Abstract
OBJECTIVE: To evaluate the current status and factors influencing the occurrence of percutaneous liver biopsy bleeding in children through a retrospective study, and to develop and validate a risk prediction model to reduce the incidence of percutaneous liver biopsy bleeding in children. METHODS: From the hospital's electronic medical record system, clinical data of the study subjects were obtained during their hospitalization. Continuous variables were described using the median (interquartile range), while categorical variables were described using frequencies, proportions, and rates. Feature variables were screened using Lasso regression, and the data were divided into training and validation sets in a 7:3 ratio. Variables with statistically significant differences were included in a binary logistic regression model, and a risk prediction model was constructed using stepwise bidirectional regression. The model was visualized using a nomogram and internally validated. The ROC curve was used to assess the model's discriminative ability, the calibration curve to evaluate its calibration, and the decision curve analysis to assess its clinical decision-making capability. RESULTS: The incidence of bleeding in this study was 13.3%, most of which were minor and did not cause serious complications. Variables with meaningful Lasso regression coefficients were included in the multivariate logistic regression analysis, and the stepwise bidirectional regression ultimately yielded seven independent influencing factors: Pre-Corticosteroid, Post Liver Transplantation, Needle Depth, ALT, PT, PLT, and GPR. These factors will be used to construct a prediction model for percutaneous liver biopsy bleeding in children. In this study, the training set AUC was 0.720, with a 95% CI of 0.675-0.765, and the validation set AUC was 0.700, with a 95% CI of 0.633-0.767. CONCLUSION: This study created and internally tested a bleeding prediction model for children undergoing percutaneous liver biopsy, demonstrating moderate discriminative ability. Additional optimization and external validation are necessary. Expanding research with larger, multi-center datasets is crucial to enhancing the model's predictive accuracy and clinical applicability.