摘要
Accurate blood pressure （BP） measurement is essential in epidemiological studies, screening programmes, and research studies as well as in clinical practice for the early detection and prevention of high BP-related risks such as coronary heart disease, stroke, and kidney failure. Posture of the participant plays a vital role in accurate measurement of BP. Guidelines on measurement of BP contain recommendations on the position of the back of the participants by advising that they should sit with supported back to avoid spuriously high readings. In this work, principal component analysis （PCA） is fused with forward stepwise regression （SWR）, artificial neural network （ANN）, adaptive neuro-fuzzy inference system （ANFIS）, and the least squares support vector machine （LS-SVM） model for the prediction of BP reactivity to an unsupported back in norrnotensive and hypertensive participants. PCA is used to remove multi-collinearity among anthropometric predictor variables and to select a subset of components, termed ＇principal components＇ （PCs）, from the original dataset. The selected PCs are fed into the proposed models for modeling and testing. The evaluation of the performance of the constructed models, using appropriate statistical indices, shows clearly that a PCA-based LS-SVM （PCA-LS-SVM） model is a promising approach for the prediction of BP reactivity in comparison to others. This assessment demonstrates the importance and advantages posed by hybrid models for the prediction of variables in biomedical research studies.
目的:准确测量血压(BP)对于流行病学研究、筛查规划、调研研究和高血压相关病变(冠心病、中风、肾衰竭等)的早期诊断及预防有重要意义。被测者体位对于准确测量血压有重要影响。血压测量指南建议测试时被测者应在后背有支撑的情况下保持坐姿,以避免血压读数偏高。本文使用混杂模型预测血压对于无支撑后背的反应。创新点:本文考虑血压正常和高血压测试者的人体预测变量(如年龄、身高、体重、体块指数和上臂周长(AC)),使用基于PCA的前向逐步回归(PCA-SWR)、基于PCA的人工神经网络(PCA-ANN)、基于PCA的自适应神经模糊推理系统(PCA-ANFIS)和基于PCA的最小方差支持向量机(PCA-LS-SVM)等模型预测血压对无支撑后背的反应。方法:使用PCA消除人体预测变量间的多重共线性,并在原始数据集中选取主元(PC)。所选主元被输入至所建立预测模型用于建模及测试。结论:通过评估合适的统计指标(确定性系数、平均平方根误差、平均绝对百分比误差),得出较之其他模型,PCA-LS-SVM对于预测血压反应较有前景。此评估也展示了混杂模型在预测生物医学领域其他参数时的重要性和先进性。
