Key result
An institutional prediction model outperformed the RESP, ECMOnet, and SOFA scores in predicting hospital mortality for adults on VV-ECMO, achieving an AUROC of 0.779 (P<0.001).
Why the study?
A general agreement on the use and timing of venovenous extracorporeal membrane oxygenation (VV-ECMO) for adult respiratory failure has not been reached due to its invasiveness and high resource usage, and existing predictive scores require validation in medium-sized cohorts.
Does an institutional prediction model improve the prediction of hospital mortality compared to published scores in adults receiving VV-ECMO for severe acute respiratory failure?
Cohort (n=107)
No
Does an institutional prediction model improve the prediction of hospital mortality compared to published scores in adults receiving VV-ECMO for severe acute respiratory failure?
Effect estimate: AUROC 0.779
p-value: p=<0.001
An institutional prediction model incorporating SOFA score, mechanical ventilation duration, and immunocompromised status outperformed existing published scores in predicting hospital mortality for adults on VV-ECMO for severe respiratory failure.
Despite a potentially effective therapy for adult respiratory failure, a general agreement on venovenous extracorporeal membrane oxygenation (VV-ECMO) has not been reached among institutions due to its invasiveness and high resource usage. To establish consensus on the timing of intervention, large ECMO organizations have published the respiratory extracorporeal membrane oxygenation survival prediction (RESP) score and the ECMOnet score, which allow users to predict hospital mortality for candidates with their pre-ECMO presentations. This study was aimed to test the predictive powers of these published scores in a medium-sized cohort enrolling adults treated with VV-ECMO for acute respiratory failure, and develop an institutional prediction model under the framework of the 3 scores if a superior predictive power could be achieved. This retrospective study included 107 adults who received VV-ECMO for severe acute respiratory failure (a PaO2/FiO2 ratio <70 mm Hg) in a tertiary referral center from 2007 to 2015. Essential demographic and clinical data were collected to calculate the RESP score, the ECMOnet score, and the sequential organ failure assessment (SOFA) score before VV-ECMO. The predictive power of hospital mortality of each score was presented as the area under receiver-operating characteristic curve (AUROC). The multivariate logistic regression was used to develop an institutional prediction model. The surviving to discharge rate was 55% (n = 59). All of the 3 published scores had a real but poor predictive power of hospital mortality in this study. The AUROCs of RESP score, ECMOnet score, and SOFA score were 0.662 (P = 0.004), 0.616 (P = 0.04), and 0.667 (P = 0.003), respectively. An institutional prediction model was established from these score parameters and presented as follows: hospital mortality (Y) = -3.173 + 0.208 × (pre-ECMO SOFA score) + 0.148 × (pre-ECMO mechanical ventilation day) + 1.021 × (immunocompromised status). Compared with the 3 scores, the institutional model had a significantly higher AUROC (0.779; P < 0.001). The 3 published scores provide valuable information about the poor prognostic factors for adult respiratory ECMO. Among the score parameters, duration of mechanical ventilation, immunocompromised status, and severity of organ dysfunction may be the most important prognostic factors of VV-ECMO used for adult respiratory failure.
No takes yet. Share an insight, caveat, or question.
Hsin et al. (2016) conducted a cohort in severe acute respiratory failure (n=107). Institutional prediction model vs. RESP, ECMOnet, and SOFA scores was evaluated on hospital mortality prediction (AUROC) (AUROC 0.779, p=<0.001). An institutional prediction model outperformed the RESP, ECMOnet, and SOFA scores in predicting hospital mortality for adults on VV-ECMO, achieving an AUROC of 0.779 (P<0.001).
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: