Table 3.
Domains of performance of aortic dissection risk prediction models.
| Author (year) | Missing data handling | Continuous variable processing method | Variable selection | Model development method | Calibration method | Validation method | Model performance | Model presentation |
|---|---|---|---|---|---|---|---|---|
| Cai, 2025 | Random forest algorithms was applied to variables with <20% missing data, whereas those with a proportion exceeding 20% were excluded. | Continuity | LASSO Cox regression analysis, univariate analysis and correlation analysis | SVM | Brier score | Hold-out validation, 10-fold cross-validation and external temporal validation | A: 0.914 (0.981, 0.920) B1: 0.853 (0.850, 0.862) B2: 0.877 (0.870, 0.898) | Risk score |
| Chen, 2023 | Excluded | Continuity | LASSO | LR, KNN, RF, GBM, SVM, MLP | Brier score and Hosmer-Lemeshow test | Hold-out validation | B: 0.828 (0.585, 0.902) | NI |
| Chen, 2021 | Features with missing values more than 20% were excluded. Missing data were assumed to be missing at random and were imputed using 10-fold multiple imputation by chained equations. | Continuity | Kendall correlation coefficient | NB, LR, DT, RF, GBDT | NI | 5-fold cross-validation | A: 0.991 (0.978, 1.000) B: 0.837 (0.766, 0.908) | NI |
| Chen, 2025 | MissForest imputation method was applied to variables with <10% missing data, whereas those with a proportion exceeding 10% were excluded. | Continuity | Univariate and multivariate binary logistic regression analyses, DT, RF, XGBoost, SVM | LR, DT, RF, XGBoost, SVM | Brier score, calibration curve and Hosmer-Lemeshow test | Hold-out validation | A: 0.842 (0.780, 0.910) B: 0.782 (0.698, 0.860) | Web-based calculator |
| Chen, 2025 | Variables with more than 20% missing data were excluded from further analysis. For variables with missing data below this threshold, multiple imputation was employed to minimize bias. | Continuity | NI | GBM, LightGBM, RF, KNN, MLP-NN, NB, LR | Brier score, calibration curve | 10-fold cross-validation | B: 0.874 (0.831, 0.918) | Web-based calculator |
| Dai, 2023 | K-nearest neighbor (KNN) method was applied to fill in the missing data. | Continuity | LR, LASSO | LR with L2, RFC, SVM, XGBoost | Brier score | 10-fold cross-validation | B: 0.800 (0.683, 0.917) | NI |
| Dong, 2021 | Excluded | Continuity | LASSO | XGBoost, AdaBoost, NB, LR, RF, KNN, K-SVM, BPNN | Calibration curve | 5-fold cross-validation and bootstrapping | A: 0.848 (0.720, 0.975) B: 0.802 (0.689, 0.915) | Nomogram |
| Guo, 2021 | Imputed values (which were combined using Rubin's rules) were used to impute variables with <10% missing data, whereas those with a proportion exceeding 10% were excluded. | Continuity | NI | LR, DT, k-NN, GNB, XGBoost | NI | 10-fold cross-validation | B: 0.927 (0.860, 0.968) | NI |
| Guo, 2022 | Multiple imputation was applied to variables with <20% missing data, whereas those with a proportion exceeding 20% were excluded. | Continuity | LASSO | COX, NN | Brier score | 10-fold cross-validation | A: 0.912 (0.866, 0.957) B: 0.899 (0.824, 0.975) | Nomogram |
| He, 2025 | Multiple imputation was applied to variables with <10% missing data, whereas those with a proportion exceeding 10% were excluded. | Continuity | CoxBoost, StepCox, RSF, LASSO | LASSO, Ridge, ElasticNet, StepCox, RSF, SuperPC, plsReox, GBM, CoxBoost, survival-SVM | Calibration curve | Hold-out validation | B1a: 0.826 (0.760, 0.891) B1b: 0.815 (0.758, 0.871) | Web-based calculator |
| Jiang, 2023 | Multiple imputation was applied to variables with <10% missing data, whereas those with a proportion exceeding 20% were excluded. | Continuity | Backward stepwise regression, LASSO, Best subset regression, RF, AdaBoost, Weighted k-NN, SVM, NN | LR, LASSO, BSS, RF, AdaBoost, Weighted k-NN, SVM, NN | NI | 5-fold cross-validation | B: 0.841 (0.771, 0.911) | Nomogram |
| Jin, 2025 | We addressed missing values through multiple imputation | Continuity | LASSO | LR, SVC, RF, XGBoost, NB, MLP | Calibration curve and Brier score | 5-fold cross-validation | A: 0.888 (0.887, 0.889) B: 0.797 (0.794, 0.800) | NI |
| Jin, 2025 | Variables with >10% missing data were excluded before further analysis. We subsequently addressed missing values in clinical variables identified during the model-building process using multiple imputation | Continuity | Univariate and multivariate analyses | LR, deep learning | Brier score | 5-fold cross-validation and external temporal validation | B2: 0.780 (0.777, 0.785) | NI |
| Lei, 2024 | Excluded | Continuity | Univariate analyses | LR, DT, RF, XGBoost | NI | Hold-out validation and external temporal validation | B1: 0.870 (0.744, 0.996) B2: 0.767 (0.658, 0.876) | Web-based calculator |
| Li, 2025 | Variables with missing data exceeding 30% were excluded. For variables with less than 30% missing data, missing values were imputed using appropriate methods: the mean was used for continuous variables, and the mode was used for categorical variables. | Continuity | LASSO | XGBoost, LR, DT, GNB, MLP, KNN, SVM | Calibration curve | 5-fold cross-validation | B: 0.96 (0.93, 0.99) | SHAP force plot |
| Li, 2025 | Variables with missing data exceeding 30% were excluded. For the remaining missing variables, imputation was performed using the MissForest package in R 4.3.3 to ensure the completeness and accuracy of the dataset. | Continuity | LASSO | RF, KNN, SVM, DT | Calibration curve | 5-fold cross-validation | B: 0.933 (0.840, 0.997) | SHAP dependency plot |
| Li, 2025 | Multiple imputation was applied to variables with <30% missing data, whereas those with a proportion exceeding 30% were excluded. | Continuity | Boruta algorithm | XGBoost, AdaBoost, KNN, LR, LightGBM, GNB, MLP, CNB, SVM | Calibration curve | Hold-out validation | A: 0.960 (0.940, 0.970) B: 0.710 (0.620, 0.800) | SHAP |
| Li, 2024 | NI | Continuity | LASSO and PCA | LR, SVM, RF, GBM, SL | Brier score | NI | A: 0.959 (0.927, 0.992) | NI |
| Li, 2022 | NI | Continuity | Recursive Feature Elimination (RFE) and 5-fold cross-validation | DT, RF, XGBoost, LightGBM, LR | Calibration curve and Brier score | Bootstrap | A1: 0.760 (0.630, 0.881) A2: 0.734 (0.623, 0.847) | NI |
| Lin, 2023 | NI | Made binary using cutoff points | forward step method | LR, CNN, RF, SVM | Calibration curve and Hosmer-Lemeshow test | Bootstrap | B: 0.990 (0.950, 0.990) | NI |
| Liu, 2024 | NI | Continuity | LASSO, SVM-RFE, RF | ANN, LR | Calibration curve | Bootstrap and 10-fold cross-validation | B: 0.916 | NI |
| Lu, 2024 | NI | Continuity | ANOVA and LASSO | XGBoost | Calibration curve | Hold-out validation and external temporal validation | A: 1.000 (1.000, 1.000) B1: 0.990 (0.966, 1.000) B2: 0.985 (0.965, 1.000) | NI |
| Luo, 2025 | NI | Continuity | elastic network, Lasso, Ridge, stepwise Cox, CoxBoost, RSF, SVM, partial least squares regression for Cox, GBM, and supervised principal components | elastic network, Lasso, Ridge, stepwise Cox, CoxBoost, RSF, SVM, partial least squares regression for Cox, GBM, and supervised principal components | Calibration curve and Brier score | Hold-out validation | A: 0.957 B: 0.882 | Web-based calculator |
| Ma, 2024 | NI | Continuity | LR | RF | NI | NI | A: 0.990 (0.966, 1.000) | NI |
| Pang, 2024 | NI | Continuity | LASSO, 10-fold cross-validation and multivariable LR | LR, SVM, KNN, RF, DT, XGBoost | NI | Hold-out validation | A: 0.877 (0.803, 0.952) B: 0.818 (0.641, 0.995) | NI |
| Pei, 2021 | Multiple imputation was applied to handle missing data. | Continuity | stepwise forward selection, LASSO and XGBoost | LR | Hosmer-Lemeshow test | Hold-out validation | B: 0.766 (0.718, 0.734) | Nomogram |
| Song, 2024 | Multiple imputation was applied to variables with <20% missing data, whereas those with a proportion exceeding 20% were excluded. | Continuity | RF | LR | Hosmer-Lemeshow test and calibration curve | Hold-out validation | A: 0.759 (0.735, 0.851) B: 0.856 (0.791, 0.922) | Nomogram |
| Sun, 2024 | Multiple imputation was applied to variables with <10% missing data, whereas those with a proportion exceeding 10% were excluded. | Continuity | LASSO | RSF, COX, SVM, XGBoost, Deepsurv | Brier score | Hold-out validation | A: 0.930 B: 0.756 | Nomogram |
| Wang, 2023 | Variables missing more than 10% of the values were excluded; for variables missing less than 10% of the values, the data were interpolated (5 times interpolation). | Continuity | univariate logistic analysis | XGBoost, LR | NI | Hold-out validation | B: 0.797 | Nomogram |
| Wang, 2023 | Excluded | Continuity | LASSO | LR, NB, DT, CART, Bootstrap | Calibration curve and Hosmer-Lemeshow test | 10-fold cross-validation | A: 0.942 B: 0.852 | Nomogram |
| Wei, 2025 | The data were preprocessed by removing features with more than 20% missing values, and the remaining missing values were added to the dataset using the predictive mean matching method (PMM) for multiple imputations. | Continuity | univariate and multivariate logistic regression analyses | SVM, GBM, NNET, XGBoost, KNN, LightGBM, CatBoost | NI | 10-fold cross-validation | A: 0.876 (0.833, 0.997) B1: 0.723 (0.610, 0.837) B2: 0.712 (0.604, 0.791) | NI |
| Wei, 2025 | The data were preprocessed by removing features with more than 10% missing values, and the remaining missing values were added to the dataset using the predictive mean matching method (PMM) for multiple imputations. | Continuity | univariate and multivariate Logistic Regression (LR) | LR, SVM, GBM, NNET, XGBoost, KNN, AdaBoost, LightGBM, CatBoost | Hosmer-Lemeshow test | 10-fold cross-validation | A: 0.888 (0.849, 0.928) B: 0.763 (0.667, 0.859) | NI |
| Wen, 2025 | For continuous variables, missing values were imputed using the mean if the data were normally distributed or the median if the data were skewed. For categorical variables, missing values were replaced with the most frequent category. If the proportion of missing values for a variable exceeded 30% of the total observations, the variable was excluded. | Continuity | LASSO | MLR, DT, RF, XGBoost, SVM, KNN, LightGBM | Calibration plots | 10-fold cross-validation | A: 0.984 B1: 0.969 B2: 0.990 | Web-based calculator |
| Wu, 2023 | NI | Continuity | NI | LR, DT, RF, XGBoost, SVM | Calibration curve | Hold-out validation | A: 0.926 (0.855, 0.997) | Nomogram |
| Xie, 2024 | NI | Continuity | LASSO | XGBoost, LR, RF, GNB, SVM, KNN | Calibration curve and Brier score | 10-fold cross-validation | A: 0.997 (0.993, 1.000) B: 0.716 (0.668, 0.854) | NI |
| Xie, 2024 | NI | Continuity | LASSO | XGBoost, LR, RF, MLP, SVM, KNN | Calibration curve and Brier score | 10-fold cross-validation | B: 0.975 (0.940, 1.000) | NI |
| Zhang, 2024 | NI | Continuity | LASSO, 10-fold cross-validation | LR, Treebag, GBM, AdaBoost | Calibration curve, Brier score | 5-fold cross-validation | B: 0.910 (0.841, 0.962) | Web-based calculator |
| Zhang, 2025 | ELM was employed to replace the missing data. | Continuity | Pearson correlation coefficients, XGBoost | XGBoost, LR, SVM, RF | NI | Hold-out validation | B: 0.869 | NI |
| Zhang, 2025 | Multiple imputation was applied to variables with ≤25% missing data, whereas those with a proportion exceeding 25% were excluded. | Continuity | LASSO | XGBoost | NI | Hold-out validation and external temporal validation | A: 0.928 (0.901, 0.956) B1: 0.919 (0.891, 0.949) B2: 0.941 (0.915, 0.967) | Web-based calculator |
| Zhao, 2021 | NI | Continuity | univariate and multivariate stepwise logistic regression analysis, the 10-fold CV LASSO | SVM, RF, NN, DNN | NI | Hold-out validation | A: 0.982 (0.967, 0.997) B: 0.964 (0.932, 0.997) | NI |
ANOVA, analysis of variance; ANN, artificial neural network; BSS, best subset selection; BPNN, back propagation neural networks; CART, classification and regression trees; CNN, convolutional neural network; CNB, categorical naive bayes; DT, decision tree; DNN, deep neural network; ELM, extreme learning machine; GBDT, gradient boosting decision tree; GBM, gradient boosting machine; GNB, Gaussian naive Bayes; KNN, K-nearest neighbors; K-SVM, kernel support vector machine; LASSO, least absolute shrinkage and selection operator; LR, logistic regression; LightGBM, light gradient boosting machine; MLP, multilayer perceptron; MLR, multiple logistic regression; NB, naive bayes; NN, neural network; NNET, neural network; PCA, principal component analysis; PMM, predictive mean matching; RF, random forest; RFE, recursive feature elimination; RSF, random survival forest; Ridge, ridge regression; SL, super learner; SVM, support vector machine; SVC, support vector classifier; SuperPC, supervised principal components; XGBoost, extreme gradient boosting.