Supplementary Material for: Performance of a diagnostic model for the presence of unruptured intracranial aneurysms in the general population

Introduction: The prevalence of unruptured intracranial aneurysms (UIAs) in the general population is 3%. Aneurysmal subarachnoid hemorrhage (aSAH) can be prevented by screening for UIAs followed by monitoring and, if needed, preventive neurosurgical or endovascular treatment of identified UIAs. Therefore, we developed a diagnostic model for presence of UIAs in the general population to help identify persons at high risk of having UIAs.
Methods: Between 2005-2015, participants from the population-based Rotterdam Study underwent brain magnetic resonance imaging at 1.5 Tesla, on which presence of incidental UIAs was evaluated. We developed a multivariable logistic regression model using candidate diagnostic markers that were selected based on the literature, including sex, age, hypertension, smoking, hypercholesterolemia, diabetes, alcohol, and their interactions. We corrected for overfitting using bootstrapping. Model performance was assessed with discrimination, calibration, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV).
Results: 5835 persons were included (55.0% women, mean age 64.9 ± 10.9 years) with a 2.2% UIA prevalence. Sex, age, hypertension, smoking, diabetes, and interactions of sex with age, hypertension, and smoking were independent diagnostic markers. The resulting model had a c-statistic of 0.65 (95% confidence interval [CI] 0.60 – 0.68) and 56% sensitivity, 52% specificity, 98% PPV, and 3% NPV for UIA presence at a cut-off value of 4%. Because of interactions with sex, additional models for men and women separately were developed. The model for men had a c-statistic of 0.70 (95% CI 0.62 – 0.78) with age, hypertension, and smoking as diagnostic markers and comparable additional performance values as for the full model. The model for women had a c-statistic of 0.58 (95% CI 0.52 – 0.63) with smoking as the only diagnostic marker.
Conclusion: Our diagnostic model had insufficient performance to help identify persons at high risk of having UIAs in the general population. Rather, it provides insight in risk factors contributing to UIA risk and shows that these may be in part sex-specific.

引言：未破裂颅内动脉瘤（UIAs）在普通人群中的患病率约为3%。通过筛查UIAs并进行监测，以及在必要时对已确定的UIAs进行预防性神经外科或血管内治疗，可以预防动脉瘤性蛛网膜下腔出血（aSAH）。因此，我们开发了一个针对普通人群中UIAs存在的诊断模型，以协助识别具有UIAs高风险的个人。研究方法：在2005-2015年间，来自基于人群的鹿特丹研究（Rotterdam Study）的参与者接受了1.5特斯拉的脑部磁共振成像检查，用以评估偶然发现的UIAs。我们基于文献，利用候选诊断标志物构建了一个多变量逻辑回归模型，包括性别、年龄、高血压、吸烟、高胆固醇血症、糖尿病、饮酒及其相互作用。我们通过自助法（bootstrapping）纠正了过拟合问题。模型性能通过区分度、校准、灵敏度、特异性、阳性预测值（PPV）和阴性预测值（NPV）进行评估。研究结果：共有5835名参与者被纳入研究（其中55.0%为女性，平均年龄64.9 ± 10.9岁），UIAs的患病率为2.2%。性别、年龄、高血压、吸烟、糖尿病以及性别与年龄、高血压、吸烟的相互作用是独立的诊断标志物。所得模型在4%的截断值下具有0.65的c统计量（95%置信区间[CI] 0.60 – 0.68），56%的灵敏度、52%的特异性、98%的PPV和3%的NPV。由于与性别的相互作用，还开发了针对男性和女性的单独模型。男性的模型具有0.70的c统计量（95% CI 0.62 – 0.78），年龄、高血压和吸烟作为诊断标志物，并具有与完整模型相当的性能值。女性的模型具有0.58的c统计量（95% CI 0.52 – 0.63），吸烟是唯一的诊断标志物。结论：我们的诊断模型在帮助识别普通人群中具有UIAs高风险的个人方面性能不足。相反，它揭示了导致UIA风险的因素，并表明这些因素可能部分与性别相关。