Collaborate

Collaboration in research

One of the most important pillars of the Center for Statistical Science is the conduct of innovative research in statistical science and data science. In order to develop this innovative research, we aim to add a new flavor to the existing research network by creating a proposed collaborative research project system in the real world by members from various fields of the Graduate School, and a system for connecting researchers in the virtual world on a global scale. or "Researchers' Connected Society.".

Collaboration between Statistical Science and Other Fields

Case Studies in Pediatric Medicine

Tanabe, Y., Araki, Y., Kinoshita, M. et al. Bayesian nonparametric quantile mixed-effects models via regularization using Gaussian process priors. Jpn J Stat Data Sci 5, 241–267 (2022).

要旨

In this study, we proposed using Bayesian nonparametric quantile mixed-effects models (BNQMs) to estimate the nonlinear structure of quantiles in hierarchical data. Assuming that a nonlinear function representing a phenomenon of interest cannot be specified in advance, a BNQM can estimate the nonlinear function of quantile features using the basis expansion method. Furthermore, BNQMs adjust the smoothness to prevent overfitting by regularization. 
We also proposed a Bayesian regularization method using Gaussian process priors for the coefficient parameters of the basis functions, and showed that the problem of overfitting can be reduced when the number of basis functions is excessive for the complexity of the nonlinear structure. Although computational cost is often a problem in quantile regression modeling, BNQMs ensure the computational cost is not too high using a fully Bayesian method. Using numerical experiments, we showed that the proposed model can estimate nonlinear structures of quantiles from hierarchical data more accurately than the comparison models in terms of mean squared error. Finally, to determine the cortisol circadian rhythm in infants, we applied a BNQM to longitudinal data of urinary cortisol concentration collected at Kurume University. The result suggested that infants have a bimodal cortisol circadian rhythm before their biological rhythms are established.

Case Study in Epidemiology

This research represents an international collaborative study conducted with Professor Kolon at the Radiation Effects Research Foundation (RERF) in Hiroshima. The study employed the RERF cohort dataset, comprising approximately 5,000 Japanese adults, which is a uniquely comprehensive resource in Japan, characterized by its early initiation in 1957 and continuous longitudinal follow-up over a span of 48 years through detailed multi-item health examinations.
Focusing on the residual structure within survival analysis frameworks, the study provided rigorous empirical evidence that recurrent extreme fluctuations in body weight significantly elevate mortality risk. This association was shown to be independent of variations in health dynamics attributable to differential radiation exposure, thereby highlighting the critical impact of long-term body weight instability on survival outcomes.  

Cologne J, Takahashi I, French B, et al. Association of Weight Fluctuation With Mortality in Japanese Adults.
JAMA New Open. 2019;2(3):e190731.

Importance Weight cycling is associated with the risk of mortality from heart disease, but many studies have not distinguished between simple nonlinear (monotone) weight changes and more complex changes that reflect fluctuations. Objective To assess whether extreme body weight variation is associated with mortality after controlling for nonlinear weight changes. Design, Setting, and Participants In this prospective clinical cohort study, 4796 Japanese atomic bomb survivors were examined in the clinic as part of a biennial health examination and research program. The study consisted of a 20-year longitudinal baseline period (July 1, 1958, to June 30, 1978) and subsequent mortality follow-up of 27 years (July 1, 1978, to June 30, 2005) Participants were initially between the ages of 20 and 49 years during the baseline period and, throughout the baseline period, had no diagnoses of cardiovascular disease (CVD) or cancer and attended at least 7 of 10 scheduled examinations. Data analysis was performed from October 16, 2015, to May 13, 2016. Exposures Residual variability in body mass index (BMI) during the baseline period. Main Outcomes and Measures Outcomes were mortality from ischemic heart disease, cerebrovascular disease, other CVDs combined, other causes (except cancer), and cancer. Root mean squared error was calculated to capture individual residual variation in BMI after adjustment for baseline BMI trends, and the association of magnitude of residual variation with mortality was calculated as relative risk. Results In total, 4796 persons (mean [SD] age, 35.0 [7.3] years at first baseline examination; 3252 [67.8%] female; mean [SD] BMI, 21.2 [2.8] at first baseline visit [20.6 (2.4) among men and 21.5 (2.9) among women]) participated in the study. During follow-up, 1550 participants died: 82 (5.3% of all deaths) of ischemic heart disease, 181 (11.7%) of cerebrovascular disease, 186 (12.0%) of other CVDs, 615 (39.7%) of cancer, and 486 (31.3%) of other causes. Magnitude of residual variation in weight was associated with all-cause mortality (relative risk, 1.25 for 1 U of additional variation; 95% CI, 1.06-1.47) and ischemic heart disease mortality (relative risk, 2.49; 95% CI, 1.41-4.38). Conclusions and Relevance The findings suggest that an association exists between weight variation and heart disease mortality and that weight loss interventions, if deemed to be necessary, should be considered carefully.

Case Study in Engineering and Cognitive Science

This study was conducted in collaboration with Professor Masashi Nishida’s laboratory at the Faculty of Informatics, Shizuoka University. It introduces functional data analysis methods into the field of sign language research, proposing a novel approach to sign motion analysis based on multivariate functional principal component analysis (MFPCA). This method aims to enhance both the analytical techniques for sign language and the development of improved sign language education systems.

Given that sign language inherently consists of continuous movements, and motion-captured sign language data are naturally characterized as high-dimensional spatiotemporal data, this study applied functional principal component analysis to extract a small number of principal components representing the temporal and spatial structure of signing movements. Notably, these results revealed motion structure features that could not be effectively captured by conventional deep learning-based approaches.
The outcomes of this research were published as follows:

Kyosuke Sakurada, Yuko Araki, Yuki Izumi, and Masashi Nishida, “Multivariate Functional Principal Component Analysis of Sign Language Data,” Journal of Human Interface Society, Vol. 22, No. 4, pp. 475–484, 2020. 

The aim of this paper is to establish a novel statistical methods for characterizing the sign language movements at multiple body parts simultaneously. The method we applied is the multivariate functional principal components analysis (MFPCA), which is capable of capturing the individual variation of sign language movements using not only palm movements but also multiple movements such as fingers, elbows, and shoulders. This method successfully captures the characteristic that sign language is composed of a combination of multiple consecutive actions. We apply MFPCA to quantify the differences in variation of the performance among ten beginner and one master of the sing languages measured at nineteen body parts. The results of MFPCA quantify the individual qualities for the sign languages by making use of multivariate function principal component scores. At the same time, MFPCA revealed which part the characteristic movement of the individual sign language appears strongly. Finally, we distinguished some words that tend to be difficult or easy to learn.