ID:98 FC-D05-1 External Validation of Existing Mortality Prediction Scores for Veno-Venous Extracorporeal Membrane Oxygenation Patients in Japan: A Retrospective Study Using a Japanese ICU Database.

Background: Veno-venous ECMO is an established treatment for acute severe respiratory failure. Several mortality prediction models of veno-venous ECMO patients have been reported. However, those models mainly focused on populations aged 30–50 years. Their transportability to Japanese ICU patients, with a high percentage of elderly patients, remains unclear. The aim of this study was to externally validate existing mortality prediction scores in veno-venous ECMO patients using data from a nationwide Japanese ICU database. Methods: We conducted a retrospective cohort study using the Japanese Intensive Care Patient Database (JIPAD), a large comprehensive database consisting of deidentified data from ICU patients across participating Japanese institutions. We extracted patients aged 18 years or above admitted to the ICUs between April 2015 and March 2024 and who received veno-venous ECMO. Patients with elective admission/surgery, venous-arterial ECMO or readmission were excluded. We externally validated five mortality prediction scores: ECMOnet, PRESERVE, PRESET, Roch, and Enger scores. The predictive accuracy was assessed using the area under the receiver operating characteristic curve (AUROC) and 95% confidence interval(CI). Result: A total of 876 patients met the inclusion criteria. The median age was 61 [interquartile range 50-70] years and in-hospital mortality was 41.3%. All the evaluated prediction scores required some parameters that were unavailable in JIPAD. With these missing data, each score was assigned a null value. The AUROCs (95% CI) for the five scores were calculated as follows: PRESET, 0.649 (0.611–0.686); PRESERVE, 0.638 (0.602–0.673); Enger, 0.632 (0.595–0.669); Roch, 0.614 (0.578–0.651); and ECMOnet, 0.607 (0.568–0.647). Conclusion: Existing VV-ECMO mortality scores showed modest discrimination in a large Japanese ICU cohort, though some parameters were unavailable in JIPAD. Development of new prediction models tailored to JIPAD is warranted.