冠狀動脈疾病是影響全球的重要心血管疾病之一，且罹患冠狀動脈疾病的患者人數都有逐漸上升的趨勢，造成許多人死亡。已有研究指出，提早預防或在患病早期接受治療能有效降低死亡率，因此，本研究使用深度學習演算法，包括長短期記憶模型 (Long Short-Term Memory, LSTM)和卷積神經網路 (Convolutional Neural Networks, CNN)，並使用長庚醫學研究資料庫從2001年至2018年10月的病患診斷和檢驗資料，建立冠狀動脈疾病時序預測模型。經資料清理後，病例組有8,818位患者 (13.1%)；對照組有58,562位患者。以CNN建立之最佳模型，ROC (receiver operating characteristic)曲線下面積為0.945±0.0024。使用診斷和檢驗兩者整合資料建立的模型，效能較單獨使用診斷或檢驗資料建立模型好 (P<0.001)。本研究也使用單一時間點資料建立CNN模型，並和使用時序資料所建立之模型比較，比較結果為使用時序資料建立之模型預測效能較佳 (P<0.001)。本研究結果表明，使用深度學習和時間序列資料建立冠狀動脈疾病預測模型能有效預測病患未來是否發生冠狀動脈疾病。;Coronary artery disease (CAD) is one of the major cardiovascular diseases affecting the global population, causing many deaths each year. Coronary artery disease is on the rise in both developed and developing countries. Studies have shown that early prevention or treatment at an early stage of the disease can effectively reduce mortality. We use deep learning algorithms, including Long Short-Term Memory (LSTM) and Convolutional Neural Networks (CNN), to develop a CAD prediction model, using data from Chang Gung Research Database from 2001/01/01 to 2018/10/31. A total of 8,818 patients had CAD while 58,562 patients did not. The best performance of CNN model has an Area under the receiver operating characteristic curve (AUC) of 0.945 and a standard deviation of 0.0024. We found that using both diagnosis and lab data has a better performance compared to using only diagnosis or lab data. We also compared the use of time series data and non-time series data in CNN model. The results show that the model using time series data has higher performance. The results of our study show that the use of deep learning and time series data to build a coronary artery disease prediction model can effectively predict whether patients will have coronary artery disease in the future.
