Usefulness of Pancreas Donor Risk Index and Pre-Procurement Pancreas Allocation Suitability Score: Results of the Polish National Study

Background

Transplantation of suboptimal pancreatic grafts often result in severe transplant pancreatitis and thrombosis, which are life-threatening complications that affect both graft and recipient survival. Hence, reliable assessment of a donor and quality of the harvested pancreas is of utmost importance. Histopathology of the graft seems an obvious solution, but steatosis and fibrosis do not correlate with donor age or BMI. Fibrosis is frequent in donors <40 years old [1]. Measurement of tissue adenosine triphosphate metabolites with magnetic resonance spectroscopy prior to transplantation is reliable but too cumbersome to be applied in clinical practice [2]. Recently, measurement of donor hemoglobin A1c levels has been used as a prognostic factor of pancreatic graft survival, but it was not added to the standard criteria [3].

To minimize the risk of complications, in 2008, Eurotransplant introduced use of the pre-procurement pancreas allocation suitability score (P-PASS) as a reliable indicator of graft quality. The model was constructed from donor age and BMI, length of ICU stay, duration of cardiac arrest, serum sodium and amylase concentrations, and catecholamine requirement. P-PASS scores over 16 are considered high risk [4]. However, subsequent studies showed an increased early complication rate but no difference in graft survival with higher P-PASS [5,6]. To overcome the problem of subjective criteria of pancreas donor assessment, in 2010 UNOS introduced a pancreas donor risk index (PDRI), calculated from a huge database of 9400 transplants [7]. The index was constructed from donor sex, age, BMI, cause of death, serum creatinine, donation after cardiac death status, and cold ischemia time. A Dutch study validated both indices in a population of 350 pancreas transplant recipients. P-PASS had no predictive value, while a PDRI over 1.24 was associated with reduced graft survival both in univariate and multivariate analysis [8]. A large UK retrospective analysis showed PDRI to be a relevant predictor for simultaneous pancreas-kidney transplantation (SPKTx) but not for pancreas transplantation alone (PTA) or pancreas after kidney (PAK) transplantation modality [9]. However, a Brazilian study [10] and a Spanish study [11] failed to confirm the usefulness of PDRI. This raises the question of reliability and repeatability of these results in other populations. To validate both indices for the Polish population, we analyzed available data of pancreata transplanted in all 4 centers across the country since 1998.

Material and Methods

STATISTICAL ANALYSIS:

Patient and graft survival were calculated according to Kaplan-Meier method (Statistica 12). To identify risk factors for pancreas graft loss within 12 months after transplantation, Cox regression analysis was performed and hazard ratios were calculated. Nonlinear regression models were constructed, probability of 12-month survival was calculated, and receiver operating characteristic (ROC) curves were drawn.

Results

Death-uncensored pancreas graft survival was 66%, 55.6% and 44.4%, at 1, 5, and 10 years, respectively. The lowest 12-month graft survival (54%) was observed in the PAK/PTA group. During the first month, 16.9% of pancreatic grafts are lost due to early complications, including early patient mortality (2.5%). Pancreas transplants that survived over 12 months came from younger donors (27.7 vs. 32 years, p<0.001), with lower BMI (23.8 vs. 24.5 kg/m2, p<0.04) and with lower P-PASS (15.7 vs. 16.4 points, p<0.03). PDRI was also lower in this group (0.944 vs. 0.992 points), but the difference was not significant (p=0.3). Cox hazard ratios were 1.039 (CI: 1.022–1.056) for donor age, 1.057 (CI: 1.005–1.111) for donor BMI, and 1.082 (CI: 1.015–1.154) for P-PASS. The confidence interval for PDRI hazard ratio was 0.798–2.158, and thus was not reliable as a prognostic factor of pancreatic graft survival. Pancreas 1-year graft survival according to P-PASS group is shown in Figure 1. Neither of the other donor-dependent factors succeeded in differentiating between surviving and non-surviving grafts.

When logistic regression models were used to calculate survival probability and ROC curves were drawn, the area under the curve (AUC) was 0.566 when P-PASS was used in estimation and 0.524 when PDRI was used. However, a simple model constructed from donor age and BMI resulted in an AUC of 0.611. Probability of death-uncensored 12-month graft survival could be calculated with Z=2.6302–0.0449×[DONOR AGE] – 0.0292×[DONOR BMI]. Graft survival estimation ROC curves of P-PASS, donor age/BMI, and PDRI prediction models are shown in Figure 2.

Discussion

Studies appraising PDRI admit its C-statistic hardly exceeds 0.5, and thus is useless in predicting graft function. It should be noted that donor age and BMI are redundant in the original PDRI equation [7], and BMI coefficient has a negative value while BMI ≥25 kg/m2 coefficient is positive, which is inconsistent. In our study, the ROC AUC for PDRI was insignificant and almost equal to 0.5. A study by Amaral et al. did not confirm correlation of PDRI with pancreas graft survival, but the quality of the study was questionable because the percentage of incomplete records that were excluded from analysis was 73% [10]. Our data is far more complete, with nearly 100% of 1-year follow-up and 27.8% of missing records. An earlier study from the Amaral group identified independent risk factors for pancreatic graft loss: recipient BMI, induction therapy, donor age, iliac venous drainage, and transplantation of the pancreas as the first graft. A risk stratification model was constructed and ROC curves were calculated, yet the model included few donor-dependent and non-modifiable variables; therefore, its applicability was limited [12]. We did not analyze recipient-dependent factors and tailored our study to allow deciding whether to accept a potential pancreas donor. A study by Blok showed PDRI over 1.24 to be a significant risk factor for graft loss [8]. We had only 74 cases of transplantation with such high PDRIs, but 12-month survival in this group was 67.6% (not significantly different from the low-PDRI group). Although PAK and PTA transplants were not analyzed separately, PDRI index in our study proved unreliable. Instead, we confirmed the usefulness of P-PASS in predicting graft survival; when interquartile differences were analyzed, a P-PASS over 18 points predicted significantly inferior outcomes. This substantially extends the previous limit of 16 points [4]. The most efficient model of prediction we built consisted of only 2 variables – donor age and BMI – and its ROC AUC was superior to P-PASS and PDRI. Although P-PASS and donor age/BMI models proved significant, C-statistics of 0.566 and 0.611 are not impressive and only slightly superior to a 50%/50% chance. Their clinical applicability should be very limited. In our national study, grafts from the worst prognosis quartiles still have over 50% chance of surviving beyond 1 year. Hence, the field of pancreases suitable for transplantation is most likely bigger than we use today, although caution and reasonable clinical judgement in accepting grafts at risk is needed.

Conclusions

Recipient- and surgeon-dependent factors were not analyzed in our study. Of course, quality of the recipient contributes to pancreatic graft survival. However, there are no risk prediction systems that specifically address this issue. Some general surgery risk prediction models can be applied with variable effectiveness [13].