术后患者脑卒中发生风险预测模型的构建与验证

Construction and validation of a predictive model for postoperative stroke risk

摘要:

目的]构建并验证术后患者脑卒中发生风险的列线图预测模型。 [方法]回顾性纳入2017年7月—2023年8月于济宁医学院附属医院接受手术治疗后发生脑卒中的358例患者,并按1∶2比例匹配同时期术后未发生脑卒中的714例患者作为对照组。使用R语言软件,按7∶3的比例随机分为训练集751例和验证集321例。将单因素Logistic回归分析有意义的因素纳入Lasso回归分析,筛选出的独立预测因子构建预测术后脑卒中发生风险的列线图模型。采用ROC曲线和校准曲线分别验证列线图预测模型的区分度和校准度；决策曲线用于分析和评价临床效用和净收益。 [结果]除年龄及总胆红素外,其他临床数据在训练集与验证集之间差异无统计学意义(P＞0.05)。将单因素Logistic回归分析有意义的因素进一步进行Lasso回归分析,结果显示,脑梗死史、术前静脉血栓栓塞(VTE)评分、白蛋白及血浆纤维蛋白原是术后脑卒中发生风险的独立影响因素。使用上述危险因素制作了列线图预测模型。训练集与验证集ROC曲线下面积分别为0.651和0.683。校准曲线显示预测概率与实际概率具有高度一致性,临床决策曲线显示列线图预测模型在0～0.61的高风险阈值概率范围内具有良好净收益。 [结论]脑梗死史、术前VTE评分高危、白蛋白降低、血浆纤维蛋白原升高与术后脑卒中发生风险相关,构建的列线图预测模型能够有效预测患者预后水平。

Abstract:

Aim To construct and validate a predictive nomogram model for the risk of postoperative stroke. Methods We retrospectively enrolled 358 surgical patients who developed stroke between July 2017 and August 2023 at the Affiliated Hospital of Jining Medical University, and matched them 1∶2 with 714 contemporaneous surgical patients without postoperative stroke. Using R software, the cohort was randomly split into a training set (n＝751) and a validation set (n＝321) in a 7∶3 ratio. Significant variables in univariate Logistic regression were included in Lasso regression to determine independent predictors, which were then used to build a nomogram model for postoperative stroke risk. Discrimination and calibration were assessed with ROC curves and calibration plots, respectively; decision-curve analysis was used to evaluate clinical utility and net benefit. Results Except for age and total bilirubin, baseline clinical data were comparable between the training and validation sets (P＞0.05). Lasso regression revealed history of cerebral infarction, preoperative VTE score, serum albumin and plasma fibrinogen as independent predictors of postoperative stroke. A nomogram model incorporating these factors was constructed. The area under the ROC curve was 0.651 in the training set and 0.683 in the validation set. Calibration plots demonstrated high agreement between predicted and observed probabilities.Decision-curve analysis indicated favorable net benefit within a risk threshold probability range of 0～0.61. Conclusion History of cerebral infarction, high risk of preoperative VTE score, decreased albumin, and increased plasma fibrinogen are related to the risk of postoperative stroke. The constructed nomogram model can effectively predict the prognosis of patients.