Perfusion can preserve a donor liver and expose useful signals. Proving that those signals improve transplant decisions is a different experiment.
A donor liver can be connected to a machine, supplied with circulating fluid and observed while the transplant team prepares for what comes next. That creates an unusually persuasive image: a vulnerable organ, outside a body, apparently demonstrating what it can do. But the image compresses several different scientific questions. Has the machine prevented additional injury? Has it revealed damage that would otherwise have remained hidden? Has its information changed a decision? And did that changed decision help a recipient?
A randomized trial published online on 26 October 2026 makes the separation especially clear. Arianna Trizzino and colleagues compared normothermic with dual-hypothermic oxygenated ex-situ perfusion in 50 livers from donation after circulatory death, following normothermic regional perfusion. Their author abstract explicitly states that the ex-situ phase was not used for additional assessment. The small sample and limited events prevent definitive conclusions about the relative efficacy of the two strategies. It is a comparison of preservation approaches in a particular sequence, not a validation of a machine-based suitability test. This account is limited to the opened abstract, not the unavailable publisher full text. Trizzino and colleagues, randomized trial.
That distinction matters more than a competition between a warm machine and a cold one. Preservation and prediction can support each other, but neither logically proves the other. A machine might protect already accepted organs without identifying which additional organs should be accepted. A measurement might predict complications without being a useful reason to discard an organ. An apparently reassuring result might be reliable only among the selected organs that a particular team was willing to transplant.
The demanding question is therefore not whether an organ looks active. It is what claim a specific experiment permits, at what point in the transplant pathway, and for which decision. Following that question reveals why perfusion research needs clinical trials, mechanistic measurements and careful accounting of organs that never reach a recipient.
Four different jobs inside one attractive word
Conceptual research map: preservation, measurement, decision and recipient follow-up require distinct evidence. Not a clinical protocol or acceptance rule.
In this article, preservation means attempting to maintain an organ's condition before implantation. Assessment means observing characteristics that may inform a decision. Selection means acting on those observations and other information. Outcome evaluation means learning what happened afterward. These are working distinctions for reading the evidence, not four neatly isolated hospital departments. One perfusion episode can participate in all of them.
Normothermic systems support activity under warm conditions; hypothermic oxygenated approaches provide oxygen while metabolism is reduced by cooling. Neither label describes the entire intervention. Where perfusion begins, what came before it, how an organ was selected and what happens after it can all distinguish two studies that appear to compare the same broad technology. Regional perfusion occurs within the donor, whereas ex-situ perfusion occurs after organ removal. The sequence must remain visible when interpreting the October trial, rather than being flattened into a generic claim about machine perfusion.
A useful way to organize the questions is to imagine four separate verbs on a research form: protect, measure, decide, follow. Under protect, the comparison is between preservation strategies. Under measure, it is between a signal and a defined later outcome. Under decide, it is between policies that use information differently. Under follow, the issue is the completeness and duration of recipient observation. Writing the verbs separately prevents a positive result beside one from silently filling in the other three.
For example, a hypothetical study might show that a preservation method reduces an early laboratory abnormality. That does not establish that its monitoring display identifies every organ with a future bile-duct problem. Conversely, a test might identify a high-risk group accurately while offering no practical benefit if the same group was already recognized by other means. An informative signal and an effective decision rule are related, but they are not synonyms.
This separation is not a criticism of the technology. It specifies the kinds of success that can be demonstrated. A preservation improvement can be valuable without becoming a universal diagnostic platform. An assessment method can be worth investigating before it is ready to determine acceptance. The scientific task is to avoid requiring one result to carry claims it was not designed to test.
The point of randomization changes the question
The Bridge to HOPE trial, published online on 27 May 2026, randomized recipients only after the liver had been inspected and accepted. The trial compared portal-venous hypothermic oxygenated perfusion after cold storage with cold storage alone. Among 219 recipients, early allograft dysfunction occurred in 20.2% of the perfusion group and 37.3% of the comparison group. Its design provides evidence about preservation among accepted grafts; it cannot, by itself, establish how many previously rejected livers an assessment policy would recover. The study also notes limitations involving its early endpoint, secondary analyses and follow-up. Reich and colleagues, Bridge to HOPE.
Consider two hypothetical randomization points. In the first design, researchers randomize an organ before the acceptance decision, allowing the assigned pathway to influence whether it is used. In the second, they randomize only after acceptance. The first can examine a broader pathway that includes selection. The second more cleanly addresses what happens to the organs already chosen. Neither is automatically the better trial. They estimate different effects.
Now imagine both reports contain a table headed “transplant outcomes.” If the early-randomization study includes only transplanted recipients in that table, the groups reaching it may differ because the intervention changed selection. Removing the unused organs from view removes part of the intervention's story. In the late-randomization study, a low discard rate is unsurprising because the acceptance decision has largely occurred already. It should not be advertised as proof of extraordinary rescue performance.
This is why a study diagram can be more revealing than a headline percentage. The reader needs to know where the pathway branches: before retrieval, after inspection, after connection to a machine or immediately before implantation. The denominator at each branch identifies who remains eligible to contribute to the next result. A preservation benefit measured downstream can be real while leaving an upstream selection question unanswered.
For evidence synthesis, the practical consequence is to compare like questions before comparing effect sizes. Pooling studies solely because they contain the word “perfusion” risks mixing a treatment of accepted grafts with a policy for finding additional grafts. Statistical sophistication cannot repair an undefined question. The intervention's position in the pathway belongs in the argument from the beginning.
Injury, use and survival are different outcomes
The 2018 randomized trial by David Nasralla and colleagues reported lower hepatocellular injury with normothermic preservation, alongside lower organ discard and longer preservation time, compared with conventional cold storage. Its abstract did not establish a corresponding significant difference in graft survival, patient survival or bile-duct complications. Those are separate outcomes, not interchangeable descriptions of one result. The abstract is the evidence available for this account; detailed subgroup explanations would require the full paper. Nasralla and colleagues, randomized preservation trial.
A hypothetical reporting example shows the problem with treating them as one success measure. Suppose Pathway A uses 60 of 100 offered organs and 54 recipients meet a specified one-year outcome. Pathway B uses 75 of 100 offered organs and 66 recipients meet it. Among transplanted recipients, the outcome proportions are 90% and 88%. Per 100 offered organs, the counts are 54 and 66. Neither calculation is wrong. They answer different questions.
These invented numbers do not establish that Pathway B is clinically preferable. The recipients might differ, the adverse outcomes might differ in severity, and patients who did not receive these organs might later receive others. The example instead shows why a slightly lower recipient-level percentage could coexist with more recipients benefiting across the offered-organ pathway. Equally, a higher utilization rate could coexist with unacceptable harm. Utilization cannot settle the evaluation alone.
An early biochemical endpoint introduces another denominator: measurements taken during a defined early period in recipients who reach that period. A survival endpoint asks about a different event over a different interval. A complication endpoint may count affected people, individual procedures or recurrent episodes. Without definitions, the same word “reduction” can hide substantially different observations.
There is also a difference between failing to demonstrate a survival difference and demonstrating identical survival. If a study is designed around a more frequent early endpoint, it may have too few deaths or graft losses to resolve a modest difference in those events. That uncertainty should not erase a demonstrated early benefit, but neither should the early benefit be promoted into an unmeasured long-term guarantee. The most useful reporting keeps each endpoint attached to its actual horizon.
A liver is not a single performance channel
A liver's metabolic activity and the condition of its bile ducts cannot be reduced to one shared gauge. In a 2018 observational perfusion series, Christopher Watson and colleagues examined different indicators relevant to hepatocellular and biliary function. The work is exploratory clinical evidence, not a universal pass–fail specification. Its importance here is the attempt to distinguish components of organ performance rather than treating visible activity as comprehensive proof of suitability. The institutional repository PDF was accessible; alternate routes were not. Watson and colleagues, observations on ex-situ liver perfusion.
The distinction can be understood without inventing a clinical threshold. Imagine an assessment with two channels: one summarizes a metabolic process, the other a different vulnerable tissue function. If the first improves while the second remains uncertain, averaging them into a single reassuring display does not resolve the uncertainty. It hides the disagreement. A composite score needs evidence that its combination rule predicts a meaningful outcome, not merely an attractive interface.
The choice of outcome determines which disagreement matters. A score optimized for an early aggregate measure could perform poorly for an uncommon later complication. A signal that contributes little to the aggregate prediction might nevertheless be important for that particular complication. This is a general property of prediction problems with multiple targets; it is not evidence that any named perfusion marker has a specific undiscovered power.
One proposed test may therefore need several separate validation questions. Can it anticipate the early functional outcome it claims to predict? Does it add useful information about a later complication? How often does it fail to produce an interpretable reading? And does the relationship persist when the workflow changes? Treating missing or ambiguous readings as if they were reassuring results would make the instrument appear more decisive than it is.
The appropriate visual metaphor is not a green light attached to an organ. It is a set of partial observations connected to different outcome questions. Some relationships may be well supported; others may be exploratory. Their uncertainty should remain visible rather than disappearing behind the singular word “viable.” In this context, a more complicated answer can be more clinically honest than a cleaner dashboard.
A convincing signal can still be a poor solitary judge
Watson and colleagues' 2022 study analyzed perfusate variables from 203 livers, of which 154 were transplanted. Several variables were associated with early allograft function in their analysis. The authors cautioned against using lactate alone as a viability indicator, and reported nonanastomotic biliary strictures despite selection incorporating bile measurements. The study supports investigating combinations of observations while preserving the limits of the selected clinical cohort. It does not authorize a general acceptance rule from one laboratory value. Watson and colleagues, predicting early allograft function.
There are two distinct reasons a solitary marker can disappoint. It may capture only one aspect of the target condition. Or it may appear informative in a development dataset but add little once the information already used by clinicians is considered. The first is a coverage problem; the second is an incremental-information problem. A study should specify which it is testing rather than presenting a strong association as a complete answer.
Suppose, purely hypothetically, that a proposed score orders ten organs from apparently lowest to highest risk. The order can look impressive even if every risk estimate is systematically too optimistic. Ranking and calibration are different properties. Saying which organ has the higher estimated risk does not establish whether that risk is 5%, 15% or 40%. Decisions involving rare adverse outcomes may be particularly sensitive to that distinction.
Now suppose a second measurement changes the order of only two organs. It might seem unimportant in an overall performance statistic, yet those two changes could occur exactly where a decision policy is uncertain. Alternatively, the added measurement might improve a global statistic without changing any decision. Neither conclusion can be read directly from a correlation coefficient. The relevant comparison is the complete assessment process with and without the additional information.
These examples are not proposed clinical rules. They explain why validation requires a declared intended use. A marker for research stratification, a marker for identifying organs that need further assessment and a marker used to support acceptance have different consequences. The evidence standard should follow the consequence of being wrong, not the visual neatness of the measurement.
Cold perfusion is not information-free
It would be a mistake to assign preservation to cold systems and assessment exclusively to warm ones. An international study led by Eden examined perfusate flavin mononucleotide, or FMN, during hypothermic oxygenated perfusion. The accessible primary-paper PDF reports a multicentre validation effort relating this mitochondrial-associated signal to later outcomes. That is evidence about a candidate measurement under cold-perfusion conditions, not a randomized demonstration that an FMN-guided acceptance policy improves outcomes. Eden and colleagues, international validation study.
A separate study by Dingfelder and colleagues, published online in October 2024 and appearing in a 2025 issue, investigated FMN measurements in 50 grafts. It connects a mechanistic interest in mitochondrial injury with clinical associations. Its background describes how interrupted oxygen supply and subsequent reoxygenation can disturb cellular energy processes. The clinical associations remain different evidence from a trial assigning decisions according to the marker. Neither study should be converted into an assurance that a particular reading makes transplantation risk-free. Dingfelder and colleagues, FMN validation.
For a candidate assay, at least three relationships have to be kept separate: what biological process contributes to the signal, how reliably the instrument measures it, and what the measured value predicts. A compelling biological explanation cannot substitute for measurement reliability. A reliable instrument cannot establish that the chosen biological process captures everything relevant to a recipient's outcome.
Consider a hypothetical comparison between two laboratories testing the same sample. If their readings differ, an apparent difference between organs could partly be a laboratory difference. Agreement studies would address that problem. If the laboratories agree but their readings predict outcomes differently across centres, the question moves to populations, workflows or other contextual differences. These are not interchangeable failures, and they call for different investigations.
A multicentre study is useful partly because it creates opportunities for such variation to become visible. It does not automatically make a marker transportable to every setting. The stronger claim requires showing what remained stable across centres and what needed adjustment. That is an argument for examining how a measurement travels, rather than treating the number of participating hospitals as a guarantee of universal applicability.
The missing outcomes of organs not transplanted
Assessment research faces a distinctive missing-data problem. For an organ that is not transplanted, the recipient outcome under transplantation is unobserved. That is not the same as an ordinary missing laboratory result. The very decision being studied can prevent the outcome of interest from being observed.
A hypothetical dataset makes the asymmetry concrete. Imagine 100 assessed organs. A team accepts 70 and declines 30. Outcomes are available for the 70 transplanted organs, and a model predicts those outcomes well. It does not follow that the same model has been validated for the 30 declined organs. Their measurements may occupy a different range, and their outcomes under transplantation remain unknown. Adding the word “all” to the model's intended population would extend it beyond the observed evidence.
Nor can the declined organs simply be counted as failures. Some might have functioned adequately, but that counterfactual was never observed. Counting every decline as a biological failure would reward agreement with existing decisions, whether or not those decisions were optimal. Counting every decline as a lost opportunity would make the opposite unsupported assumption.
This creates a useful distinction between predicting what clinicians decided and predicting what would have happened after transplantation. A model can be excellent at the first while providing little new information about the second. If the training label is acceptance, it may learn institutional habits, available resources or established caution as much as organ biology. That may describe practice, but it should not be presented as an independently validated measure of suitability.
These are original methodological deductions from the structure of the decision problem, not reported findings about the performance of a named clinical model. They explain why prospective assessment studies need transparent pathways for eligibility, measurement, acceptance and follow-up. The unanswered outcomes cannot be conjured into existence by a more complex algorithm. Researchers can narrow uncertainty through carefully designed studies, but they must first identify which part of the population their observed outcomes actually represent.
Rescue evidence needs a counterfactual, not a slogan
The VITTAL study, published in 2020, investigated normothermic assessment of livers that had been declined for transplantation. It was a prospective single-arm study, not a randomized trial. Its results demonstrate clinical use of selected organs within that assessment pathway, while early outcomes and later biliary complications require separate interpretation. The evidence is important without implying that every previously declined liver can be recovered or that perfusion alone explains every outcome. Mergental and colleagues, VITTAL.
The VITTAL report provides a concrete reason to retain both clocks. Twenty-two of 31 assessed livers were transplanted, with 100% patient and graft survival at 90 days. During a median follow-up of 542 days, four recipients developed biliary strictures requiring retransplantation. Reporting only the early survival figure would omit a consequential later part of the same study. VITTAL follow-up.
The word “rescue” can refer to several different counterfactuals. An organ might have been discarded under the previous local policy. It might have been accepted elsewhere. It might have been accepted by the same team if different information or logistics had been available. Those possibilities do not make the transplant less consequential, but they change what the technology can be credited with accomplishing.
Imagine a hypothetical programme that reports 20 additional transplants after introducing perfusion. During the same period, it also expands its retrieval team, changes referral arrangements and becomes willing to consider a different donor population. The count of 20 is a valid descriptive observation if correctly recorded. Attributing all 20 to the biological action of perfusion would require an additional argument. The machine, the information it supplies and the organizational changes may be inseparable parts of the implemented programme.
A pathway-level evaluation can acknowledge that package rather than pretending only the machine changed. It can ask whether the whole programme produced better results for the relevant population. A mechanistic evaluation asks a narrower question about preservation or measurement. Both are legitimate, but they should not borrow each other's causal certainty.
The most informative rescue report would also make unsuccessful assessments visible. If readers see only transplanted cases, they cannot determine how many organs were evaluated, why others were declined or how often the assessment process failed to resolve uncertainty. Reporting the full pathway protects the meaning of the successful cases. It turns an impressive series into evidence that another programme can interrogate rather than merely admire.
The machine can change the thing it measures
There is another conceptual complication: perfusion is not simply a camera pointed at a static organ. It is an intervention occurring while measurements are taken. An observed trajectory can reflect the organ's starting condition, its response to the intervention and features of the measurement environment. Separating those contributions is a scientific question, not something the display answers automatically.
This can be illustrated without assigning any clinical parameter. Suppose two hypothetical organs begin with different values of a measured signal. After an intervention, their values converge. One interpretation is that the intervention reduced a meaningful biological difference. Another is that the measurement became less sensitive to that difference under the new conditions. A third is that both processes occurred. The trajectory alone does not choose among them.
Likewise, a declining concentration is not automatically identical to a declining total amount. In a deliberately abstract arithmetic example, ten units of a substance in one litre produce ten units per litre. The same ten units in two litres produce five units per litre. That calculation is not an explanation of any particular trial result, and it does not imply that clinical teams ignore fluid accounting. It demonstrates why a measured concentration must be interpreted with the system in which it was measured.
A useful research record therefore distinguishes the observation from the interpretation. “The signal changed” is an observation. “The organ repaired itself” is a mechanistic explanation. “The recipient will do well” is a prediction. “The organ should be accepted” is a decision. Each step requires evidence that the previous step does not automatically contain.
This matters when comparing early and late measurements. A later reading may be more informative because it captures a response, but it may also be conditional on the organ remaining in the assessment process long enough to produce it. Excluding early stops from analysis can alter the population being described. Time is not merely another column in the data; it can determine which observations become available and which cases disappear from the comparison.
A clinical endpoint can answer a narrower question well
The DHOPE-DCD randomized trial, published online in February 2021, tested hypothermic oxygenated machine perfusion in livers from donation after circulatory death. It used a clinically specific biliary outcome rather than relying only on a general impression of organ activity. Its author abstract reports fewer nonanastomotic biliary strictures in the perfusion group. That result belongs to the tested donor population, intervention and follow-up; it is not a finding about every organ or every perfusion arrangement. van Rijn and colleagues, DHOPE-DCD trial.
The primary horizon in DHOPE-DCD was six months. The abstract reports strictures in 6% of machine-perfusion recipients and 18% of controls, with 78 transplanted recipients in each group. The reported risk-ratio confidence interval was 0.14 to 0.94. Those details identify both the observed contrast and its uncertainty; the result is not a lifetime estimate. DHOPE-DCD endpoint and results.
The methodological lesson is that a narrow outcome can be more revealing than an expansive slogan. “Better organ quality” is difficult to evaluate unless quality is defined. A specified complication, assessed over a specified interval, gives the comparison a target. It also makes the residual uncertainty clearer: a benefit for that complication does not settle every other outcome.
Suppose an imaginary trial reduces one complication but leaves another unchanged. A composite endpoint combining them can dilute the apparent improvement. Alternatively, a composite dominated by a frequent, less severe event can appear favourable even when a rare, severe event is unresolved. This does not make composite endpoints inherently misleading. It means their components deserve inspection, especially when readers are tempted to translate the result into an undifferentiated claim of safety.
Recurrent events add another choice. Ten procedures in two recipients and one procedure each in ten recipients produce the same procedure count but a different distribution of burden. Counting affected recipients answers a different question from counting interventions. A comprehensive interpretation might need both, together with the severity and timing of the events. The relevant measure depends on what aspect of benefit the study set out to establish.
For readers, the discipline is straightforward: keep the noun after the percentage. A reduction in a named complication is a reduction in that complication, within the study's conditions. It need not be inflated into a universal outcome to be valuable. Precision preserves the achievement instead of making it vulnerable to a claim the research never supported.
A transportable machine is also a workflow
The OCS Liver PROTECT randomized trial, published online in January 2022, evaluated portable normothermic blood-based perfusion beginning at the donor hospital. Its clinical outcomes and utilization analyses concern that device-supported pathway, not merely a temperature category. The trial is useful evidence for a particular intervention package; it does not establish that every arrangement called normothermic perfusion produces identical results. Markmann and colleagues, OCS Liver PROTECT.
A workflow can change what is possible even when its components are familiar. In a hypothetical comparison, one programme begins its intervention earlier and moves the organ while it is supported; another introduces an intervention only after transport. If outcomes differ, the comparison includes timing and organization as well as the technology itself. Calling both programmes “machine perfusion” conceals precisely the distinctions that might explain the difference.
An implementation study should therefore make its clock explicit. Relevant intervals might include the period before connection, the intervention period and the transition to implantation. This article does not propose acceptable limits for any interval. The point is analytical: a total duration can hide very different sequences. Two pathways lasting the same number of hours need not expose an organ to the same conditions.
Resource accounting needs the same specificity. A device cost is not the entire cost of a programme, just as an avoided hospital day is not automatically a cash saving of the same amount everywhere. Staffing, transport, laboratory support and the consequences of complications belong to different budgets. A claim of economic benefit requires an actual evaluation in the relevant setting, rather than multiplying a clinical result by an assumed price.
There is also a practical distinction between what a trial can deliver with dedicated support and what an adopting centre can reproduce. That is not a reason to discount a controlled trial. It is the reason implementation evidence has a separate job. The trial estimates an effect under its conditions; subsequent work tests how the pathway behaves when those conditions vary. Neither form of evidence replaces the other.
What a registry can reveal, and what it cannot recover
A National Organ Perfusion Registry report published online on 31 August 2026 described 545 normothermic-perfusion liver transplants at nine US centres during 2022–2024. It recorded variation in procurement and perfusion practices that broad transplant datasets may not capture. This was a descriptive recipient cohort, not randomized evidence comparing every offered organ with an alternative pathway. Its calendar is also important: a 2026 publication reports earlier clinical activity, not 545 procedures newly performed that year. Cortez and colleagues, National Organ Perfusion Registry.
A registry's strength is not that it recreates randomization after the fact. It can expose variation, reveal uncommon events, document implementation and generate questions that a smaller controlled study could not answer. But a recipient registry begins downstream of the decision to transplant. Unless linked to additional records, it does not contain the full population of assessed and declined organs.
Consider a hypothetical centre comparison. Centre A uses perfusion for relatively straightforward cases, while Centre B reserves it for difficult cases. A crude comparison could make Centre B look worse even if its pathway is effective. Reversing the patterns could make the technology look better for reasons unrelated to its effect. Adjustment can address recorded differences, but unrecorded reasons for selection remain a limitation.
An especially valuable addition would be the decision context: why perfusion was chosen, what information was available before it, whether the decision changed and which alternatives were feasible. These fields would not magically remove confounding. They would make the causal question more explicit and help identify comparisons that are plausible rather than merely convenient.
Longitudinal consistency also matters. If a programme changes how it defines an event or captures follow-up, an apparent trend can combine clinical change with recording change. A registry should preserve enough version information to distinguish them. The goal is not administrative perfection for its own sake. It is to prevent a measurement system from quietly becoming a moving target while its output is interpreted as a biological trend.
Why the kidney comparison belongs in a liver story
A randomized kidney-transplant trial by Slagter and colleagues was published online on 24 December 2025, ahead of its May 2026 issue. In 80 recipients, the tested addition of normothermic perfusion to an existing hypothermic-perfusion pathway did not improve the prespecified early outcome. The finding concerns that addition, comparator and population. It is neither evidence that all kidney perfusion is ineffective nor a reason to transfer a liver result to kidneys. Slagter and colleagues, randomized kidney trial.
Here the primary endpoint was immediate graft function: 16 of 41 recipients in the added-perfusion group and 19 of 39 in the comparison group met it. The reported odds-ratio confidence interval extended from 0.28 to 1.64. This is not a precise estimate of no difference; it is a result with substantial uncertainty around the tested contrast. Slagter trial results.
The comparison guards against a category error. A technology family can include interventions with different mechanisms, starting points and clinical targets. If a beneficial component is already present in the comparison group, a trial of an added component asks about its incremental contribution. That is a different experiment from comparing the entire combined pathway with no such component.
A simple hypothetical analogy within trial design makes this precise. Let Pathway A contain steps X and Y, and Pathway B contain X alone. The randomized contrast estimates the added effect of Y in the presence of X. It does not estimate the effect of X, and it does not directly estimate Y used alone. If the result is inconclusive, neither “the whole pathway works” nor “the whole pathway fails” follows automatically.
Organ-specific outcomes add another boundary. A common engineering vocabulary does not create a common clinical endpoint. Even where two organs are supported by circulating fluid, their relevant functional demands and complications must be evaluated on their own terms. The appropriate level of generalization is therefore earned, not assumed from the shared presence of pumps and tubing.
For research reporting, this suggests a more useful sentence pattern: the tested strategy, compared with the stated alternative, changed or did not establish a change in the named outcome for the enrolled population. That sentence is less spectacular than “perfusion works.” It also contains the information another team needs to decide whether the evidence addresses its question.
The next decisive experiment is about a decision
Preservation studies and assessment studies have already produced important evidence. The remaining challenge is not to replace them with one enormous trial, but to connect their claims carefully. If a new marker is proposed specifically to change acceptance decisions, the decisive evaluation must eventually examine a policy that uses it, not only its association with outcomes among organs selected by existing practice.
A hypothetical policy study would begin by declaring the population to which the policy applies and the alternative decision process. It would record how often the new information changes a decision, which organs are transplanted and what happens to recipients. It would also retain the cases where the result is missing, ambiguous or ignored. Excluding those cases would describe an idealized policy rather than the one actually implemented.
A further distinction concerns the person waiting, rather than the organ being evaluated. In a hypothetical decision analysis, declining an offered organ does not place that person into an outcome-free category. They may remain waiting, receive another transplant or experience a change in health. A study confined to recipients of the evaluated organs cannot, on its own, estimate the complete effect of the policy on everyone to whom those organs could have been offered.
That observation does not justify accepting an organ with an uncertain prognosis. It identifies the alternative that a decision analysis must represent. “Transplant now” and “continue through the waiting pathway” are not the same comparison as “transplant a perfused organ” and “transplant a conventionally preserved organ.” A useful evaluation specifies which comparison it has actually observed and which additional assumptions would be needed to connect it to the broader allocation question.
The outcome need not be reduced to a single winner–loser score. Recipient survival, complications, organ use and the burden of assessment can be reported together. Their trade-offs require explicit interpretation. A programme that uses more organs but changes the distribution of complications should not be summarized by whichever denominator makes its headline strongest.
This is a proposal about evidence architecture, not a recommendation for a clinical acceptance protocol. The studies discussed here do not license readers to judge an individual organ or choose a treatment. They show why such decisions require specialist teams and why apparently intuitive claims about suitability still need well-designed evaluation.
The organ on the machine is doing something scientifically valuable: making preservation and observation possible during a period that would otherwise offer less information. What it is not doing is narrating its own future. The future belongs to the recipient, the whole transplant pathway and outcomes that must actually be observed. Perfusion earns confidence when those connections are demonstrated—not when the vividness of the machinery makes them seem unnecessary.