台灣人口結構呈現高齡化社會,因醫療水準和衛生環境的改變,使得人口年齡慢慢趨升。銀屑病伴隨睡眠障礙是一種慢性疾病,慢性疾病在老年人口是主要死亡的主要原因,心血管疾病在台灣十大死因中之一的疾病,研究顯示出高血壓、糖尿病、高血脂、酒精中毒的條件是公認的危險因素。 本研究以近年某醫療機構資料庫中銀屑病伴隨睡眠障礙患者為研究對象,透過文獻探討與醫師訪談,篩選出增加罹患心血管疾病風險之重要因子後,運用人工智慧中粒子群最佳化演算法、基因邏輯斯迴歸演算法與交叉熵演算法計算因子權重並分別結合倒傳遞類神經網路與支援向量機以建構六種預測模型與三種評估系統,評估是否有罹患心血管疾病的風險。研究結果顯示,雖以粒子群最佳化演算法結合案例式推理評估系統有較高之準確度,但傅利曼檢定結果顯示三種演算法之權重值對於計算相似度不存在顯著差異,且準確率與ROC曲線下面積皆達到83%及0.812以上,皆適合作為評估系統之權重值計算;預測模型部分,雖然粒子群最佳化演算法結合支援向量機有較高之準確率,十疊驗證後平均準確率為89.73%,ROC曲線下面積為0.871,但六種預測模型經傅利曼統計檢定後,並不存在顯著差異。本研究結果能提供醫療機構及臨床工作者,作為輔助診斷之參考依據,給予病患正確治療使能減輕病患疾病負擔。
Taiwan’s population structure presents an aging society. Due to the changed medical standard and health environments, the population age has gradually risen. Psoriasis with accompanying sleep disorder is a chronic disease. Chronic disease is the main cause of death for the elderly population. Cardiovascular disease is one of the ten leading causes of death in Taiwan. Research shows that hypertension, diabetes, hyperlipidemia, and alcoholism are conditions recognized as risk factors. In this study, psoriasis patients with accompanying sleep disorder from the recent database of an anonymous medical institution were adopted as research participants. Through literature reviews and interviews with physicians, the important factors contributing to increased risk of cardiovascular disease were screened. Using the algorithm of artificial intelligence such as particle swarm optimization algorithm, genetic logistic regression algorithm, and cross entropy algorithm, the factor weights were calculated. The back propagation neural network and support vector machine were conjunctively used to construct six predictive models and three assessment systems to evaluate the risk of contracting cardiovascular disease. Research results show that although the particle swarm optimization algorithm combined with the case-based reasoning assessment system derive at higher accuracy, the Friedman’s test results show that the weights of the three algorithms produce no significant differences on the similarity calculation. Moreover, the accuracy and the area under the ROC curve reach above 83% and 0.812% respectively, indicating both are suitable for computing the assessment system weight. For the predictive model part, although the particle swarm optimization algorithm coupled with support vector machine produce higher accuracy, the ten-fold cross validation shows the average accuracy of 89.73% and the area under the ROC curve of 0.871. However, the six predictive models after the Friedman’s test show no significant differences. The research results shall be provided to medical institutions and clinical workers as references in aided diagnosis and for correct treatment to patients that will help lessen their disease load.