Contributed Poster Presentations: Isolated Statisticians

This article aims to study efficient/trace optimal designs for crossover trials with multiple responses recorded from each subject in the time periods. A multivariate fixed effects model is proposed with direct and carryover effects corresponding to the multiple responses. The corresponding error dispersion matrix is chosen to be either proportional or the generalized Markov covariance type, permitting the existence of direct and cross-correlations within and between the multiple responses. The corresponding information matrices for direct effects under the two types of dispersions are used to determine efficient designs. The efficiency of orthogonal array designs of Type I and strength 2 is investigated for a wide choice of covariance functions, namely, Mat(0.5), Mat(1.5) and Mat(∞). To motivate these multivariate crossover designs, a gene expression dataset in a 3×3 framework is utilized. 

In recent years, electronic medical record data has been increasingly utilized, and there is a growing need for trajectory and classification of electrolyte concentration data obtained from patients' routine blood drawings.
In this study, Group-Based Trajectory Modeling (GBTM) was applied to medical record data of patient in the terminal phase of the disease in Japan.
It is important to predict the prognosis and sudden change of condition of terminally ill patients, and the classification of trajectories and their eventual inclusion in the medical record will be useful for patient prediction.
The study utilized blood collection data from 4,981 terminally ill patients admitted to a Japanese university hospital between 2008 and 2016.
Focusing on potassium levels, which are known to decrease significantly at the end of life, GBTM was applied. The five groups that best fit the BIC were: very low [1] (7.2%), low [2] (32.5%), normal [3] (15.6%), high [4] (32.9%), and very high [5] (11.8%).
 

The limiting spectral distribution (LSD) of high-dimensional Kendall's correlation matrix and tests of independence based on this matrix have been studied in the literature when the observations are absolutely continuous with respect to Lebesgue measure, and are independent and identically distributed. Here the investigation is focused on the LSD of Kendall's correlation matrix under much weaker assumptions which accommodate discrete and/or non-identical distributions, and also identify the limit distribution by using free probability. A graphical test of independence was also proposed under these assumptions. 

The decrease in demand and dropout rates in undergraduate programs in exact and technological sciences are problems that have implications for society, affecting the job market and potentially leading to a shortage of professionals. Statistical and machine learning methods can contribute to a better understanding of this phenomenon and assist in the design of actions at universities. The objective of this work is to outline the student profile and identify factors for dropout, completion, and retention in undergraduate programs. The research was conducted on a technological sciences campus, including students in engineering, computing, and mathematics teaching. Data from the academic system were used. Exploratory analysis and data visualization techniques were applied to construct the profile. Logistic regression and tree-based models were used to identify the success factors in program completion. The partial results indicate that the student profile varies according to the program. In mathematics and engineering programs, the dropout rate is high and first-phase courses have a strong impact on student dropout,indicating the need for specific actions aimed at incoming students. 

Statistical test for high-dimensional means under missing observations seems to be very rare in the literature. We propose a new two-sample test for high-dimensional means based on independent observations with missing values. The critical region of the proposed test is based on a bootstrap estimate of the sample quantiles of the proposed test statistic. Unlike the existing tests, this test does not require any distributional assumptions or any particular correlation structure of the covariance matrices. We establish the Gaussian approximation result for the proposed test statistic which is a non-trivial extension of the two-sample Gaussian approximation result with no missing values. The rate of accuracy of the bootstrap approximation of the sample quantile of the proposed test statistic is also derived. This Gaussian approximation result and the accuracy of the bootstrap estimators together provide the theoretical guarantees on the size and power of the proposed test.