2 and OI.  An important factor for the response to HFOV may be played by the proportion of patients with pulmonary versus extra-pulmonary ARDS. Indeed, in this study we show that the largest change in PaO 2 /FiO 2 post HFOV was seen for extra-pulmonary ARDS whereas little difference in PaO 2 /FiO 2 was seen in pulmonary ARDS. This is consistent with the findings reported by Pachl et al. that show that oxygenation and recruitment during HFOV are more pronounced in patients with extra-pulmonary ARDS [37]. However, Pachl et al. studied changes in oxygenation under normocapnic conditions; therefore, no data on the different behavior of PaCO 2 in the two types of ARDS are available for comparison with our data. The other important finding of our study is the effect of HFOV on PaCO 2 . In our study, in contrast to other reports [19,26], the PaCO 2 decreased significantly throughout the duration of HFOV in parallel to an increase in PaO 2 /FiO 2 despite similar settings of frequency, power and amplitude. We found that in survivors there was both an increase in PaO 2 /FiO 2 and a decrease in PaCO 2. In addition, patients who had at least a 38% increase in their PaO 2/ /FiO 2 (as identified by the ROC curve), also showed greater reductions in PaCO 2 allowing for a reduction in delta pressure. Patients with a greater disease severity (higher PaCO 2, lower compliance and worse gas exchange), showed a higher rate of clearance in PaCO 2 during the first six hours of HFOV. These changes in physiological variables have been described in patients with severe ARDS and higher potential for lung recruitment [38].
The increase in intra-thoracic pressure generated during a SRM could have caused a reduction in cardiac output and pulmonary blood flow, leading to a decrease in venous admixture and to an apparent improvement in PaO 2 /FiO 2 in the absence of true alveolar recruitment [39,40]. However, this mechanism seems less likely as an explanation for the changes seen in our study, as the cardiac output and oxygen delivery were unchanged following the SRM, and therefore the combined improvement in PaO 2 /FiO 2 and PaCO 2 leads us to speculate that HFOV facilitated lung recruitment in a manner similar to that described for patients responding to prone positioning [41].
The 30-day mortality in this study was 56%, comparable to the mortality rate reported in other uncontrolled studies (61.7% [23,42], 66% [18], 53% [17]) but higher than the studies using HFOV as primary intervention (43% [25] and 37% [26]) and studies of trauma patients [16]. Multivariate analysis shows that changes in PaO 2 / FiO 2, age, days with ARDS prior to HFOV and baseline pH are independent predictive factors of mortality. Lung injury is positively associated with duration of mechanical ventilation in both animal and human studies: increased lung injury is associated with reduced likelihood of pulmonary recruitment

Conclusions
In conclusion, this study shows that HFOV is effective in improving oxygenation in some adults with ARDS, particularly when instituted early. This study also shows that changes in PaO 2 /FiO 2 are sensitive criteria to predict survival and the change in PaCO 2 may identify patients with a greater proportion of recruitable lung more likely to benefit from HFOV. Patients who do not show improvement in PaO 2 /FiO 2 ratio and oxygenation index within six hours on commencing HFOV should be considered for extracorporeal support. These data are of potential value in aiding decision making.
Further randomized controlled trials powered to detect a difference in survival between HFOV strategies are expected. Interpretation of these comparisons will also need to take into consideration the number, the duration, and the type of recruitment maneuvers carried out during HFOV and CMV (for example, slow stepwise RMs, as employed in this study versus traditional RMs, with continuous positive airway pressure (CPAP) of 40 cmH 2 O for 40 seconds). The identification of patients likely to benefit from HFOV and the identification of physiological variables associated with the potential for lung recruitment will prove essential to ensure the best use of HFOV in adults with ARDS.

Key messages
• Changes in PaO 2 /FiO 2 early during HFOV are sensitive criteria to predict survival.
• Patients who do not show improvement in the PaO 2 /FiO 2 ratio and OI within three hours should be considered for alternative treatment (for example, extracorporeal support).
• The identification of patients likely to benefit from HFOV and the identification of physiological variables associated with the potential for lung recruitment will prove essential to ensure the best use of HFOV in adults with ARDS.