Development of risk reduction for health promotion strategy based on data-mining

由于日本最近的疾病结构发生了变化，为预防与生活方式有关的疾病，出台了诸如“健康日本21”或“健康促进计划”等卫生政策。在这种情况下，预计不仅对疾病进行筛查，而且对减少风险也将在卫生政策中发挥重要作用。风险降低模型的建立需要健康状况的早期变化和纵向变化的评价。然而，普通的风险分析方法对于早期变化的发现存在一定的局限性。本研究采用基于数据挖掘的非线性方法对风险降低策略进行评估。遗传算法和免疫系统算法也包括在本项目中。对ICU获得性感染的病因分析采用流行病学分析方法（如比例危害模型、神经网络模型）进行比较。比较集合单点模型、多点模型和比例风险模型的ROC曲线下面积和分类精度。结果表明，多点模型优于比例风险模型和集合单点模型。对于风险降低模型的开发，非线性模型似乎更适合于数据挖掘策略。数据库的设置似乎对正确的知识获取也很重要。Harell等人提出的一致性指数将是对某些模型进行优先级比较的考虑事项。

Because of the recent change in disease structure in Japan, health policies such as Healthy Japan 21 or Health Promotion low were introduced for he prevention of lifestyle-related diseases. In this circumstance, not only screening to diseases but also risk reduction are expected to be valuable in health policy. Development of risk reduction model requires the early change of health status and evaluation of longitudinal change. However, the ordinary risk analysis methods have the limitation for the detection of the early changes.This study conducted the non-linear methods based on data-mining for assessment of the risk reduction strategies. Genetic algorism and immune-system algorism were also included in this projects. Comparison of epidemiological analysis such as proportinal hazard model, neural network modes were performed for the etiological analysis for ICU acquired infections. Aggregate single point model, multiple point model and proportional hazard model are compared in the area under the curve of ROC and classification accuracy. This result showed the multiple point model were superior to proportional hazard model and aggregate single point model. For development of risk reduction model, non-linear model seems to be appropriate for data-mining strategies. The cimposition of the database also seems to be important for the proper knowledge acquisition. Concordance index proposed by Harell et al will be matter under consideration for the comparison of the priority to some models.