Orthotopic liver transplant (OLT) is the main therapeutic approach for patients with a severe liver disease but insofar as clinical results of transplant have steadily improved, a shortage of donors has emerged as a serious problem, evidencing the necessity for the best possible donor-recipient matching. With a rising demand for organs and a steady improvement in immunosuppression strategies, more and more marginal donors and recipients will be transplanted,1 thus the number of dysfunctional allografts might increase; therefore clear criteria to set limits of reasonable acceptance of marginal donors and recipients is paramount to achieve the best performance in OLT policies. In this scenario, identifying those risk factors that signal early liver allograft dysfunction (EAD) can give us some indications for the best donor-recipient matching according to their outcome, but would also allow us a prompt diagnosis and apply early measures that could diminish the intensity or even arrest the development of EAD.2–6

Defining EAD after OLT is a difficult task due lack of consensus among researchers.6–9 Almost all criteria proposed to date rely on the determination of widely used markers of hepatic function, such as transaminases, coagulation factors, bilirubin, ammonia and/or lactate,10 but there is a lack of agreement about which ones should be applied. Moreover, some criteria also include less common parameters, such as the rate of elimination of molecules cleared by the liver (e.g., indocyanine green or C-Methacetin).11–13 Different classifications define diverse grades of dysfunction (some as dichotomous,12,14–25 some as progressively graded4,26–30 and finally a few as a numeric value31–33) based on different intervals of these parameters.

In a search of the literature using PubMed regarding liver allograft dysfunction criteria, we retrieved 38 definitions since 1985, most of them including some but not all of the aforementioned parameters, besides the presence of overt clinical evidence of primary non-function. This lack of agreement explains why, depending on the criterion selected, EAD after OLT is reported to appear with a broad range (8–29%), and the same applies to primary non-function (1–7%).7

Our aim was to evaluate the capability of some of these criteria, specifically those developed by Makowka et al.,26 Ardite et al.,27 Nanashima et al.,23 Dhillon et al.29 and Pareja et al.,33 for detecting EAD using as comparator the United Network for Organ Sharing (UNOS) criterion for primary non-function34 and their relationship with outcome (hospital mortality and 2-year survival) in our population.

During the study, 272 patients were admitted to our unit after OLT, of them, 19 showed exclusion criteria leaving us 253 patients for analysis. Main characteristics are presented in Table 2.

Mortality in our series was 17.4% (44 patients): 12 (4.7%) in the early postoperative course at the ICU, 7 (2.8%) in the hospital after ICU discharge and 25 (9.9%) during 2-year follow-up after hospital discharge. Cause for postoperative (in-hospital) mortality was acute rejection in 1 patient, primary non-function in 4, vascular complications in 4 and multi-organ failure after a complicated ICU course in 10 (one secondary to infection). During the two-year follow-up period, causes of mortality were motivated by graft problems in 14 cases: rejection in 2 (8%) patients, vascular complications in 5 (20%), a recurring disease in 6 (4%) and biliary complication in 1 (4%). None of the patients with past medical history of hepatocellular carcinoma died.

According to UNOS criterion, we diagnosed 35 cases of severe EAD (13.8%), but by Makowka they would have been 16 (6.3%), by Ardite 27 (10.7%), by Nanashima 52 (20.6%), by Dhillon 78 (30.8%), and by MEAF criterion 34 cases (13.4%). In order to assess the concordance between those criteria and UNOS, we calculated Cohen's Kappa statistic, that showed only fair concordance for all of them: Nanashima 0.55 (0.39–0.68), Ardite 0.49 (0.47–0.50), Makowka 0.46 (04.29–0.6), Dhillon 0.44 (0.32–0.56) and MEAF 0.41 (0.25–0.58).

Regarding the univariate analysis, the relationship between EAD diagnosed by each criterion and ICU length of stay, hospital length of stay and mortality are depicted in Table 3.

We next performed a multivariate logistic regression and a ROC curve analysis for each one of the EAD diagnostic criteria looking for an independent relationship between these and ICU or Hospital mortality, and the results are presented in Table 4. This table shows only the OR for each criterion analyzed but in the final step of all the models age, gender, comorbidities, etiology of end-stage liver disease and MELD were excluded and only chronic kidney disease, APACHE II at admission; acute kidney injury during the postoperative period and previous OLT remained in he model. Goodness-of-fit was tested with the Hosmer–Lemeshow test (p 0.43). Finally, we analyzed 2-year survival, and the results are presented in Table 4 and Fig. 2. Table 4 shows only the HR for each criterion analyzed but in the final step of the Cox regression analysis for each model, only etiology of end-stage liver disease and acute kidney injury during the postoperative period remained.

Figure 2.

Impact of early graft dysfunction on two-year survival. One Kaplan–Meier curve is presented for each diagnostic method applied for early graft dysfunction. EAD=early allograft dysfunction. Comparisons by Log-rank test, HR stands for Hazard ratio (95% confidence interval).

(0.34MB).

We designed the present study to assess the performance of various criteria to diagnose EAD and their relationship with outcome. Diagnosing EAD is a challenge because of a large number of criteria proposed, as a consequence, its epidemiology is not well characterized yet. To further complicate this scenario, studies comparing EAD criteria are scarce and whilst most of them intend to validate new sets of variables, almost none has established comparisons among existing criteria, and no study has encompassed them all.

Incidence of EAD reported by several research groups differs from ours: primary non-function by UNOS criterion was 13.8% in our population; Makowka26 reported poor function in 17.5% against 6.3% in our population, and Ardite27 reported a 19% of severe initial graft injury, whereas in our series just 10.7% were detected. Closer results were found for Nanashima23 (reporting 18.3% “initial poor graft function” against ours 20.6%) or Dhillon29 (26% of “initial poor graft function” against ours 30.8%). Our results disclose a huge variability in EAD diagnosis depending on the criteria used, as it is also pointed out in a recent review by Chen.7

Several factors could account for these differences: the fact that they were single-center studies or with a retrospective design, or perhaps the advances in the care for OLT, as the incidence of EAD and other complications seems to be decreasing despite marginal donor use.37

In our current settings of a limited number of donors, the appearance of EAD after OLT implies a double problem: a complicated postoperative course with risk for the receptor, and the possibility that in a different recipient this complication would not have developed.38 Thus, it is necessary to accurately define risk factors for EAD but this is difficult due to lack of consensus among researchers about how to define it.6–8 Obviously, this lack of consensus is detrimental to the design of research strategies that could help clinicians to decide and treat these patients, hampering, in the end, our capability to improve the transplant process.

Our main challenge when comparing these diagnostic sets was how to measure their performance. As definite criteria – either analytical or pathological – are lacking, we decided to analyze their capability to predict early and late outcome (e.g., length of stay or mortality) insofar as there is a clear relationship between EAD and poor outcome. This relationship has been confirmed in our population, where we found an increased length of stay and mortality for those patients developing EAD, irrespective of the diagnostic set used (Table 3).

According to our results, the UNOS, MEAF, Makowka, Ardite, and Nanashima criteria showed a good performance in predicting ICU mortality and all of them performed well predicting ICU length of stay. This was expected as most of these criteria were developed for the first postoperative week, despite the poor concordance shown by weak Kappa values, which points to a different set of patients detected by each criterion. Nevertheless, the capability to discriminate ICU mortality was poor according to ROC curve for all the criteria.

When addressing in-hospital mortality, all the criteria except MEAF showed a poor relationship. In our cohorts, MEAF performed well for in-hospital mortality, in agreement with the results reported in the original publication and in an external validation cohort.33,36

Among all the criteria analyzed in this study, the UNOS, Makowka, and MEAF criteria showed the better relationship with an early and late outcome and among them the MEAF score outperformed. Similar results have been reported from one study comparing the MEAF score with the widely used criteria proposed by Olthoff et al.39

We must acknowledge some limitations of our study. As in all previous studies addressing EAD, criteria were based on clinical and laboratory parameters that, in some way or another, mark a group of patients with a poorer outcome, but corresponding pathological findings (that could definitely ascertain that EAD is responsible for the clinical picture) are lacking. Despite this, taking into account different criteria already used in clinical practice and comparing their performance, even when lacking definitive confirmation of EAD, can be of aid in defining which of these tools behave better in our population to predict morbidity and mortality.

In addition, a limitation of this study is the single-center setting, which restricts the external validity of our results. Being a study about diagnostic criteria this aspect is critical, but the OLT patient and the transplant process are well characterized, and our team has large experience in its management. Criteria for graft allocation, patient selection, surgical approach and immunosuppressive regimes are widely agreed and there is scarce variability among centers, so our data can be reasonably extrapolated to other groups. Moreover, comparison of these criteria in different populations helps to precise the validity of these tools.

Part of the problem when studying EAD is the relatively low incidence, making desirable multicenter studies with larger and more varied populations. Also, in order to make sound comparisons, we included as EAD only the severe cases for Makowka, Ardite, Dhillon, and MEAF criteria; if we had also included patients with mild dysfunction, comparisons would not have been clinically consistent, and results would have been even poorer with those criteria. Taken these facts into account, we believe that our results are fair and mark the path for new studies on this topic.

Another possible drawback is that one group of patients was excluded from the study, namely acute liver failure. Our decision was motivated by the possibility of a different behavior of markers of cytolysis. Taking into account that all the criteria included in the study rely heavily on transaminases levels, inclusion of those patients would have potentially biased the results as they present with a serious derangement of liver function parameters before surgery and a different profile of cytolysis markers in the first hours after the transplant can be expected.35 Thus, we believe that the exclusion of these group assured a more homogeneous profile within our population and strengthened the results.

Still other aspect that can be challenged in our protocol is the inclusion of a score that was published in the last stages of patient's recruitment, namely the MEAF criterion. Even when this decision breaks the integrity of the “prospective” condition of the study, we opted for this strategy because, based on the first reports published, this was a promising diagnostic tool that needed external validation, and the set of parameters that compound the criterion had been prospectively registered in all our patients. Moreover, all the variables included in this criterion are routinely used and the laboratory procedures are standardized and have not suffered significant variations during the period of study. Thus, we assumed that this approach did not compromise the validity of our conclusion and could even add usefulness to the study, as it helped to evaluate this new diagnostic tool.

In summary, when comparing different severe EAD criteria against UNOS criterion we found that all show poor in-between concordance. MEAF criterion showed in our population the best performance to predict short and long term mortality and if these results are confirmed in wider population studies, this could be a useful tool for the characterization of EAD after OLT.

All author participated in writing and revising the paper. H.G.M.E, S.P.G, and B.F.J.E contributed to design, the performance of research, data collection, and analysis. A.V.M.D and L.S.R contributed to the performance of research and data collection.

This work has not received external funding.

None of the authors declare any kind of conflict with the contents of the present manuscript.