Young-Geun Choi’s research while affiliated with Sungkyunkwan University and other places

Background In studies with small sample sizes, the use of nonparametric methods is generally recommended for statistical analysis. However, various studies continue to employ parametric analysis without verifying the assumption of a normal distribution. Objectives To assess the current utilization of parametric and nonparametric methods for primary outcomes, as well as the reporting of normality assumptions, in phase 3 clinical trials with small sample sizes. Methods All phase 3 trials registered on ClinicalTrials.gov until September 12, 2023, were collected. After undergoing a two-step selection process, only publications with a sample size per group of less than 30, involving two or more groups, and specifying the statistical methods used to compare the means or medians of primary outcomes between groups were selected. Statistical methods were categorized as nonparametric, parametric, and either parametric or nonparametric. The reporting of normality assumptions was also evaluated. Results A total of 317 studies were assessed in this study. Among these studies, 164 (51.7%) studies conducted parametric analysis, and 111 (35.0%) studies employed nonparametric analysis; however, 42 (13.2%) studies conducted parametric or nonparametric analysis without specifying which method was used. In addition, 63.1% of the total studies did not report normality assumptions. Specifically, within the subset of studies with parametric analysis, 70.1% of studies did not report normality assumptions. Conclusions This research demonstrated that most studies with small sample sizes employed parametric analysis without reporting normality assumptions. The findings emphasize the need for increased awareness of and compliance with statistical principles in the analysis of phase 3 clinical trials with limited sample sizes.

Mixed data refers to a mixture of continuous and categorical variables. The clustering problem with mixed data is a long-standing statistical problem. The latent Gaussian mixture model, a model-based approach for such a problem, has received attention owing to its simplicity and interpretability. However, these approaches are prone to dimensionality problems. Specifically, parameters must be estimated for each group, and the number of covariance parameters is quadratic in the number of variables. To address this, we propose “regClustMD,” a novel model-based clustering method that can address sparse dependence among variables. We consider a sparse latent Gaussian mixture model, assuming that the precision matrix between variables has sparse nonzero elements. We propose maximizing a penalized complete log-likelihood using the Monte Carlo expectation-maximization (MCEM) algorithm. We demonstrate our method through a simulation study and real-world data examples.