Analysis of risk factors for hemorrhage during CT-guided lung biopsy based on a random forest model

Abstract

Objective To systematically analyze and identify key risk factors for postoperative pulmonary hemorrhage using a combination of the random forest (RF) model and traditional logistic regression analysis, so as to provide data support for clinical practice. Methods This study included patients who underwent needle biopsy of lung masses from January 2020 to December 2023 in the Department of Interventional Therapy, Cancer Hospital, Tianjin Medical University. There were 844 cases, including 387 males and 457 females, ranging in age from 39 to 82 years. Clinical data and puncture-related characteristics were collected, including tumor size, puncture depth, puncture angle, presence of emphysema, lesion location in the lung, body position during puncture, whether the puncture passed through the interlobar fissure, and the number of punctures. The RF model was used to rank the importance of all variables, identifying those with the highest predictive value. Subsequently, a multivariate logistic regression model was applied to the top-ranked important variables to further evaluate their independent impact on postoperative pulmonary hemorrhage. Results The RF model results showed that tumor size and puncture depth had the highest importance in predicting the risk of postoperative pulmonary hemorrhage. Multivariate logistic regression analysis further confirmed that smaller tumor size (HR:0.980, 95% CI:0.971-0.989, P<0.05) was significantly associated with a lower risk of hemorrhage, while greater puncture depth (HR:1.146, 95% CI:1.063-1.235, P<0.05) was closely related to a higher risk of hemorrhage. Additionally, other factors such as puncture angle, age, lesion location in the lung and presence of emphysema showed some influence but did not reach statistical significance in the multivariate analysis. Conclusion This study successfully identified tumor size and puncture depth as independent risk factors for postoperative pulmonary hemorrhage by combining the RF model with multivariate logistic regression analysis. The application of the RF model improved the accuracy of feature selection, allowing us to focus on the most contributory predictive variables. These findings provide important support for preoperative risk assessment, suggesting that clinicians should prioritize these key factors in preoperative evaluations to develop safer and more effective surgical plans, thereby reducing the risk of postoperative hemorrhage and other complications.

Keywords

• biopsy
• hemorrhage
• random forest（rf）
• risk prediction