建立與分析衰弱症的血壓波與皮膚導電度資料庫並發展預測模型

Abstract

預防醫學概念的興起使得檢測相形重要，非侵入性的測量相較傳統之侵入性檢測提供了更舒適及安全的診斷方式，同時可以通過分析大量醫療數據來預測疾病，因此受到了廣泛的關注。本研究旨在利用兩種非侵入性儀器—測量血壓波之脈診儀及測量皮膚導電度之經絡能量分析儀，以衰弱症作為疾病標的，進行早期預測並分析可能的衰弱機制。衰弱是一種臨床生理狀態，當個體處於此狀態下時，易遭受與健康相關之不良事件。根據Fried衰弱表型，衰弱臨床表徵包括自述疲憊感、低身體活動量、步行速度緩慢、手握力減弱和非刻意的體重減輕。 本研究為一項橫斷式研究，於同一時間點招募衰弱、衰弱傾向及非衰弱三群老年人，測量其血壓波、皮膚導電度和進行衰弱相關評估。本試驗共招募了311位研究參與者，其中非衰弱162人，衰弱傾向130人，衰弱19人。通過多元線性迴歸之分析，初步確定了皮膚導電度中第七（脾經）及第十（膀胱經）個穴位點，血壓波中第零（心包經）及第九個諧波（三焦經）為與衰弱相關之重要因子，且透過勝算比之分析，了解皮膚導電度及血壓波與衰弱呈現負相關性，當皮膚導電度及諧波分量較高時，罹患衰弱的風險會降低。最後，利用邏輯式迴歸及支持向量機(SVM)進行衰弱的分類，皮膚導電度的預測效果在邏輯式迴歸及SVM分別達到66.6%及61.7%的準確度，血壓波則分別達到65.0%及63.8%的準確度，模型預測效果較低，未來需增加數據量或進行資料強化，並嘗試更多不同種類的機器學習演算法，以建立預測效果良好的診斷模型。 本研究發現皮膚導電度及血壓波可做為衰弱的潛在指標，同時也可作為衰弱臨床快速診斷的依據，並透過建立預測模型，達到及早預防之目的。

Nowadays, the rise of preventive healthcare concept highlights the importance of measurement. Non-invasive measurement provides a more comfortable and safer diagnosis than invasive measurement. The combination of non-invasive measurement and the analysis of large amounts of medical data to predict diseases has garnered widespread attention. In this study, we aim to use frailty as a disease model to investigate the application of two non-invasive devices, a Pulse sensor for measuring pulse wave and a Meridian Energy Analysis Device for measuring skin conductance, for early prediction and analyzing the possible frailty mechanism. Frailty is a clinical state in which individuals are prone to experience adverse health-related events. According to Fried’s frailty phenotype, the clinical syndromic definition of frailty includes self-reported exhaustion, low physical activity, slow walking speed, reduced grip strength, and unintentional weight loss. Here, we conducted a cross-sectional clinical trial with frail, pre-frail, or non-frail participants and measured their pulse wave, skin conductance, and frailty-related assessment. The number of eligible participants is 311, including 162 non-frail, 130 pre-frail, and 19 frail subjects. Through multiple linear regression analysis, the 7th (spleen meridian), and 10th (bladder meridian) acupoints in skin conductance, as well as the 0th (pericardium meridian) and 9th (sanjiao meridian) harmonics in pulse wave would determine the progress of frailty. Furthermore, analysis of odds ratios revealed a negative correlation between skin conductance, pulse wave, and frailty. Higher levels of skin conductance and harmonic were associated with a reduced risk of frailty. Finally, logistic regression and support vector machine (SVM) were used for frailty classification. The prediction accuracy of skin conductance reached 66.6% with logistic regression and 61.7% with SVM, while pulse wave prediction accuracy reached 65.0% with logistic regression and 63.8% with SVM. The model prediction performance was relatively low, indicating the need for increasing the data volume or data augmentation, and exploring more diverse machine learning algorithms to establish a more effective diagnostic model. This study claims that skin conductance and pulse waves can serve as potential indicators of frailty, providing a basis for rapid clinical diagnosis of frailty. By establishing predictive models, early prevention objectives can be achieved.