The year 2008 marks the 25th anniversary of the National Institutes of Health Consensus Development Conference on liver transplantation (LTx), in which the life-saving procedure was recognized as an established means of restoring vital functions of the liver in patients who have sustained an irreversible hepatic injury.1 Since then, LTx has been widely accepted as a therapy that prolongs survival and restores health in the recipient. While many subsequent surgical and medical innovations led to dramatic improvement in the outcome of LTx, efficient, yet equitable means to allocate donated organs has become an important challenge to the transplant community.
Over the years, two schools of thought have emerged as a potential guiding principle in determining organ allocation priorities. The first approach, sometimes referred to as the urgency principle, advocates performing LTx in patients with the highest level of sickness, thereby rescuing them from certain death.2, 3 The other approach, the so-called utility principle, focuses on maximizing the outcome of LTx.4 If one follows the first principle, organs are to be allocated in the order of disease severity, realizing that the transplant outcome in the very sickest patients may not be as favorable as other patients with adequate physiologic reserve. In the latter approach, organs are used in patients who have the highest likelihood of achieving long-term survival in an optimum level of health, sometimes at the expense of forgoing patients who are desperately ill.
During the Clinton administration, the policy makers in the United States made an unequivocal decision to adopt the first, namely, the sickest-first, principle, in part as a response to the public demand for more transparent allocation system of the liver.5 Introduction of organ allocation on the basis of Model for End-Stage Liver Disease (MELD) score was the most noticeable event in the implementation of the policy. Despite the initial concerns that the sickest-first policy would lead to a dramatic decrease in post-LTx survival and/or massive increase in the economic cost, there is a wide consensus that the current policy has contributed to the reduction in wait-list mortality, while post-LTx outcome remains robust at a reasonable level of expenditure.
The MELD-based system in LTx appears to have energized policy-making bodies for other organ transplantation to utilize quantitative, objective mathematical measures to achieve equitable allocation. In kidney transplantation, a new policy proposal is being considered in which candidates will be ranked (in part) by the calculated benefit of transplanting a given organ. This calculation will consist of (1) estimation of survival with transplantation, (2) estimation of survival without transplantation (thus remaining on hemodialysis), and (3) subtraction of (2) from (1). This concept has been termed “life years from transplant”.6 Estimation of survival with and without transplantation can be made based on predictive models consisting of variables derived from the candidate and the donor.
A similar calculation has been performed in liver transplantation. Merion et al. examined wait-list and post-transplant mortality among 12,996 adult patients wait-listed in 2001 and 2003.7 The benefit of survival was calculated by comparing survival with and without LTx in several tiers of MELD. For example, for a patient with a MELD score of 40, the risk of death after undergoing LTx was 96% lower than that of continuing to wait. On the other hand, recipients who underwent LTx at a lower MELD score faced a much higher risk of death than those who remained on the wait-list. When the time-frame of 1 year was used in the analysis, the benefit of LTx was higher among patients with higher MELD, whereas patients with low MELD would be better off postponing the procedure.
These data raise the question whether the LTx allocation policy must be revised utilizing the benefit framework. Under a system in which the estimated benefit is used as the main criterion to determine priority, available organs will be allocated to the candidate with the largest expected benefit. Although this is a highly rational and admirable compromise between the urgency and utility principles, it does represent a significant paradigm shift. The ultimate result will be that some patients with the highest MELD score on the list will be bypassed. In a sense, the system will formularize futile transplantation and shifts the decision whether the patient on the top of the list is too sick to be transplanted from the hands of transplant physicians and surgeons to a numerical scoring system.
Whether benefit-based organ allocation in LTx may be successful will depend on the degree to which the estimation of the benefit may be made precisely and accurately. Precision of a prediction model evaluates whether the prediction can be made reproducibly (e.g., with a narrow confidence interval). For example, in predicting waitlist mortality, MELD has been shown to be sufficiently precise such that one can be confident that a patient with a higher MELD faces a higher risk of mortality than a counterpart with a lower score. For example, the concordance statistic for MELD in predicting waitlist mortality has been as high as 90%, indicating that 90% of the time, a patient with higher MELD dies before another with a lower MELD. In contrast, predicting the outcome following liver transplantation has been proven difficult, and the c-statistics reported in studies that attempted the prediction have been in the 60% range (as compared to 50% for a coin toss), even when all available information at the time of transplantation, including donor characteristics, was taken into account. Although there may still be predictors that have not been identified (e.g., genetics), random events in the perioperative period, which are by definition not predictable, significantly reduce the ability to confidently estimate post-LTx outcome. In order for the benefit model to be successful, the confidence with which post-LTx outcome can be predicted must improve.
Accuracy refers to the ability for a prediction model to provide correct estimates. A model may be precise but not accurate; for example, it was never asked of MELD as a metric for organ allocation to provide accurate survival probability. For the purpose of organ allocation, precision was all that was necessary in order that patients can be ranked in the correct order. In that context, it does not matter whether the estimated survival probability for a given MELD is exactly what would be observed or is systematically off by a substantial margin. In contrast, in the case of the benefit model, it is critical that the prediction of the mortality before and after transplant is accurate, so that the difference between the two may be calculated. Thus, the models to be used for the benefit estimation must meet a much higher standard than MELD was ever able to.
Regardless of the mathematical details of the benefit model, the difficulty to estimate post-transplant outcome precisely and accurately is confounded by the selection process inherent in the observational transplant data. For example, Merion et al. documented that the 1-year mortality in patients whose pre-LTx MELD was higher than 40 (calculated MELD up to 68) was only 8% higher than those with MELD between 30 and 39. The key fact to note here is that those patients in the highest tier of MELD were transplanted in spite of their high MELD. In other words, those patients represent a carefully selected group who were judged to have a reasonable chance to survive the procedure (pointing to the effectiveness, by and large, with which futile LTx has been avoided under the current system). It is highly unlikely that the same outcome will be replicated if all patients with MELD of 40 or higher were to be given a transplant. At the other end of the spectrum are patients who had LTx with a low MELD (e.g., <15). Some of those patients had reasons to undertake the risk of proceeding with LTx despite their low MELD score. These biases lead to underestimation of the risk (overestimation of the benefit) at the high end of MELD and overestimation of the risk (underestimation of the benefit) at the low end of MELD. The selection process embedded in the data cannot be disentangled by statistical adjustments and, thus, the resulting bias cannot be estimated.
Finally, compared to kidney transplantation, the stake is higher with LTx because of the lack of replacement therapy. The penalty of guessing wrong in LTx is, one way or the other, death of a candidate, compared to kidney patients who have a fall-back option of returning to dialysis. Further, outcome after kidney transplantation tends to be more predictable by pretransplant factors, such as diabetes and dialysis.
In summary, incorporating the expected benefit of LTx, in principle, represents a logical approach toward efficient and equitable use of donated organ to maximize the societal gain from the scarce resources. However, significant challenges exist in estimating benefits of transplantation and implementing prediction models for patient outcomes. Careful consideration of the benefits (and disadvantages) of the benefit model is requested.