Dealing with Sepsis poses a critical challenge in healthcare and necessitates rapid and well-adapted treatment responses. Corticosteroids have been used as a treatment but individual-level effects vary widely. This study aims at improving treatment efficacy by leveraging machine learning techniques to predict patients' sensitivity to corticosteroids. We use two comprehensive datasets of sepsis patients to evaluate two distinct model configurations. These configurations employ Logistic Regression and Random Forest algorithms, both with and without class balancing using the Synthetic Minority Over-sampling Technique for Nominal and Continuous (SMOTE-NC) data augmentation to address mixed data types. Our findings consistently demonstrate that Random Forest models, particularly when paired with appropriate class balancing techniques, outperform other model configurations in predicting corticosteroid sensitivity using both datasets individually and combined. Notably, incorporating SMOTE-NC significantly enhances model performance, underscoring the importance of appropriately addressing imbalanced datasets in predictive modeling.

The purpose of this paper is to extend standard finite mixture models in the context of multinomial mixtures for spatial data, in order to cluster geographical units according to demographic characteristics. The spatial information is incorporated into the model through the mixing probabilities of each component. To be more specific, a Gibbs distribution is assumed for prior probabilities. In this way, assignment of each observation is affected by neighbors' cluster and spatial dependence is included in the model. Estimation is based on a modified EM algorithm which is enriched by an extra, initial step for approximating the field. The simulated field algorithm is used in this initial step. Simulation studies are also provided to examine the ability of the methodology to properly cluster the data and reveal the true parameters, while focus is given in the performance of the approximated BIC on recovering the true number of clusters. The presented model will be used for clustering municipalities of Attica with respect to age distribution of residents. The results of the analysis have revealed eight distinct clusters. Each cluster includes municipalities with similar age structure of the population, as for example group of municipalities with excess of young adult people because of the universities in these regions.

Mendelian randomization (MR) is an epidemiological tool that is used to infer the causal relationship between an exposure and an outcome through genetic variants as instrumental variables (IVs) for the exposure. The horizontal pleiotropy can cause biased inference in MR. We conduct a mixture of the pleiotropic variant detection method, called the Mixture of Horizontal Pleiotropy Detection (MPD), for considering the horizontal pleiotropy on correlated variants in two-sample summary data. The MPD uses the inverse variance weighted (IVW) regression without an intercept to identify the invalid IVs from a group of the IVs with directional pleiotropy, while using the Egger regression with an intercept to identify the invalid IVs from a group of the IVs with balanced pleiotropy. The MPD is consequently more detailed for searching for the correlated variants associated with the outcome conditionals on the exposure, in comparison with the current pleiotropic variant detection methods using the IVW- or Egger-based regression without or with an intercept in MR. We use the simulated data and analyze the real-world data with the exposure, low-density lipoprotein cholesterol, on the outcome, coronary artery disease, to investigate the finite-sample properties of the MPD on the pleiotropic variant detection.

This paper proposes a variable selection method that combines instrumental variables and adaptive Elastic Net penalty for the spatial panel quantile autoregressive (SPQAR) model with fixed effects. This method can effectively identify key variables, estimate spatial effects, and address collinearity among variables while controlling for individual fixed effects. This paper gives the variable selection algorithm and establishes the large-sample properties of the penalized estimators. Numerical simulation results show that the adaptive Elastic Net method outperforms existing approaches in terms of estimation accuracy and variable selection precision, particularly under conditions of high collinearity and non-normal disturbances. Finally, this method is applied to analyze the impact of 13 explanatory variables on agricultural carbon emissions in China at different quantile levels. Both simulation and empirical results demonstrate the feasibility and effectiveness of the proposed method.

In this paper, we introduce statistical modeling strategies for assessing the effects of socioeconomic factors such as poverty, income level, and income inequality on COVID-19 mortality across the different phases of the pandemic. In doing so, we consider Bayesian spatial, spatio-temporal and non-spatial models, and discuss relevant inference results. Our findings indicate that deteriorating socioeconomic factors lead to higher mortality rates when we accurately account for spatial effects across neighboring units. In addition, we investigate the effects of temporal variations in socioeconomic covariates on relevant spatial units over time. We provide insights that can be useful for policy makers and public health decision makers. Our numerical analysis focuses on publicly available data merged from various federal as well as state level sources, with an emphasis on the state of California.

Graph-based clustering methods partition data by leveraging neighborhood relationships among data points. These methods are often sensitive to the predefined affinity matrix. To achieve adaptive and optimal clustering, recent studies have introduced techniques that learn adaptive neighbors automatically. However, these methods still depend on an essential parameter: the number of neighbors, which must be specified in advance. To address this limitation, this paper proposes a novel objective function that incorporates a density measure into a self-weighted adaptive neighbor clustering framework. Extensive analysis and experimental results show that the proposed approach exhibits reduced sensitivity to the number of neighbors and achieves superior performance compared to several established clustering algorithms. Furthermore, it preserves the fundamental advantages of self-weighted adaptive neighbor clustering.

Ideal point estimation is a widely used statistical method for understanding the preferences of elected representatives in political science, statistics, and social sciences. However, protest votes – where individuals deliberately obscure their true ideal points to express dissatisfaction with their own political party – present a significant challenge to the accuracy of this method. In this paper, we first examine the impact of protest votes on ideal point estimation, demonstrating that they introduce substantial attenuation bias that leads to the misrepresentation of extreme legislators as moderates. After establishing the importance of this issue, we propose a novel method that corrects the bias stemming from protest votes, thereby allowing researchers to obtain more accurate estimates of legislators’ ideal points. Our method detects and masks votes suspected to be protest votes within a Bayesian framework, reducing the bias introduced by such votes in posterior inference of ideal points. We demonstrate the effectiveness of our proposed method in addressing the attenuation problem caused by protest votes using both simulated scenarios and real-world roll-call data.

Misclassification of correlated binary responses may occur in clinical and epidemiological studies, resulting in biased and/or inefficient parameter estimation. We extend existing generalized estimating equation (GEE) approaches to allow for differential misclassification via two sets of estimating equations for the analysis of error-prone correlated binary outcomes when internal validation data are available. One set of estimating equations uses logistic regression to model the misclassification process with the internal validation data, using subject characteristics to estimate the differential sensitivities and specificities of the error-prone response in relation to the gold standard. The second set of estimating equations models the mismeasured binary response by leveraging the subject-specific estimates of sensitivity and specificity. The subject-specific covariates need not be identical in the two sets of estimating equations. We present analysis of longitudinal assessments of bacterial vaginosis from the HIV Epidemiology Research Study (HERS), compare the proposed population-average approach based on GEE with a subject-specific one based on a full-likelihood mixed-effects analysis, and discuss the differing parameter interpretations. We also present results from simulated data with a misclassified binary outcome analyzed with this GEE approach.

Testing for treatment effects without adjusting for the covariates may lead to erroneous conclusions. In many situations, ignoring variance heteroscedasticity can have serious consequences. In this article, new test procedures are proposed to test the homogeneity of treatment effects in a one-way ANCOVA model with heteroscedastic error variances. Most of the existing studies on this problem deal with non-parametric or semi-parametric situations. In a fully parametric setting, earlier one plug-in test statistic has been proposed using bootstrap approach. In this study, we develop the likelihood ratio test (LRT) and a test utilizing pair-wise differences between treatment effects. The parametric bootstrap is employed for determining the critical points. Extensive simulation studies demonstrate that the proposed tests perform better than the existing tests in some situations. The applicability of these approaches is illustrated using three datasets. Some ‘R’ functions are made available in ‘GitHub’ for easy computation.