This retrospective, multicenter, longitudinal cohort study included pediatric kidney transplant recipients enrolled in the CERTAIN registry. The CERTAIN registry collects detailed longitudinal clinical and laboratory data and applies rigorous validity checking procedures ¹ . Participation in the CERTAIN registry was approved by the ethics committee at each center. Informed consent was obtained from the parents or legal guardians prior to enrollment, with assent from patients when appropriate for their age. This general ethics approval for participation in the registry fully covers the use of the collected data for this registry-based study. This general ethics approval for the registry fully covers the use of the collected data for the present study. The time points of data collection, and the corresponding time intervals were as follows: baseline (pre-transplant), at months 1, 3, 6, 9, 12 and every 6 months thereafter up to 5 years post-transplant. One graft per patient was analyzed. A more detailed description of the CERTAIN registry with respect to data quality and completeness can be found in the Supplementary Appendix.

Patients were eligible for the study if they received a kidney transplant before the age of 21 years. Exclusion criteria were early graft nephrectomy within the first week after transplantation, primary hyperoxaluria as primary kidney disease, combined transplants, or early recurrence of focal segmental glomerulosclerosis post-transplant (Figure 1).

Data from both deceased and living donor transplants in children transplanted between March 1997 and August 2022 were included. The study was designed, analyzed and reported according to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines ² . Patients who experienced DGF are referred to as the “DGF group,” and the others as the “no-DGF group”. We focused on risk factors occurring within the first 24 h post-transplant to allow early identification of at-risk patients who might benefit from pharmacologic intervention. The following post-transplant parameters were selected based on evidence from the available literature [9–11] and/or clinical plausibility to be evaluated in recipients during the first 24 h after kidney transplantation surgery: post-transplant complications requiring re-operation, rate of change in recipient’s serum creatinine, systolic and diastolic blood pressure, central venous blood pressure, need for vasopressor therapy, urine production, furosemide therapy, and start of calcineurin inhibitor therapy within the first 24 h after kidney transplantation (Table 1). Parameters not documented in the CERTAIN registry minimum data set were documented in a study-specific electronic case report form designed specifically for this project. In addition, we assessed perioperative parameters specific to this study, namely, “type of donor kidney (left/right)”, “type of arterial vessel anastomosis (aorta/iliac)”, and “type of venous vessel anastomosis (cava/iliac externa/iliac communis/other)” (Table 1). Finally, we developed a DGF risk assessment model based on the analyzed parameters. We report this model according to the guidelines of the Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD) statement (see the Supplementary Appendix) [15]. In a subcohort of 680 patients (n = 56 in the DGF group, n = 624 in the no DGF group), we investigated whether pre-transplant factors that might influence the occurrence of DGF post-transplant could further enhance the predictive accuracy of the developed prediction model (Supplementary Figure 1). The following pre-transplant parameters were included in this extended analysis: pre-transplant dialysis mode, donor type, donor sex, recipient age, duration of cold ischemia time, donor hemodynamic instability (see definition below) and number of HLA-DR mismatches (Supplementary Table 1). No patient was treated following a steroid avoidance protocol.

The variable “complications after kidney transplantation requiring reoperation” included at least one of the following events: thrombosis (DGF group, n = 5; no DGF group, n = 2), bleeding (DGF group, n = 3; no DGF group, n = 4), or vascular stenosis (DGF group, n = 0; no DGF group, n = 2) within 24 h post-transplant. If any of these three complications occurred, the patient was classified as having a post-transplant complication. There were no other reasons for reoperation within the first 24 h post-transplant. “Immediate urine production” was defined as urine production by the graft, either during transplant surgery or in the first 24 h thereafter. Urine production was assessed in n = 552 patients (79.5%) by direct drainage through a catheter stenting the graft ureter (n = 507 patients, 79.8% in the DGF group and n = 45 patients, 76.3% in the no DGF group). Supplementary Table 2 shows the type of graft ureteral stent used. The variable “vasopressor therapy” had four categories, which were “none,” “(nor)epinephrine,” “dopamine,” and “other vasopressors.” If the patient received more than one vasopressor including (nor)epinephrine, they were classified as treated with (nor)epinephrine. If the patient received combination therapy that included dopamine but not (nor)epinephrine, the patient was classified as receiving dopamine. If a vasopressor other than (nor)epinephrine or dopamine was used, the patient was classified in the “other vasopressor” category. In n = 17 patients, the parameter: “other vasopressor” included dobutamine. The variable “rate of change in recipient’s serum creatinine” was mathematically modeled using Formula 1, assuming an exponential decay process: 1 − recipient ′ s   serum   crea   pretransplant   mg dl − recipient ′ s  serum crea at timepoint  T  posttransplant mg dl   recipient ′ s  serum crea  pretransplant mg dl (T) − 1 (1)

Definitions: T = time point of recipient’s serum creatinine measurement within the first 24 h after transplant surgery [hours]. The time point of graft reperfusion after transplantation surgery was defined as T = 0; the time point 24 h after reperfusion was defined as T = 24. Crea = creatinine.

In the subgroup analysis of pre- and post-transplant predictors of DGF the variable “donor hemodynamic instability” was defined as the presence of at least one of the following parameters: absence of spontaneous cardiac activity, occurrence of hypotensive episodes as documented in CERTAIN, treatment of the donor with norepinephrine and/or dopamine.

The risk assessment model was developed using multiple logistic regression based on forward selection of individual variables. Patient and graft characteristics were described by mean and standard deviation for continuous variables and relative and absolute frequencies for categorical variables. First, the peri- and post-transplant parameters assessed in this study were compared between DGF and no-DGF patients using univariable t-tests or chi-squared tests, as appropriate. Second, these parameters were analyzed using the variable selection process. To develop the risk assessment model, patients were randomly assigned to a training and a test cohort in a 2:1 ratio to ensure equal prevalence of the outcome „occurrence of DGF”. Multiple imputation was then performed on the training data. Imputation was based on the fully conditional specification method of Van Buuren et al. [16] and performed with the “mice” package (using 20 imputations). Forward selection was applied to the imputed data sets for logistic regression modeling. Forward selection based on 1000-fold subsampling was performed on 63.2% (1-1/e) of the cohort for each imputed data set. Variables selected for the final model had to be both selected in all imputed data sets for the full training cohort and also to be confirmed in the subsampling process. The final model was built by pooling the respective sub-models derived from the multiple imputation process. To control for interdependence of model parameters and to provide transparency following the guidelines of the TRIPOD statement [15], we calculated adjusted odds ratios for each parameter selected for the model. Internal validation of the final model was performed (i) only on complete cases with no missing data in the test cohort, (ii) on each imputed dataset (same procedure as for the training cohort), and (iii) on a combined multiple imputed data set, which was derived by averaging. Validation was assessed by the area under the receiver operating characteristic curve (ROC-AUC). Model calibration was assessed using calibration plots. All analyses were performed with R version >4.0.0.

We screened n = 816 patients. After applying the exclusion criteria, n = 694 children remained for further analysis (Figure 1). The overall rate of DGF in the entire patient cohort was 59 out of 694 patients (8.5%); the rate of DGF after deceased donation was 11.8% (55 out of 468 patients), after living donation 1.77% (4 out of 226 patients) (P < 0.001). Table 2 shows the baseline characteristics of patients with and without DGF and the corresponding donor characteristics. Patients with DGF were more likely to be male, less likely to have at least one HLA-DR mismatch, and less likely to have undergone preemptive transplantation. 93.2% of patients with DGF had received a transplant from a deceased donor and 6.8% from a living donor. Cold ischemia time in patients with DGF (821 ± 360 min) was 34.6% longer (P < 0.001) than in patients without DGF (610 ± 424 min). Supplementary Table 3 shows the timepoint of initiation of post-transplant dialysis within the first 7 days after transplant surgery in the DGF cohort.

Table 1 shows the analyzed peri- and post-transplant risk factors for DGF. Post-transplant complications requiring reoperation were significantly more frequent in the DGF group (12%) than in the no-DGF group (1%, P < 0.001). The rate of decline in recipient’s serum creatinine (calculated by Formula 1, see above) during the first 24 h post-transplant was significantly slower in the DGF group (−0.008 ± 0.025 = 0.8% ± 2.5% decline per hour) than in the no-DGF group (−0.064 ± 0.092 = 6.4% ± 9.2% decline per hour; P < 0.001). The frequency of vasopressor use was significantly different between the DGF and the no-DGF groups. Norepinephrine and epinephrine were used more frequently in the DGF group (47%) than in the no-DGF group (21%, P < 0.001), whereas dopamine was used more frequently in the no-DGF group (21%) than in the DGF group (16%, P < 0.001). Immediate urine production was significantly more frequent in the no-DGF group (83%) than in the DGF group (31%, P < 0.001). Several measures of central venous pressure were significantly higher in the DGF group than in the no-DGF group, whereas the measures systolic blood pressure, diastolic blood pressure, and mean arterial pressure (MAP) were comparable between groups (Table 1).

The total patient population was divided 2:1 into a training cohort (n = 466 patients) and a validation cohort (n = 228) (Figure 1). The DGF risk assessment model was trained in the training cohort using logistic regression analysis on the parameters listed in Table 1. The following combination of four key post-transplant parameters was selected for the DGF risk assessment model by the forward selection process: (i) occurrence of post-transplant surgical complications, (ii) immediate urine production, (iii) rate of change in recipient’s serum creatinine, (iv) initial calcineurin inhibitor therapy (Table 3). In addition, the significance of the association of each of these four post-transplant parameters with the occurrence of DGF was confirmed by calculating adjusted odds ratios (Table 3). Post-transplant surgical complications and slower rate of decline in recipient’s serum creatinine were associated with increased odds of DGF, whereas immediate urine production and initial calcineurin inhibitor were associated with decreased odds of DGF.

In the training cohort, the corresponding ROC-AUC of the final model was 0.9043. In a second step, the imputation procedures generated several validation cohorts with imputed data for parameters with missing data. The mean ROC-AUC of the model used on these imputed datasets was 0.8765. After merging the imputed datasets, an imputed validation cohort was generated with a ROC-AUC of 0.9104 in the model. In comparison, the ROC-AUC of the validation cohort without imputed data was 0.878 (Table 4; Figure 2). The corresponding risk of developing DGF based on post-transplant parameters can be calculated with the model using the following formula (Formula 2) together with the coefficients written below and shown in Table 3. Linear Score = 2.316 + Variable  A + Variable  B + Variable  C + 0.233 × V a r i a b l e   D P   DGF = 1 1 + e − L inear   Score   (2)

Definitions: -

P (DGF) = probability of developing DGF predicted by the DGF risk assessment model (range: 0–1)

Abbreviations: DGF, delayed graft function. An online calculator based on Formula 2 will be available on the CERTAIN website at the time of publication of this article.

The calibration of this final pediatric DGF risk assessment model was evaluated using calibration plots. The corresponding calibration plot is shown in Figure 2. The model showed the following performance in terms of calibration: Intercept: 0.47; slope: 0.96. Due to the low prevalence of DGF in the study cohort, there are some deviations at very low and high predicted probabilities. Additionally, the confidence intervals for the intercept (−0.24 – 1.19) and slope (0.51–1.40) are relatively broad. Although the C-statistic of 0.88 indicates that the model has high discriminative power, at very high predicted probabilities (>0.8), the model appears to slightly overestimate the probability of the event (Figure 2).

In the extended analysis of combined pre- and post-transplant risk factors for DGF, we used the same statistical methods and procedures as used in the main analysis (described above). We were able to demonstrate that the inclusion of additional pre-transplant parameters did not improve the predictive performance of the primarily developed prediction model based on post-transplant parameters alone, as identified in the main analysis (see above). Supplementary Table 1 shows all pre- and post-transplant parameters examined in this extended analysis. For the combined model, only the number of HLA-DR mismatches was additionally selected by the forward selection procedure (Supplementary Table 4). The adjusted odds ratio for the selected pre-transplant parameter “HLA-DR mismatches” was not statistically significant (Supplementary Table 4). The AUC (0.8782 in the validation cohort without imputed data) of this combined pre and post-transplant model was comparable to the AUC (0.878 in the validation cohort without imputed data) of the developed post-transplant prediction model. Supplementary Figure 2 shows the calibration plot of the combined pre -and post-transplant pediatric DGF risk prediction model.

Sensitivity analysis after exclusion of the variable ‘initial calcineurin inhibitor therapy: To assess the impact of the predictor ‘start of calcineurin inhibitor therapy within 24 h post-transplant’ on model performance, we repeated the forward selection procedure as in the main analysis. Then, we excluded the CNI variable post-selection and refitted the model. The three remaining predictors were highly significant (p < 0.001 for each parameter) and had regression coefficients comparable to those in the main model (surgical complications: 1.574 vs. 1.694; immediate urine production: −2.049 vs. −2.408; rate of change in serum creatinine: 0.233 vs. 0.233). The corresponding adjusted odds ratios were also comparable (surgical complications: 4.827 [95% CI: 1.147–21.35] vs. 5.443 [1.236–26.21]; immediate urine production: 0.128 [0.055–0.277] vs. 0.09 [0.033–0.211]; rate of change in serum creatinine: 1.263 [1.112–1.462] vs. 1.262 [1.107–1.467]; Supplementary Table 5). The discriminative performance of this three-variable model remained good, closely resembling that of the four-variable model (ROC-AUCs: 0.882 vs. 0.904 in the training cohort; 0.891 vs. 0.878 in the validation cohort with complete cases; a mean of 0.885 vs. 0.877 across imputed validation datasets; 0.917 vs. 0.910 in the merged imputed validation cohort; Supplementary Table 6.