Estimating Pediatric Acute Respiratory Distress Syndrome Outcomes From Oxygenation Variables.
E. Killien,J. Zimmerman
DOI: https://doi.org/10.1097/CCM.0000000000002969
IF: 8.8
2018-04-01
Critical Care Medicine
Abstract:654 www.ccmjournal.org April 2018 • Volume 46 • Number 4 Pediatric acute respiratory distress syndrome (ARDS) remains a significant antecedent of childhood morbidity and mortality worldwide (1, 2). Quality interventional trials are needed to identify evidence-based interventions to define best practice regarding critical care for these patients. It makes sense to identify the population that is most likely to derive benefit from the treatment under investigation: some patients may be too moribund for any intervention to be effective, whereas others may be minimally affected by the illness or injury and will survive intact regardless and may be exposed to potential adverse effects of the intervention without potential benefit (3, 4). It has previously been established that discrimination of mortality in pediatric ARDS based on oxygenation variables is improved after 24 hours compared with at ARDS onset (5–7). Ideally, discrimination of the subject population should occur early in the disease process, however, in order for the intervention to have the greatest chance of being effective (8). Accordingly, identifying patients at high risk for poor outcomes while also shortening the timeframe for ARDS screening and trial recruitment would be desirable. The article by Yehya et al (9) in this issue of Critical Care Medicine describes an observational cohort study that examined oxygenation variables (Pao 2 /Fio 2 and oxygenation index [Fio 2 × mean airway pressure × 100]/Pao 2 ) to predict pulmonary-focused outcomes (duration of mechanical ventilation, ventilator free days, probability of extubation) and global outcomes (mortality) among two cohorts of children with ARDS. Data for the derivation cohort were gleaned from a database at Children’s Hospital of Philadelphia (CHOP), whereas data for the validation cohort were obtained from a similar but independent database maintained at Children’s Hospital of Los Angeles (CHLA). It is important to stress that ventilation strategies and use of ancillary therapies were not protocolized for either of these cohorts. It is unclear if mechanical ventilator–related outcomes are reliable in the absence of ventilator weaning and extubation readiness testing (and perhaps sedation weaning) protocols, as multiple variables besides natural history of the pulmonary disease process and interventions are operative. The investigators first ascertain that no relationship exists between oxygenation variables at ARDS onset and mortality in either cohort (although discrimination improved when the cohorts were combined with added power). Some patients initially categorized with ARDS will improve quickly, perhaps with a little lung recruitment strategy; indeed, in this cohort, over half of patients only had mild ARDS or no longer met ARDS criteria by 12 hours. These patients would be poor candidates for additional experimental intervention, just adding noise to the outcome signal. Next, the investigators demonstrated a statistically significant dose-response relationship between Pediatric Acute Lung Injury Consensus Conference (PALICC) oxygenation index categories and risk of mortality at 12 and 24 hours after ARDS diagnosis (increasing mortality with worsening oxygenation, Fig. 1 in [9]). These findings provide face validity for use of oxygenation index to identify an appropriate experimental cohort with risk for the gold standard patient-centered outcome, mortality. Table 1 in (9) provides a summary of areas under the receiver operating characteristic (AUROC) curve indicating that between 6 and 24 hours, both Pao 2 /Fio 2 (Berlin criteria) and oxygenation index (PALICC criteria) predict statistically relevant risks for mortality. However, the AUROC for Pao 2 /Fio 2 and oxygenation index were only 0.57–0.62 and 0.60–0.62, respectively. For this type of analysis, AUROC of 0.70–0.80 is considered acceptable discrimination (10). With AUROC less than this, considerable lack of sensitivity and/or specificity must exist, suggesting that oxygenation variables alone do not consistently identify the appropriate population for an ARDS experimental intervention. At 12 and 24 hours, calibration for mortality was poor in the CHLA validation cohort, with r less than or equal to 0.70. The fact that mortality at CHLA was 23.1% as compared to 14.6% at CHOP with both cohorts exhibiting similar Pediatric Risk of Mortality (PRISM) scores underscores the fact that multiple variables (not quantified in this investigation) influence pediatric ARDS mortality. In fact, most children succumbing from ARDS do not die of refractory hypoxemia (11, 12); specifically, for this cohort, only 2.8% died from this cause. On the other hand, ventilator-free days, ventilator days among survivors, and probability of tracheal extubation all demonstrated improved association with the oxygenation variable severity at 12 and 24 hours, with better *See also p. 619.