Clinical Utility of Next-generation Sequencing in Hematologic Malignancies: Integrating Molecular Diagnosis, Measurable Residual Disease, and Infection Diagnostics in Immunocompromised Patients
Year: 2026
Type: article
Abstract: Chronic kidney disease (CKD) remains a major health burden worldwide. Identifying people at risk earlier may allow timely intervention and help slow disease progression. However, many machine-learning studies focus mainly on accuracy and give less attention to model stability, formal statistical comparisons, calibration, and clinical interpretability. We compared Logistic Regression, Random Forest, and XGBoost for CKD prediction and examined how the models compared in calibration, interpretability, and estimated clinical utility. We analyzed the UCI CKD dataset (400 records, 24 predictors). A stratified 80:20 training-test split was created, and all preprocessing steps were embedded within the modeling pipeline to prevent information leakage. Performance was assessed using accuracy, precision, recall, F1-score, and ROC-AUC, along with repeated stratified five-fold cross-validation (10 repetitions). We also estimated bootstrap confidence intervals, examined calibration, generated SHAP-based explanations, and performed decision-curve analysis. Model differences were tested with the Friedman test, followed by paired Wilcoxon signed-rank tests with Holm adjustment. Random Forest and XGBoost produced no classification errors in the held-out test set, whereas Logistic Regression misclassified one CKD case, yielding 98.8% accuracy. In repeated cross-validation, mean accuracy was 99.5% for Logistic Regression, 99.1% for Random Forest, and 98.3% for XGBoost (Friedman p < 0.001). SHAP analyses consistently identified hemoglobin, specific gravity, serum creatinine, packed cell volume, and red blood cell count as among the strongest predictors. All three models performed strongly, although Logistic Regression offered greater simplicity and interpretability. These findings require confirmation in larger, independent clinical cohorts before the models can be considered for routine use.
Source: Acta Innovations
Authors: Ola Alkhuffash, Mariem Baazaoui
Language: English
Field: Medicine
Domain: Health Sciences
Open Access status: diamond
OpenAlex ID: W7225188674