Abstract
Introduction:
This study aims to construct an efficient random forest predictive model to address the knowledge limitations in predicting deep vein thrombosis (DVT) among tibial plateau fracture (TPF) patients. DVT, being a critical and potentially life-threatening consequence, often necessitates intricate clinical management.
Materials and Methods
This investigation retrospectively examined adult patients who underwent surgical intervention for tibial plateau fractures in our institution from June 2020 to December 2023. Among the 562 patients who underwent surgical intervention, 231 were included in the study cohort, subsequently divided into training and testing cohorts in a 70:30 ratio. The training cohort utilized the R software to construct a random forest predictive model, which was then validated in the testing group. Furthermore, logistic regression analysis was conducted in this study to acquire feature selection variables and the area under the curve (AUC), evaluating the credibility and discriminative capacity of the random forest algorithm.
Results
As for the discriminative capacity of the random forest predictive model, in the training cohort, the 95% confidence interval (CI), area under the curve (AUC), sensitivity, specificity, F1, and Balanced Accuracy were respectively (0.9775, 1), 1.0000, 1.0000, 1.0000, 1.0000, and 1.0000. Correspondingly, in the testing cohort, these metrics were (0.7326, 0.9176), 0.901, 0.8696, 0.8261, 0.7843, and 0.8478.
Conclusion
Utilizing the discerningly chosen " important " variables, this study employed the random forest algorithm to craft a predictive model, demonstrating exceptional discriminative prowess. These identified " important " variables, serving as predictive factors, aid clinicians in the identification of high-risk DVT patients. This, in turn, facilitates timely intervention, fortification of care, and enhancement of prognostic outcomes for patients with fractures.