Pediatric Acute Respiratory Distress Syndrome: Increase the Positive End-Expiratory Pressure?

The goals of mechanical ventilation in acute respiratory distress syndrome (ARDS) are to maintain adequate oxygenation and ventilation while simultaneously avoiding injurious stress and strain. Standard maneuvers to improve arterial oxygen saturation are delivery of higher Fi_O2 and increasing positive end-expiratory pressure (PEEP) to improve ventilation and perfusion matching and recruit collapsed alveoli. PEEP has the added benefit of potentially reducing ventilator-induced lung injury by diminishing atelectrauma and minimizing exposure to potentially toxic high Fi_O2, if balanced appropriately with the risk for alveolar overdistension (1, 2). Substantial research has been performed on PEEP optimization, but the current standard of care for adults with ARDS derives from the ARDS Network (ARDSNet) trials employing a fixed titration table of Fi_O2 and PEEP that was generated by expert consensus (3, 4). Three large trials examined higher versus lower PEEP (4–6), and although no individual trial showed a difference in outcomes, a meta-analysis suggested that in moderate-severe ARDS, higher PEEP was associated with improved survival (7). A recent trial of alveolar recruitment and higher PEEP for ARDS, however, showed an increased mortality in the intervention arm that runs counter to the trend of improved survival with higher PEEP (8).

Although pediatric ARDS (PARDS) shares common pathology and pathophysiology with adult ARDS, data from adult trials may not always translate to younger populations: infants have markedly reduced chest wall elastance compared with adults, and elastin and collagen composition of lungs change with age (9, 10). Prospective studies of optimal PEEP levels in PARDS are sparse, leading to variability in the relative reliance on Fi_O2, PEEP, and other treatments for hypoxemia in the absence of clear evidence (11, 12).

In this issue of the Journal, Khemani and colleagues (pp. 77–89) examined previously collected data from 1,134 mechanically ventilated children with PARDS to learn more about the association between PEEP levels and mortality (13). The authors performed a thoughtful and detailed analysis in the largest study to date to explore PEEP in a pediatric population, strengthened by its size and inclusion of multiple centers. They used four datasets: two from Children’s Hospital Los Angeles (CHLA), one from Children’s Hospital of Philadelphia (CHOP), and one from the Collaborative Pediatric Critical Care Research Network. Because the Network dataset contained fewer descriptive variables, the most informative part of the study derived from the two hospitals (CHLA and CHOP).

Higher PEEP and higher Fi_O2 reflect PARDS severity and are individually associated with mortality, so the authors compared the administered PEEP to the PEEP recommended by the original ARDSNet PEEP–Fi_O2 table (3) for the administered Fi_O2, using the median values in the first day of PARDS. In these cohorts, there was a striking tendency to use 8–10 cm H₂O of PEEP without much escalation across the spectrum of Fi_O2, meaning that children with less severe PARDS were likely to receive PEEP at or above that recommended by the ARDSNet table, whereas those with more severe PARDS were likely to receive PEEP lower than recommended. This results in an association between lower-than-recommended PEEP with higher severity of lung injury, and consequently an association with higher mortality. The question is whether the finding persists when accounting for confounders.

Using multivariable modeling controlling for hospital and markers of disease severity (Pa_O2:Fi_O2 ratio, Pediatric Risk of Mortality III score, immunodeficiency, stem cell transplant, driving pressure, and inhaled nitric oxide [iNO]), the authors found a doubling of odds of death among children managed with PEEP less than recommended by the ARDSNet table compared with children managed with recommended or higher PEEP. If borne out in prospective research, the findings would suggest that clinicians should increase the PEEP to at least that recommended by the ARDSNet table. However, caution is advised, as there are important limitations to the current study.

The first limitation is the nonrandom application of PEEP and other therapies, a problem that is inherent to the observational study design. Correctly accounting for multiple confounders that are closely related and interdependent, (e.g., PEEP application modifies the Pa_O2:Fi_O2 ratio and pulmonary compliance [14, 15], and the Pa_O2:Fi_O2 ratio and airway pressures are likely to influence the decision to initiate iNO and other ancillary therapies) is challenging in a retrospective, observational study. The authors’ efforts to control for severity of lung injury with Pa_O2:Fi_O2 ratio, driving pressure, and iNO help reduce confounding, but residual confounding likely persists.

The second limitation derives from the fact that the data sets used for the multivariable modeling come from just two hospitals. Although neither hospital used a formal protocol for PARDS, there was more frequent use of iNO and alternative ventilation modes and higher PEEP at CHOP compared with CHLA. The CHLA cohort also had a higher incidence of immunodeficiency (26.3% vs. 19.2%) and a higher mortality rate (22.4% vs. 14.4%). A formal test of interaction between hospitals and the association of PEEP (relative to recommended) with mortality was negative (P = 0.4), but analyses of interaction terms tend to have low power. When the model was separately derived on the CHLA and CHOP cohorts, the doubling of risk for mortality for lower-than-recommended PEEP persisted in the CHLA cohort, but the point estimate for the CHOP cohort went in the opposite direction, albeit with a wide confidence interval (0.87; 95% confidence interval, 0.32–2.34 points). This suggests that the association of lower-than-recommended PEEP to increased mortality is driven by the CHLA cohort. The authors also performed an adjusted risk model with a propensity score for use of lower-than-recommended PEEP, and belonging to the CHLA cohort is the variable with the highest odds ratio in the propensity score. One must wonder whether unaccounted-for hospital-level differences such as the timing of initiation of nonconventional ventilator modes, threshold for use of iNO, and potentially use of other nonprotocolized therapies uncontrolled for in the multivariable model could explain the difference in the direction of association between mortality and lower-than-recommended PEEP between hospitals.

In the absence of prospective trials to guide PEEP selection in PARDS, the investigation by Khemani and colleagues is a welcome addition to the evidence base, but because of its limitations, it should not yet inform clinical care. Nonetheless, having demonstrating a variability in use of PEEP for PARDS that appears to be independently associated with mortality, the authors establish a clear need for prospective studies of PEEP in PARDS.