Scholarlink

<p>Fever is common after open craniotomy, and early distinction between infectious and non-infectious causes remains challenging in the neurosurgical intensive care unit. We evaluated early risk stratification of postoperative fever at initial onset (postoperative day ≥ 3) using conventional statistics and interpretable machine learning to estimate infection probability before microbiological confirmation. This retrospective cohort study included patients who underwent open craniotomy at a single tertiary neurosurgical center between January 2021 and December 2023. Patients with preoperative fever, early postoperative death, or non-craniotomy procedures were excluded. Postoperative fever was defined as a body temperature ≥ 38.0 °C occurring after postoperative day 3. Etiology was classified as infectious or non-infectious based on Centers for Disease Control and Prevention/National Healthcare Safety Network criteria and expert consensus. Variable selection was performed using a random forest algorithm, followed by multivariable logistic regression. Machine learning models—including logistic regression, extreme gradient boosting, and categorical boosting—were trained to predict fever occurrence and etiology. SHapley Additive exPlanations (SHAP) were used to assess model interpretability. Of 1,419 patients screened, 584 met inclusion criteria. Fever occurred in 316 patients (54.1%), of whom 144 (45.6%) had infectious fever. Risk factors for fever included lower preoperative Glasgow Coma Scale scores, longer surgical duration, larger craniotomy size, and specific pathologies. Infectious fever was associated with older age, external ventricular drainage, blood transfusion, and prolonged hospitalization. SHAP analysis identified surgical pathology, neurological status, incision length, and body mass index as major predictors. Among the models, categorical boosting showed the highest predictive performance. Post-craniotomy fever is common and often non-infectious. Combining regression analysis with interpretable machine learning models enabled effective identification of risk factors and classification of fever etiology. This approach may support timely clinical decisions and reduce unnecessary antibiotic exposure in neurocritical care.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Discriminating infectious and non-infectious fever after open craniotomy: a combined statistical and machine learning approach

  • 期刊论文

  • 发表日期：

    2026年3月26日

查看全文
  • Hyo Jeong Kim,
  • Jeonghwan Kim,
  • Jinhoo Seok,
  • Haewon Roh,
  • Joonho Byun,
  • Wonki Yoon,
  • Jong Hyun Kim,
  • Taek-Hyun Kwon,
  • Inyong Jeong,
  • Hwamin Lee,
  • Hyunjun Jo,
  • Jin Gu Yoon

摘要

Fever is common after open craniotomy, and early distinction between infectious and non-infectious causes remains challenging in the neurosurgical intensive care unit. We evaluated early risk stratification of postoperative fever at initial onset (postoperative day ≥ 3) using conventional statistics and interpretable machine learning to estimate infection probability before microbiological confirmation. This retrospective cohort study included patients who underwent open craniotomy at a single tertiary neurosurgical center between January 2021 and December 2023. Patients with preoperative fever, early postoperative death, or non-craniotomy procedures were excluded. Postoperative fever was defined as a body temperature ≥ 38.0 °C occurring after postoperative day 3. Etiology was classified as infectious or non-infectious based on Centers for Disease Control and Prevention/National Healthcare Safety Network criteria and expert consensus. Variable selection was performed using a random forest algorithm, followed by multivariable logistic regression. Machine learning models—including logistic regression, extreme gradient boosting, and categorical boosting—were trained to predict fever occurrence and etiology. SHapley Additive exPlanations (SHAP) were used to assess model interpretability. Of 1,419 patients screened, 584 met inclusion criteria. Fever occurred in 316 patients (54.1%), of whom 144 (45.6%) had infectious fever. Risk factors for fever included lower preoperative Glasgow Coma Scale scores, longer surgical duration, larger craniotomy size, and specific pathologies. Infectious fever was associated with older age, external ventricular drainage, blood transfusion, and prolonged hospitalization. SHAP analysis identified surgical pathology, neurological status, incision length, and body mass index as major predictors. Among the models, categorical boosting showed the highest predictive performance. Post-craniotomy fever is common and often non-infectious. Combining regression analysis with interpretable machine learning models enabled effective identification of risk factors and classification of fever etiology. This approach may support timely clinical decisions and reduce unnecessary antibiotic exposure in neurocritical care.