Table 2.
Analysis for the BSI-affected patients treatment research.
| Ref. # | Reference Paper | Dataset | Data Preprocessing/Mechanism | Evaluation Methods/Algorithms | Outcome/Explanation Type |
|---|---|---|---|---|---|
| [21] | Burnham, J.P. et al. (2018) | 430 patients | Chi-squared/Fisher exact test, Student t test /Mann–Whitney U/Global | Multivariate Cox proportional hazards models | Kaplan–Meier curves and p-values/ante-hoc |
| [22] | Beganovic, M. et al. (2019) | 428 patients | Chi-square/ Fisher exact test for categorical variables, and t test/ Wilcoxon rank for continuous variables./Global | Propensity scores (PS) using logistic regression with backward stepwise elimination and Cox proportional hazards regression model. | p-values./ante-hoc |
| [23] | Fiala, J. et al. (2019) | 757 patients | Generalized estimating equations (GEE) and Poisson regression models/Global | Logistic regression models, Cox proportional hazards (PH) regression models | p-value before and after adjustment/ante-hoc |
| [24] | Fabre, V. et al. (2019) | 249 patients | χ2 test and Wilcoxon rank sum test/Local | multivariable logistic regression for propensity scores | Weighted by the inverse of the propensity score and 2-sided p-value/ante-hoc |
| [25] | Harris, P.N.A. et al. (2018) |
391 patients | Charlson Comorbidity Index (CCI) score, multi-variate imputation/Global | Miettinen–Nurminen method (MNM) or logistic regression. | A logistic regression model, using a 2-sided significance level |
| [26] | Delahanty, R.J. et al. (2018) | 2,759,529 patients | 5-fold cross validation/Local | XGboost in R. | Risk of Sepsis (RoS) score, Sensitivity, Specficity and AUROC/post-hoc |
| [27] | Kam, H.J. et al. (2017) | 5789 patients | Data imputation and categorization./Local | Multilayer perceptron’s (MLPs), RNN and LSTM model. | Accuracy and AUROC/post-hoc |
| [28] | Taneja, I. et al. (2017) | 444 patients | Heatmaps, Riemann sum, categories and batch normalization/Global | Logistic regression, support vector machines (SVM), random forests, adaboost, and naïve Bayes. | Sensitivity, Specificity, and AUROC/ante-hoc |
| [29] | Oonsivilai, M. et al. (2018) | 243 patients | Z-score, the Lambda, mu, and sigma (LMS) method. 5-fold cross-validated and Kappa based on a grid search/Global | Decision trees, Random forests, Boosted decision trees using adaptive boosting, Linear support vector machines (SVM), Polynomial SVMs, Radial SVM and k-nearest neighbours (kNN) | Comparison of perfor-mance rankings, Calibration, Sensitiv-ity, Specificity, p-value and AUROC/ante-hoc |
| [30] | García-Gallo, J.E. et al. (2019) | 5650 patients | Least Absolute Shrinkage and Selection Operator (LASSO)/Local | Stochastic Gradient Boosting (SGB) | Accuracy, p-values and AUROC/post-hoc |