This study used data extracted from the UK Transplant Registry (maintained by NHS Blood and Transplant), following approval from the Pancreas Advisory Group. All adult (≥18years) patients who underwent SPK transplantation in the UK from 1st April 2007 to 31st March 2019 were included. Patients who received a normothermic-regional perfusion preserved graft, recipients who underwent re-transplantation, recipients who received an induction agent other than Alemtuzumab or Basiliximab, and recipients with missing outcome data were excluded. One centre exclusively uses anti-thymocyte globulin as their induction agent, therefore all patients from this centre were excluded as numbers were too few for any meaningful analysis.

Induction immunosuppression was with either Alemtuzumab or Basiliximab, according to local centre protocols. Maintenance immunosuppression typically comprised tacrolimus, mycophenolate mofetil, and corticosteroids (tapered according to centre-specific protocols), but detailed maintenance regimens were not uniformly available in the dataset.

Demographic and clinical variables included donor and recipient age, sex, ethnicity, blood group, donor type, cytomegalovirus (CMV) status, cause of death, body mass index, creatinine at retrieval, recipient diabetes aetiology, dialysis modality, sensitisation at transplantation, waiting time to transplant, cold ischaemic time, and transplant era (grouped into three time periods by financial year: 2007–2011, 2011–2015, and 2015–2019). HLA mismatches were recorded separately for each of the following loci: A, B, Cw, DR, and DQ. NHS Blood and Transplant defined HLA mismatch groups was also included for comparison (grouped into three categories: Level 1 & 2 due to small numbers of Level 1 transplants, Level 3 and Level 4, Supplementary Table S1).

The primary outcome was pancreas graft survival at 10-year post-transplant. Pancreas graft survival at 1-year post-transplant, and additionally kidney graft survival and patient survival at both horizons were analysed as secondary outcomes. Graft survival was defined as the time from transplant to graft failure, censoring for death with a functioning graft. Patient survival was defined as the time from transplant to patient death. Patients lost to follow-up were censored at the date of last known status.

Acute pancreas graft rejection rates at 3- and 12-month post-transplant were obtained. For follow-up which was ‘Not reported’, it was assumed that no rejection episodes occurred.

Demographic and clinical factors are summarised by induction therapy group to describe the cohort. Categorical variables are presented as counts and percentage, and comparisons between the groups made using Fisher’s exact test or Chi-Squared test. Continuous variables are reported as medians and interquartile range (IQR), and comparisons made using the Student’s t-test.

Kaplan-Meier survival analysis was used to estimate the unadjusted survival for each outcome (pancreas graft, kidney graft and patient mortality). Survival curves were stratified by induction therapy group, and log-rank p-values allowed for comparisons between the groups.

Observations with missing data for the outcome or exposure variables were excluded from the analysis. To retain the sample size when multivariable modelling, missing data for the covariates were addressed using simple imputation. For categorical variables, the most frequently occurring category was imputed while continuous variables were imputed using median values. A comparison of this analysis and a complete-case analysis for the primary outcome was performed.

Separate Cox proportional hazards models were built for each survival outcome (pancreas graft, kidney graft, and patient survival). Each model was stratified by transplant centre to account for centre-specific baseline hazard functions. Stepwise backward selection was used, with variables retained if p < 0.10. Continuous variables were assessed for non-linearity using natural cubic splines and included in that form where appropriate.

A final model was created for each outcome incorporating any covariate found to be significant at any time horizon. The proportional hazards assumption was assessed using cumulative hazard function plots. Model checking involved assessing covariate functional form by Martingale residuals and checking for influential observations using df betas and likelihood displacement. Individual HLA locus models and induction therapy covariates were added to the model to evaluate their significance. The final model for analysis of the primary outcome was assessed for multicollinearity using the Variance Inflation Factor.

Further exploratory analyses of these outcomes at 3 and 5 years were performed to assess consistency with the results at other time horizons; however, these results are not presented in this paper. An interaction term between mismatch at HLA-DQ and induction therapy was included in the multivariable model for the primary outcome to assess potential effect modification. A 1-month landmark survival analysis was performed for the unadjusted pancreas graft survival, excluding any patients who died or were censored within the first month of transplantation.

All analyses were performed using SAS Enterprise Guide version 7.13 (SAS Institute Inc., Cary, NC).

During the study period, 1897 adult patients underwent first-time SPK transplantation. The final analysis cohort consisted of 1705 SPK graft recipients after excluding recipients transplanted with a graft preserved normothermic regional perfusion (n = 26), recipients who received an induction agent other than Alemtuzumab or Basiliximab (n = 142), and recipients with missing outcome data (n = 24).

The donor and recipient demographic factors for the whole cohort and by induction agent are presented in Table 1 (with further data in Supplementary Table S2). Median recipient age was 43 years (IQR 36–49 years), 720 recipients were female (42.2%), and 70 recipients had Type II diabetes (4.1%). At the time of transplantation, 682 recipients were pre-dialysis (40.0%), and 1,019 recipients (59.8%) were on either haemodialysis (n = 627, 36.8%) or peritoneal dialysis (n = 392, 23.0%). In 1,625 recipients (95.3%), sensitisation was less than 85%. The median waiting time to transplantation was 1.16 years (IQR 0.55–1.17 years), and 1,385 recipients received a graft from a brainstem-death donor (81.2%). Alemtuzumab was used as the induction agent in 1,196 recipients (70.1%) compared to 509 recipients who received Basiliximab (29.9%). The proportion of donation after circulatory death (DCD) graft recipients was greater with Alemtuzumab compared to Basiliximab (20.2% vs. 15.5%, P = 0.025).

HLA mismatch distribution varied across loci (Table 2). At the HLA-A locus, 14.8% of recipients were fully matched, 59.8% had one mismatch, and 25.5% had two. For HLA-B, most recipients had either one (47.2%) or two mismatches (48.3%), with complete matches in only 4.5%. HLA-Cw matching was similar: 13.6% had no mismatches, 50.0% had one, and 36.4% had two. A single HLA-DQ mismatch occurred in 58.9%, whereas 34.8% had no mismatches and 6.3% had two mismatches. At the HLA-DR locus, 9.0% were fully matched, while 56.8% and 34.2% had one or two mismatches, respectively. Mismatches were similar across induction agent groups.

The unadjusted estimated pancreas, kidney, and patient survival were high over the study period. Pancreas graft survival was 88.6% (95% CI 87.0%–90.0%) at 1-year post-transplant and 72.7% (95% CI 70.1%–75.0%) at 10 years. Kidney graft survival was estimated to be higher with rates of 96.3% (95% CI 95.3%–97.1%) and 76.7% (95% CI 74.2%–79.1%) at 1 and 10 years respectively. Patient survival followed a similar trend, with estimated survival rates of 97.2% (95% CI 96.2%–97.9%) at 1 year, and 75.3% (95% CI 72.6%–77.8%) at 10 years. Most of the graft losses and deaths occurred in the first year following transplantation, with relative stability thereafter.

Unadjusted pancreas graft survival, based on induction agent, is presented in Figure 1, with some evidence to suggest significantly better 10-year survival (p = 0.012) in the Alemtuzumab group (74.6%, 95% CI 71.5%–77.3%) compared to Basiliximab (68.4%, 95% CI 63.6%–72.6%). However, after excluding graft failures that occurred in the first 30 days post-transplant, there was no evidence of a difference in long term pancreas graft survival between the groups, as shown in Supplementary Table S3.

Multivariable modelling of predictors of pancreas graft survival at 1- and 10-year post-transplant are presented in Table 3. Several factors were associated with pancreas graft survival at 10-year. Increasing donor age and longer cold ischaemic time were significantly associated with a higher risk of graft failure (p < 0.0001 and p = 0.015 respectively), and older recipient age was associated with a reduced risk of graft failure (p < 0.0001). Donor CMV positivity was associated with increased risk of graft loss when compared to CMV negative donors (HR 1.25, 95% CI 1.02–1.53, p = 0.030). A sensitivity analysis for handling missing data is presented in Supplementary Table S4, comparing the 10-year pancreas graft survival model estimates for the simple imputation cohort and a complete-case cohort, but the pattern of results was similar between the cohorts.

Regarding immunological factors, HLA mismatches at most loci were not significantly associated with pancreas graft outcomes. However, one HLA-DQ mismatch was associated with improved long-term pancreas graft survival at 10 years (HR 0.75, 95% CI 0.59–0.94, p = 0.033), compared to zero mismatches. This finding was not observed for two DQ mismatches, however the sample size in this category is small (6.3%) presenting high uncertainty in the true effect. HLA-B mismatch status was identified as a predictor of graft loss at 10-year (global p = 0.008), however the confidence intervals for one and two mismatches crossed 1.0, therefore the true effect remains uncertain. No significant associations were identified for mismatches at HLA-A, -Cw, or -DR. Induction therapy type (Alemtuzumab vs. Basiliximab) was not significantly associated with pancreas graft survival at 10 years when adjusted for other factors.

An interaction term between mismatch at HLA-DQ and induction therapy was included in the multivariable model for 10-year pancreas graft survival to explore effect modification. There was no evidence of effect modification by induction therapy (p = 0.71) (Supplementary Table S5).

After adjusting for confounders, there were no significant associations between HLA mismatches at any locus and pancreas graft survival at 1 year and induction therapy was also not significantly associated with this outcome, as all estimates were consistent with no effect. At 3-month post-transplantation, there was a non-significant difference in the incidence of at least one acute pancreas graft rejection episode when comparing Alemtuzumab with Basiliximab (6.1% vs. 4.0%, p = 0.06). There was no difference when comparing HLA-B mismatch status (p = 0.34) or -DQ status (p = 0.99). Follow-up data for acute rejection rates at 12 months post-transplant was available for 1,544 (91%) patients, and was no difference when comparing Alemtuzumab with Basiliximab (6.8% vs. 7.5%, p = 0.60), HLA-B mismatch status (p = 0.17) or -DQ status (p = 0.30).

Unadjusted kidney graft survival is presented in Figure 2, with no evidence to suggest a difference in 10-year graft survival (p = 0.601) when comparing Alemtuzumab (76.7%, 95% CI 73.6%–79.4%) with Basiliximab (77.0%, 95% CI 72.3%–81.0%). Predictors of kidney graft survival at 1- and 10-year post-transplant are presented in Table 4. Kidney graft survival improved significantly, with the most recent era (2015–2019) associated with the lowest risk of graft failure at 10 years (HR 0.62, 95% CI 0.44–0.87, p = 0.008). Donor age showed a significant non-linear association with graft loss at 10 years (p < 0.0001). Higher donor creatinine at retrieval was associated with increased graft failure risk at 10 years (HR 1.29, 95% CI 1.07–1.56, p = 0.008). Non-white donor ethnicity and female recipient sex were significantly associated with increased risk of graft failure at 10 years (HR 1.52, 95% CI 1.04–2.22, p = 0.031; and HR 1.46, 95% CI 1.16–1.85, p = 0.002, respectively).

Recipient dialysis modality also impacted outcomes, with haemodialysis (vs. no dialysis) significantly associated with worse kidney graft survival at 10 years (HR 1.41, 95% CI 1.09–1.83; p = 0.005). In contrast, peritoneal dialysis was not associated with excess risk when compared to no dialysis at 10 years (HR 0.89, 95% CI 0.64–1.23).

HLA mismatches at all loci were not significantly associated with kidney graft loss at 10 years, when adjusted for other factors. Similarly, induction therapy had no significant impact on kidney graft survival in adjusted models. The same conclusion was observed for 1-year kidney graft survival, as neither the association with HLA mismatch nor induction therapy was significant after adjusting for other factors.

Unadjusted patient mortality is presented in Figure 3, with no evidence to suggest a difference in 10-year patient survival (p = 0.699) when comparing Alemtuzumab (75.5%, 95% CI 72.3%–78.4%) with Basiliximab (75.2%, 95% CI 70.1%–79.5%). Predictors of patient survival at 1- and 10-year post-transplant are presented in Table 5. Increasing recipient age was a predictor of higher mortality risk with significant association at 10 years (HR 1.04, 95% CI 1.02–1.05, p < 0.0001). Cold ischaemic time was significantly associated with increased 1-year mortality (HR 1.13, 95% CI 1.02–1.24, p = 0.020), but not at 10-year. Longer waiting time to transplant also emerged as a significant non-linear predictor of mortality at 1- and 10-year post-transplant (p ≤ 0.033.) Neither transplant era, donor age, donor CMV status, nor dialysis modality were significant predictors of early mortality.

HLA mismatches at all loci and induction therapy regimen were not significantly associated with patient mortality at either timepoint, when adjusted for other factors.