Supplementary Material for: Two Years with COVID-19: The Electronic Frailty Index Identifies High-Risk Patients in the Stockholm GeroCovid Study

<b><i>Introduction:</i></b> Frailty, a measure of biological aging, has been linked to worse COVID-19 outcomes. However, as the mortality differs across the COVID-19 waves, it is less clear whether a medical record-based electronic frailty index (eFI) that we have previously developed for older adults could be used for risk stratification in hospitalized COVID-19 patients. <b><i>Objectives:</i></b> The aim of the study was to examine the association of frailty with mortality, readmission, and length of stay in older COVID-19 patients and to compare the predictive accuracy of the eFI to other frailty and comorbidity measures. <b><i>Methods:</i></b> This was a retrospective cohort study using electronic health records (EHRs) from nine geriatric clinics in Stockholm, Sweden, comprising 3,980 COVID-19 patients (mean age 81.6 years) admitted between March 2020 and March 2022. Frailty was assessed using a 48-item eFI developed for Swedish geriatric patients, the Clinical Frailty Scale, and the Hospital Frailty Risk Score. Comorbidity was measured using the Charlson Comorbidity Index. We analyzed in-hospital mortality and 30-day readmission using logistic regression, 30-day and 6-month mortality using Cox regression, and the length of stay using linear regression. Predictive accuracy of the logistic regression and Cox models was evaluated by area under the receiver operating characteristic curve (AUC) and Harrell’s C-statistic, respectively. <b><i>Results:</i></b> Across the study period, the in-hospital mortality rate decreased from 13.9% in the first wave to 3.6% in the latest (Omicron) wave. Controlling for age and sex, a 10% increment in the eFI was significantly associated with higher risks of in-hospital mortality (odds ratio = 2.95; 95% confidence interval = 2.42–3.62), 30-day mortality (hazard ratio [HR] = 2.39; 2.08–2.74), 6-month mortality (HR = 2.29; 2.04–2.56), and a longer length of stay (β-coefficient = 2.00; 1.65–2.34) but not with 30-day readmission. The association between the eFI and in-hospital mortality remained robust across the waves, even after the vaccination rollout. Among all measures, the eFI had the best discrimination for in-hospital (AUC = 0.780), 30-day (Harrell’s C = 0.733), and 6-month mortality (Harrell’s C = 0.719). <b><i>Conclusion:</i></b> An eFI based on routinely collected EHRs can be applied in identifying high-risk older COVID-19 patients during the continuing pandemic.

引言：衰弱作为生物学衰老的衡量指标，已被证实与更不良的新型冠状病毒感染（COVID-19）预后相关。然而，由于不同新冠疫情波次间的死亡率存在差异，我们此前为老年人群开发的、基于医疗记录的电子衰弱指数（electronic frailty index, eFI）能否用于住院新冠患者的风险分层，目前尚无定论。研究目的：本研究旨在探讨衰弱与老年新冠患者的死亡率、再入院率及住院时长的关联，并对比该电子衰弱指数与其他衰弱及共病评估工具的预测准确性。研究方法：本研究为回顾性队列研究，使用瑞典斯德哥尔摩9家老年病诊所的电子健康档案（electronic health records, EHRs），纳入2020年3月至2022年3月期间收治的3980例新冠患者（平均年龄81.6岁）。衰弱评估采用针对瑞典老年患者开发的48项条目电子衰弱指数、临床衰弱量表（Clinical Frailty Scale）及医院衰弱风险评分（Hospital Frailty Risk Score）；共病情况采用查尔森共病指数（Charlson Comorbidity Index）进行评估。我们分别采用logistic回归分析院内死亡率及30天再入院率，采用Cox回归分析30天及6个月死亡率，采用线性回归分析住院时长。分别通过受试者工作特征曲线下面积（area under the receiver operating characteristic curve, AUC）及Harrell一致性指数（Harrell’s C-statistic）评估logistic回归模型与Cox回归模型的预测准确性。研究结果：整个研究周期内，院内死亡率从第一波疫情的13.9%降至最新一波（奥密克戎毒株流行）的3.6%。在控制年龄与性别混杂因素后，电子衰弱指数每升高10%，与院内死亡率（优势比=2.95；95%置信区间：2.42~3.62）、30天死亡率（风险比[HR]=2.39；95%置信区间：2.08~2.74）、6个月死亡率（HR=2.29；95%置信区间：2.04~2.56）的更高风险显著相关，同时与更长的住院时长（β系数=2.00；95%置信区间：1.65~2.34）相关，但与30天再入院率无显著关联。电子衰弱指数与院内死亡率的关联在各疫情波次间均保持稳健，即使在疫苗推广后亦是如此。在所有评估工具中，电子衰弱指数对院内死亡率（AUC=0.780）、30天死亡率（Harrell一致性指数=0.733）及6个月死亡率（Harrell一致性指数=0.719）的区分度最佳。研究结论：基于常规采集的电子健康档案构建的电子衰弱指数，可用于在新冠疫情持续大流行期间识别高风险老年新冠患者。