基于模糊神经网络的老年人体质综合评价与预测模型研究

The constitution was a reflection of a country’s comprehensive national power, and the elderly viewed health as the first need of lives; so studies on the evaluation of elderly constitution were showed more urgently. How to accurately assess the constitution? The researchers at home and abroad have put forward many measuring methods at present. However, these methods were more focused on physiological indexes such as body shape, function and athletic ability and so on, failed to comprehensively embody the relationship between many domains including physiology, psychology, society, et al. and the constitution and less the measuring standard of constitution for the elderly. Therefore, the idea that the constitution will be evaluated and forecasted by the fuzzy neural network model was proposed innovatively on the basis of using mathematical models to predict and evaluate disease at past. In this study, reliable and recognized for good factors of the constitution among the elderly will be screened through systematic reviews and follow-up studies of large populations in city communities, then looked as input variables; ultimately, the fuzzy neural network model which will be applied to detect in the community elderly population is established. This research not only makes up for the deficiency of the current measuring methods of human body constitution, but also possesses an important value for promoting researches in Geriatrics. Especially, it will provide an important foundation in assessing the inherent essence of the elderly's constitution scientifically and comprehensively, predicting and preventing chronic diseases and establishing the criterion of constitution for the elderly.

体质是一个国家综合国力的体现，老年人更是把健康作为生活的第一需要，因此老年体质评价的研究越显迫切。如何准确评估体质，目前国内外提出了众多的测量方法，但这些方法多侧重于形态、机能和运动能力等生理指标，未能全面体现生理、心理、社会等多领域与体质的关系，且少有针对老年人群的体质测量标准。为此，申请者在以往数学模型预测和评估疾病的基础上，创新性地提出利用模糊神经网络（FNN）模型来评价和预测体质。本研究拟通过文献系统评价和社区大样本人群的随访研究，筛选出老年体质可信度高、公认度好的相关因子（形态、机能、运动能力和身体素质、心理和社会适应能力等）为输入变量，建立适用社区老年人群检测的FNN模型。 本研究不仅弥补了目前人体体质测量方法的不足，而且对推动老年医学的研究具有重要价值，尤其对科学、全面评估老年人体质的内在实质、预测和预防慢病，为该人群体质判断标准的建立提供重要依据。

我国正面临着前所未有的快速老龄化现实，体质被认为是老年人高生活质量的关键先决条件，如何正确评估体质并加以干预，对缓解老年健康问题、促进健康老龄化具有重要的现实意义。本研究采用文献搜集、Delphi、层次分析、探索性因子分析等方法，筛选出老年人体质可信度高、公认度好的相关因子，形成 “形态、机能、运动能力和身体素质、心理和社会适应能力”五维度体质指标体系，在此基础上，分别构建老年人体质测评和预测的曲线方程、模糊神经网络（FNN）、随机森林（RF)和支持向量机（SVM)模型，并应用于社区。主要结果：（1）构建的曲线方程以Cubic和Compound模型决定系数（R2）最高，曲线方程为：yˆ男=330.040-6.096x+0.032x2，yˆ女=180.524-2.155x+0.006x2;（2）体质评价的机器学习模型中，FNN、RF和SVM模型的准确度均>0.65，且FNN模型的预测效果优于RF和SVM模型，其中男性准确度为0.703,　95%CI[0.662　-0.741]，女性为0.754，95%CI[0.720-0.786]。本研究依据量表的个体评分和曲线方程模型，可较直观的得出某个体的体质水平位于相同年龄人群的百分顺位和初步判断老年人的体质等级，便于社区 60 岁以上个体的体质综合评估和随访监测；所建立的机器学习模型具有客观、稳定、适用社区大样本老年人群检测的优点，尤其对科学、全面评估老年人体质的内在实质、预测和预防慢病，为老年人体质判断标准的建立提供重要依据。