Research Highlights

A Machine-learning Approach to Human Ex Vivo Lung Perfusion Predicts Transplantation Outcomes and Promotes Organ Utilization Sage AT, Donahoe LL, Shamandy AA, et al. Nat Commun 2023;14:1187. Lungs from marginal donors are frequently discarded, significantly contributing to prolonged wait times. Recent ex vivo lung perfusion (EVLP) techniques, along with the ability to measure many parameters simultaneously during EVLP, have allowed for an improved assessment of such marginal organs, with the potential to increase utilization.1 However, a standardized algorithm for an accurate assessment predicting organ quality and transplant outcome has been lacking, contributing to inconsistent acceptance. The development of such an effective algorithm may thus support a more effective use of EVLP, potentially benefitting a wide range of lung transplant programs. In their study published in Nature Communications, Sage et al2 developed a machine learning (ML) model, which they called “InsighTx” that feeds clinical EVLP data into a decision tree–based ML technique called “eXtreme Gradient Boosting (XGBoost).” Clinical EVLP data used included physiological measurements (ie, gas exchange, compliance, airway pressure), biochemical (eg, glucose and lactate levels, pH, acid–base chemistry), imaging (ie, radiographic images, bronchoscopy), and biological information (ie, cytokines, chemokines). The authors trained the model with >500 EVLP cases from their own center (Toronto General Hospital). At the training stage, the model was instructed to predict 3 clinical outcomes: (1) donor lung discard, (2) donor lung transplantation and extubation <72 h, and (3) donor lung transplantation and extubation ≥72 h. The training process achieved an overall area under the receiver operating characteristic curve (AUROC) of 79% ± 3%. Next, the authors validated the model with 97 (test set 1) and 124 (test set 2) EVLP cases, again recruited from their own center. At the validation stage, each EVLP case was assigned a predicted outcome based on the highest probability derived from InsighTx, which was then used for model performance analysis. The InsighTx model achieved an overall AUROC of 75% ± 4% and 85% ± 3% for test sets 1 and 2 in predicting all 3 clinical outcomes. Moreover, the InsighTx model performed even better in predicting outcome no. 1 (donor lung discard) with an AUROC of 88% ± 4% (test set 1) and 95% ± 2% (test set 2), and outcome no. 2 (transplanted and extubated in <72 h) with an AUROC of 76% ± 6% (test set 1) and 83% ± 4% (test set 2), respectively. Importantly, the InsighTx model performed significantly better in predicting lung discard or rapid extubation posttransplant based on the area under the precision-recall curve. Moreover, when the investigators incorporated specific recipient features (age, sex, body mass index, status, and transplant indication) into the InsighTx model, they found an additional improvement by ~6% in AUROC. Thus, at least in theory, the application of the InsighTx model may result in a 13% improvement in support of proceeding to transplant of lungs that are empirically discarded. In addition, the model provided a 13-fold increase in the overall odds of a favorable transplant outcome. These assessments may provide a particular advantage for EVLP inexperienced centers (<100 total EVLP cases). The current study represents the first important step in the direction of incorporating ML for the decision to accept lungs for transplant in the era of EVLP. To determine the true utility of such an algorithm as stated by the authors, it seems of paramount importance to perform a prospective, multicenter trial comparing the InsighTx model with the current best available multivariable model predicting transplant outcomes. A comparable approach predicting outcomes following kidney transplant has recently been published. In this study, the authors determined that their ML model did not outperform a standard Cox-Based Prognostication System.3 Therefore, continuous improvements in using ML models will be extremely important to enhance the current AUROCs for lungs or any other transplant. Nevertheless, the model presented by Sage et al is timely. Continued refinement with the incorporation of data based on newer perfusion techniques designed for organ repair, regeneration, and immune modifications, optimizing donor–recipient matching beyond immunological matching will be extremely beneficial. Pig-to-human Heart Xenotransplantation in Two Recently Deceased Human Recipients Moazami N, Stern JM, Khalil K, et al. Nat Med. 2023;29:1989–1997. Heart failure is a leading cause of mortality, which is as high as 50% by 5 y after diagnosis despite maximal medical therapy. A profound shortage continues to limit access to life-saving organ transplantation. Xenotransplantation presents an intriguing and potentially game-changing avenue in addressing the organ shortage crisis. The use of genetically modified pigs has shown promise in overcoming the hurdle of hyperacute rejections. Recently, an orthotopic heart xenotransplant has been performed with a patient/graft survival of nearly 2 mo.1 In addition, xeno-kidney transplants in brain-dead individuals have recently been reported.2 Here, Moazami et al3 used cardiac xenografts from 10 gene-edited (10GE) pigs performing 2 heart transplants into brain-dead recipients using surgical techniques comparable with those in human cardiac allotransplantation. The performance of these xenografts was monitored for a total duration of 66 h. Xenografts used had 4 pig genes knocked out: Alpha-Gal, Beta4GalNT2, CMAH, and growth hormone receptor; 6 human genes had been knocked in: CD55, CD46, CD47, human hemeoxygenase-1, human endothelial protein C receptor, and human thrombomodulin. These genetic modifications aimed to reduce the antigenicity in addition to complement-mediated injury while limiting inflammation and thrombosis of cardiac xenografts. The complement inhibitor eculizumab was added to a standard immunosuppression, which included rabbit antithymocyte globulin induction and maintenance with daily methylprednisone and mycophenolate mofetil for the duration of the study. The cardiac xenograft worked immediately and effectively during the short duration of the observation period. Moreover, 10GE xenografts did not show hyperacute rejections with serial and explant biopsies, demonstrating an absence of antibody-mediated rejection. Interestingly, hemodynamic parameters and inflammatory indices differed between the first and second xenografts. The postoperative ejection fraction of the first compared with the second graft was 45% versus 75%, whereas the vasopressor requirement was overall higher for the first recipient. At explant, C4d deposition in myocytes of the first xenograft was noted (commonly seen in heart transplant myocyte injury unrelated to hyperacute rejection), whereas both xenografts showed subendocardial hemorrhage, a likely consequence of ischemia–reperfusion injury. Differences in cardiac performance were attributed to the size mismatch in the first xenograft. At the same time, the second recipient required continuous renal replacement at the onset of the study. Knockout of the growth hormone receptor leads to a smaller heart size than predicted on the basis of pig size and weight, thus challenging size matching. A variable expression of inserted transgenes added to the complexity of the 10GE pig hearts. Furthermore, the authors were able to investigate the transmission of zoonotic pathogens during the observation period via serial measurements of both porcine cytomegalovirus and porcine endogenous retroviruses, both of which remained negative. Moving forward, longer follow-up and surveillance will be required to establish the risk of swine pathogen transmission risk. Overall, this study was successful in demonstrating the feasibility of investigating immunological and infectious risks of xenotransplantation by using a model of 10GE xeno hearts transplanted into brain-dead recipients. Further work is needed to study xenograft function beyond the 66-h duration of the study, assessing the role of CD40-based immunosuppression in controlling adaptive immunity and promoting longer-term xenograft function.

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.