Skip to main content

Table 3 Subgroup analysis of model performance and feature importance in living vs. Deceased donor transplantation

From: Development of explainable artificial intelligence based machine learning model for predicting 30-day hospital readmission after renal transplantation

Characteristic

Overall cohort (n = 588)

Living donor recipients (n = 500)

Deceased donor recipients (n = 87)

P-value*

Clinical outcomes:

Readmission rate within 30 days (%)

88.9%

88.4%

92.0%

0.430

Mean hospital length of stay (days)

4.0 ± 4.1

3.9 ± 4.0

4.6 ± 4.6

0.259

Graft rejection episodes

110 (18.7%)

69 (13.8%)

41 (47.1%)

0.000

Model Performance Metrics:

AUC (95% CI)

0.837 (0.802–0.872)

0.787 (0.738–0.836)

0.762 (0.685–0.839)

N/A

Sensitivity

0.388

0.402

0.412

N/A

Specificity

0.72

0.69

0.71

N/A

Accuracy

0.796 ± 0.050

0.778 ± 0.061

0.783 ± 0.058

N/A

Precision

0.629 ± 0.090

0.654 ± 0.082

0.643 ± 0.097

N/A

F1-Score

0.469 ± 0.105

0.453 ± 0.101

0.498 ± 0.112

N/A

Feature Importance (SHAP Contribution %):

Length of hospital stay

20.6%

24.5%

20.1%

N/A

Post-transplant systolic BP

17.2%

21.2%

16.7%

N/A

Pre-transplant BMI

9.8%

2.6%

12.6%

N/A

Pre-transplant diastolic BP

3.7%

1.5%

2.9%

N/A

Post-transplant BMI

3.2%

5.4%

3.3%

N/A

Pre-transplant HbA1c

1.5%

9.9%

3.5%

N/A

Post-transplant eGFR

2.6%

8.8%

2.6%

N/A

  1. Notes: * p-values compare living vs. deceased donor groups using appropriate statistical tests: t-test or Mann-Whitney U test for continuous variables depending on distribution normality; Chi-square or Fisher’s exact test for categorical variables. N/A indicates comparison was not performed for this metric