Clinical characteristics

A total of 2080 pediatric patients under the age of 12 underwent liver transplantation (LT) at the Liver Surgery Department of Renji Hospital, which is affiliated with the Shanghai Jiaotong University School of Medicine, between the years 2006 and 2019. To enhance the reliability of our follow-up data, we excluded information from 38 foreign pediatric patients and 10 orphaned children. Consequently, our analysis was based on a cohort of 2032 pediatric patients.

Since 2006, the number of pediatric LTs performed at Renji Hospital has been increasing year by year, especially for living donor LT. From 2017 to 2019, the number of pediatric LTs remained at approximately 400 cases annually, and the number of living donor LT cases increased much more dramatically than the number of deceased donor LT cases over the years (Fig. 1a). Supplementary Fig. 1 depicts the geographic distribution of all recipients who have undergone LT at our center.

Fig. 1

Number of pediatric liver transplantations from 2006 to 2019 at Renji Hospital and total survival rate for the allograft and patient using the Kaplan–Meier curve. The y-axis indicates the graft survival rate, and the x-axis indicates the number of years after liver transplantation (a). Percentages of recipient survival among the total recipients at different years are shown in the table in the middle (b). DDLT deceased donor liver transplantation, LDLT living donor liver transplantation

The details of the donor and recipient characteristics are summarized in Table 1. This study included 2032 pediatric LT recipients with a median [interquartile range (IQR)] age of 8 months (6–17 months) and body weight of 7.7 kg (6.6–10 kg); there were 1067 (52.50%) girls and 965 (47.50%) boys. Overall, 211 (10.40%) patients were assessed as having comorbid heart disease; 728 (35.80%) had portal hypertension; 244 (12.00%) experienced gastrointestinal bleeding; 579 (28.50%) had cholangitis, and 1131 (55.65%) were diagnosed with ascites. According to ultrasound results, the direction of the portal vein flow was toward-liver in 1413 cases (69.5%) and ex-liver in 332 cases (16.4%), with data missing for the remaining 287 cases (14.1%). The operation methods recorded included 100 cases of split LT (4.9%), 276 cases of orthotopic LT (13.6%), and 1656 cases of living donor LT (81.5%). The median (IQR) follow-up time was 2.69 years (1.45–4.27), and the total number of deaths recorded until 2019 was 170 (8.3%).

Table 1 Donor and recipient characteristics for pediatric patients undergoing liver transplantation

The indications for pediatric LT are listed in Supplementary Table 1. Among the 2032 cases, there were 1801 cases of cholestatic liver disease (88.6%) and 1647 cases of biliary atresia (81.1%). There were 154 cases of metabolic diseases (7.6%), 28 cases of tumor diseases (1.4%), 24 cases of retransplantation, 13 cases of acute liver failure, and 12 cases of vascular diseases. All patients received immunosuppressive therapy following the operation, with cyclosporine administered in 282 cases and tacrolimus in 1750 cases.

Postoperative prognosis and univariable and multivariable Cox regression analyses

Among the 2032 patients who underwent pediatric LT, the 1-, 3-, 5-, and 10-year graft survival rates were 93.3%, 90.9%, 89.9%, and 87.3%, respectively. The 1-, 3-, 5-, and 10-year overall survival rates were 94%, 92%, 91%, and 89%, respectively (Fig. 1b).

The results of the univariable and multivariable Cox regression analyses are shown in Table 2. According to the univariable analysis, weight, age, diagnosis, operation type, and all five preoperative comorbidities were selected for the multivariable analysis. Moreover, preoperative laboratory measurements, including serum albumin, serum total bilirubin, prothrombin time, INR, and IL-1β were screened. Other features, including the portal vein flow direction, spleen thickness and diameter, and graft-to-recipient weight ratio along with the BMI of the donors, were also included as independent risk factors in the following analysis.

Table 2 Univariable and multivariable Cox regression analyses for predicting allograft survival after pediatric liver transplantation

The adjusted odds ratios and their precision indicated that lower body weight (< 7.2 kg), age ≥ 10 years, heart disease, cholangitis, direction of portal vein flow, spleen thickness (≥ 27 mm), retransplantation, split-liver transplantation, and higher levels of total bilirubin (≥ 5.3 mg/dL) significantly increased the risk of allograft dysfunction. The survival curves of significant covariates in each group are demonstrated in Supplementary Fig. 2, panels A-I.

Development and validation of the allograft survival after pediatric liver transplantation nomogram

Based on the multivariable Cox regression results listed in the previous section and combined with a stepwise model selection and clinical consideration, 10 variables were selected and used to construct a visualized nomogram model (Fig. 2) to predict the survival rate of allografts. Considering that IL-1β is not a routine preoperative test in some transplantation centers, we also created a model without IL-1β. Each variable has a corresponding score on the point scale axis of the nomogram. The scores of the different variables are presented in Supplementary Table 2. By adding the scores of different variables, the total score of ASPELT can be easily calculated, and the survival probability of a graft can be subsequently determined. The C-indices of the prediction nomogram, ASPELT, for predicting graft survival at 1, 3, and 5 years were 0.776, 0.757, and 0.753, respectively, whereas those of the version without IL-1β (i.e., ASPELT/IL-1β) were 0.774, 0.751, and 0.749, respectively (Supplementary Table 3).

Fig. 2

Nomogram predicting graft survival after a pediatric liver transplantation. To determine the number of points received for each level, a line is drawn upward on each variable axis. Points from every categories were summed together to acquire a total point, which will be marked on the “Total Points” axis. From each point, a line is drawn downward to the survival axes to determine the likelihood of 1-, 3-, or 5-year survival of an allograft. IL interleukin, GS graft survival, LT liver transplantation

The results of internal validation are illustrated by the AUC with the 95% CI and calibration plots. The bootstrap validation exhibited a significant prediction accuracy for both ASPELT and ASPELT/IL-1β. The AUCs for predicting graft survival at 1, 3, and 5 years were 73.27% (95% CI 68.25–77.76%), 67.49% (95% CI 62.27–72.31%), and 60.80% (95% CI 54.87–66.43%), respectively, for ASPELT and 73.44% (95% CI 68.43–77.92%), 67.67% (95% CI 62.58–72.37%), and 61.15% (95% CI 55.25–66.74%), respectively, for ASPELT/IL-1β (Fig. 3a–c). The calibration plots created using bootstrap resampling also suggested that the 1-, 3-, and 5-year predicted survival rates agreed with the actual survival rates (Fig. 4a–c).

Fig. 3

ROC curves for ASPELT, ASPELT/IL-1β, PELD, and Child–Pugh scores. Curves present the scores at (a) 1 year, (b) 3 years, and (c) 5 years posttransplantation. ROC receiver operating characteristic, ASPELT allograft survival after pediatric liver transplantation, IL interleukin, PELD Pediatric End-Stage Liver Disease, AUC area under the receiver operating characteristic curve, CI confidence interval

Fig. 4

Calibration curves for allograft survival after pediatric liver transplantation scores. Calibration curves showing allograft survival at a 1 year, b 3 years, and c 5 years posttransplantation. The nomogram-predicted probability of OS is plotted on the x-axis, actual OS is plotted on the y-axis. Grey line showed the ideal model where predicted probability fully aligned with the actual probability. OS overall survival

Comparison with other prediction models

According to the results of the ROC and decision curve analyses, our prediction model outperformed the existing prediction scoring systems, PELD and Child–Pugh, at all three predicting points (Figs. 3a–c and 5a–c). Figure 5a–c indicates that the use of ASPELT and ASPELT/IL-1β had a higher net benefit than the other two models over a wide range of threshold probabilities.

Fig. 5

Decision curve analysis for ASPELT. The curves show the analyses at a 1 year, b 3 years, and c 5 years posttransplantation. The y-axis is the net benefit, and the x-axis denotes the threshold probability. ASPELT allograft survival after pediatric liver transplantation, PELD Pediatric End-Stage Liver Disease, IL interleukin, GS graft survival

Clinical use

The ASPELT scores showed a normal distribution (Fig. 6a). Based on the final score, the cutoff values to determine the risk of poor allograft outcomes were estimated to be at scores of 146 and 196. Recipients with a score below 146 were assigned to the low-risk group, those with a score of 146–196 to the median-risk group, and those with a score above 196 to the high-risk group (Supplementary Table 4). With an increase in risk, the graft survival rate at 1, 3, 5, and 10 years after the operation decreased significantly (P < 0.001; Fig. 6b). The ASPELT model can also be accessed on the Internet or via a web application at https://aspelt.shinyapps.io/ASPELT/.

Fig. 6

Distribution of ASPELT scores and survival rates based on the ASPELT score. ASPELT allograft survival after pediatric liver transplantation, LT liver transplantation