Performance of Multiple Risk Assessment Tools to Predict Mortality for Adult Respiratory Distress Syndrome with Extracorporeal Membrane Oxygenation Therapy: An External Validation Study Based on Chinese Single-center Data

BACKGROUND: There has been no external validation of survival prediction models for severe adult respiratory distress syndrome (ARDS) with extracorporeal membrane oxygenation (ECMO) therapy in China. The aim of study was to compare the performance of multiple models recently developed for patients with ARDS undergoing ECMO based on Chinese single-center data. METHODS: A retrospective case study was performed, including twenty-three severe ARDS patients who received ECMO from January 2009 to July 2015. The PRESERVE (Predicting death for severe ARDS on VV-ECMO), ECMOnet, Respiratory Extracorporeal Membrane Oxygenation Survival Prediction (RESP) score, a center-specific model developed for inter-hospital transfers receiving ECMO, and the classical risk-prediction scores of Acute Physiology and Chronic Health Evaluation (APACHE) II and Sequential Organ Failure Assessment (SOFA) were calculated. In-hospital and six-month mortality were regarded as the endpoints and model performance was evaluated by comparing the area under the receiver operating characteristic curve (AUC). RESULTS: The RESP and APACHE II scores showed excellent discriminate performance in predicting survival with AUC of 0.835 (95% confidence interval [CI], 0.659–1.010, P = 0.007) and 0.762 (95% CI, 0.558–0.965, P = 0.035), respectively. The optimal cutoff values were risk class 3.5 for RESP and 35.5 for APACHE II score, and both showed 70.0% sensitivity and 84.6% specificity. The excellent performance of these models was also evident for the pneumonia etiological subgroup, for which the SOFA score was also shown to be predictive, with an AUC of 0.790 (95% CI, 0.571–1.009, P = 0.038). However, the ECMOnet and the score developed for externally retrieved ECMO patients failed to demonstrate significant discriminate power for the overall cohort. The PRESERVE model was unable to be evaluated fully since only one patient died six months postdischarge. CONCLUSIONS: The RESP, APCHAE II, and SOFA scorings systems show good predictive value for intra-hospital survival of ARDS patients treated with ECMO in our single-center evaluation. Future validation should include a larger study with either more patients’ data at single-center or by integration of domestic multi-center data. Development of a scoring system with national characteristics might be warranted.