- Author: Graeme MacLaren, Hwa Jin Cho, Luregn J Schlapbach
- Date: 2019/05/01
- Journal: Pediatric Critical Care Medicine;20(5);490-491
- PMID: 31058786
- PMCID:
- DOI: 10.1097/PCC.0000000000001910
Abstract
Extracorporeal membrane oxygenation (ECMO) is a resource-intensive and costly form of life support which is being applied to increasingly complex patients (1). The decision to cannulate is still made on a case-by-case basis, but there is considerable variation in initiation criteria between clinicians, institutions, and regions, reflecting the paucity of evidence with which to guide best practice. This lack of standardization represents a barrier to the benchmarking of ECMO services. Any tool which is simple, quick, and uses readily available clinical information to predict the likelihood of survival could be very helpful in the care of potential ECMO candidates (2). However, current prediction models are imperfect, lack sufficient discrimination to be used in individual patients, and are often only validated in specific disease states such as acute respiratory distress syndrome rather than in all ECMO patients (3–6).
In this issue of Pediatric Critical Care Medicine, Bailly et al (7) present a new model to predict the survival of children managed with ECMO. The strength of their approach was the use of an existing, prospective, high-quality dataset from an earlier study which detailed ECMO in 514 children from eight North American centers over a 22 month period (8), incorporating simple clinical variables collected within 12 hours prior to initiation of ECMO. Significant predictors of mortality included age, indication for ECMO (cardiac; respiratory; and extracorporeal cardiopulmonary resuscitation), bloodstream infection (BSI) prior to ECMO, underlying disease (congenital diaphragmatic hernia; and meconium aspiration syndrome), baseline acidosis, and coagulation disturbances.
The principal advantages of the study by Bailly et al (7) were that it was based on a relatively large, multicenter cohort including children of all ages from premature newborns through to adolescents; it included any patient on ECMO irrespective of the disease or indication; and the data collected had more detail on individual organ dysfunction than many earlier reports, which generally relied on the Extracorporeal Life Support Organization Registry. The model performed equally well or better than previous pediatric studies in the field but, as the authors rightly highlighted, it cannot be used to assess patients’ ECMO candidacy because the study did not include a comparison group of patients without ECMO. Other limitations relate to the potential colinearity of variables included in the model such as international normalized ratio and activated partial thromboplastin time; the amount of missing data; and the absence of international centers in the study. Of note, a surprisingly high portion of patients—approximately one-third—received roller pump ECMO. Consequently, although the controversy over the putative superiority of centrifugal versus roller pump ECMO continues (9–11), the results of the study by Bailly et al (7) may not be generalizable to centers in Europe or Asia, where centrifugal pump technology was universally adopted well over a decade ago.
Interestingly, the risk factor with the highest associated effect size for mortality in the model was BSI prior to ECMO. Did this reflect the underlying complexity of the patient cohort, who may have had prolonged hospital stays or immune dysfunction which predisposed them to nosocomial infection? Or did it perhaps reflect the relatively high mortality when ECMO is used in sepsis? As the authors speculated, ECMO can immediately improve cellular oxygen delivery during cardiac arrest in most situations but does so less readily or consistently in septic shock because of mitochondrial dysfunction and increased metabolic demand. However, only 27 of the 514 patients had BSI prior to ECMO, so it was a relatively uncommon clinical scenario. As the study by Bailly et al (7) did not validate the model in an independent external dataset, there is a risk of model overfitting, and very few events may have a disproportionate effect on overall performance. Acknowledging these limitations, the approach by Bailly et al (7) represents a promising start toward more robust risk adjustment in pediatric ECMO regardless of the underlying disease.
Currently, most ECMO practices in pediatric critical care lack high-grade evidence of benefit, and substantial patient and disease heterogeneity have been used to explain the high practice variability within and between institutions. Rapid, easily available models for risk adjustment in children considered for or treated with ECMO are a fundamental requirement to risk-stratify patients for enrollment in pediatric ECMO interventional trials and to benchmark practice. At present, the main benchmark in ECMO used internationally is unadjusted mortality based on basic groupings such as age, ECMO cannulation strategy, and indication. In contrast, benchmarking of risk-adjusted mortality is considered best practice in most ICU settings. However, several barriers need to be overcome to develop robust benchmarking in ECMO. First, the area under the curve of the model by Bailly et al (7) is still substantially lower than risk scores such as the Pediatric Index of Mortality which have been established for general PICU patients. Second, generalizability will require the creation of much larger international databases that are powered to reliably validate novel scores (12). Third, given that many ECMO centers have less than 50 patients per year, applying risk adjustment to such small patient groups will call for the creation of more innovative, adaptive strategies. Finally, although mortality represents an objective endpoint, there is growing realization of the need to examine other endpoints such as long-term neurodevelopmental outcomes (13).
In summary, although the proposed pediatric ECMO prediction model requires prospective, external validation, the authors should be congratulated for providing an important stepping-stone toward future prediction models that may be used to quantify the risks and benefits of initiating ECMO rather than continuing conventional care.