建立75歲以上亞洲老年人之五年心血管風險評估工具及其確效

Abstract

研究背景 心血管疾病是全球老年人的首要死因，因此極需早日找出高風險病人並給予藥物或非藥物治療。然而目前心血管風險預測的研究族群大多為中壯年人，因老年人的風險因子與中壯年有所不同，故現行的評估工具無法適用於老年人；此外，現今常用的心血管風險量表主要建立在西方高加索族群，不同種族間會因為先天差異而造成疾病風險不同，因此為亞洲老年人建立一個心血管風險評估工具是必要且重要的。 研究目的 本研究欲為亞洲75歲以上的男性及女性老年人分別建立五年心血管風險預測工具並進行確效。 研究方法 此為一回溯性研究，使用之資料庫為台大醫院整合資料庫中於2006-2017年間曾具三高疾病或心臟血管疾病診斷之病人資料，並與全民健康保險資料庫串聯以獲得追蹤期間完整的用藥紀錄和疾病診斷。收納對象為2008年2月至2012年12月間曾有台大醫院就診紀錄且就醫年紀為75歲以上的病人，排除過去曾經有心血管疾病診斷者。研究族群依男、女性分組後再各自隨機以80%、20%比例分為建立模型組和確效模型組，研究終點為廣義心血管事件，包含心因性死亡、中風、冠心症、周邊動脈疾病、心衰竭住院事件和冠狀動脈成形手術，納入分析的潛在風險因子為年齡、身體質量指數、血脂、高血壓、糖尿病、自體免疫疾病、其它慢性疾病和使用藥品。模型建立分為兩步驟：首先，利用雙變項分析檢視單一風險因子與心血管事件的關聯性，P < 0.25的變項則進入Cox比例風險模型，利用逐步選擇法篩選出最適模型。最終模型建立後，我們使用C-statistic檢視模型鑑別力，並比較預測風險及實際風險的數值接近程度來確認模型校準力，另外，本研究也將利用確效組的模型表現來檢視預測模型的外推性，並進行相關的敏感性分析。 研究結果 本研究共納入12,174位病人，平均追蹤年限7.4年 (四分位差5.4-9.2)，有2,779位病人發生終點事件，女性終點事件發生率略低於男性 (35/1,000 vs. 40/1,000 人年)。屬於心血管傳統風險因子的年紀、吸菸史、HDL-C (女性) 及TG (男性) 仍留在最終模型，此外，衰弱、慢性腎臟病和心房顫動會顯著增加男、女性心血管風險，在女性預測模型中，全身性紅斑狼瘡、重症肌無力、轉移性癌症和失智與心血管疾病有顯著關聯性，若使用抗凝血藥物或aspirin也會增加女性心血管風險，而statins類藥品在女性中顯示能降低14%心血管風險。 預測模型的鑑別力C-statistic在女性、男性分別為0.66和0.64，且使用類別型變項的模型鑑別力與使用連續型變項的模型並無顯著差異。確效模型組與建立模型組有相似的鑑別力，但校準力較差 (P < 0.05)。敏感性分析中顯示若將原研究終點排除心衰竭、出血性中風及冠狀動脈成形手術後，女性預測模型AUC則從0.65上升為0.66，單一重大心血管事件如心肌梗塞、心因性死亡之女性模型AUC可各自達到0.70和0.72，然而男性模型AUC於敏感性分析中皆並未大幅改善。 結論 本研究為男、女性的亞洲老年人找出心血管風險因子，並分別建立心血管風險預測工具，雖然此預測工具僅具普通的模型表現，但仍能做為輔助醫療人員臨床決策的參考。期許未來研究持續找出重要但尚未量測的風險因子，並將此風險預測工具套用在其他亞洲地區或國家的老年人以驗證其外推性。

Background Cardiovascular diseases (CVD) are the leading cause of death in older adults globally. It is important to identify high-risk patients so that non-pharmacological therapy or medications can be initiated as early as possible. However, most existing risk prediction tools were derived from middle-aged populations and thus are inappropriate for older adults since the cardiovascular (CV) risk factors could be different in the elderly. Moreover, risk scores currently in use were mainly developed in Caucasians. Cardiovascular risk differs between ethnicities due to indigenous difference. It is crucial to develop a risk assessment tool for older adults in the Asian population. Objectives The aim of this study was to develop and validate a gender-specific 5-year CV risk tool in individuals aged 75 years or older. Methods In this retrospective study, we used the Integrated Medical Database of National Taiwan University Hospital (NTUH-iMD) for patients who had a diagnosis of hypertension, diabetes, dyslipidemia or CV disease between 2006 and 2017. The NTUH-iMD was linked to the National Health Insurance Research Database (NHIRD) to obtain the complete health information during follow-up. Patients who visited the National Taiwan University Hospital between Feb 2008 and Dec 2012 were first included, and those aged 75 years or above and without a history of CVD were further identified. Study subjects was divided by gender and randomly split into a derivation cohort (80%) and validation cohort (20%). Gender-specific risk assessment models were developed to predict composite CVD, an endpoint consisting of CV death, strokes, coronary heart diseases, peripheral artery diseases, hospitalization of heart failure, and cardiac revascularization. Potential risk factors analyzed included age, body mass index, lipid profiles, hypertension, diabetes, autoimmune diseases, other chronic comorbidities, and medications. A two-step approach was used: bivariate analysis was first performed to evaluate the unadjusted association between each risk factor and the outcome, and variables with P < 0.25 were then included in the stepwise selection process of Cox proportional hazards regression to derive the final model. After the prediction models were developed, we assessed discrimination of the model by C-statistic and evaluated calibration by comparing the predicted and observed survival. Validation cohort was used to test the model’s external validity. Sensitivity analyses were also conducted. Results There were 12,174 patients included in the final analysis. With a median follow-up of 7.4 years (interquartile range: 5.4-9.2), 2,779 patients developed the composite CVD outcome. The incidence rate of cardiovascular events in women was lower than in men (35/1,000 vs. 40/1,000 person-years). Traditional risk factors remaining significant in the final model were age, smoking, HDL-C, and TG. Additionally, frailty, chronic kidney disease, and atrial fibrillation substantially increased the cardiovascular risk in both women and men. Systemic lupus erythematosus, myasthenia gravis, metastatic cancer, dementia and administration of oral anticoagulants and aspirin were the other significant factors associated with increased risk of CVD in women’s model. On the other hand, statins showed a protective effect that women users had a 14% reduced risk of CVD. The C-statistic for prediction models were 0.66 (women) and 0.64 (men). Discrimination of the models having categorical variables was found to be comparable to containing continuous variables. In the derivation cohort, model discrimination was similar but had poorer calibration than in the validation cohort (P < 0.05). Sensitivity analyses showed that when excluding heart failure, hemorrhagic stroke, and revascularization from the composite outcome endpoint, AUC of the women’s model increased from 0.65 to 0.66. The AUC of women’s model increased to 0.70 and 0.72 when the risk of individual major cardiovascular events such as MI and CVD death was predicted respectively. However, the AUC of men’s model did not improve in the sensitivity analyses. Conclusions In this study, cardiovascular risk factors were identified, and the gender-specific CV risk assessment tool was developed and validated to quantify risk in Asian older adults. Although the model performance was only fair, this risk assessment tool can still facilitate the clinical decision making by health professionals. Further research is needed to continuously identify important but unmeasured risk factors, and this model should be applied in other Asian regions or countries to confirm the external validity.