Building a better mouse trap in liver retransplantation risk prediction: Is reSOFT the end or the beginning?

The manuscript by Terry et al1 in Liver Transplantation introduces the Survival Outcomes Following Liver Retransplantation (reSOFT) score, a predictive model for 3-month survival after liver retransplantation. By analyzing 6660 adult liver retransplants over a 21-year period, the authors derived a scoring system that categorizes patients into low-risk, moderate-risk, and high-risk groups. Their findings underscore the significance of 14 recipient factors, 1 donor factor (age), and 1 operative factor (cold ischemia time) that significantly impacted 3-month patient survival. The predictive model by Terry et al offers a structured approach to decision-making in an inherently high-risk population primarily using recipient characteristics. These data should also be considered in the context of clinically relevant factors in daily transplant practice and how they should be integrated with current preservation technologies. This work arrives at a pivotal moment in transplantation, where the adoption of advanced clinical perfusion technology, including normothermic machine perfusion (NMP) and normothermic regional perfusion (NRP), is revolutionizing organ preservation, utilization, and patient outcomes.2–4 Although this contribution improves upon previously derived retransplant outcome prediction models, several critical questions remain unanswered. The reSOFT model provides a framework for considering relevant clinical data points and augmenting clinical decision-making in liver retransplantation. However, it is critical to understand how certain risk factors mediate the observed outcomes, as well as what is missing. For example, the study found that HCC was protective from mortality following retransplant. Since cancer inherently does not protect from mortality, the effect of this variable was likely mediated by the provision of MELD exception points that allow for earlier organ offers and shorter waiting times throughout much of the 21-year timeframe of the cohort. An important omission in this study’s source data is granular data regarding specific indications for retransplantation. No 2 liver retransplant cases are necessarily equal, and the lack of cause-specific data for retransplantation constrains the model from risk-stratifying across distinct clinical scenarios. Future studies, perhaps using a multicenter cohort approach with deeper clinical data, should incorporate more clinical detail. This would help enhance reSOFT’s performance as a surgical decision aid. As a heuristic, reSOFT may provide direction in improving retransplant candidate selection in its current form. However, it is impossible to consider the selection of donors and candidates for liver retransplantation today without considering the impact of advanced clinical perfusion. Specific questions remain after considering reSOFT’s data through this lens. How will advancements in organ preservation impact retransplantation survival? Should perfusion-adjusted risk factors be incorporated into predictive models? Now that thousands of liver transplants have occurred with the use of advanced perfusion in the last few years, it is abundantly clear that traditional predictive models based on clinicodemographic variables must be reconsidered. Other models, such as the UK DCD (donation after circulatory death) risk score, Liver DRI (Donor Risk Index), BAR (Balance of Risk), and SOFT (Survival Outcomes Following Transplantation), were useful based on previous practice patterns. The data used to derive these models are from the pre-perfusion era. The potential mitigation of cold ischemic preservation injury demands a reassessment of the clinical norms we utilize to match donors and recipients. While reSOFT gives us a framework to consider patient candidacy, additional data derived from advanced clinical perfusion, whether from NMP, NRP, or both, may help fine-tune decision-making, and may even redefine perceptions of risk in liver retransplant cases. The authors acknowledge that reSOFT must be considered in a larger context. Donation after circulatory death (DCD) allograft status was not a significant predictive factor for 3-month outcome in reSOFT, contrary to what has been observed in previous studies.5,6 While extended-criteria donor livers offer a strategy to address the liver shortage, these grafts have been avoided for high-risk liver recipients due to risks of complications, graft loss, and death.7 DCD may not have been identified as a risk factor for reSOFT outcomes due to statistical power limitations—DCD grafts may have been underutilized in favor of optimal donation after brain death (DBD) grafts for retransplantation.8 Efforts to transplant these higher-risk grafts into high-risk recipients have shown some success, but optimal utilization remains constrained by clinical, technical, and logistical challenges.9 Importantly, DCD livers are ever-present on the daily organ offer docket considered by US programs, now constituting more than 40% of deceased donors. Advanced perfusion may empower teams to utilize these and other extended-criteria donor allografts for liver retransplantation, which may become a clinical necessity. Recent studies indicate that NMP can facilitate DCD transplantation in high-risk recipients—including those undergoing retransplantation and patients with MELD scores ≥35, PVT, or advanced age (≥70 y). Reported survival rates are as high as 96.2%.10 These outcomes may not have been predicted by traditional risk prediction tools and scoring systems. Liver retransplantation requires a significant depth of clinical and technical expertise to execute successfully. The reSOFT model provides a data-driven fortification of surgical intuition. However, neither the reSOFT model nor other liver transplant outcomes prediction models, such as BAR or SOFT, provide dynamic, real-time predictive analytics to assist surgeons in real-time with a particular candidate and donor liver. Future decision tools to aid in risk prediction must incorporate traditional clinicodemographic variables from transplant datasets and perfusion biomarkers from NMP, NRP, or both. Novel statistical methodologies may be required to develop these tools but could refine survival predictions. These next-generation models will have transformative utility as bedside tools but could also shape organ allocation and distribution schemes, increase organ use, and offer more access to liver transplants. Terry and colleagues should be commended for building a better mouse trap for liver retransplantation risk prediction. Previous tools did not perform as well. These cases are clinically challenging and nuanced. The reSOFT model has the most relevance in candidate selection. Given that the model arrives in our perfusion era, the model’s precision may be limited due to the mitigation of donor risk factors and the ability to extend non-ischemic preservation. Next-generation outcome prediction tools must integrate more granular clinical data and perfusion-specific information. The technology to accomplish this exists, and we should seize this moment to enhance how we think about risk in liver transplantation.