The trial evaluated the effect of NMP availability on “functional utilisation”: the proportion of donor liver offers resulting in a functioning transplant at 12 months. To identify livers at greatest risk of non-use, a novel, regression-based score was developed using UK Transplant Registry (UKTR) data to estimate the probability of non-utilisation of individual livers offered for transplant (SDC1). This algorithm, the Donor Utilisation Index (DUI), identified the 60% of donors livers at greatest risk of decline for inclusion in the study [13].

PLUS was originally conceived as a randomised controlled trial, with eligible liver offers randomised to having NMP available or not. During sequential investigators’ meetings, surgeons in most centres became increasingly unwilling to randomise high-risk liver cohorts (particularly DCD) to SCS, prefering some form of machine preservation. The study was therefore revised to a “threshold-crossing” design, in which a real-world control cohort, selected consecutively according to the same objective inclusion/exclusion criteria to minimise bias, was used a priori to estimate the counterfactual (SCS cohort). A threshold was set for the primary outcome prior to recruiting a prospective single-arm study cohort (NMP available cohort) [14]. The lower bound of the 95% confidence interval for effect size had to exceed the pre-defined efficacy threshold to demonstrate superiority.

The study protocol was prospectively registered (16th November 2021; ISRCTN11552402). Ethical approval was granted by the South Central – Oxford C Research Ethics Committee (ref 21/SC/0297). Patients on the liver transplant list, liver recipients and family members were consulted during study design. A patient representative was part of the trial management group and trial steering committee, advising on conduct and dissemination.

Eligible liver offers were DBD or DCD livers from UK donors aged 16 years or over, with a DUI >0.27, offered to any of the 7 UK liver transplant centres. This DUI threshold was selected as it identified the 60% most poorly utilised livers, with utilisation of 26% compared to 85% in livers with a DUI ≤0.27. Eligible offers were predominantly DCD livers (79%), reflecting lower utilisation compared to DBD livers. Offers from donors with HIV or hepatitis C were excluded, as were livers undergoing any other form of ex-situ machine perfusion or where the participating centre was unable to offer NMP at the time due to logistical reasons (unavailability of device, consumables or trained staff). Livers from donors undergoing NRP were eligible, with planned adjustment for NRP use in analysis. All eligible livers were enrolled to ensure comparability between cohorts.

Eligible recipients were adults (≥18 years) on the elective or super-urgent transplant waiting list. Those declining NMP use through local consent policy, receiving a split liver or reduced liver or multiorgan transplant were excluded.

To estimate the counterfactual, the primary outcome was assessed in a real-world control cohort where NMP was not routinely available, derived from the UKTR between 1st January 2018 and 31st December 2019 (SDC2). Identical inclusion/exclusion criteria were applied to this cohort as to the prospective study cohort, with stratified random sampling used to exclude a small proportion of patients to mimic the consent process used in the prospective cohort. A predefined efficacy threshold of at least 40 additional functional transplants per year was set, equivalent to an odds ratio of 1.22 for functional utilisation. This threshold was agreed by all seven participating centres to represent the minimum clinically-important increase in transplant rates that would justify the increased cost and logistical complexity of NMP use.

A prospective cohort was recruited between 11th April 2022 and 3rd April 2023, with NMP made available for all eligible liver offers. Participating centres were informed of NMP availability at the time of offer, and the impact of availability on the primary outcome was assessed by comparing the two cohorts against the pre-defined efficacy threshold.

NMP was used at the discretion of the implanting surgeon. All participating centres had access to an OrganOx metra device and disposable kits at no additional cost, with limited per-case funding provided for staff time. NMP could be used either from the donor centre (device-to-donor) or following SCS and transport to the recipient centre (back-to-base) at the preference of the implanting team. The OrganOx metra device was used according to the current instructions for use and local protocols, with a minimum of 4 h and maximum of 24 h of perfusion time.

The primary outcome was defined as the proportion of livers with “functional utilisation”: transplantation of an offered liver with 1-year graft survival. This ensures that any increase in utilisation does not occur at the cost of inferior graft outcomes.

Pre-specified secondary outcomes included: proportion of non-utilisation, median biochemical liver function in the first 7 days post-transplant [bilirubin, GGT, ALT, INR, lactate (recorded during high-level care only)], and 12 months post-transplant (bilirubin, ALP, ALT, AST), median Model for Early Allograft Function score (MEAF), 12-month graft and patient survival, proportion of transplants with primary non-function, median length of ITU and hospital stay, proportion of recipients requiring renal replacement therapy after transplant and proportion of recipients experiencing adverse events in the first 3 months and 12 months after transplant. Outcomes for both cohorts were collected from the UK Transplant Registry (UKTR), with additional outcomes collected in study case report forms for the prospective cohort over a 12-month follow-up period. Unexpected serious adverse events during the 12-month follow-up period were collected for the prospective cohort only. A health-economic analysis has been undertaken and will be presented in a separate manuscript.

The sample size assumed a 50% increase in 1-year functional utilisation from 21.8% in the control cohort to 32.7% in the prospective NMP available cohort, representing 135 additional successful transplants per year. Allowing for cross-over for those in the prospective cohort not receiving NMP the study was powered to detect an increase in functional utilisation to 31.2%. To achieve 90% power using a 2.5% one-sided significance level to detect an increase of at least 40 additional transplants per year, 1,035 liver offers were required. This allowed for up to 25% loss due to logistical constraints, and 10% loss due to post-recruitment exclusions and loss to follow-up. Protocol v1.0 planned a smaller sample size of 799 liver offers, but this was revised in v2.0 to allow for higher loss to follow-up and NMP unavailability for logistical reasons. An embedded feasibility study assessed recruitment rates and NMP use at 6 months, but there was no formal interim outcome analysis.

Primary analysis was by modified intention-to-treat, where livers unable to be perfused for logistic reasons were excluded, along with any found to be ineligible after study entry (once recipient characteristics were known). Analyses were adjusted for donor type and use of NRP (DBD, DCD with NRP, DCD without NRP) as a fixed effect as these both have an important effect on organ utilisation.

Primary outcome results, from logistic regression models, are reported as odds ratios (OR) with 95% confidence intervals (CI), and p-values for the cohort effect under the null hypothesis of no difference. Pre-planned subgroup analyses for the primary outcome investigated the effect of donor type, donor liver index (DLI, a UK-derived algorithm that predicts transplant outcome based on donor parameters [15]) and DUI.

Sensitivity analyses and secondary outcome analyses are described in SDC3.

All analyses were performed in SAS v9.4M5.

The flow of liver offers through the study is shown in Figure 1. The control cohort comprised 2,465 liver offers where SCS was the routine method of preservation and NMP was unavailable. The prospective study cohort comprised 1,056 liver offers. Of this cohort, 983 met the eligibility criteria for the modified intention-to-treat analysis. 542 control offers and 278 study offers resulted in transplants.

Donor, liver and recipient characteristics were comparable between the two cohorts (Tables 1–3) and median follow-up time post-transplant was 365 days for both cohorts. For those livers resulting in a transplant, there was an increase in the proportion of DCD livers in the prospective cohort (61.2% vs. 51.5%), indicating an increased willingness to transplant DCD livers where machine perfusion is available. Use of NRP was also higher (29.0% vs. 12.4%), reflecting changes in practice as the prospective cohort was more recent than the control cohort.

In the prospective study cohort NMP was used for 31.1% of eligible livers that arrived at a transplant centre (12.4% of all livers offered). Of those where NMP was used, four livers (3.3%) used a device to donor approach, and 116 (96.7%) used a back to base approach. The approach was unknown for two livers (Table 2). There was considerable variation in NMP use between centres (SDC4) and the key reasons for non-use of NMP related to (i) livers being deemed untransplantable; (ii) surgeon deemed the liver clinically suitable for transplant without perfusion; and (iii) perfusion kit/staff unavailability (SDC10).

The primary outcome of functional utilisation was significantly higher in the prospective study cohort (NMP available) (25.2% vs. 19.6%, OR: 1.26, 95% CI: 1.04–1.53), representing 47 additional functional transplants per year (95% CI 8–90 transplants) (Figure 2). However, the lower bound of the 95% confidence interval did not exceed the pre-defined threshold of 40 additional liver transplants per year, therefore the intervention is not considered superior under the threshold-crossing design. The difference in functional utilisation was primarily due to increased organ utilisation in the cohort with NMP available (27.8% vs. 22.0%) with smaller improvements in patient (95.1% vs. 92.9%) and graft (90.8% and 89.1%) survival. Reasons for non-use of livers transported to the recipient centre for the prospective cohort are shown in SDC12; the most common reason for non-use in NMP treated livers was poor function on the device (78.1%).

Subgroup analyses demonstrated a statistically significant interaction when looking at the impact of NMP availability by donor type (p = 0.0245), with a greater difference in functional utilisation observed in the DCD group (OR: 1.70, 95% CI: 1.36–2.12), exceeding the efficacy threshold. However, this analysis did not include adjustment for NRP, which is not used for DBD donors. In a post hoc analysis which only included DCD donors and with risk adjustment for NRP the OR was 1.37 (95% CI 1.08–1.73; p = 0.0089), not exceeding the efficacy threshold. In livers from DCD donors, post hoc analysis demonstrated no significant interaction between the impact of NMP and use of NRP (p = 0.8861).

There were no statistically significant interactions between NMP availability and the DUI (p = 0.3087) or DLI (p = 0.0833), although for some subgroups the lower bound of the 95% confidence interval did exceed the efficacy threshold (Figure 2). A per-protocol analysis also did not exceed the efficacy threshold (OR: 1.18, 95% CI: 0.96–1.44). Two secondary, pre-specified propensity score matched analyses were also undertaken. The first mirrored the primary analysis and assessed the effect of NMP availability on functional utilisation of offered livers, and this also did not exceed the efficacy threshold (OR: 1.14, 95% CI: 0.93–1.39) (Figure 3). The second propensity score matched analysis instead assessed the impact of NMP usage (rather than availability) on functional utilisation. Post-hoc, this analysis was amended to avoid the bias whereby only retrieved livers could have received NMP and therefore analysed only retrieved livers in the prospective cohort, and compared between those where NMP was used, and propensity score matched livers where NMP was not used. This analysis did exceed the efficacy threshold, indicating that NMP does increase functional utilisation of retrieved livers when it is used (post hoc, OR: 2.18, 95% CI: 1.35–3.51) (Figure 3). The sample size was lower for the propensity score matched analyses than for the primary analysis.

There were no statistically significant differences in biochemical liver function tests at 12 months post-transplant, graft and patient survival, primary non-function, length of hospital stay, length of ITU stay, and need for renal replacement therapy (SDC5). Sensitivity analyses of graft and patient survival outcomes adjusting for recipient risk factors did also not result in statistically significant differences between the two cohorts (SDC6).

Biochemical liver function on days 1–7 and MEAF score were only available for the prospective cohort with a mean MEAF score of 4.5 (SDC8).

A larger proportion of organs were retrieved in the prospective study cohort (NMP available) compared to the retrospective control cohort (SCS) (45.8% vs. 33.7%). The proportion of organs retrieved but not transplanted was also higher in the prospective study cohort (38.2% vs. 32.9%, respectively). Following adjustment for donor type and NRP, this yielded an adjusted odds ratio for organ non-utilisation of 1.33 (95% CI: 1.04–1.71). This was statistically significant at the 5% level (p = 0.0226) (SDC9).

There was one positive perfusate culture reported. Using data collected through the UKTR the proportion of recipients experiencing infections (41.0% vs. 22.9%, respectively), treated biopsy-proven rejection episodes (19.6% vs. 16.2%, respectively), and biliary complications requiring intervention (8.7% vs. 5.1%, respectively) in the first 3 months post-transplant was higher in the retrospective control cohort (SCS) than the prospective study cohort (NMP available). Slightly lower proportions were observed in the retrospective control cohort (SCS) for vascular complications, (9.0% vs. 12.7%) and reoperations (6.4% vs. 9.0%) at 3 months post-transplant. However, data completeness varied, so no formal statistical comparisons were performed (SDC5).

SAEs were only reported in the prospective study cohort (NMP available). There were 156 SAEs reported in 94 recipients (36.9% of the cohort) (SDC7). 62 SAEs were reported for cases where NMP was used, but only 1 was reported to be linked to errors in device function and use, and only 4 possibly related to the intervention.