| 摘要: | 背景：心血管疾病(CVD)是全球的領先死因之一，而動脈粥狀硬化是CVD發生的主因。其初期病變常是無症狀的，導致疾病的預防及治療無法及時。因此，能夠在病變初期便篩檢出個案，並給予適當預防治療是十分重要的。冠狀動脈鈣化分數(CACS)用來定義心血管的粥狀硬化程度，被證實比其他傳統風險計算公式有更好的預測能力。然而測量的昂貴性使其無法成為一項普遍的篩檢項目。有研究指出，ㄧ些較易取得的指標顯示出了與CACS的相關性。若能使用這些指標辨識出CACS>0的患者，便可透過更簡單的篩檢找出需介入治療的患者。本研究之研究目的為：1. 透過機器學習方法建立冠狀動脈鈣化(CAC)的預測模型，並找出最高預測能力的模型。2. 評估重要影響因子的不同組合下，出現冠狀動脈鈣化的機率。3. 探討冠狀動脈鈣化與周邊血管硬化及頸動脈硬化情形的相關性。
研究方法：本研究使以病歷回溯方式納入2017/3/2至2022/10/31間，至新光醫院進行心血管健康檢查者。最終共1023位個案納入研究。自變項為健康檢查結果值，並以類別變項及連續變項兩種方式放入模型中。依變項為冠狀動脈鈣化(CACS>0)。建置羅吉斯迴歸模型(Logistic regression model)、隨機森林模型(Random Forest model)、支援向量機(Support vector machine model)三種模型並以精確度(Accuracy)、ROC曲線下面積(AUC)、特異度(Specificity)、敏感度(Sensitivity)判斷模型表現度，表現最優者即為最終模型。不同因子組合的冠狀動脈鈣化機率以最佳類別變項模型計算。
研究結果：1023位參與者中，456位有冠狀動脈鈣化(CACS>0)。在所有模型當中，自變項為連續變項的隨機森林模型顯示出了最好的AUC。年齡、性別、收縮壓、醣化血色素為較重要的變項。在所有自變項組合當中，≧50歲且醣化血色素及收縮壓皆異常的男性有冠狀動脈鈣化的機率為99%。冠狀動脈鈣化與周邊血管硬化、頸動脈硬化有正相關性。
結論：在所有模型當中，自變項為連續變項之隨機森林模型有最好的預測能力。所有自變項當中，年齡為最重要的自變項。建議≧50歲且醣化血色素或收縮壓任一異常的族群進行進一步的冠狀動脈鈣化檢查。冠狀動脈鈣化與周邊血管硬化、頸動脈硬化有正相關性，建議冠狀動脈鈣化的族群亦要注意自身的周邊血管及頸動脈健康情況。此研究提供了機器學習方法在冠狀動脈鈣化分數中的應用，未來可將此預測模型運用至健檢族群中，進一步增進冠狀動脈鈣化檢查的效益。
Objective: Cardiovascular disease (CVD) is a leading cause of mortality worldwide. Atherosclerosis, the primary underlying cause of CVD, often remains asymptomatic in its initial stages, resulting in missed opportunities for early intervention in at-risk patients. Therefore, identifying patients in the early stages of atherosclerosis is crucial. The Coronary Artery Calcium Score (CACS) is a well-established metric for assessing the risk of atherosclerotic cardiovascular disease. However, the high cost of CACS measurement limits its routine clinical application. Research has identified various easily obtainable indicators associated with CACS. By identifying patients with a CACS greater than 0, more feasible methods can be employed to recognize individuals needing intervention. The aims of this study are (1) to develop a machine learning-based prediction model to estimate coronary artery calcium using health screening data; (2) to calculate the probability of coronary artery calcium based on different combinations of important factors; (3) to explore the correlation between coronary artery calcification and the levels of peripheral vascular sclerosis and carotid artery sclerosis.
Methods: A total of 1,023 participants who underwent a cardiovascular health examination at Shin Kong Hospital between 2017 and 2022 were retrospectively included in this study. The independent variables consisted of health screening outcomes, included in the model as continuous and categorical variables. The dependent variable was coronary artery calcium (CACS > 0). The machine learning models used in this study were logistic regression, random forest, and support vector machine. Model performance was evaluated using accuracy, area under curve (AUC), specificity, and sensitivity.
Results: Among the 1,023 participants, 456 had coronary artery calcium. The continuous random forest model demonstrated the best performance of all models. Age, sex, HbA1c, and systolic blood pressure (SBP) were the top four important variables. Males aged ?50 with abnormal HbA1c or diabetes and abnormal SBP or hypertension had the highest probability of having coronary artery calcium (99.8%).
Conclusion: The continuous random forest model demonstrated the best predictive performance among all models. Age was the most significant predictor. It is recommended that those aged?50 who have abnormal HbA1c and SBP should undergo coronary artery calcium screening. Coronary artery calcification positively correlated with peripheral and carotid artery sclerosis. Individuals with coronary artery calcification also need to pay attention to the health of peripheral and carotid arteries. Our model could be used to identify those who need to do screening for coronary artery calcification among the health examination population if the characteristics of their population are similar to ours. |
