Multi-Center Analysis of Novel and Established Variables Associated with Successful Human Islet Isolation Outcomes

Abstract

Islet transplantation is a promising therapy used to achieve glycometabolic control in a select subgroup of individuals with type I diabetes. However, features that characterize human islet isolation success prior to transplantation are not standardized and lack validation. We conducted a retrospective analysis of 806 isolation records from 14 pancreas processing laboratories, considering variables from relevant studies in the last 15 years. The outcome was defined as post-purification islet equivalent count, dichotomized into yields ≥ 315,000 or ≤ 220,000. Univariate analysis showed that donor cause of death and use of hormonal medications negatively influenced outcome. Conversely, pancreata from heavier donors and those containing elevated levels of surface fat positively influence outcome, as did heavier pancreata and donors with normal amylase levels. Multivariable logistic regression analysis identified the positive impact on outcome of surgically intact pancreata and donors with normal liver function, and confirmed that younger donors, increased body mass index, shorter cold ischemia times, no administration of fluid/electrolyte medications, absence of organ edema, use of University of Wisconsin preservation solution, and a fatty pancreas improves outcome. In conclusion, this multi-center analysis highlights the importance of carefully reviewing of all donor, pancreas, and processing parameters prior to isolation and transplantation.

Introduction

Pancreatic Islet Transplantation (PIT) has been shown to be an effective therapeutic strategy to achieve glycometabolic control in a subset of patients diagnosed with type I diabetes mellitus(1). A recent report confirmed that in subjects achieving insulin independence at least once after the first islet infusion, graft function persisted in approximately 75% of these individuals 40 months post follow-up(2). However, PIT is still an experimental procedure, due in part to the complexities surrounding human islet isolation and transplantation techniques, activation of innate and adaptive immunity, and early loss of transplanted cells(3, 4). Furthermore, the ultimate goal of insulin independence, alongside a diminishing incidence of secondary complications, remains elusive for many of those with labile disease(1, 5).

Overall pancreas quality has been documented as a major determinant of islet isolation yield prior to transplantation. Factors such as donor age(6-10), extended cold ischemia time (CIT)(6-9), and history of diabetes(6, 8, 9) have consistently been found to negatively impact isolation outcome. In contrast, organs from brain-dead donors with elevated Body Mass Index(BMI) and increased pancreas weight(6, 7, 9, 11-15) have been documented to be among the best predictors of successful isolation outcome. Other variables such as donation after cardiac death (DCD)(16, 17), type of enzyme used in the digestion process(9, 10, 14, 18), cause of death(6, 7), and organ preservation methods(19-26) are still controversial with respect to their role in isolation outcome.

However, a standard definition of isolation success remains a matter of debate. In 1996, Lakey et al. defined a successful isolation as a post-purification count above 100,000 Islet Equivalents (IEQs), a standard also used in a study conducted ten years later(6, 27). Recently, others have increased the cut-off point for success to 250,000 IEQ (10, 14). However, no widespread consensus has emerged as to what constitutes a successful isolation, which may help to explain why findings from previous analyses range from consistent to contradictory.

Although the body of literature addressing this topic is considerable, most research has been restricted to single center experiences, a limited number of observations, non-comparable standards of success, and divergent inter-center methodologies that tend to restrict overall applicability. Moreover, current United Network of Organ Sharing (UNOS) allocation policy for PIT is generally limited to organs from donors above 50 years of age with BMI greater than 30 kg/m² when suitable recipients for solid pancreas transplant are not identified at the local, regional, or national level (28). Therefore, we conducted the present study to establish reliable markers of successful isolation yield and identify factors associated with outcome using data from first and largest multi-center database of its kind, collected by the Islet Cell Resource (ICR) Center Consortium. The ICR Consortium has been in operation for 8 years, with data from 14 centers, resulting in a robust set of isolation and merged donor data, to allow for a multi-centered evaluation of factors associated with successful islet yield.

Materials and Methods

Human Pancreatic Islet Isolation Records

A total of 1,122 pancreas islet isolation records were entered into the ICR database from December 3, 2004 to October 27, 2008 for human pancreata obtained between July 22, 2001 through October 25, 2008 from 14 laboratories across the U.S. Of those, 1,094 records were linked to pancreas donor data from UNOS. IEQ counts for isolations at the post-purification stage were available for only 1,023 records.

Selection and Definition of Donor and Isolation Variables

A literature search was performed using PubMed and combined the use of Medical Subject Heading (MeSH) terms and free form text searches to identify original studies, from January 1, 1994 to January 31, 2009, that analyzed the influence of pancreas donor and/or processing variables on isolation outcome. A flow chart outlining our search strategy is provided as Supplemental Figure 1. Thirty one articles were identified(6, 7, 9, 10, 12-14, 18, 20, 23, 24, 27, 29-47). Additional variables were obtained from review of the ICR/UNOS datasets, including donor personal and medical history, management after admittance into a clinical facility, factors related to pancreas procurement, preservation, handling and data related to technical aspects of islet production. A definition for each variable used is provided as Supplemental Table 1. All data for variables analyzed were solely obtained from the ICR/UNOS datasets and does not include the use of results from other previously published works.

Study Outcome

The range and distribution of post-purification IEQs were evaluated to help identify “low” and “high” yielding isolations (Figure 1). Isolations were grouped into quintiles as follows: 1^st quintile (667-145,968 IEQs), 2^nd quintile (146,413-221,612), 3^rd quintile (221,684-312,758), 4^th quintile (312,774-432,921), and 5^th quintile (432,929 – 1,132,083). Isolations with IEQs in the middle quintile (n=217) were excluded from further analysis because of known variability in IEQ measurements both within and between centers(48). Exclusion of the middle quintile reduces potential intra/inter-variability in IEQ measurements and helps strengthen the validity of any association with a successful yield. Based on this approach, a low-yielding (unsuccessful) isolation was defined as those records with IEQ counts ≤ 220,000 (n=402) and high-yielding (successful) as those with IEQ counts ≥ 315,000 (n=404). A total of 806 records remained and were analyzed to help determine predictors that may increase the odds of a high yield.

Post-purification purity and viability data were considered in defining the study outcome, but not used due to missing data in 124 and 617 records, respectively. Although purity was excluded as a measure of isolation success, a comparison between high and low yielding preparations was generated using a two-sample t-test and an Ftest to examine the normality of the variable (F-test p-value not reported).

Univariate and Multivariable Analysis

Percentages were reported for all categorical variables. The measure of central tendency was described using either mean (± 1 SD) or median (min, max) for all continuous variables, depending on whether a variable was normally distributed. Coding of continuous variables to categorical groupings was based on distribution of the variable for age (data not shown) and established laboratory reference values for kidney, liver, and pancreas assays. To test for center effect, the laboratory with the largest number of isolations reported in the database was selected as the baseline group.

All variables were included for univariate logistic regression (LR) model testing. Chi-square p-values, the corresponding odds ratio (OR), and the 95% confidence interval (95% CI) are reported for all categorical variables. In the case of continuous variables, the OR and 95% CI are reported in units defined using the inter-quartile range between the 75^th and 25^th percentiles, allowing for a direct comparison of OR values among all continuous and categorical predictors. The profile likelihood method was used to generate the reported statistical measures.

To test for simultaneous effects of multiple predictors, all variables found to be significant from univariate LR analysis with a p-value of <0.20 were considered in defining an optimal multivariable model. If a variable was not found to be statistically significant in the analysis, but previously shown to be significant in 2 or more studies, it was also included in the model for testing. A matrix of significant pair-wise correlations was generated to identify possible collinear variables, defined as any two terms with a p-value <0.05 and a Pearson's correlation coefficient of ≥0.80. Multi-class categorical variables were independently screened for co-linearity using a chi-square test. All variables were added into the model and the factor with the highest p-value was removed. This process was continued until only those variables with p-values of <0.10 remained. Interaction terms were generated using a priori hypotheses for greater-than-additive effects on the dependent variable and included organ intact by procurement team, preservation solution by CIT, donor medications received by amylase test, collagenase provider by age, and procurement team by CIT. Regression diagnostics were used to examine influential isolation records and outliers. The Hosmer-Lemeshow test was used to fit the model. This test may not be appropriate in all cases(49), but demonstrated to perform sufficiently in datasets with a large sample size, presence of continuous variables, and lack of significant interaction terms(50), as is reflected in our data. All statistical analysis was performed using SAS software version 9.1.3 12(SAS Institute, Cary, NC).

Results

A total of 806 records were analyzed to help determine predictors that increase the odds of achieving isolation success.

Purity

A statistically significant difference in post-purification purity was found between high (n=349, 63.1% ± 21.3) and low (n=333, 59.7% ± 22.6) yielding isolations (p=0.0396).

Univariate Analysis

Several donor demographic variables were found to correlate significantly with successful isolation outcome (Table 1). Relative to younger donors, pancreata from individuals over 47.5 years of age were less likely to result in high yielding isolations (OR=0.61, p=0.0004). Cause of death was shown to be important (overall p=0.032); in particular, individuals succumbing to head trauma were less likely to result in high yielding isolations compared to those in the cerebrovascular/stroke group (OR=0.69, p=0.015). Although mechanism of death was only marginally significant (p=0.053), the results were complementary to that of cause of death; relative to the blunt injury group, pancreata from donors in the intracranial hemorrhage/stroke group were more likely to result in high yielding isolations (OR=1.50, p=0.020). An increase in donor weight improved the odds of obtaining a high yielding isolation (OR=2.12 per an increase equal to the inter-quartile range value of 27kg, p<0.0001), as did an increase in BMI (OR=1.82 per an increase of 8.09kg/m², p<0.0001). A pancreas from a DCD donor was somewhat less likely to result in a high yield isolation (OR=0.46, p=0.058).

Table 1. Organ Donor Demographic Variables Affecting Islet Isolation Success.

Name	≤220,000 IEQs		≥315,000 IEQs		Odds Ratio1	95% CI	p-value
	N	% or Mean (±1SD)	N	% or Mean (±1SD)
Age (in yrs)
00.1 - 47.5	182	44%	233	56%	-
>47.5 – 70.8	220	56%	171	44%	0.61	(0.46, 0.80)	0.0004
Gender
Female	192	53%	170	47%	-
Male	210	47%	234	53%	1.26	(0.95, 1.66)	0.11
Weight2 (kg)	402	81.4 (±19.2)	404	93.0 (±22.3)	2.12	(1.74, 2.61)	<0.0001
BMI2 (kg/m2)	402	27.8 (±5.8)	404	30.8 (±7.0)	1.82	(1.51, 2.22)	<0.0001
Cause of Death							0.032
Cerebrovascular/Stroke	209	47%	236	53%	-
Head Trauma	156	56%	121	44%	0.69	(0.51, 0.93)	0.018
Other2	34	45%	42	55%	1.09	(0.67, 1.79)	0.72
Mechanism of Death							0.053
Blunt Injury	110	58%	81	42%	-
Intracranial  Hemorrhage/Stroke	212	48%	234	52%	1.50	(1.07, 2.11)	0.020
All other categories3	68	47%	76	53%	1.52	(0.98, 2.35)	0.06
Donation after cardiac death (DCD)
No	383	49%	395	51%	-
Yes	19	68%	9	32%	0.46	(0.20, 1.00)	0.058
History of Diabetes
No	390	49%	398	51%	-
Yes	10	63%	6	37%	0.59	(0.20, 1.60)	0.31
Insulin Dependent
No	5	45%	6	55%	-
Yes	4	100%	0	0%	NA	NA	NA
History of Hypertension
No	257	51%	250	49%	-
Yes	143	49%	148	51%	1.06	(0.80, 1.42)	0.67
Heavy Alcohol Use
No	287	55%	233	45%	-
Yes	65	55%	54	45%	1.02	(0.68, 1.53)	0.91
History of Cigarette Use
No	241	49%	248	51%	-
Yes	158	51%	152	49%	0.94	(0.70, 1.24)	0.64
Cocaine Use
No	354	50%	350	50%	-
Yes	42	47%	47	53%	1.13	(0.73, 1.77)	0.58
Other Drug Use
No	298	50%	293	50%	-
Yes	95	48%	104	52%	1.11	(0.81, 1.54)	0.51

¹Odds ratios, confidence intervals and p-values were calculated using univariate logistic regression. Dashes indicate baseline category.

²Odds ratio and 95% CI for continuous variables were reported between 75^th and 25^th percentiles.

³Other reported causes of death include anoxia (n=65), CNS tumor (n=7), bacterial meningitis (n=1), cardio-pulmonary arrest (n=1), Cerebral Edema (n=1), and spontaneous cranial bleed (n=1).

⁴Other mechanisms of death include gunshot wound (n=72), cardiovascular (n=36), drug intoxication (n=12), asphyxiation (n=8), natural causes (n=7), drowning (n=5), and seizure (n=4).

Overall organ function and the number and type of medication given to the donor just prior to donation affected isolation outcome (Table 2). An increase in the number of medications (irrespective of their type) administered simultaneously to the donor less than 24 hours prior to cross clamp reduced the odds of obtaining a high yielding isolation (OR=0.87 per an increase of 7 medications, p=0.027). When medications were classified into groups according to mechanism of action, pancreata from donors receiving hormonal medications were less likely to result in high yielding isolations (OR=0.61, p=0.002). Moreover, pancreata from donors requiring fluid and electrolyte resuscitation showed a decrease in the likelihood of a successful outcome (OR=0.58, p=0.006). A list of medications for each of the 6 groups is provided as Supplemental Table 2. Lastly, pancreata from donors with amylase levels of ≤120 u/L were more likely to result in high yielding isolations (OR=1.51, p=0.016).

Table 2. Influence of Donor Medications and Organ Function on Islet Isolation Outcome.

Name	≤220,000 IEQs		≥315,000 IEQs		Odds Ratio1	95% CI	p-value
	N	% or Mean (±1SD)	N	% or Mean (±1SD)
Number of Donor Man Meds Received2	402	6.0 (±3.4)	404	5.5 (±3.2)	0.87	(0.76, 0.98)	0.027
Anesthetics Received3
No	362	50%	363	50%	-
Yes	40	49%	41	51%	1.02	(0.65, 1.62)	0.93
Antibiotics Received3
No	232	48%	248	52%	-
Yes	170	52%	156	48%	0.86	(0.65, 1.14)	0.29
Cardiovascular Meds Received3
No	48	44%	61	56%	-
Yes	354	51%	343	49%	0.76	(0.51, 1.14)	0.19
Hormonal Meds Received3
No	97	41%	138	59%	-
Yes	305	53%	266	47%	0.61	(0.45, 0.83)	0.002
Fluids/Electrolytes Received3
No	326	48%	356	52%	-
Yes	76	61%	48	39%	0.58	(0.39, 0.85)	0.006
Other Meds Received3†
No	70	43%	91	57%	-
Yes	332	51%	313	49%	0.73	(0.51, 1.03)	0.07
Total Number of Donated Organs4							0.98
1-2	12	50%	12	50%	-
3-4	272	50%	276	50%	1.02	(0.44, 2.32)	0.97
5-6	118	50%	116	50%	0.98	(0.42, 2.30)	0.97
Kidney
Creatinine >1.0 mg/dL	74	51%	72	49%	-
Creatinine ≤1.0 mg/dL	324	50%	327	50%	1.04	(0.72, 1.49)	0.84
Liver Test 1
Total Bilirubin >1.0 mg/dL	64	49%	67	51%	-
Total Bilirubin ≤1.0 mg/dL	330	50%	328	50%	0.95	(0.65, 1.38)	0.79
Liver Test 2
SGOT >36 u/L	220	53%	199	47%	-
SGOT < 36 u/L	174	47%	198	53%	1.26	(0.95, 1.67)	0.11
Liver Test 3
SGPT >27 u/L	243	53%	219	47%	-
SGPT < 27 u/L	151	46%	178	54%	1.31	(0.99, 1.74)	0.06
Pancreas Test 1
Lipase >50 u/L	124	51%	117	49%	-
Lipase ≤50 u/L	245	49%	252	51%	1.09	(0.80, 1.48)	0.58
Pancreas Test 2
Amylase >120 u/L	109	58%	80	42%	-
Amylase ≤120 u/L	267	48%	295	52%	1.51	(1.08, 2.10)	0.016

¹Odds ratios, confidence intervals and p-values were calculated using univariate logistic regression. Dashes indicate baseline category.

²Represents a total for the number of different medications administered to the donor less than 24 hrs prior to cross clamp. Odds ratio and 95% CI for continuous variables were reported between 75^th and 25^th percentiles.

³Categories are not mutually exclusive and represent re-grouping by the authors to reduce 295 medications into manageable classes based on drug mechanism of action. See Supplemental Table 2.

^†Medications with broad action placed into this category

⁴Organs that were donated included kidney (left, right, or both), lung (left, right, or both), pancreas, heart, liver, and intestine.

Once the pancreas was recovered, factors related to morphology and surgical handling of the organ were also shown to influence isolation outcome (Table 3). The type of solution used to preserve the pancreas following organ procurement was found to be statistically significant (overall p=0.0023). Although this difference was not seen between the use of a University of Wisconsin (UW) alone vs. two-layer method (TLM; p=0.39), the use of Histidine-Tryptophan-Ketoglutarate (HTK) reduced the odds of obtaining a high yielding isolation (OR=0.42, p=0.0005). In addition, pancreata classified as edematous and those with prolonged CIT greatly reduced the odds of obtaining a high-yielding isolation (OR=0.49, p=0.003; OR=0.82 per increase of 4.9 hrs, p=0.012; respectively). Relative to pancreata with clean or light surface fat, those with moderate or heavy surface fat were nearly twice as likely to result in high-yielding isolations (OR=1.75, p=0.0006). Likewise, pancreata with moderate or heavy fat infiltration were more likely to result in high yielding isolations relative to those organs with no or patchy fat (OR=1.63, p=0.002). An increase in the pre-distention weight of the pancreas also improved the odds of obtaining high yields (OR=1.55 per increase of 38.5 g, p<0.0001).

Table 3. Pancreas Factors Tested for Contribution to Islet Isolation Outcome.

Name	≤220,000 IEQs		≥315,000 IEQs		Odds Ratio1	95% CI	p-value
	N	% or Mean (±1SD)	N	% or Mean (±1SD)
Pancreas Preservation Solution Used							0.0023
UW Alone	221	46%	263	54%	-
Two-layer	100	49%	103	51%	0.87	(0.62, 1.20)	0.39
Histidine-Trypophan-Ketoglutarate	56	67%	28	33%	0.42	(0.26, 0.68)	0.0005
Pancreatic Surface Fat
Clean or Light	129	58%	94	42%	-
Moderate or Heavy	211	44%	269	56%	1.75	(1.27, 2.42)	0.0006
Fatty Infiltration
None or Patchy	190	56%	147	44%	-
Moderate or Heavy	140	44%	176	56%	1.63	(1.19, 2.22)	0.002
Pre distention pancreas weight3 (g)	402	95.5 (±31.5)	404	106.7 (±31.8)	1.55	(1.30, 1.85)	<0.0001
Organ Intact
No	44	58%	32	42%	-
Yes	347	48%	370	52%	1.47	(0.91, 2.38)	0.12
Organ Damage
No	310	48%	331	52%	-
Yes	78	55%	65	45%	0.78	(0.54, 1.12)	0.18
Organ Edema
No	313	47%	352	53%	-
Yes	55	65%	30	35%	0.49	(0.30, 0.77)	0.003
Pancreas Procurement Team
Remote	267	50%	268	50%	-
Local	110	48%	121	52%	1.10	(0.81, 1.50)	0.53
Cold Ischemia Time2† (hrs)	374	8.7 (±4.8)	376	7.7 (±4.8)	0.82	(0.70, 0.95)	0.012

¹Odds ratios, confidence intervals and p-values were calculated using univariate logistic regression. Dashes indicate baseline category.

²Odds ratio and 95% CI for continuous variables were reported between 75^th and 25^th percentiles.

^†Cold Ischemia time was defined as the time from cross clamp to the start of dissection by the pancreas processing laboratory.

No technical isolation factors that were found to significantly improve the odds of obtaining a high yielding isolation(Table 4).

Table 4. Influence of Pancreas Processing Factors on Isolation Outcome.

Variable Name	≤220,000 IEQs		≥315,000 IEQs		Odds Ratio1	95% CI	p-value
	N	%, Mean (±1SD), or Median (min, max) 2	N	%, Mean (±1SD), or Median (min, max) 2
Collagenase Provider							0.10
Roche	272	48%	293	52%	-
Serva	88	57%	66	43%	0.70	(0.49, 1.00)	0.048
Sigma	21	57%	16	43%	0.71	(0.25, 1.38)	0.31
Pre-distention pancreas temp3 (°C)	41	2.1 (0.2, 15.2)	26	2.1 (0.4, 6.7)	0.71	(0.36, 1.25)	0.28
Post-distention pancreas temp3 (°C)	14	5.1 (3.0, 32.0)	12	4.0 (4.0, 14.4)	0.79	(0.32, 1.53)	0.51
Digestion Time3* (mins)	397	19.3 (±10.4)	401	18.7 (±10.4)	0.95	(0.84, 1.07)	0.40
Dilution Times3** (mins)	303	37.5 (±18.6)	307	39.9 (±19.6)	1.14	(0.97, 1.36)	0.12
Total2 (Digestion and Dilution; mins)	303	56.9 (±17.1)	307	59.1 (±18.9)	1.14	(0.96, 1.35)	0.14
Purification Method							0.21
Continuous	340	49%	351	51%	-
Discontinuous	49	58%	36	42%	0.71	(0.45, 1.12)	0.14
Both	10	40%	15	60%	1.45	(0.65, 3.38)	0.37
Type of Gradient Used							0.22
Bicoll (Ficoll)	273	49%	288	51%	-
Euroficoll	33	56%	26	44%	0.75	(0.43 1.28)	0.29
Optiprep	69	57%	53	43%	0.73	(0.49, 1.08)	0.11
More than one of the above	11	37%	19	63%	1.64	(0.78, 3.62)	0.20
Other	13	45%	16	55%	1.17	(0.55, 2.51)	0.69

¹Odds ratios, confidence intervals and p-values were calculated using univariate logistic regression. Dashes indicate baseline category.

²Number and % are reported for all categorical variables. Use of mean or median for continuous factors based on if variable was normally distributed.

³Odds ratio and 95% CI for continuous variables were reported between 75^th and 25^th percentiles.

^*Also known as Phase I or digestion phase. Begins when the collagenase is first circulated through the pancreas digestion chamber.

^**Also known as Phase II or dilution phase. Begins with influx of albumin into pancreas digestion chamber to stop enzymatic action of collagenase.

Multivariable Analysis

Variables selected for further analysis were assessed using a multivariable logistic regression model to determine if a multi-factorial influence on isolation outcome existed. Although history of diabetes had been found to be significant in 2 or more previous studies, it was not included in the multivariable model for testing due to small sample response size. Additionally, mechanism of death was found to be collinear with cause of death and excluded from testing; likewise, weight was found to be collinear with BMI and also excluded from multivariable model testing.

The final model yielded a Hosmer-Lemeshow p-value of 0.42 indicating a good fit of the model (Table 5). Several univariate factors that improved the odds of obtaining high yielding isolations also increased the odds when examined in a multivariable model, including elevated BMI (OR=2.26, p<0.0001), normal liver SGPT levels (OR=1.59, p=0.020), and pancreas fat infiltration (OR=1.81, p=0.003). Likewise, extended CIT (OR=0.77, p=0.016), the use of fluids/electrolyte medications (OR=0.55, p=0.032), pancreata from elderly donors (OR=0.57, p=0.005), and pancreas edema (OR=0.47, p=0.011) were all found to decrease the odds of obtaining a high yielding isolation. Use of an organ preservation solution other than UW alone and TLM also decreased the odds of obtaining a high yield isolation in the univariate and multivariable analyses (OR=0.31, p=0.0007).

Table 5. Multivariable logistic regression model of factors influencing islet isolation success.

Variable	N	Odds Ratio1	95% CI	p-value
BMI 2	526	2.26	(1.72, 3.01)	<0.0001
Cold Ischemia Time 2	526	0.77	(0.62, 0.95)	0.016
Fluids/Electrolytes Received
No	448	-
Yes	78	0.55	(0.31, 0.94)	0.032
Pancreas Function
Amylase > 120 u/L	128	-
Amylase < 120 u/L	398	1.48	(0.95, 2.31)	0.08
Liver Function
SGPT > 27 u/L	319	-
SGPT ≤ 27 u/L	207	1.59	(1.08, 2.35)	0.020
Age
0.1 – 47.5	270	-
>47.5 – 70.8	256	0.57	(0.39, 0.84)	0.005
Organ Edema
No	456	-
Yes	70	0.47	(0.26, 0.83)	0.011
Fatty Infiltration
None or Patchy	271	-
Moderate or Heavy	255	1.81	(1.22, 2.69)	0.003
Organ Intact
No	58	-
Yes	468	1.95	(1.07, 3.62)	0.031
Pancreas Preservation Solution Used				0.003
UW Alone	328	-
Two-Layer	143	0.94	(0.60, 1.46)	0.77
All Other Solutions	55	0.31	(0.15, 0.59)	0.0007

¹Odds ratios, confidence intervals and p-values were calculated using multivariable logistic regression. Dashes indicate baseline category.

²Odds ratio and 95% CI for continuous variables were reported between 75^th and 25^th percentiles.

However, there were three variables that were found to be significant in only one form of the analysis. First, an intact pancreas was not shown to be significant in the univariate analysis, but nearly doubled the odds of obtaining a high yielding isolation in the multivariable model (OR=1.95, p=0.031). Second, amylase levels below 120 U/L was found to be an important factor in the univariate analysis, but failed to show any statistically significant influence in obtaining high yields in the multivariable model (OR=1.48, p=0.082). Finally, center effect (see Supplemental Table 3) was found to be significant in the univariate analysis (overall p<0.0001), for only 4 of the 13 isolation laboratories compared to the baseline facility, but failed to show any statistical significance in the multivariate model (overall p=0.09).

Discussion

In this study, we analyzed datasets made available to us through the ICR and UNOS, including a subset of factors reported in the literature over the past 15 years. Our objective was to identify variables associated with a dichotomized outcome of islet isolation by using a unique and extensive collection of heterogeneous data generated by 14 laboratories.

Several factors were shown to consistently influence post-purification isolation outcome in both the univariate and multivariable analysis. Our study is the first to show that when grouping donor medications by mechanism of action, fluid/electrolyte administration less than 24 hours prior to cross-clamp negatively impacts isolation outcome. We did not find any collinear variables that might help to explain this observation, nor was there any statistically significant interaction terms. The use of hormonal medications likewise negatively influenced isolation outcome, although only univariately. In a related, but earlier study, Fiedor et al. found that the hemodynamic and metabolic status of the patient were variables significantly associated with islet yield, but not fluid intake(31). Nonetheless, it is possible that certain combinations of drugs, beyond our broad groupings, have synergistic or antagonistic effects yet to be identified.

Pancreata from older donors were found to negatively impact isolation outcome. These findings complement an earlier report, where islet recovery and purity both improved when using pancreata from younger donors(8), but contradict later studies in which pancreata from older donors resulted in high yielding isolations vs. those from younger donors or at the extremes of life(6, 7, 9, 10). It is hypothesized that the collagen composition of the extracellular matrix of the pancreas changes with age, thus impacting collagenase affinity during the digestion phase of the isolation process(6, 27). This may partially help to explain why no clear consensus has emerged as to ideal age boundaries for optimal donors.

We also observed a detrimental impact of CIT on isolation yield, along with a progressive decline in the odds of success when pancreata were subject to prolonged cold storage periods. This observation is in agreement with previous studies in which extended CIT inversely correlated with organ quality, post-isolation yield and graft function(6-9). Furthermore, the type of solution used to preserve the pancreas during transportation was also identified as a significant variable associated with isolation outcome. There is substantial controversy regarding pancreas preservation due to the lack of agreement in the literature and the costs associated with each method. In 2002, a series of separate studies indicated that the use of TLM for pancreas preservation was superior to UW alone and improved post-isolation islet recovery(21, 22, 26). However, recent reports challenge these findings and indicate that there is no real benefit in using TLM preservation (20, 24, 51). In our study, we found that the TLM does not significantly change the odds of obtaining a high-yielding isolation when compared to UW alone. However, we did find a significant negative correlation between isolation success and the use of HTK when compared to UW alone. Data available on pancreas preservation for islet isolation with HTK is scarce, but Brandhorst et al. reported in 1995 that preservation with UW and HTK was equivalent for the purpose of islet isolation(19), a finding that was later reproduced by Salehi et. al in a study analyzing data from 96 human isolations(52). Conversely, Stewart et. al recently reported a large multicenter experience and concluded that pancreas preservation with HTK for solid organ pancreas transplant was associated with decreased graft survival and early graft loss(53), further supporting our observation that HTK preservation was inferior to both UW and TLM for islet isolation purposes.

Pancreata from high BMI donors were found to positively influence isolation outcome. This observation is in universal agreement with previous reports showing that isolations using pancreata from overweight and obese donors are consistently superior in yield(7-11, 13, 18, 35, 54). Moreover, we found that an increase in both donor and pancreas weight positively influenced isolation outcome univariately. Interestingly, although pancreas weight is not known prior to recovery, others have established a direct relationship between donor and pancreas weight, using it to estimate organ weight from donor information for pancreas selection purposes (12).

A number of related factors influenced post-isolation outcome in the multivariable analysis, but not consistently shown to do so univariately. Of note, we found the use of intact pancreata and those containing moderate to heavy surface fat or with moderate to heavy fat infiltration increases the likelihood of isolation success, but the presence of organ edema to negatively impact outcome. With the exception of organ edema, these observations are in agreement with previous studies(7, 9, 14, 54). Pancreas edema has been shown to be associated with increased post-purification yield and purity, affecting outcome by enhancing enzymatic action during the digestion phase(55); however, our data show that the presence of edema diminished by half the probability of a favorable outcome, an observation supported by Mahler et. al. who found that fibrotic and edematous pancreata result in inferior post-isolation yields(56). This might be explained by the observed relationship between pancreas edema and sub-optimal preservation, as well as inadequate collagenase diffusion after intraductal distention in the edematous pancreas(23). Additionally, normal liver and pancreas function tests were also found to positively influence isolation outcome. Although these results are in general agreement with available data(7, 57), our study is the first to show that both normal liver and pancreas function tests are important in isolation outcome within a multivariable model.

Despite being cited as relevant elsewhere, we were not able to confirm the relationship to outcome for several variables of interest. Contrary to previous reports(6, 9, 18, 58), we failed to see an effect on outcome for pancreata recovered using a local vs. remote procurement team. Furthermore, we did not find that variables relating to the donors medical history to be significant, although noted as important elsewhere(7, 18) Moreover, it was striking that none of the pancreas processing variables had an effect on outcome. In our study the type of enzyme used, digestion period, and purification methods did not modify the odds of obtaining a high yielding isolation, despite being extensively described as being significant in other publications (9, 10, 14, 18, 27). Although manufacturer lot number was available for nearly all isolation records (n=789 of 806), data on enzymatic activity (n=0), collagenase (n=77) and neutral-protease (n=43) concentrations were limited. Additionally, beyond the standard certificate of analysis, there is no consensus on biochemical assays proposed to characterize the collagenase and neutral-protease product(59, 60), thus limiting the collection and analysis of this kind of data in a multi-center setting.

Our results indicated that the use of DCD donors negatively impacts isolation outcome, a noteworthy, but marginally significant, finding given that such pancreata have been successfully used for PIT(42). Closely related to DCD status, cause of death was found to be a determinant to isolation success, as donors who were declared brain-dead following traumatic injury were less likely to result in a successful isolation. Brain death is characterized by severe hemodynamic, metabolic and immunological alterations that compromise adequate perfusion and oxygenation of abdominal organs. O'Gorman et al. described a similar association between cause of death, especially with abdominal injury, and a decreased score in their screening process leading to a reduced possibility of a successful outcome (7). We also observed that statistical significance for mechanism of death closely mirrored cause of death. The main difference in the analysis was the reference group used for baseline comparison. Although the findings were complementary, the definitions for each category within a variable were slightly different (see Supplemental Table 1) and may help explain the difference in statistical significance.

Center effect was not a significant variable in the multivariable model. Although we considered analyzing other factors that might account for center effect (such as the number of years in operation, years of staff/technician experience, and turnover rate) such data was not available to us for examination and precluded a more extensive analysis.

Post-purification IEQ count was selected as the outcome since it was documented in nearly all records analyzed and has been regarded as the most important criterion for transplant suitability, along with preparation purity and viability. However, islet quantity does not necessarily reflect preparation quality, limiting the significance of our data in the transplant scenario. Additional research is needed in order to link donor and organ characteristics to in-vivo islet function, and avoid confounding variables related to the isolation procedure itself. Therefore, until better quality indicators are tested, validated, and commonly reported, islet yield remains the single best and most widely available metric to determine isolation outcome.

In summary, this multi-center analysis reveals that non-edematous intact pancreata with moderate to heavy fat infiltration from younger, obese, hemodynamically stable donors with normal liver function increase the odds of achieving post-purification islet isolation success. Minimization of CIT and the use of UW solution or the TLM, but not HTK, for pancreas preservation also improve the odds of isolation success. The data derived from this study serves as a reference, along with established methodologies(61) and regulatory guidelines(62-64), to help screen candidate donors, potentially lowering overall pancreas processing costs, maximizing efficient use of limited resources, and improving islet extraction and transplantation rates which may ultimately lead to superior post-transplant outcomes.