Epidemiological isk assessment based on structural equations with latent variables

由于初级卫生保健的重要性，风险评估引起了人们的极大兴趣。风险评估通常基于从流行病学数据库中获取的知识。统计分析已经成为从数据库中获取知识的标准方法。然而，这种分析对于评估非线性条件似乎还不够合适。本项目旨在将具有潜在变量的结构方程引入到非线性条件下的流行病学风险评估中。流行病学数据收集自富山研究(富山县儿童队列研究)。从3岁开始计算肥胖儿童在小学发生的风险。身体成分正常。比较了隐变量结构方程模型、Logistic模型和神经网络模型的预测结果。结构方程模型显示了利用潜在变量进行有效分析的优势。根据拟合度指数、调整后的拟合度指数、Akaike的信息准则对模型进行了评价。结果表明，潜变量应作为观测变量的乘积加入到方程中。结构方程也应该从医学因果关系的角度进行评估，这种结构方程表现出了足够的表达非线性关系和专业知识的能力。然而，与Logistic和神经网络模型相比，结构方程中的个体风险估计是困难的。

Risk assessment has attracted much interest according to the importance of primary health care. Risk assessment is usually based on the knowledge acquisition from epidemiological database. Statistical analysis has been standard method for getting knowledge from database. However, this kind of analysis seems to be not enough appropriate to evaluate the non-linear conditions. This project aims to introduce the structural equations with latent variables to epidemiological risk assessment in the non-linear conditions.The epidemiological data was gathered from the TOYAMA study(cohort study of children in Toyama Prefecture). The risk was calculated for the occurrence of the obese child in the elementary school from 3 y.o. normal body composition. The results were compared between structural equations with latent variables, logistic model and neural network model. The structural equation model showed the advantage of effective analysis by using the latent variables. The models were evaluated based on goodness of fit index, adjusted goodness of fit index, Akaike's information criterion. The results indicated the latent variables should be incorporated into the equations as the products of observed variables. Structural equations should be also evaluated from the viewpoints of medical causal relationships.This structural equations showed enough abilities to express the non-linear relationships and the expertise. However, estimation of individual risk was difficuit in the structural equations compared with logistic and neural network models.