Development and evaluation of a dynamic nomogram model for intraoperative blood transfusion decision-making

Abstract

Objective: By gathering data on patients with intraoperative blood transfusion and investigating the factors influencing intraoperative blood transfusion in patients, we aimed to construct a dynamic nomogram decision-making model capable of continuously predicting the probability of intraoperative blood transfusion in patients. This was done to explore a new mode of individualized and precise blood transfusion and to guide doctors to make timely and reasonable blood transfusion decisions and save blood resources.

Methods: Data of surgical patients in our hospital from 2019 to 2023 were collected. Among them, 705 patients who had blood transfusions were the experimental group, and 705 patients without intraoperative blood transfusions were randomly selected as the control group. Preoperative and intraoperative indicators of 1,410 patients were collected. 80% of the data set was used as the training set and 20% as the test set. In the training set, independent risk factors affecting intraoperative blood transfusion in patients were obtained through Lasso regression and binary logistic regression analysis, and the regression model was established and validated.

Results: Through Lasso regression with cross-validation and binary logistic regression analysis, the independent risk factors affecting patients' intraoperative blood transfusion decision-making were determined as ASAs (III) (OR = 3.009, 95% CI = 1.311-6.909), surgical grading (IV) (OR = 3.772, 95% CI = 1.112-12.789), EBL (OR = 1.003, 95% CI = 1.002-1.004), preHGB (OR = 0.932, 95% CI = 0.919-0.946), LVEF (OR = 1.063, 95% CI = 1.028-1.098), Temp (OR = 57.14, 95% CI = 9.740-35.204), preAPTT (OR = 1.147, 95% CI = 1.079-1.220), and preDD (OR = 1.127, 95% CI = 1.048-1.212). The area under the curve (AUC) of the receiver operating characteristic curve (ROC) of the training set was 0.983, p < 0.05. By calculating the Jordon index, when the Jordon index reached its maximum value, the corresponding diagnostic probability threshold was 0.515. When the model predicted that the probability threshold was 0.515, the sensitivity was 0.939 and the specificity was 0.964. These independent risk factors were introduced into R statistical software to fit the intraoperative blood transfusion decision-making dynamic nomogram model. The goodness of fit test of the model for the training set was χ ² = 111.85, p < 0.01, and the AUCs of the training set and the testing set were 0.983 and 0.995, respectively, p < 0.05. The calibration curve showed that the predicted probability of the model was in good agreement with the observed probability. Clinical decision curves (CDA) and clinical impact curves were plotted to evaluate the clinical utility of the model and the net benefit of the model.

Conclusion: Variables were screened by Lasso regression, the model was developed by binary logistic regression, and a dynamic nomogram model for intraoperative transfusion decision-making was successfully fitted using R software. This model had good goodness of fit, discrimination, and calibration. The model can dynamically and instantaneously predict the probability of blood transfusion and its 95% confidence interval under the current patient indicators by selecting the patient's independent risk factors in the drop-down mode during the operation. It can assist doctors in making a reasonable blood transfusion decision quickly and save blood resources.

Keywords: Lasso regression; blood transfusion decision-making; nomogram modeling; peroperative; transfusion.

Figure 1

Lasso regression path diagram.

Figure 2

Lasso regression cross-validation results.

Figure 3

Forest plot of intraoperative blood transfusion risk.

Figure 4

Multi-indicator independent diagnostic ROC curve.

Figure 5

Multi-indicator combined diagnostic ROC curve.

Figure 6

Intraoperative blood transfusion dynamic nomogram model.

Figure 7

Model discrimination between training and test set.

Figure 8

Training set and test set model calibration curve.

Figure 9

Clinical decision curve (left) and clinical impact curve (right) of the training set.