Construction of a logistic model and GBM model for infection after spinal canal resection of intraspinal tumors.

Abstract

Objectives: To identify risk factors for postoperative infection following intraspinal tumor (IT) resection, and to construct predictive models using a Logistic regression model and gradient boosting machine (GBM) algorithms.

Methods: A retrospective study was conducted on 136 patients who developed postoperative infections after IT resection at Ziyang Central Hospital from November 2013 to October 2024. Logistic regression and GBM models were developed using R 4.3.2.

Results: Logistic regression analysis identified age >55 years, type II diabetes, operation time >3.3 h, interleukin-6 (IL-6)>5.5 ng/L, and procalcitonin (PCT)>0.3 ug/L as independent risk factors for infection after IT resection (P<0.05). The logistic regression equation was: Logit (P) = -12.238 + 2.081 × Age + 1.118 × Type II diabetes + 1.381 × operation time + 2.131 × IL-6 + 1.843 × PCT. In the GBM model, the relative importance of variables was: age (23.13011), IL-6 (22.98775), type II diabetes (18.57776), PCT (17.86779), and operation time (17.73660). The areas under the ROC curves (AUC) was 0.886 for the logistic model and 0.907 for the GBM model. Calibration curves demonstrated good agreement between predicted and observed infection rates in both models.

Conclusions: The identified risk factors and the predictive models offer valuable tools for early identification and prevention of postoperative infection following IT resection.