A radiographic score for human donor lungs on ex vivo lung perfusion predicts transplant outcomes

Background Ex vivo lung perfusion (EVLP) is an advanced platform for isolated lung assessment and treatment. Radiographs acquired during EVLP provide a unique opportunity to assess lung injury. The purpose of our study was to define and evaluate EVLP radiographic findings and their association with lung transplant outcomes. Methods We retrospectively evaluated 113 EVLP cases from 2020–21. Radiographs were scored by a thoracic radiologist blinded to outcome. Six lung regions were scored for 5 radiographic features (consolidation, infiltrates, atelectasis, nodules, and interstitial lines) on a scale of 0 to 3 to derive a score. Spearman's correlation was used to correlate radiographic scores to biomarkers of lung injury. Machine learning models were developed using radiographic features and EVLP functional data. Predictive performance was assessed using the area under the curve. Results Consolidation and infiltrates were the most frequent findings at 1 hour EVLP (radiographic lung score 2.6 (3.3) and 4.6 (4.3)). Consolidation (r = −0.536 and −0.608, p < 0.0001) and infiltrates (r = −0.492 and −0.616, p < 0.0001) were inversely correlated with oxygenation (∆pO2) at 1 hour and 3 hours of EVLP. First-hour consolidation and infiltrate lung scores predicted transplant suitability with an area under the curve of 87% and 88%, respectively. Prediction of transplant outcomes using a machine learning model yielded an area under the curve of 80% in the validation set. Conclusions EVLP radiographs provide valuable insight into donor lungs being assessed for transplantation. Consolidation and infiltrates were the most common abnormalities observed in EVLP lungs, and radiographic lung scores predicted the suitability of donor lungs for transplant. Ex vivo lung perfusion (EVLP) is an advanced platform for isolated lung assessment and treatment. Radiographs acquired during EVLP provide a unique opportunity to assess lung injury. The purpose of our study was to define and evaluate EVLP radiographic findings and their association with lung transplant outcomes. We retrospectively evaluated 113 EVLP cases from 2020–21. Radiographs were scored by a thoracic radiologist blinded to outcome. Six lung regions were scored for 5 radiographic features (consolidation, infiltrates, atelectasis, nodules, and interstitial lines) on a scale of 0 to 3 to derive a score. Spearman's correlation was used to correlate radiographic scores to biomarkers of lung injury. Machine learning models were developed using radiographic features and EVLP functional data. Predictive performance was assessed using the area under the curve. Consolidation and infiltrates were the most frequent findings at 1 hour EVLP (radiographic lung score 2.6 (3.3) and 4.6 (4.3)). Consolidation (r = −0.536 and −0.608, p < 0.0001) and infiltrates (r = −0.492 and −0.616, p < 0.0001) were inversely correlated with oxygenation (∆pO2) at 1 hour and 3 hours of EVLP. First-hour consolidation and infiltrate lung scores predicted transplant suitability with an area under the curve of 87% and 88%, respectively. Prediction of transplant outcomes using a machine learning model yielded an area under the curve of 80% in the validation set. EVLP radiographs provide valuable insight into donor lungs being assessed for transplantation. Consolidation and infiltrates were the most common abnormalities observed in EVLP lungs, and radiographic lung scores predicted the suitability of donor lungs for transplant.