Development of statistical methods utilizing longitudinal data and time to event data effectively.

针对风险因素对疾病发展的影响，提出了一种能够检测其影响的非线性的建模算法，并通过仿真研究验证了该算法的有效性。我们发现使用联合建模可以减少回归稀释偏差的影响，即联合建模对于观察性研究可能是有用的。在营养流行病学中，通过生物标志物的降阶回归提取的饮食模式的可变性很大，因此该模式的概括性可能是有问题的。此外，我们还进行了许多流行病学研究，并在国际期刊上发表了许多论文。

For effect of risk factors on development of diseases, we proposed a modeling algorithm which can detect nonlinearity of the effect and confirmed performance of it by a simulation study. We showed that using joint modeling can reduce effect of regression dilution bias, i.e., joint modeling may be useful for observational studies. In nutrition epidemiology, variability of dietary patterns extracted by using reduced rank regression with biomarkers is large so generalizability of the pattern may be problematic. Moreover, we conducted many epidemiological studies, and many papers were published in international journals.