研究目的：本研究以2005年到2007年間，由一般正常血壓狀態轉為高血壓之樣本為對象，利用相關文獻所提及之高血壓危險因子，並利用網狀致病模式的觀點，使用決策樹演算法，尋找出造成高血壓的危險因子結構，並發展台灣民眾高血壓風險分類準則，以提供未來公共衛生領域的參考依據，介入未曾發病的人群中以達到預防保健之功能。研究方法：本研究之研究對象來源採用國家衛生研究院與國民健康局共同進行之2005年「國民健康調查（National Health Interview Survey, NHIS）」及衛生署中央健康保險局2005年~2007年之「健保資料庫（National Health Insurance Research Database, NHIRD）」串聯兩資料庫之資料，並依研究目的進行資料篩選，最後以描述性統計、CHAID演算法、Logistic迴歸進行研究分析。研究結果：本研究樣本7,548人，利用相關文獻所提及之高血壓危險因子作為研究變項，在第一階段之CHAID分類中找到形成高血壓之主要因子為是否年齡55歲以上、是否過重或肥胖以及4類規則，再將這些規則作為第二階段CHAID分類之目標變項，並找出形成高血壓之次要因子為教育程度、性別、運動習慣、嚼檳榔習慣、飲酒習慣及11類規則，最後在第三階段之CHAID分類將第二階段之規則重新分類得到形成高血壓生活型態之規則3類。最後透過Logistic迴歸分析來檢驗這些規則的預測力，做為本研究之結果。結論：本研究之目的是要尋找出造成高血壓的危險因子結構，並發展血壓風險分類準則，得到之結果為55歲以上肥胖者，且為低教育程度者則為具有高血壓風險族群、55歲以下非肥胖者，且高教育程度者則為低高血壓風險族群，在其他類別方面仍無法正確判斷，這一方面可待後續研究繼續探討。 Research purpose: This study aims to apply decision tree approaches to build a model of hypertension risk factors recognition. Based on the perspective of web of causation on hypertension, we adopted the hypertension risk factors in the literatures as the structure nodes with the application of creating the decision tree approach.Methods: Research sample were enrolled based on the 2005 National Health Interview Survey (NHIS) whose participants with the newly diagnosed hypertension patients between 2005 and 2007. The data were collected by combining the NHIS sample and medication record of National Health Insurance Research Database (NHIRD) between 2005 and 2007. Data was analyzed by applying descriptive statistic approaches, CHAID algorithm, and Logistic regression approach. Results: A totally 7,548 participants were enrolled in the model. In the first step, we found 4 major rules of classifying the hypertension. The major risk factors were the age with a cut-point of 55 years old and BMI with the cut-point of overweight and obesity. In the second step, we found 11 sub-major rules classifying the 4 major rules with risk factors of education level, gender, regular physical activities, betel-chewing behavior, and drinking behavior. In the third step, we applied the CHAID algorithm with the sub-major rules as the structure nodes to create 3 rules of hypertension recognition. Finally, the model was evaluated by Logistic regression approach. Conclusions: With this model we conclude that the structure of recognizing hypertension risk factors in Taiwan include two major rules. The first rule is patients aged above 55 years old with obesity and lower education level can be predicated as higher risk group. The second rule is patients aged less than 55 years old with higher education level can be predicted as lower risk group.
