大面积脑梗死急性期患者颅内压增高风险的预测模型构建及验证

Construction and validation of a predictive model for the risk of increased intracranial pressure in patients with acute phase of massive cerebral infarction

摘要:

目的 探讨大面积脑梗死急性期患者颅内压（ICP）增高风险的预测模型构建及验证。方法 选取2019年1月—2021年9月西安交通大学第一附属医院收治的102例大面积脑梗死急性期患者，根据是否出现ICP增高将患者分为ICP增高组（63例）和非ICP增高组（39例）。通过回顾性分析两组患者的临床数据，包括年龄、性别、病史、神经功能评分及治疗情况，利用多因素一般Logistic回归模型分析ICP增高的危险因素，并构建预测模型，通过受试者工作特征（ROC）曲线、使用Bootstrap方法进行1 000次重复采样以验证模型的预测效能。结果 ICP增高组年龄大于非ICP增高组，体温、氧分压（PaO₂）、二氧化碳分压（PaCO₂）、格拉斯哥昏迷量表（GCS）评分均低于非ICP增高组（P <0.05），心率、合并高血压、脑梗死面积≥ 2 cm²、收缩压、舒张压、ICP、美国国立卫生研究院卒中量表评分（NIHSS）评分、里士满躁动-镇静评分、呕吐占比、脑水肿占比和脑电频率变慢占比均高于非ICP增高组（P <0.05），入院延迟时间长于非ICP增高组（P <0.05）。两组性别、体质量指数、合并冠心病、合并糖尿病、神志清醒、瞳孔异常、颅内感染、抗凝药物、溶栓治疗比较，差异均无统计学意义（P >0.05）。多因素一般Logistic回归分析结果显示：PaCO₂［O^R=0.792（95% CI：0.673，0.933）］、ICP［O^R=1.061（95% CI：1.026，1.097）］、NIHSS评分［O^R=1.231（95% CI：1.073，1.413）］、呕吐［O^R=6.220（95% CI：1.086，36.639）］、脑水肿［O^R=39.888（95% CI：4.865，327.050）］、入院延迟时间［O^R=6.517（95% CI：1.661，25.574）］均是ICP增高的危险因素（P <0.05）。ROC曲线结果显示，模型预测ICP增高的曲线下面积为0.989（95% CI：0.976，1.000），约登指数为0.911，敏感性为93.7%（95% CI：0.845，0.985），特异性为97.4%（95% CI：0.917，1.000），表明模型有较好的区分能力。校准曲线和Bootstrap法的自校验均显示模型具有良好的预测准确性和一致性。结论 构建的模型能有效预测大面积脑梗死急性期患者ICP增高的风险。这一模型有助于临床医生早期识别高风险患者，及时采取适当的干预措施，从而改善患者的预后。

Abstract:

Objective To develop and validate a predictive model for assessing the risk of increased intracranial pressure in patients experiencing the acute phase of massive cerebral infarction.Methods We retrospectively analyzed clinical data from 102 patients with massive cerebral infarction treated in the First Affiliated Hospital of Xi'an Jiaotong University between January 2019 and September 2021. Patients were classified into an intracranial hypertension group (63 cases) and a non-intracranial hypertension group (39 cases) according to the presence of intracranial hypertension. Clinical data of the two groups of patients including age, sex, medical history, neurological function scores, and treatment, were respectively analyzed. The risk factors for increased intracranial pressure were identified using multivariable Logistic regression analysis, based on which a predictive model was established. The performance of the predictive model was evaluated via receiver operating characteristic (ROC) curve analysis and validated through 1 000 bootstrap resamples.Results Compared with the non-intracranial hypertension group, the age was greater, the body temperature, PaO₂, PaCO₂ and GCS scores were lower, the heart rate, the proportion of patients with hypertension, the proportion of patients with infarct area ≥ 2 cm², systolic blood pressure, diastolic blood pressure, ICP, NIHSS scores, RASS scores, incidence of vomiting, incidence of brain edema, and the rate of electroencephalograph frequency slowing were higher, and the delay to admission was longer in the intracranial hypertension group. There were no statistically significant differences between the two groups in terms of the sex composition, body mass index, presence of coronary heart disease, presence of diabetes mellitus, the level of consciousness, pupil abnormalities, intracranial infection, anticoagulant use, and thrombolytic therapy (P >0.05). The results of multivariable Logistic regression analysis showed that higher PaCO₂ [O^R = 0.792 (95% CI: 0.673, 0.933) ], ICP [O^R = 1.061 (95% CI: 1.026, 1.097) ], and NIHSS scores [O^R = 1.231 (95% CI: 1.073, 1.413) ], presence of vomiting [O^R =6.220 (95% CI: 1.086, 36.639) ] and cerebral edema [O^R = 39.888 (95% CI: 4.865, 327.05) ], and longer delays to admission [O^R = 6.517 (95% CI: 1.661, 25.574) ] were all risk factors for increased intracranial pressure (P < 0.05). The ROC curve analysis revealed that the area under the curve of the predictive model was 0.989 (95% CI: 0.976, 1.000), with a Jordan index of 0.911, a sensitivity of 93.7% (95% CI: 0.845, 0.935), and a specificity of 97.4% (95% CI: 0.917, 1.000), indicative of good discriminative ability of the model. Both the calibration curve and the self-validation using the Bootstrap method demonstrated that the model had good predictive accuracy and consistency.Conclusions The predictive model effectively forecasts the risk of increased intracranial pressure in patients during the acute phase of massive cerebral infarction. By allowing clinicians to early identify high-risk patients, this tool facilitates timely interventions that can potentially improve patient outcomes.