Fig 1.
Workflow of the ML-based strategy for preoperative blood preparation.
Abbreviations: ANN = artificial neural network, AUC = Area under the ROC curve, DT = Decision tree, GBC = gradient boosting classifier, MAE = Mean Absolute Error, ML = machine learning, NB = naïve Bayes, NPV = negative predictive value, PPV = positive predictive value, R2 = R-squared, RF = random forest, RMSE = Root Mean Squared Error, SVM = support vector machine.
Fig 2.
Screenshot of the ML-based application for preoperative blood preparation.
The tool is used via QR code or https://neurosxpsu.shinyapps.io/crossmatch/. To use the web application, input the new patient’s parameters and press the red bottom for the number of PRC units, then input the calculated PRC units and press the yellow bottom for the number of FFP units.
Table 1.
The strategy of preoperative blood preparation.
Table 2.
Cost of blood product preparation per unit.
Table 3.
Baseline characteristics of the second cohort (2020–2021, N = 414).
Fig 3.
Comparison of the effectiveness index for packed red cell preparation among strategies by quarters.
(A) Crossmatch to transfusion ratio, (B) transfusion probability, (C) transfusion index. Abbreviations: C/T ratio = Crossmatch to transfusion ratio, Ti = transfusion index, Tp = transfusion probability, Ref = effectiveness criteria for each index, S1 = Machine learning-based strategy, S2 = Clinical trial-based strategy, S3 = Routine-based strategy.
Table 4.
Crossmatch to transfusion ratio, transfusion probability, transfusion index of packed red cell.
Fig 4.
Cost and cost difference of preoperative blood product preparation among strategies.
Abbreviations: S1 = Machine learning-based strategy, S2 = Clinical trial-based strategy, S3 = Routine-based strategy.