International Journal of Statistics in Medical Research

Extreme Heterogeneity in Global Prevalence Meta-Analyses: Evaluating Current Practices and Exploring Bayesian Alternatives - an Umbrella Review

Víctor Juan Vera-Ponce — 2026-01-30

Introduction: Global prevalence meta-analyses often exhibit extreme heterogeneity (I² > 90%), yet criteria designed for clinical trials, where homogeneity is desirable, continue to be applied without recognizing that in prevalence studies, variability reflects real differences between populations.

Objective: To document the magnitude of heterogeneity in global prevalence meta-analyses, evaluate the methodological strategies employed for its exploration and management, and explore through illustrative application how Bayesian methods—rarely employed in prevalence meta-analyses—compare with standard frequentist approaches.

Methods: Umbrella review conducted according to PRIOR guidelines. Systematic search in SCOPUS for systematic reviews with global/worldwide prevalence meta-analyses published between 2015-2025. Data were extracted on I², statistical models, subgroup analyses, sensitivity analyses, meta-regression, and prediction intervals. Three meta-analyses were randomly selected for illustrative Bayesian re-analysis using hierarchical models with weakly informative priors, and the results were compared with those from frequentist approaches.

Results: Of 53 included meta-analyses, 52 (98.1%) presented I²≥75%, 47 (88.7%) I²≥90%, and 34 (64.2%) I²>99%. Management strategies showed a decreasing implementation rate: subgroup analyses (96.2%), sensitivity analyses (64.2%), meta-regression (34.0%), and prediction intervals (5.8%). Among studies with I²≥75%, 63.5% provided explicit justification for proceeding with pooling. The illustrative Bayesian analysis of three randomly selected studies demonstrated excellent concordance with frequentist estimates (differences <0.1%), while providing additional uncertainty quantification for heterogeneity parameters unavailable from standard approaches.

Conclusions: Extreme heterogeneity constitutes the norm in global prevalence meta-analyses. The underutilization of prediction intervals (5.8%) and meta-regression (34.0%) represents the critical gap for improving statistical rigor. An exploratory Bayesian analysis demonstrated concordance with frequentist estimates, while providing additional uncertainty quantification. This illustrates that alternative methods are feasible, though their value lies primarily in specific scenarios rather than routine application. Prevalence-specific frameworks should recognize high heterogeneity as an expected characteristic requiring comprehensive exploration rather than elimination.

Mortality Prediction and Survival Estimation in Dialysis Patients Using Logistic and Cox Regression with Machine Learning Feature Selection

Vajala Ravi — 2026-01-30

Mortality remains high among patients undergoing maintenance dialysis for end-stage renal disease (ESRD). Identification of key mortality predictors is paramount for improving prognosis and guiding care. Recent advances in machine learning (ML) offer potential to enhance risk stratification beyond traditional statistical models. This study compares feature selection methods—LASSO, Random Forest, and Gradient Boosting—in predicting mortality risk among dialysis patients, integrating logistic regression and Cox proportional hazards modelling. Retrospective data from 224 ESRD patients on maintenance haemodialysis were analysed. Thirty-three clinical and demographic variables were evaluated. Feature subsets were generated using ML algorithms and used for building predictive models. Model performance was assessed via discrimination (AUC), accuracy, sensitivity, specificity, and survival prediction concordance index (C-index). LASSO-selected features yielded an AUC of 0.82 and C-index of 0.81, demonstrating strong discriminatory ability. Random Forest showed highest AUC (0.85) but lower sensitivity. Gradient Boosting offered balanced sensitivity and specificity with an AUC of 0.81. The parsimonious common-feature model (dialysis session frequency, diabetes) achieved the best survival discrimination (C-index 0.83). Full models with all variables demonstrated moderate performance, highlighting potential overfitting. Key mortality predictors included dialysis adequacy, diabetes status, respiratory comorbidities, and hemodynamic parameters. Machine learning–aided feature selection enhances mortality risk prediction in dialysis patients. Parsimonious models focusing on consistent predictors may optimize clinical applicability. These findings support integrating ML and traditional regression approaches to refine prognostic tools and inform personalized care strategies in ESRD.

A Hybrid Bayesian-PDE Constrained Optimization Framework for High-Dimensional Image Reconstruction

Mushtaq K. Abdalrahem — 2026-01-30

High-dimensional image reconstruction problems in fields such as medical imaging, astrophysics, and remote sensing are typically ill-posed inverse problems affected by noise, under sampling, and imperfections in the physical forward model. Traditional methods for resolving these conflicts suffer from an inherent trade-off: pure physics-based PDE-constrained models impose physical consistency but are deterministic and do not represent uncertainty, and fully Bayesian models provide principled uncertainty quantification but tend to become computationally intractable in very high-dimensional spaces. In response to these challenges, we propose the Hybrid Bayesian - PDE Constrained Optimization Framework. The Hybrid Bayesian - PDE Constrained Optimization Framework leverages the physical fidelity of PDE-based forward models with the expressive capability of Bayesian inference to model uncertainty. The reconstruction problem is cast as an optimization problem whereby a variational or hierarchical Bayesian prior is combined with a PDE-constrained data fidelity term, and the optimization objective is solved by an efficient stochastic variational optimization scheme. Experiments using a representative CT example and MRI datasets demonstrated how the hybrid methods provided (i) better reconstructions, preserving fine structures for substantially undersampled data and robust performance in noise when compared to the pure physics model along with providing (ii) clinically meaningful pixel-wise uncertainty maps. These results support the view that the proposed hybrid method provides a principled, computationally efficient, reliable approach to the challenge of solving large-scale inversion problems, while addressing the fundamental limitations of both deterministic, physics-based methods and probabilistic Bayesian inversion.

HLA-DQ Allele Carriers as Genetic Risk Factors for Pulmonary Tuberculosis: A Meta-Analysis

Bellytra Talarima — 2026-02-04

Several studies have shown that pulmonary tuberculosis (PTB) is a major global health issue, affecting various countries around the world. Susceptibility to the disease has been reported to be influenced by host genetic variables, including Human Leukocyte Antigen (HLA) class II genes, specifically HLA-DQ. Despite the association, studies on the relationship between HLA-DQ allele carriers and the risk of PTB are still not consistent among various populations. This meta-analysis aims to assess the association between carrier status (phenotype frequency) of HLA-DQA1 and HLA-DQB1 alleles and susceptibility to pulmonary tuberculosis. Several HLA-DQ alleles were examined, and the pooled effect size estimates were calculated based on carrier (phenotype) frequencies of HLA-DQA1 and HLA-DQB1 alleles, using odds ratios (ORs) and 95% confidence intervals (CI). A total of 21 high-quality studies (NOS ≥7) were included in the review, with 25,896 controls and 3,927 cases. The results showed that the risk of PTB was significantly increased by allele carriers of HLA-DQA1*01:01 (OR = 1.79; 95% CI: 1.22–2.62) and HLA-DQA1*03:01 (OR = 1.64; 95% CI: 1.08–2.48). Risk factors for HLA-DQB1 allele carriers were also found to be the *02:01 (OR = 1.36; 95% CI: 1.04–1.79), *05:03 (OR = 1.35; 95% CI: 1.01–1.80), and *06:01 (OR = 1.41; 95% CI: 1.00–1.97) allele carriers. Meanwhile, HLA-DQA1*02:01, *04:01, *05:01, and *06:01 allele carriers had protective effects. The majority of analyses found no indications of publication bias. This meta-analysis demonstrates that carrier status of specific HLA-DQA1 and HLA-DQB1 alleles is significantly associated with susceptibility to pulmonary tuberculosis

Circular Statistical Analysis of Emergency Department Admissions

Emadeldin I.A. Ali — 2026-02-13

Health-care administrators face ongoing challenges managing emergency department (ED) operations, particularly in understanding how patient arrival trends fluctuate within the 24-hour day. Although prior research has examined the times at which patients seek emergency care, most of these studies have used simple statistical methods that do not account for time as a periodic variable. As a result, many significant time-of-day patterns may not be detected. We use circular statistics on 142,005 hourly emergency department admissions at a large hospital in Iowa from January 2014 to August 2017. Overall, the pattern of ED visits presents an anisotropic distribution that is statistically significant according to both Rayleigh and Kuiper tests. Patient arrival times show a circular mean in the early to mid-afternoon, a marked late-afternoon modal peak, and a diffuse distribution across the day. Adjusted circular probability models such as the von Mises distribution, cardioid distribution, and the wrapped normal distribution perform significantly better than the circular uniform model when the AIC, BIC, CAIC, and HQIC criteria are considered. The circular summary charts help in understanding the various trends observed in the time series graph. In pointing out how the use of a circular method is a mathematically appropriate and more interpretable approach for describing trends related to admissions on an hourly basis, this piece of research also points out the benefits of such a method as being a useful tool for health-care planners.

Comparative Evaluation of Inception V3 and ResNet 50 for Pneumonia Prediction

Anuja Bokhare — 2026-02-13

Pneumonia is a fatal respiratory infection that has become the leading cause of death among many people across the world. Its widespread has grabbed great attention making it a major topic for research under various domains. Its severity has led to the development of systems that can predict whether a patient has chances of being diagnosed with pneumonia or not, this is also called as computer aided diagnosis. However, current study intends to identify an Artificial Neural Network (ANN) model that has been able to provide the highest accuracy when it comes to predicting this life-threatening condition. The prediction was initially done with Machine learning techniques but with the introduction of ANN, it was observed that there are models that provided higher accuracy than the ML models. This study investigates how the concept of deep learning which is a vital part of ANN makes use of one of its most efficient models including Inception V3 and ResNet 50 for the prediction of pneumonia and compare their performance to suggest a better solution to the problem. Results indicate that ResNet50 offers clinically meaningful improvements in sensitivity and specificity, supporting its role as a decision-support tool for early pneumonia detection.