基于决策树和反向传播神经网络模型的不同状态年龄人群生活行为抗衰老方案研究  

Study on anti-aging regime of life behavior of different state age groups based on decision tree and BP-neural network model

作　　者：杨善岚 吴雪琴 涂嘉欣 黄河浪[1] 朱祥 张朗朗 邓莉芳 吴磊[1] YANG Shan-lan;WU Xue-qin;TU Jia-xin;HUANG He-lang;ZHU Xiang;ZHANG Lang-lang;DENG Li-fang;WU Lei(School of Public Health of Nanchang University,Jiangxi Province Key Laboratory of Preventive Medicine,Nanchang,Jiangxi 330006,China)

机构地区：[1]南昌大学公共卫生学院,江西省预防医学重点实验室,江西南昌330006

出　　处：《现代预防医学》2023年第3期517-524,560,共9页Modern Preventive Medicine

基　　金：国家重点研发项目(2020YFC2002901);国家自然科学基金(81960620,81360446);南昌大学大学生创新创业训练计划项目(S202210403016)。

摘　　要：目的 筛选不同状态年龄人群日常生活行为,拟合DT与BP-ANN模型并比较预测性能,构建生活行为抗/缓衰老方案。方法 使用PPSHAS量表筛选状态年龄“明显年轻”和“明显老”等不同人群并开展抗衰老因素现况调查;以不同状态年龄组人群的衰老有关变量拟合以上两模型并比较预测性能,选择综合性能最优的模型以构建抗衰老方案。结果902名南北居民测得衰老度得分为(50.04±9.77)分,状态年龄得分为(67.67±17.47)分,其中明显轻者占26.39%,尤其是65岁后占比超过明显老者,在75~79年龄达到顶峰(45.16%);从59个因素中筛选得饮食、饮酒、饮茶、运动、睡眠、吸烟情况、兴趣爱好、护肤品等25个有意义的变量,经主成分分析获得9个不同载荷的主成分;训练集和验证集中结果可得,DT模型的准确率、Youden指数、Kappa指数、AUC均高于BP-ANN模型,且AUC差值均有统计学意义(P<0.05);DT模型构建的抗衰老方案提示,拥有1~2个兴趣爱好、静态行为时长≤8 h/d、睡眠时间≤8 h/d、午睡时间≤60 min/d对抗衰老有益(贡献率为71.43%~98.40%)。结论 中老年人应有1~2个健康的兴趣爱好,静态行为时长不超过8h/d;构建抗衰老方案时应综合考虑睡眠时长和午睡时间,将睡眠时间压缩至8 h/d以下、午睡时间压缩至60min/d以内,并搭配60 min/d以内的轻度体育锻炼。Objective To screen the daily life behaviors of different state age groups, fit the decision tree(DT) and back propagation artificial neural network(BP-ANN) models, and compare the prediction performance, to construct a life behavior anti-aging regime. Methods The PPSHAS scale was used to screen the state age of different groups, such as “obviously younger” and “obviously older”, and collect the information of anti-aging factors in age groups of different states. The above two models were fitted with the state age of “obviously younger” and “obviously older” as output variables, the prediction performance of the two was compared, and the better one was selected to construct anti-aging regime. Results The aging score of 902 residents in the north and the south was 50.04±9.77, and the state age score was 67.67±17.47, of which 26.39%were obviously younger, especially after the age of 65, the proportion exceeded that of the obviously older, and reached the peak(45.16%) between the ages of 75 and 79 years. Twenty-five significant variables such as diet, drinking, tea drinking,exercise, sleep, smoking, hobbies, and skin care products were selected from 59 factors, and 9 principal components with different loads were obtained by principal component analysis. The results of training set and verification set showed that the accuracy, Youden index, Kappa index, and AUC of DT model were higher than those of BP-ANN model, and the difference of AUC was statistically significant. The anti-aging regime constructed by the DT model suggested that having 1 to 2hobbies, static behavior time ≤ 8h/d, sleep time ≤ 8h/d, and nap time ≤ 60min/d were beneficial to anti-aging(the contribution rate was 71.43%-98.40%). Conclusion Middle-aged and elderly should have 1 or 2 healthy hobbies, and the duration of static behavior should not exceed 8h/d. When constructing an anti-aging regime, the duration of sleep and nap time should be taken into consideration, and the sleep time should be compressed to less than 8 h/d an

关 键 词：老年保健 衰老度测量 决策树 BP神经网络 现况调查 

分 类 号：R181[医药卫生—流行病学]