Recursive partitioning analysis of factors determining infection after intracranial tumor surgery

Abstract

Objectives: Several factors are commonly associated with the occurrence of post-operative infection after craniotomy. However, the risk factors associated with tumor surgery have been less intensively investigated. The aim of the present study was to analyze the risk factors for infection and categorize patients according to risk rate. Methods: In this study, we retrospectively evaluated 987 adult patients consecutively submitted to craniotomy for tumor resection. The primary outcome was the occurrence of infection within 30 days after surgery. The following independent variables were assessed: age, gender, surgery duration, length of hospital stay prior to surgery, reoperation, body mass index, serum albumin, hemoglobin, lactic dehydrogenase, smoking, diabetes, corticoid use, preoperative chemotherapy, previous irradiation, elective or urgent indication for surgery, supra or infratentorial lesion location, and tumor histology. We performed a recursive partitioning analysis to assess the relative importance of these variables in predicting infection. Results: The model returned a 3-level classification: 1. CSF-leakage (relative contribution 70%), 2. Emergency surgery indication (18%), and 3. Tumor histology (8%). Additionally, partitioning clustered together 3 risk groups: 1. CSF-leakage group (probability of infection 72.5%), 2. No CSF-leakage and urgent surgery (mean probability 18.1%); and 3. no CSF-leakage and no urgent surgery (3.4%). The misclassification rate was 4.5%, the overall specificity and sensitivity were 99.6% and 75.5%, respectively, and the area under the ROC-curve was 0.6908. Conclusion: Our analysis indicates that technical and treatment-related factors are significantly more relevant than patient- or disease-related factors in determining the risk of postoperative infection.