神经外科术后颅内感染Nomogram模型的建立与验证

Construction and validation of a Nomogram model of intracranial infection after neurosurgery

摘要:

目的 探讨神经外科术后患者颅内感染危险因素，建立并验证Nomogram预测模型。 方法 回顾性分析2019年1月1日—2022年12月31日南京某医院978例神经外科手术患者资料，经logistic单因素和多因素分析筛选独立危险因素。通过Lasso回归筛选建模变量，采用logistic回归构建Nomogram模型并进行内部验证，采用受试者工作特征（ROC）曲线、校准曲线、决策曲线测评模型效果。 结果 978例神经外科手术患者中，293例发生术后颅内感染，医院感染发病率为29.96%。感染组和非感染组患者的年龄、性别、冠心病、脑梗死、糖尿病和高血压的比例等基本情况比较，差异均无统计学意义（均P>0.05）。logistic多因素分析结果显示，术后颅高压、发热、血常规中性粒细胞百分比升高，脑脊液浑浊、潘氏试验阳性、葡萄糖浓度降低、脑脊液/血清葡萄糖比值异常、微生物培养阳性，以及未留置脑室外引流管（EVD）、留置腰大池置管引流管（LD）、使用免疫抑制药物和手术时间长是神经外科术后患者颅内感染的独立危险因素（均P < 0.05）。通过Lasso回归筛选15个变量，经共线性筛查、缺失数据插补（随机森林法）、检查两两交互作用项后最终纳入14个变量建模。建立Nomogram预测模型，其ROC曲线下面积、灵敏度、特异度和准确度分别为0.885、0.578、0.896、0.704，对模型进行内部验证，建模组和验证组效果相近，校准曲线和决策曲线同时提示模型有良好的预测能力。 结论 本研究构建的神经外科术后颅内感染Nomogram预测模型，预测指标科学且容易获取，稳定性与可靠性强，具有较高的应用价值和较广的适用范围，可为神经外科术后颅内感染的判定提供参考。

Abstract:

Objective To explore the risk factors for intracranial infection in patients after neurosurgery, construct and validate a Nomogram prediction model. Methods Data of 978 patients who underwent neurosurgery in a hospital in Nanjing from January 1, 2019 to December 31, 2022 were retrospectively analyzed. Independent risk factors were screened through logistic univariate and multivariate analyses. Modeling variables were screened through Lasso regression. A Nomogram model was constructed and internally validated by logistic regression. Effectiveness of the model was evaluated with receiver operating characteristic (ROC) curve, calibration curve and decision curve. Results Among 978 patients underwent neurosurgery, 293 had postoperative intracranial infection, with an incidence of healthcare-associated infection of 29.96%. There was no significant difference in age, gender, proportion of coronary heart disease, cerebral infarction, diabetes and hypertension between the infected group and the non-infected group (all P>0.05). Multivariate logistic analysis showed that postoperative intracranial hypertension, fever, increased neutrophil percentage in blood routine examination, turbid cerebrospinal fluid, positive Pan's test, decreased glucose concentration, abnormal ratio of cerebrospinal fluid/serum glucose, positive microbial culture, absence of indwelling external ventricular drainage tubes, presence of indwelling lumbar cistern drainage tubes, use of immunosuppressive agents, and long duration of surgery were independent risk factors for postoperative intracranial infection in patients who underwent neurosurgery (all P < 0.05). Fifteen variables were screened out through Lasso regression. Fourteen variables were finally included for modeling after collinear screening, missing data imputation (random forest method) and checking pairwise interaction items. A Nomogram prediction model was constructed, with the area under ROC curve, sensitivity, specificity, and accuracy of 0.885, 0.578, 0.896, and 0.704, respectively. Internal validation of the model was conducted. The modeling and validation groups presented similar effects. The calibration curve and decision curve also indicated that the model had good predictive efficacy. Conclusion The constructed Nomogram prediction model for postoperative intracranial infection after neurosurgery is scientific, and the prediction indicators are easy to obtain. The model presents with high stability, reliability, and application value, thus can provide reference for the assessment of postoperative intracranial infection after neurosurgery.