RT Journal Article
SR Electronic
T1 Big data reveals deep associations in physical examination indicators and can help predict overall underlying health status
JF bioRxiv
FD Cold Spring Harbor Laboratory
SP 855809
DO 10.1101/855809
A1 Haixin Wang
A1 Ping Shuai
A1 Yanhui Deng
A1 Jiyun Yang
A1 Shanshan Zhang
A1 Yi Yin
A1 Lin Wang
A1 Dongyu Li
A1 Tao Yong
A1 Yuping Liu
A1 Lulin Huang
YR 2019
UL http://biorxiv.org/content/early/2019/11/26/855809.abstract
AB Because of lacking of the systematic investigation of correlations between the physical examination indicators (PEIs), currently most of them are independently used for disease warning. This results in very limited diagnostic values of general physical examination. Here, we first systematically analyzed the correlations between 221 PEIs in healthy and in 34 unhealthy states in 803,614 peoples in China. We revealed rich relevant between PEIs in healthy physical status (7,662 significant correlations, 31.5% of all). However, in disease conditions, the PEI correlations changed. We further focused on the difference of these PEIs between healthy and 35 unhealthy physical status, 1,239 significant PEI difference were discovered suggesting as candidate disease markers. Finally, we established machine learning algorithms to predict the health status by using 15%-16% PEIs by feature extraction, which reached 66%-99% precision predictions depending on the physical state. This new encyclopedia of PEI correlation provides rich information to chronic disease diagnosis. Our developed machine learning algorithms will have fundamental impact in practice of general physical examination.