運用縱貫性體檢資料做機器學習演算預測國軍代謝症候群
Abstract
研究背景及目的：在現行定期體檢及體能測驗要求下，國軍官兵代謝症候群情況嚴重嗎?在智慧醫療時代，採用機器學習演算法從國軍官兵服役期間歷次的健康檢查資料序列，預測未來代謝症候群發生機率。
研究方法：選取100年5月至107年12月間，至三軍總醫院松山分院參加國軍年度體檢之20歲以上國軍官兵，連續8年參加體檢者共1,636人。運用羅吉斯回歸、決策樹、K-近鄰演算法、支持向量機及樸素貝葉斯等機器學習演算法，及運用深度學習演算法的簡單長短期記憶神經網路、堆疊長短期記憶神經網路、雙向長短期記憶神經網路進行預測。
研究結果：研究對象100-107年代謝症候群比率自100年時15.5%，增加至107年時25.2%；利用過往7年體檢結果的雙向長短期記憶神經網路模型，在預測未來有無代謝症候群模型效能最佳，模型正確率為85.6%、精準率為72.0%、召回率為73.1%、F值為71.1%。
研究結論：善用歷次健康檢查之大數據，就能據以推動精準公共衛生預防保健服務，例如將機器學習演算法置於體檢資訊系統，透過系統定期收集資料及不斷更新模型，未來於不同型式的體檢結果查詢時，能以多元方式提供醫師及國軍官兵作為促進健康生活型態的參考。
Background and objectives：Under the current requirements of regular physical examinations and physical fitness tests, is the situation of metabolic syndrome among army serious? In the era of smart medical care, machine learning algorithms are used to predict the future occurrence of metabolic syndrome from the sequence of medical examination data of army during their service. Methods：Selected from May 2011 to December 2018, the Army over the age of 20 who participated in the Army’s annual physical examination at the Tri-Service General Hospital Songshan Branch, a total of 1,636 people participated in the physical examination for 8 consecutive years. Machine learning algorithms such as Logistic regression, decision tree, K-nearest neighbor algorithm, support vector machine, and naive Bayes, and deep learning algorithms such as Vanilla long short-term memory neural networks, Stacked long short-term memory neural networks, and Bidirectional long short-term memory neural network for prediction. Results：The 2011-2018 metabolic syndrome rate of the research subjects increased from 15.5% in 2011 to 25.2% in 2018; the bidirectional long short-term memory neural network model based on the physical examination results of the past 7 years is the best in predicting the future presence of metabolic syndrome. The model accuracy is 85.6%, precision is 72.0%, recall is 73.1%, and F-measure is 71.1%. Conclusions：Making good use of the big data of previous health examination data can be used to promote precision public health preventive health care services. For example, machine learning algorithms are placed in the physical examination information system, and the system regularly collects data and continuously updates the model. In the future, different types of physical examination results when inquiring, it can provide doctors and military officers and soldiers in multiple ways as a reference for promoting a healthy lifestyle.
Background and objectives：Under the current requirements of regular physical examinations and physical fitness tests, is the situation of metabolic syndrome among army serious? In the era of smart medical care, machine learning algorithms are used to predict the future occurrence of metabolic syndrome from the sequence of medical examination data of army during their service. Methods：Selected from May 2011 to December 2018, the Army over the age of 20 who participated in the Army’s annual physical examination at the Tri-Service General Hospital Songshan Branch, a total of 1,636 people participated in the physical examination for 8 consecutive years. Machine learning algorithms such as Logistic regression, decision tree, K-nearest neighbor algorithm, support vector machine, and naive Bayes, and deep learning algorithms such as Vanilla long short-term memory neural networks, Stacked long short-term memory neural networks, and Bidirectional long short-term memory neural network for prediction. Results：The 2011-2018 metabolic syndrome rate of the research subjects increased from 15.5% in 2011 to 25.2% in 2018; the bidirectional long short-term memory neural network model based on the physical examination results of the past 7 years is the best in predicting the future presence of metabolic syndrome. The model accuracy is 85.6%, precision is 72.0%, recall is 73.1%, and F-measure is 71.1%. Conclusions：Making good use of the big data of previous health examination data can be used to promote precision public health preventive health care services. For example, machine learning algorithms are placed in the physical examination information system, and the system regularly collects data and continuously updates the model. In the future, different types of physical examination results when inquiring, it can provide doctors and military officers and soldiers in multiple ways as a reference for promoting a healthy lifestyle.
Description
Keywords
機器學習, 深度學習, 代謝症候群, 縱貫性體檢資料, 國軍, machine learning, deep learning, metabolic syndrome, longitudinal medical examination data, army