Abstract: Objective  To analyze the risk factors for delirium in patients with subarachnoid hemorrhage (SAH) in the neurology intensive care unit and to construct a risk prediction model. 
Methods  A prospective cohort study design was used to include SAH patients continuously admitted to the neurology intensive care unit of Beijing Tiantan Hospital, Capital Medical University from January 1, 2017, to January 1, 2020. After admission, a unified standard protocol was used to collect demographic information, disease-related data, surgical treatment information, and other data. According to the results of the confusion assessment method for the intensive care unit (CAM-ICU), patients were divided into the delirium group and the non-delirium group. Differences in data between the two groups were compared. The independent risk factors were screened using multivariate logistic regression. A prediction model was constructed based on risk factors, and the diagnostic efficacy of the model was evaluated using the AUC and the Hosmer-Lemeshow (H-L) goodness of fit test. 
Results  A total of 252 SAH patients were included, and 42 (16.7%) experienced delirium. The results of multivariate logistic regression analysis showed that drinking history (OR 4.005, 95%CI 1.538-10.432, P<0.001), coronary artery heart disease history (OR 6.758, 95%CI 2.048-22.301, P<0.001), mechanical ventilation (OR 7.816, 95%CI 2.305-26.500, P<0.001), use of analgesics (OR 2.832, 95%CI 1.093-7.334, P=0.032), use of sedatives (OR 21.896, 95%CI 8.397-57.092, P<0.001), and hypoalbuminemia (OR 3.047, 95%CI 1.121-8.282, P=0.029) were independent risk factors for delirium in patients with SAH in the neurology intensive care unit. The prediction model was Logit (P)=-4.553+1.388×drinking history+1.911×coronary artery heart disease history+2.056×mechanical ventilation+1.041×use of analgesics+3.086×use of sedatives+1.114×hypoalbuminemia. The AUC value of the model was 0.890 (95%CI 0.829-0.951, P<0.001). The Hosmer-Lemeshow test χ2=3.579, P=0.733. The maximum Youden index was 0.695, corresponding to a sensitivity of 76.2%, a specificity of 93.3%, and a prediction accuracy of 89.7%. 
Conclusions  Drinking history, coronary artery heart disease history, mechanical ventilation, use of analgesics, use of sedatives, and hypoalbuminemia are independent risk factors for delirium in patients with SAH in the neurology intensive care unit. The constructed prediction model based on the factors above has good performance.

Key words: Subarachnoid hemorrhage; Neurology intensive care unit; Delirium; Risk factor; Prediction model

摘要： 目的 分析神经重症监护病房蛛网膜下腔出血（subarachnoid hemorrhage，SAH）患者发生谵妄的危险因素，并构建风险预测模型。 
方法 采用前瞻性队列研究设计，连续纳入2017年1月1日—2020年1月1日收治于首都医科大学附属北京天坛医院神经重症监护病房的SAH患者。入院后，采用统一化标准方案收集患者人口学、疾病情况、手术治疗等临床信息。根据重症监护病房患者意识模糊评估量表（confusion assessment method for the intensive care unit，CAM-ICU）的评估结果，将患者分为谵妄组及非谵妄组，对比两组间资料的差异，利用多因素logistic回归筛选独立危险因素，根据危险因素构建预测模型，计算模型的AUC并使用Hosmer-Lemeshow拟合优度检验评价模型的诊断效能。
结果 共纳入SAH患者252例，42例（16.7%）发生了谵妄。多因素logistic回归分析结果显示饮酒史（OR 4.005，95%CI 1.538～10.432，P＜0.001）、冠心病史（OR 6.758，95%CI 2.048～22.301，P＜0.001）、机械通气（OR 7.816，95%CI 2.305～26.500，P＜0.001）、使用镇痛药物（OR 2.832，95%CI 1.093～7.334，P=0.032）、使用镇静药物（OR 21.896，95%CI 8.397～57.092，P＜0.001）、低蛋白血症（OR 3.047，95%CI 1.121～8.282，P=0.029）是神经重症监护病房SAH患者发生谵妄的独立危险因素。预测模型为Logit（P）=-4.553＋1.388×饮酒史＋1.911×冠心病史＋2.056×机械通气＋1.041×使用镇痛药物＋3.086×使用镇静药物＋1.114×低蛋白血症。该模型AUC值为0.890（95%CI 0.829～0.951，P＜0.001）；Hosmer-Lemeshow检验χ2=3.579，P=0.733；最大约登指数为0.695，对应的敏感度为0.762，特异度为0.933，预测准确率为89.7%。
结论 饮酒史、冠心病史、机械通气、使用镇痛药物、使用镇静药物以及低蛋白血症是神经重症监护病房SAH患者发生谵妄的独立危险因素，基于以上因素构建的预测模型性能良好。

关键词: 蛛网膜下腔出血; 神经重症监护病房; 谵妄; 危险因素; 预测模型