ABSTRACT
Background
Pancreas non‐utilization affects 25%–30% of deceased donor offers in the US. The European‐derived Pre‐Procurement Pancreas Suitability Score (P‐PASS) has performed inconsistently, is not validated in North America, and requires less accessible inputs, highlighting need for a robust prediction tool for non‐utilization.
Methods
We developed a predictive model for pancreas non‐utilization using procured offers from the US SRTR (2003–2023). A multivariable logistic model used variables selected by causal diagram, LASSO, and univariable analyses. Model performance was evaluated using discrimination (AUC), calibration, and net reclassification improvement (NRI) versus a modified P‐PASS. Internal validation used 1000 bootstrap resamples. Sensitivity analyses included recent‐era offers, an all‐comer cohort including non‐procured offers, and excluding non‐donor‐related non‐utilizations.
Results
Among 30 757 pancreas offers, 26% were non‐utilized. The model incorporated ten factors: older age, male sex, higher BMI, hypertension, gastrointestinal disease, smoking, donation after cardiac death, stroke as cause of death, elevated terminal creatinine, and abnormal lipase. This model achieved moderate discrimination (AUC 0.699; optimism‐corrected 0.698), accurate calibration, Brier score 0.090, and 9.2% NRI. Sensitivity analyses confirmed robust performance, including AUC 0.8 among all‐comer offers.
Conclusions
This model provides a practical, data‐driven tool that improves identification of high‐risk pancreata and may support more consistent, efficient utilization decisions. By demonstrating modifiable center‐level variation and outperforming P‐PASS in its first North American external validation, our findings offer a strong evidence base for targeted quality‐improvement and policy initiatives to enhance pancreas transplant activity.
Keywords: discard, donation, non‐utilization, pancreas transplant, prediction model
Abbreviations
- AUC
Area under the curve
- BMI
Body mass index
- CIT
Cold ischemic time
- DCD
Donation after cardiac death
- HRSA
Health Resources and Services Administration
- ICU
Intensive care unit
- IQR
Interquartile range
- NRI
Net reclassification improvement
- OPTN
Organ Procurement and Transplantation Network
- OR
Odds ratio
- PDRI
Pancreas Donor Risk Index
- P‐PASS
Pre‐Procurement Pancreas Suitability Score
- SD
Standard deviation
- TRIPOD
Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis
1. Introduction
Pancreas transplantation provides critical survival and quality‐of‐life benefits to patients with diabetes mellitus [1, 2, 3]. Despite these benefits and high prevalence of diabetes, pancreas transplant rates remain low and have not improved over recent years, partly due to persistently high pancreas non‐utilization rates [4, 5]. In the United States, pancreas non‐utilization rates after procurement were as high as 28.7% in 2022 and remained elevated at 23.4% in 2023 [4]. Similarly, about half of pancreas offers were non‐utilized without retrieval in the United Kingdom and Eurotransplant regions, with an additional 30%–38% non‐utilized after initially being accepted [6, 7]. High non‐utilization rates have similarly affected other solid organ transplant types, with root‐cause analyses revealing the important role of donor risk factors, increased procurement of marginal donors, complexities in donor‐recipient matching, logistical barriers, and heterogeneous clinical practices across regions, centres, and clinicians [8, 9, 10, 11, 12, 13, 14]. Such comprehensive evaluations in pancreas transplantation, however, are more limited.
Risk factors for pancreas non‐utilization vary widely, often reflecting jurisdiction‐specific criteria, such as strict donor age or body mass index (BMI) thresholds, which frequently lead to automatic declines regardless of overall donor quality [6, 15, 16]. Registry studies indicate that declines predominantly result from donor‐ or organ‐specific factors, with the latter more commonly cited after organ‐retrieval and visual inspection, while recipient‐related or logistical factors are less common [6, 16, 17, 18, 19]. However, considerable heterogeneity persists, illustrated by inconsistent application of a region's risk criteria and eventual successful transplantation of pancreata previously declined for donor or organ concerns [6, 17]. Notably, pancreata were typically placed after a median of three to four offers in the UK and Eurotransplant regions; in the latter, those initially refused for “trauma” or donor age were later successfully transplanted in 48% and 32% of cases, respectively [6, 17]. Although analyses have largely been descriptive, identified donor risk factors include older age, obesity (BMI > 30), longer intensive care unit (ICU) admissions, donation after cardiac death (DCD), and stroke as cause of death [4, 6, 15, 16, 17, 18, 19]. Organ‐specific concerns include elevated serum glucose or pancreas enzymes, fatty infiltration, procurement‐related surgical injuries, and prolonged ischemic times [6, 18, 19, 20]. The Pancreas Donor Risk Index (PDRI) integrates ten donor variables to predict 1‐year pancreas graft survival with moderate discrimination (c‐statistic 0.67), but less so in validation studies, and higher scores are linked to pancreas non‐utilization [7, 21, 22].
Given these limitations, there is a clear need to better understand pancreas non‐utilization risk factors and develop predictive tools to guide prospective decision‐making. This need aligns with current policy priorities of the United States Organ Procurement and Transplantation Network (OPTN), which has identified improving pancreas procurement and utilization efficiency as a critical area for action [4]. Effective predictive models could optimize pancreas utilization by increasing acceptance of suitable organs while reducing unnecessary resource expenditures on organs unlikely to be transplanted. Currently, the Pancreas Pre‐procurement Allocation Suitability Score (P‐PASS) is the only validated predictive tool for pancreas acceptance, derived from a retrospective Eurotransplant cohort of 2175 donors (2002–2005) [22, 23]. Predictors in the final P‐PASS model, selected by expert consensus, include donor age, BMI, duration of ICU stay, duration of cardiac arrest, serum sodium, serum amylase/lipase, and inotrope use and dosing [23]. Although originally developed as a continuous score, P‐PASS was ultimately dichotomized into high‐ versus low‐risk categories [23]. The original study reported good calibration (observed‐to‐expected ratio 1.06), but notably did not report discrimination metrics and lacked internal validation [23]. Two studies have since performed external validation in European cohorts, reporting lower O/E ratios of 0.69 to 0.72, and only one study reporting moderate discrimination (c‐statistic 0.68) where P‐PASS was outperformed by PDRI [7, 24]. Crucially, the P‐PASS is not yet externally validated in North American populations, underscoring the urgency for improved, regionally relevant predictive tools.
Using a large US registry cohort of pancreas donor offers procured for possible transplantation from 2003 to 2023, we aimed to (i) identify risk factors for pancreas non‐utilization; (ii) externally validate P‐PASS; (iii) derive and internally validate a parsimonious predictive model for pancreas non‐utilization among procured pancreata, to be used by healthcare professionals and organ procurement organizations; and (iv) compare our model's performance to P‐PASS.
2. Methods
2.1. Study Design and Population
This cohort study consisted of all consecutive deceased donor pancreas organ offers procured for possible pancreas transplantation in the US from first January 2003 to first January 2023. This study used data from the Scientific Registry of Transplant Recipients (SRTR). The SRTR data system includes data on all donor, wait‐listed candidates, and transplant recipients in the US, submitted by the members of the Organ Procurement and Transplantation Network (OPTN). The Health Resources and Services Administration (HRSA), US Department of Health and Human Services provides oversight to the activities of the OPTN and SRTR contractors. Follow‐up was recorded at 6 months, 1 year, and annually until re‐transplant, death, lost to follow‐up, or study end.
The study was approved by the University Health Network Research Ethics Board.
2.2. Outcomes
The primary outcome of interest was pancreas non‐utilization, a binary variable. We assessed predictors of pancreas non‐utilization and reported contemporary epidemiologic descriptions of non‐utilization by timing, region, and etiology.
2.3. Candidate Predictors
All candidate predictors were donor characteristics present at time of organ offer. Baseline characteristics included demographic (age, sex, race, blood group), anthropometric (weight, height, body mass index), comorbidities (history of diabetes mellitus, hypertension, cardiac disease, gastrointestinal disease, cancer), infectious risk (clinical infection, high‐risk per Centre for Disease Control criteria, cytomegalovirus, Epstein‐Barr virus, hepatitis B virus, hepatitis C virus, human immunodeficiency virus, syphilis), and substance use (smoking, alcohol, cocaine, intravenous drugs). Although donor age and BMI do not have absolute cutoffs for utilization as per current OPTN policies, deceased donors >50 years old or with BMI >30 kg/m2 have modified allocation rules and are less common. Peri‐donation characteristics included donation after cardiac death status, cause of death, mechanism of death, cardiac arrest, resuscitation duration, inotrope support, and transfusions. Organ function parameters from bloodwork included pancreas enzymes (lipase, amylase), kidney (serum creatinine, proteinuria), liver function (aminotransferases, bilirubin), and metabolic indices (hemoglobin A1c, serum sodium, pH). Logistical factors included ischemic time (warm, cold), referral timing (weekday, year), and donation center volume quartile. Geographic region was represented by US state.
2.4. Statistical Analysis
Descriptive statistics included categorical variables presented as frequencies/percentages and compared with the Chi‐squared test. Continuous variables were presented as mean with standard deviation (SD) and compared with the Student's t‐test when normally distributed, and otherwise as median with interquartile range (IQR) and compared with Wilcoxon‐rank test when non‐normally distributed.
Annual incidence of pancreas non‐utilization was depicted on a histogram to assess temporal trends. Reasons for non‐utilization were reported. PDRI was calculated using donor age, sex, Black or Asian race, BMI >24 kg/m2, height >173 cm, DCD, stroke as cause of death, serum creatinine >2.5 mg/dL, and cold ischemic time (CIT) >12 h. As CIT was unavailable for non‐utilized organs, this value was set to 12 h (null effect). Density plots visualized distributions between non‐utilized versus transplanted organs.
We used univariable logistic regression analyses to assess associations between candidate predictors and pancreas non‐utilization. Candidate predictors were included in a multivariable model, with predictor selection informed by a combination of p value > 0.10 in univariable models, LASSO selection, and a priori hypothesized causal relationships outlined in a directed acyclic graph (causal diagram) (Supplemental Methods, Figure S1). To assess regional variation in pancreas acceptance practices, geographic region was included as a random effect in the multivariable model, and also as a fixed effect in a univariable model to avoid overfitting.
A parsimonious predictive model was developed with stepwise removal of predictors that did not reach statistical significance. We followed the TRIPOD (Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis) statement for reporting the development and validation of a multivariable prediction model [25] (Table S1). Accuracy of the final prediction model was assessed with discrimination (area under the curve (AUC) of receiver operating curve (ROC)), visual examination of calibration plots, and the Brier score. Internal validity of the final parsimonious model was confirmed by bootstrapping with 1000 datasets, permitting optimism‐corrected AUC.
A sensitivity analysis refitted the primary model omitting DCD status as a predictor to assess stability of other risk factors in light of evolving DCD recovery practices. Another analysis excluded organs ultimately non‐utilized for reasons unrelated to donor characteristics at organ offer, as these outcomes would not be expected to be captured by donor‐based prediction models (Supplemental Methods). Another sensitivity analysis was done in a more contemporary cohort, restricted to pancreas offers after first January 2013. Finally, the model was applied to an all‐comer cohort of all deceased donor pancreas offers, including organs that were not procured.
Predicted probabilities of pancreas non‐utilization were calculated from the model's linear predictor (including the intercept) using the logistic transformation, thereby operationalizing the model as a continuous prognostic index. Prognostic index was used to establish categorical high‐ versus low‐risk categories for pancreas non‐utilization defined by exploratory thresholds. Because acceptable risk thresholds are context‐dependent, we presented exploratory probability cutoffs to illustrate trade‐offs rather than proposing a universal threshold for procurement or utilization decisions. At each threshold, we reported the proportion of non‐utilized organs captured within the high‐risk group (prognostic index above the threshold), model AUC, and net reclassification improvement (NRI) compared to a modified P‐PASS. The modified P‐PASS omitted number of days in ICU and randomly sampled 2 or 3 points for donors receiving inotropes, as number of ICU days and inotrope doses were not captured in the SRTR dataset. Pancreas offers with scores of 14 or higher were considered high‐risk. Another modified P‐PASS was assessed, where number of points for ICU days was randomly sampled, but was not retained as it did not improve model AUC. NRI was used to quantify whether the new model more accurately reclassified pancreata into higher or lower non‐utilization risk categories compared with modified P‐PASS, the only previously validated prediction model for pancreas utilization.
Finally, regional variation was assessed in a sub‐cohort with known donation centre geographic location, which was included in the parsimonious multivariable model as a fixed effect.
We used marginal standardization of the prediction model, including donation centre volume as a fixed effect, to estimate counterfactual discard probabilities if all pancreata were managed at higher‐volume centers (Q2–Q4) versus if all were managed at lowest‐volume centers (Q1).
Two‐sided p value < 0.05 was considered statistically significant. Statistical analyses were conducted using R Studio (Version 2023).
3. Results
3.1. Study Population
Among 187 786 all‐comer pancreas offers, the analyzed population comprised 30,757 procured deceased donor pancreas organ offers (Figure S2). Mean donor age was 26.1 ± 11.2 years and 68% were male. Donor and organ characteristics are detailed in Table 1.
TABLE 1.
Baseline demographics of procured donor pancreas offers in overall cohort (n = 30 757).
| Characteristic | Number | |
|---|---|---|
| Demographic | ||
| Age (years) | 30 757 | 26.1 (11.2) |
| Male sex (%) | 30 757 | 20 843 (67.8%) |
| Race (%) | ||
| White | 24 463 (79.5%) | |
| Black | 5391 (17.5%) | |
| Asian | 571 (1.9%) | |
| Indigenous | 166 (0.5%) | |
| Blood group (%) | 30 757 | |
| A | 11 075 (36.0%) | |
| AB | 488 (1.6%) | |
| B | 3382 (11.0%) | |
| O | 15 812 (51.4%) | |
| Anthropometric | ||
| Weight (kg) | 30 757 | 71.7 (19.2) |
| Height (cm) | 30 757 | 169.8 (18.9) |
| Body mass index (kg/m2) | 30 757 | 24.3 (4.7) |
| Comorbidities | ||
| Diabetes mellitus (%) | 30 660 | 51 (0.2%) |
| Hypertension (%) | 30 626 | 2383 (7.8%) |
| Myocardial infarction (%) | 29 922 | 267 (0.9%) |
| Gastrointestinal disease (%) | 28 801 | 835 (2.9%) |
| Cancer (%) | 30 667 | 285 (0.9%) |
| Infectious risk | ||
| Clinical infection (%) | 30 093 | 17 497 (58.1%) |
| Blood infection (%) | 30 757 | 2375 (7.7%) |
| High‐risk CDC criteria (%) | 28 183 | 3797 (13.5%) |
| Cytomegalovirus seropositive (%) | 30 617 | 17 989 (58.8%) |
| Epstein‐Barr virus seropositive (%) | 22 565 | 20 283 (9.0%) |
| Hepatitis B surface antigen positive (%) | 30 704 | 25 (0%) |
| Hepatitis C virus antibody positive (%) | 30 717 | 223 (0.7%) |
| Human immunodeficiency virus positive (%) | 30 185 | 1 (0%) |
| Syphilis (%) | 30 422 | 122 (0.4%) |
| Substance use | ||
| Smoking history (%) | 30 476 | 3478 (11.4%) |
| Heavy alcohol use (%) | 27 789 | 2684 (9.7%) |
| Cocaine use (%) | 30 286 | 4080 (13.5%) |
| History of other drugs (%) | 30 456 | 13 078 (42.9%) |
| Tattoos (%) | 30 705 | 12 389 (40.4%) |
| Peri‐donation | ||
| Donation after cardiac death (%) | 30 756 | 1089 (3.5%) |
| Stroke as cause of death (%) | 30 756 | 5453 (17.7%) |
| Cardiac arrest (%) | 30 755 | 3049 (9.9%) |
| Cardiac arrest after death (%) | 30 168 | 1987 (6.6%) |
| Cardiac arrest duration (h) | 1753 | 17.8 (15.3) |
| Inotrope support (%) | 30 669 | 15 518 (50.6%) |
| Organ function bloodwork | ||
| Lipase (U/L) | 30 203 | 67.0 (130.8) |
| Lipase abnormal (%) | 30 203 | 2796 (9.3%) |
| Amylase (U/L) | 30 317 | 115.5 (196.9) |
| Amylase abnormal (%) | 30 317 | 8401 (27.7%) |
| Hemoglobin A1c (%) | 9005 | 5.3 (0.4) |
| Peak terminal serum creatinine (mg/dL) | 30 756 | 1.1 (0.9) |
| Proteinuria (%) | 30 482 | 11 500 (37.7%) |
| Aspartate aminotransferase (AST) (U/L) | 30 698 | 85.9 (230.0) |
| AST abnormal (%) | 30 698 | 20 218 (65.9%) |
| Alanine aminotransferase (ALT) (U/L) | 30 695 | 78.4 (222.2) |
| ALT abnormal (%) | 30 695 | 14 486 (47.2%) |
| Liver enzymes abnormal (%) | 30 695 | 23 438 (76.4%) |
| Total bilirubin (mg/dL) | 30 676 | 0.99 (1.2) |
| Total bilirubin abnormal (%) | 30 676 | 6646 (21.7%) |
| Serum sodium (mEq/L) | 28 199 | 147.0 (9.0) |
| Serum sodium abnormal (%) | 28 199 | 18 191 (64.5%) |
| Blood pH | 27 980 | 7.41 (0.12) |
| Logistics | ||
| Cold ischemic time (h) | 15 472 | 11.2 (5.4) |
| Warm ischemic time (h) | 5471 | 2.8 (7.8) |
| Referral on weekend (%) | 30 757 | 10 289 (33.5%) |
| Recovery on weekend (%) | 30 757 | 8047 (26.2%) |
| Transplant center volume quartile | 30 757 | |
| Q1 | 2459 (8.0%) | |
| Q2 | 4980 (16.2%) | |
| Q3 | 8529 (27.7%) | |
| Q4 | 14 789 (48.1%) | |
| Risk score | ||
| Pancreas donor risk index score | 30 754 | 1.30 (0.48) |
| Kidney donor risk index score | 30 757 | 0.38 (0.09) |
Among procured pancreas organs, 8048 (26.2%) pancreata were non‐utilized during the study period, and this was relatively stable over time (Figure S3). Where reasons for non‐utilization were reported, these are listed in Table S2. Median PDRI was higher for non‐utilized pancreas organs (1.33 vs. 1.12; p < 0.001), but density plots showed substantial overlap (Figure S4).
3.2. Risk Factors for Pancreas Non‐Utilization
We investigated risk factors for pancreas non‐utilization in univariable analyses; ischemic times were omitted due to high missingness (Supplemental Methods). A multivariable logistic regression model among complete cases (n = 25 412 pancreata; 6598 non‐utilized) identified increased odds of pancreas non‐utilization when donors were older, male, White or Indigenous, had higher BMI, history of hypertension, gastrointestinal disease, or smoking, stroke as cause of death, higher peak terminal serum creatinine, or abnormal lipase (Table 2). The strongest risk factor for non‐utilization was DCD status (OR 3.13; 95% CI 2.60, 3.77). Odds of non‐utilization were lower when donors were classified as high‐risk by CDC criteria or were offered to higher volume donation centers (vs. the lowest volume quartile) (Table 2). Non‐utilization was minimally higher in recent years and unaffected by weekend versus weekday timing of donor offers (Table 2).
TABLE 2.
Multivariable logistic regression model of risk factors for pancreas non‐utilization (n = 25 412 pancreas offers).
| Risk factor | Reference | Odds ratio | 95% CI | p value |
|---|---|---|---|---|
| Age (years) | 1.04 | 1.04, 1.04 | <0.001 | |
| Male sex | Female | 1.11 | 1.04, 1.19 | 0.002 |
| Race | Black | |||
| White | 1.23 | 1.13, 1.34 | <0.001 | |
| Indigenous | 1.85 | 1.24, 2.73 | 0.002 | |
| Blood group | Blood group A | |||
| AB | 1.14 | 0.89, 1.45 | 0.3 | |
| B | 0.99 | 0.89, 1.09 | 0.77 | |
| O | 1 | 0.93, 1.06 | 0.87 | |
| BMI (kg/m2) | 1.09 | 1.08, 1.09 | <0.001 | |
| History of hypertension | None | 1.5 | 1.34, 1.68 | <0.001 |
| History of diabetes | None | 2.04 | 0.95, 4.48 | 0.07 |
| History of gastrointestinal disease | None | 1.35 | 1.14, 1.59 | <0.001 |
| History of myocardial infarction | None | 1.15 | 0.83, 1.59 | 0.39 |
| History of cancer | None | 1.23 | 0.91, 1.65 | 0.17 |
| History of smoking | None | 1.12 | 1.01, 1.25 | 0.03 |
| History of heavy alcohol use | None | 1.09 | 0.99, 1.21 | 0.08 |
| History of cocaine use | None | 1.01 | 0.92, 1.11 | 0.76 |
| High‐risk by CDC criteria | Low‐risk | 0.91 | 0.82, 0.99 | 0.03 |
| Clinical infection | None | 1.04 | 0.97, 1.11 | 0.28 |
| DCD status | NDD | 3.13 | 2.60, 3.77 | <0.001 |
| Stroke as cause of death | Non‐stroke cause of death | 1.19 | 1.09, 1.30 | <0.001 |
| Cardiac arrest | None | 1.04 | 0.92, 1.17 | 0.55 |
| Inotrope support | None | 1.05 | 0.99, 1.12 | 0.12 |
| Peak terminal serum creatinine (mg/dL) | 1.17 | 1.13, 1.21 | <0.001 | |
| Serum sodium (mmol/L) | 1 | 0.99, 1.00 | 0.06 | |
| Abnormal lipase | Normal | 1.12 | 1.01, 1.25 | 0.03 |
| Abnormal liver enzymes | Normal | 1.06 | 0.99, 1.14 | 0.11 |
| Donation center volume | Q1 | |||
| Q2 | 0.7 | 0.62, 0.80 | <0.001 | |
| Q3 | 0.68 | 0.60, 0.76 | <0.001 | |
| Q4 | 0.84 | 0.75, 0.94 | 0.002 | |
| Recovery year | 1.01 | 1.00, 1.01 | 0.02 | |
| Recovery on weekend | Weekday | 1.03 | 0.96, 1.11 | 0.36 |
Bold values denote conventional statistical significance.
When included as a random effect in the multivariable model, variance due to donation centre US state was low overall (intra‐class coefficient 5.6%), but there was more marked variation when individual US states were considered (Figure S5). Odds of pancreas non‐utilization varied across US states when included as fixed effects in a univariable model (Figure S6)
3.3. Predictive Model for Pancreas Non‐Utilization
We first assessed discrimination of the associative multivariable model, which was moderate (AUC 0.702). A parsimonious predictive model was then sought, considering risk factors that achieved statistical significance in the associative model. This predictive model included predictors: donor age, male sex, BMI, history of hypertension, gastrointestinal disease, smoking, DCD status, stroke as cause of death, terminal serum creatinine, and abnormal lipase (Table 3). There was preserved moderate discrimination with AUC 0.699, confirmed by internal validation with bootstrap resampling, yielding an optimism‐corrected AUC 0.698 (95% CI 0.690, 0.706).
TABLE 3.
Parsimonious predictive model for pancreas discard (n = 25 412 pancreas offers).
| Risk factor | Reference | Odds ratio | 95% CI | p value |
|---|---|---|---|---|
| Age (years) | 1.04 | 1.04, 1.05 | <0.001 | |
| Male sex | Female | 1.09 | 1.02, 1.17 | 0.01 |
| BMI (kg/m2) | 1.08 | 1.08, 1.09 | <0.001 | |
| History of hypertension | None | 1.47 | 1.31, 1.63 | <0.001 |
| History of gastrointestinal disease | None | 1.34 | 1.14, 1.58 | <0.001 |
| History of smoking | None | 1.11 | 1.00, 1.22 | 0.04 |
| DCD status | NDD | 3.33 | 2.90, 3.83 | <0.001 |
| Stroke as cause of death | Non‐stroke cause of death | 1.17 | 1.08, 1.27 | <0.001 |
| Maximum terminal serum creatinine (mg/dL) | 1.19 | 1.15, 1.24 | <0.001 | |
| Abnormal lipase | Normal lipase | 1.16 | 1.05, 1.28 | 0.005 |
| Intercept: −4.55 (SE 0.096) | ||||
Bold values denote conventional statistical significance.
Visual inspection of calibration plots showed good agreement between predicted and observed pancreas non‐utilization across the range of predicted probabilities (Figure 1). The Brier score was 0.090.
FIGURE 1.
Calibration plot for observed versus predicted probability of pancreas discard.
3.4. Risk Categories and Comparison With a Modified P‐PASS Score
A modified P‐PASS score was calculated in 25 689 cases, including 8367 cases classified as high‐risk, with increased odds of non‐utilization (OR 2.48; 95% CI 2.34, 2.63). The model demonstrated modest discrimination with AUC 0.604.
Using a 70th percentile cutoff of predicted probabilities to dichotomize risk, 30% of cases were classified as high risk for non‐utilization with our new model. Compared with the modified P‐PASS classification, the new model achieved overall net reclassification improvement of 9.2% (1082 pancreata). Specifically, among non‐utilization cases (n = 6336), 905 cases were upgraded from low to high risk while 589 were downgraded—a net gain of 5.0% (316 pancreata). In the transplant group (n = 18 249), 2243 cases were correctly reclassified downward from high to low risk, compared with 1477 that were incorrectly upgraded, yielding a net gain of 4.2% (766 pancreata).
3.5. Sensitivity Analyses
When DCD was omitted as a predictor, effect estimates of remaining predictors were similar to primary analysis, with slight reduction in discrimination (AUC 0.687) (Table S3).
In a sub‐cohort of 13 038 pancreas offers since 2013, the modified P‐PASS score demonstrated an AUC of 0.573, whereas our model achieved an AUC of 0.663, and NRI of 9.7%.
Another analysis examined 27 103 pancreas offers after excluding organs ultimately non‐utilized for reasons unrelated to donor characteristics at the time of organ offer (Figure S7). In this sub‐cohort, the modified P‐PASS score yielded an AUC of 0.627 compared with an AUC of 0.732 for our model, with a corresponding NRI of 11.1%.
Finally, we examined an all‐comer cohort of 187 786 pancreas offers, including organs that were not procured. Among these offers, 22 709 pancreata were transplanted while 165 077 were non‐utilized. The modified P‐PASS had an AUC of 0.677, while our model achieved an optimism‐corrected AUC of 0.804, and NRI of 11.7%.
Under the observed mix of donation centers, 6862 pancreases were discarded. Counterfactual predictions suggested ∼95 more utilized pancreata if all centers had performed like Q2–Q4 centers, whereas ∼1188 fewer pancreata would have been utilized if all centers followed the practice patterns of the lowest volume Q1 centers.
4. Discussion
In this large, contemporary registry cohort of deceased donor pancreas donor offers in the United States, we identified significant donor and organ risk factors associated with pancreas non‐utilization, derived a parsimonious predictive model, and externally validated a modified P‐PASS score. Our analysis revealed a pancreas non‐utilization rate of 26.2%, relatively stable over two decades, with considerable variability by region and donation center. Key donor predictors for non‐utilization included older age, higher BMI, male sex, history of hypertension, gastrointestinal disease, smoking, DCD status, stroke as cause of death, higher terminal serum creatinine, and abnormal lipase. Our parsimonious predictive model demonstrated moderate discrimination (optimism‐corrected AUC 0.698) with good calibration and significantly outperformed a modified P‐PASS score, particularly after excluding non‐utilizations unrelated to donor characteristics
Our predictive model achieved moderate discrimination (optimism‐corrected AUC 0.698) and good calibration for predicting pancreas non‐utilization in procured offers, substantially outperforming the modified P‐PASS (AUC 0.604), with an NRI of 9.2%. Several factors may explain this improvement. Firstly, P‐PASS was derived in a European cohort (2002–2005), which differs from our contemporary North American cohort (2003 to 2023) in donor population, practice patterns, and temporal context [23]. Although prior Eurotransplant validations studies reported moderate performance, the original derivation study for P‐PASS did not report discrimination metrics, limiting meaningful comparisons [7]. Further, P‐PASS was developed to predict non‐utilization across all donor offers—including those not procured—while our model focused on post‐procurement non‐utilization [23]. This distinction is important, as organ‐specific concerns often emerge during intra‐operative visual inspection of the pancreas, a subjective but key step in decision‐making [26]. These organ‐specific concerns have traditionally been considered less predictable from donor characteristics alone [26]. Indeed, a sensitivity analysis excluding non‐utilizations unrelated to donor factors demonstrated improved discrimination (AUC 0.732) and NRI of 11.1%. Model development in this enriched post‐procurement cohort may have enabled inclusion of donor variables that better predict both donor‐ and organ‐specific reasons for decline, an area of growing interest given the need to reduce unnecessary procurements. Moreover, when applied to a broader all‐comer cohort that included both procured and non‐procured offers, our model maintained strong performance (optimism‐corrected AUC 0.804) and demonstrated an NRI of 11.7% over the modified P‐PASS (AUC 0.677). These findings support the model's utility across the full spectrum of the offer process, with potential to guide both procurement and acceptance decisions.
We identified multiple key risk factors for pancreas non‐utilization. Older donors and those with higher BMI have higher risk of fatty infiltration [27]. These same factors, alongside male sex, hypertension, smoking history, DCD, and stroke as cause of death, predispose donors to underlying vascular disease. Donor gastrointestinal disease history may indicate increased anatomical complexity if there was prior surgery, bowel pathology related to inflammatory bowel disease, or undiagnosed pancreatic pathology including chronic pancreatitis. Elevated serum creatinine could suggest impaired donor hemodynamics, raising concerns of associated pancreatic ischemia. We categorized donor lipase as abnormal versus normal, given the weak correlation of the magnitude of lipase with the degree of pancreatic injury, and non‐pancreatic causes of lipase elevation [28, 29, 30]. Additional risk factors for non‐utilization included White or Indigenous race, warranting further exploration of underlying socioeconomic factors beyond the scope of this analysis. Interestingly, CDC high‐risk designation correlated with lower odds of non‐utilization, possibly due to favorable donor characteristics, overly inclusive criteria, or selection forces at play in already‐procured pancreata. In contrast to the original P‐PASS score, serum sodium levels and inotrope support were not significantly associated with non‐utilization in our cohort, potentially reflecting differences in data granularity on inotropes [23]. Higher‐volume donation centers demonstrated higher pancreas utilization, reinforcing other benefits of institutional and professional experience including successful pancreas procurement and longer graft survival [31, 32, 33]. Conversely, lower‐volume centers may be disincentivized from accepting marginal organs due to remuneration structures tied to short‐term outcomes, while organ offer acceptance metrics are less emphasized in the current system. Pancreas utilization varied widely across geographic regions, similar to prior studies in other SOT types [10]. Our counterfactual analyses suggest that aligning practice with higher‐volume donation centers could meaningfully reduce discards, underscoring the value of system‐level quality improvement and cross‐center learning. Lastly, we noted a subtle recent increase in non‐utilization rates, reinforcing ongoing concerns about pancreas utilization [4].
Our findings have several important clinical and policy implications. First, the development and validation of our predictive model offer opportunities to proactively identify pancreata at high risk of non‐utilization, enabling targeted allocation to centers with greater risk tolerance or expertise with more marginal organs. Identifying key risk factors for pancreas non‐utilization provides a foundation to refine donor acceptance criteria and promote consistency in organ evaluation practices, reducing variability and unnecessary organ non‐utilizations. Educational initiatives addressing misconceptions around donor characteristics, such as CDC high‐risk status, could further enhance pancreas utilization. Additionally, policy strategies, including revised remuneration structures or centralized procurement networks, may incentivize smaller centers to accept higher‐risk organs. At a population level, the observed NRI corresponds to over 1000 pancreata more appropriately risk‐stratified compared with the modified P‐PASS, supporting potential value of this model for utilization‐focused quality improvement rather than implying guaranteed increases in transplantation. Excellent model performance among the all‐comer cohort of pancreas offers may also suggest a role for model integration into decision‐making around mobilization of resources for organ procurement, which is especially of value for resource‐limited healthcare systems seeking optimized transplant system performance. Collectively, these interventions could be leveraged to meaningfully improve pancreas transplant rates and outcomes within logistical system constraints. Although P‐PASS was intended to achieve many of these aims, its suboptimal performance in this contemporary North American cohort underscores the need for external validations such as this. Our model provides an internally validated practical tool with fewer and more readily available data elements. Following external validation of this new model, future knowledge translation efforts could include implementation of this model as an online risk calculator to facilitate real‐time clinical use.
Our study has several strengths. We leveraged a large, contemporary national registry with mandatory reporting that spanned two decades, providing a robust dataset reflective of real‐world pancreas procurement practices across diverse US centers. Our rigorous methodological approach included derivation of a parsimonious predictive model with optimism‐corrected performance metrics and sensitivity analyses reinforcing our model's robustness. The model is practical, including easily measured variables which are routinely available to intended users of healthcare providers and organ procurement organizations during the pancreas offer process. To our knowledge, this is the first external validation of P‐PASS in a North American cohort and the largest study to comprehensively evaluate pancreas non‐utilization risk factors. We used validated statistical approaches, including net reclassification improvement and discrimination assessments, to compare models, demonstrating superior discrimination of our model over P‐PASS. Additionally, we used a directed acyclic graph to facilitate variable selection to consider variables of interest in current clinical paradigms, and leading to more generalizable models. Whereas P‐PASS used a number of data elements that are less readily accessible or require ongoing updates (e.g., number of days in ICU and inotrope type and dosing), our final model is also practical due to parsimony and accessible data inputs.
Despite these strengths, several limitations should be acknowledged. Restricting our analysis to procured pancreata may have introduced selection bias by excluding organs declined before procurement, potentially underestimating the impact of certain risk factors. The registry lacked detailed data on intraoperative findings and the rationales for non‐utilization, limiting our ability to assess surgeon‐driven decision‐making at the time of organ evaluation. The study is further limited by the quality and granularity of registry data, including the binary capture of inotrope support and absence of length of ICU admission, limiting direct comparability with prior studies on P‐PASS [34, 35]. Nonetheless, the model demonstrated moderate and consistent predictive performance. Overlapping donor risk distributions also suggests variable acceptance practices. Future studies using single‐OPO or transplant‐center designs that allow for greater clinical and operational granularity, including recovery team experience, hospital course, and concurrent organ recovery, may further clarify contributors to pancreas non‐utilization. Organ utilization decisions are also shaped by organ‐specific supply‐demand dynamics; in pancreas transplantation, lower contemporary transplant volumes and shorter waiting times relative to other organs may permit greater selectivity, potentially contributing to persistently high non‐utilization rates. Future studies are needed to conduct external and prospective validations of our model, explore its use at the time of initial donor offer to optimize pancreas utilization, and investigate associations with post‐transplant outcomes—including graft survival and whether organs predicted to be at high risk of non‐utilization would nevertheless have acceptable graft survival if transplanted.
Our findings provide insights into pancreas non‐utilization, identifying key donor risk factors and demonstrating the limitations of existing predictive tools. We developed and validated a parsimonious predictive model that significantly outperformed a modified P‐PASS score and maintained robust discrimination in sensitivity analyses. These results support the integration of a more precise, data‐driven approach to organ offer evaluation, with the potential to improve pancreas utilization and transplant rates. Future work should focus on prospective validation of our model and its application at the time of donor offer to better inform acceptance decisions. Policy and educational initiatives aimed at reducing variability in donor acceptance practices, addressing misconceptions about non‐utilization risk, and incentivizing transplant centers to consider more marginal pancreata may further optimize utilization. Together, these efforts could help bridge the gap between pancreas donation and transplantation, ultimately improving outcomes for patients with insulin‐dependent diabetes.
Author Contributions
Christie Rampersad: Participated in research design, performance of the research, data analysis, and writing of the paper. Jason Bau: Participated in research design and writing of the paper. Jeffrey Schiff: Participated in writing of the paper. Trevor Reichman: Participated in writing of the paper. S. Joseph Kim: Participated in research design, performance of the research, data analysis, and writing of the paper.
Funding
C.R. is supported by a Kidney Research Scientist Core Education and National Training Program (KRESCENT) salary award co‐funded by the Kidney Foundation of Canada, the Canadian Society of Nephrology, and the Canadian Institutes of Health Research (CIHR), and a fellows research grant funded by Ajmera Transplant Centre at University Health Network. The remaining authors declare that they have no relevant financial interests.
Conflicts of Interest
The authors declare no conflicts of interest.
Protocol and Registration
Study protocol registration was not required for this retrospective study.
Patient and Public Involvement
None.
Supporting information
Supporting File1: Ctr70507‐sup‐0001‐figuresS1‐S7
Supporting File2: Ctr70507‐sup‐0002‐tableS1.Pdf
Supporting File3: Ctr70507‐sup‐0003‐tablesS2‐S3.docx
Supporting File4: Ctr70507‐sup‐0004‐SuppMat.Docx
Acknowledgments
The data reported here have been supplied by the Hennepin Healthcare Research Institute (HHRI) as the contractor for the Scientific Registry of Transplant Recipients (SRTR). The interpretation and reporting of these data are the responsibility of the author(s) and in no way should be seen as an official policy of or interpretation by the SRTR or the US Government.
Data Availability Statement
The data supporting the findings of this study were obtained from the Scientific Registry of Transplant Recipients (SRTR) dataset and are not able to be shared directly by the authors. Analytic code may be provided upon reasonable request.
References
- 1. Esmeijer K., Hoogeveen E. K., van den Boog P. J. M., et al., “Superior Long‐Term Survival for Simultaneous Pancreas‐Kidney Transplantation as Renal Replacement Therapy: 30‐Year Follow‐Up of a Nationwide Cohort,” Diabetes Care 43, no. 2 (2020): 321–328. [DOI] [PubMed] [Google Scholar]
- 2. Gruessner R. W. G. and Gruessner A. C., Transplantation of the Pancreas (2023). [Google Scholar]
- 3. Gruessner R. W. and Gruessner A. C., “The Current State of Pancreas Transplantation,” Nature Reviews Endocrinology 9, no. 9 (2013): 555–562. [DOI] [PubMed] [Google Scholar]
- 4. Kandaswamy R., Stock P. G., Miller J. M., Handarova D., Israni A. K., and Snyder J. J., “OPTN/SRTR 2023 Annual Data Report: Pancreas,” American Journal of Transplantation 25, no. 2S1 (2025): S138–S192. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Organization WH . Diabetes. 2024, (accessed March 10, 2025) https://www.who.int/news‐room/fact‐sheets/detail/diabetes.
- 6. Cornateanu S. M., O'Neill S., and Dholakia S., “Pancreas Utilization Rates in the UK—An 11‐Year Analysis,” Transplant International 34, no. 7 (2021): 1306–1318. [DOI] [PubMed] [Google Scholar]
- 7. Kopp W. H., de Vries E., de Boer J., et al., “Donor Risk Indices in Pancreas Allocation in the Eurotransplant Region,” Transplant International 29, no. 8 (2016): 921–929. [DOI] [PubMed] [Google Scholar]
- 8. Lentine K. L., Smith J. M., Lyden G. R., et al., “OPTN/SRTR 2023 Annual Data Report: Kidney ,” American Journal of Transplantation 25, no. 2S1 (2025): S22–S137. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Kwong A. J., Kim W. R., Lake J. R., et al., “OPTN/SRTR 2023 Annual Data Report: Liver ,” American Journal of Transplantation 25, no. 2S1 (2025): S193–S287. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Mohan S., Chiles M. C., Patzer R. E., et al., “Factors Leading to the Discard of Deceased Donor Kidneys in the United States,” Kidney International 94, no. 1 (2018): 187–198. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Aubert O., Reese P. P., Audry B., et al., “Disparities in Acceptance of Deceased Donor Kidneys Between the United States and France and Estimated Effects of Increased US Acceptance,” JAMA Internal Medicine 179, no. 10 (2019): 1365–1374. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Torabi J., Todd R., van Leeuwen L. L., et al., “A Decade of Liver Transplantation in the United States: Drivers of Discard and Underutilization,” Transplant Direct 10, no. 6 (2024): e1605. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Valapour M., Lehr C. J., Schladt D. P., et al., “OPTN/SRTR 2023 Annual Data Report: Lung ,” American Journal of Transplantation 25, no. 2S1 (2025): S422–S489. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Wisel S. A., Borja‐Cacho D., Megna D., Adjei M., Kim I. K., and Steggerda J. A., “Organ Utilization Rates From Non‐Ideal Donors for Solid Organ Transplant in the United States,” Journal of Clinical Medicine 13, no. 11 (2024): 3271. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Mensink J. W., de Vries K. M., Huurman V. A. L., Pol R. A., Alwayn I. P. J., and Braat A. E., “Risk Analysis of Extended Pancreas Donor Selection Criteria,” Pancreatology 19, no. 7 (2019): 994–999. [DOI] [PubMed] [Google Scholar]
- 16. Mandwar M., Sharma A., Singh S., et al., “Reasons for Low Utilization of Pancreas for Transplantation From Deceased Donors: Analysis of Data From a Single Centre in India,” Transplantation Proceedings 52, no. 6 (2020): 1858–1859. [DOI] [PubMed] [Google Scholar]
- 17. Loss M., Drewitz K. P., Apfelbacher C. J., et al., “Why Offered Pancreases Are Refused in the Allocation Process‐A Descriptive Study Using Routine Data From Eurotransplant,” Transplantation 95, no. 9 (2013): 1134–1141. [DOI] [PubMed] [Google Scholar]
- 18. Drewitz K. P., Loss M., Loss J., and Apfelbacher C. J., “Predictors of Non‐transplantation of Adult Donor Organs–An Observational Study Using Routine Data From Eurotransplant,” BMC Health Services Research 14 (2014): 584. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. Ausania F., Drage M., Manas D., and Callaghan C. J., “A Registry Analysis of Damage to the Deceased Donor Pancreas During Procurement,” American Journal of Transplantation 15, no. 11 (2015): 2955–2962. [DOI] [PubMed] [Google Scholar]
- 20. de Boer J. D., Kopp W. H., Ooms K., et al., “Abdominal Organ Procurement in the Netherlands—An Analysis of Quality and Clinical Impact,” Transplant International 30, no. 3 (2017): 288–294. [DOI] [PubMed] [Google Scholar]
- 21. Axelrod D. A., Sung R. S., Meyer K. H., Wolfe R. A., and Kaufman D. B., “Systematic Evaluation of Pancreas Allograft Quality, Outcomes and Geographic Variation in Utilization,” American Journal of Transplantation 10, no. 4 (2010): 837–845. [DOI] [PubMed] [Google Scholar]
- 22. Ling J. E. H., Coughlan T., Polkinghorne K. R., and Kanellis J., “Risk Indices Predicting Graft Use, Graft and Patient Survival in Solid Pancreas Transplantation: A Systematic Review,” BMC Gastroenterology 21, no. 1 (2021): 80. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23. Vinkers M. T., Rahmel A. O., Slot M. C., Smits J. M., and Schareck W. D., “How to Recognize a Suitable Pancreas Donor: A Eurotransplant Study of Preprocurement Factors,” Transplantation Proceedings 40, no. 5 (2008): 1275–1278. [DOI] [PubMed] [Google Scholar]
- 24. Rodriguez‐Villar C., Paredes D., Roque R., et al., “Predictive and Comparative Study Between Clinic Consensus Document for Pancreas Acceptance and Predictive Value of Preprocurement Pancreas Allocation Suitability Score (P‐PASS),” Transplantation Proceedings 51, no. 2 (2019): 365–368. [DOI] [PubMed] [Google Scholar]
- 25. Collins G. S., Reitsma J. B., Altman D. G., and Moons K. G., “Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD): The TRIPOD Statement,” BMJ 350 (2015): g7594. [DOI] [PubMed] [Google Scholar]
- 26. Fridell J. A. and Stratta R. J., “Expanding the Pancreas Donor Pool,” Current Transplantation Reports 1 (2014): 100–112. [Google Scholar]
- 27. Catanzaro R., Cuffari B., Italia A., and Marotta F., “Exploring the Metabolic Syndrome: Nonalcoholic Fatty Pancreas Disease,” World Journal of Gastroenterology 22, no. 34 (2016): 7660–7675. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28. Rau C. S., Wu S. C., Chien P. C., et al., “Identification of Pancreatic Injury in Patients With Elevated Amylase or Lipase Level Using a Decision Tree Classifier: A Cross‐Sectional Retrospective Analysis in a Level I Trauma Centre,” International Journal of Environmental Research and Public Health 15, no. 2 (2018): 277. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29. Mahajan A., Kadavigere R., Sripathi S., Rodrigues G. S., Rao V. R., and Koteshwar P., “Utility of Serum Pancreatic Enzyme Levels in Diagnosing Blunt Trauma to the Pancreas: A Prospective Study With Systematic Review,” Injury 45, no. 9 (2014): 1384–1393. [DOI] [PubMed] [Google Scholar]
- 30. Hameed A. M., Lam V. W., and Pleass H. C., “Significant Elevations of Serum Lipase Not Caused by Pancreatitis: A Systematic Review,” HPB (Oxford) 17, no. 2 (2015): 99–112. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31. Alhamad T., Malone A. F., Brennan D. C., et al., “Transplant Centre Volume and the Risk of Pancreas Allograft Failure,” Transplantation 101, no. 11 (2017): 2757–2764. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32. Kopp W., van Meel M., Putter H., et al., “Centre Volume Is Associated With Outcome After Pancreas Transplantation Within the Eurotransplant Region,” Transplantation 101, no. 6 (2017): 1247–1253. [DOI] [PubMed] [Google Scholar]
- 33. Lam H. D., Schaapherder A. F., Kopp W. H., Putter H., Braat A. E., and Baranski A. G., “Professionalization of Surgical Abdominal Organ Recovery Leading to an Increase in Pancreatic Allografts Accepted for Transplantation in the Netherlands: A Serial Analysis,” Transplant International 30, no. 2 (2017): 117–123. [DOI] [PubMed] [Google Scholar]
- 34. Massie A. B., Kucirka L. M., and Segev D. L., “Big Data in Organ Transplantation: Registries and Administrative Claims,” American Journal of Transplantation 14, no. 8 (2014): 1723–1730. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35. Yu M., King K. L., Husain S. A., et al., “Discrepant Outcomes Between National Kidney Transplant Data Registries in the United States,” Journal of the American Society of Nephrology 34, no. 11 (2023): 1863–1874. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Supporting File1: Ctr70507‐sup‐0001‐figuresS1‐S7
Supporting File2: Ctr70507‐sup‐0002‐tableS1.Pdf
Supporting File3: Ctr70507‐sup‐0003‐tablesS2‐S3.docx
Supporting File4: Ctr70507‐sup‐0004‐SuppMat.Docx
Data Availability Statement
The data supporting the findings of this study were obtained from the Scientific Registry of Transplant Recipients (SRTR) dataset and are not able to be shared directly by the authors. Analytic code may be provided upon reasonable request.