A prediction model for alzheimer’s disease severity using gut microbiota and neuroimaging

Alzheimer’s disease (AD) is a progressive neurodegenerative disorder requiring precise severity assessment for clinical decision-making. Conventional evaluation methods are often subjective, and single-domain biomarkers insufficiently reflect the complex pathophysiology of AD. Emerging evidence regarding the “gut-brain axis” provides a novel perspective, suggesting that integrating multi-dimensional indicators may enhance predictive accuracy. This study aimed to develop and validate a predictive model for AD severity by integrating gut microbiota profiles with neuroimaging features, thereby establishing an objective tool for assessing disease progression risk and guiding individualized interventions. In this cross-sectional study, 340 patients with AD were included and randomly assigned to a training set ( n = 238) or a validation set ( n = 102) at a 7:3 ratio. Baseline demographic characteristics, gut microbiota metrics, neuroimaging findings, and fluid biomarkers were collected. Disease severity was stratified using the Clinical Dementia Rating (CDR) scale (mild: CDR 0–0.5; moderate-to-severe: CDR ≥ 1.0). Predictor selection was performed using univariate analysis combined with Least Absolute Shrinkage and Selection Operator regression. A predictive model was subsequently constructed via multivariate logistic regression. Model performance was assessed using the area under the receiver operating characteristic curve (AUC), calibration curves, and decision curve analysis (DCA). Univariate analysis showed significant intergroup differences in Gut microbiota α-diversity, Faecalibacterium relative abundance, Escherichia-Shigella relative abundance, Plasma p-tau181 concentration, Plasma neurofilament light chain (NfL) concentration, Hippocampal volume, and Mean whole-brain cortical thickness ( P < 0.05). Multivariate logistic regression identified Gut microbiota α-diversity, Faecalibacterium relative abundance, Hippocampal volume, and Mean whole-brain cortical thickness as independent protective factors against AD severity, whereas Escherichia-Shigella relative abundance, Plasma p-tau181 concentration, and Plasma NfL concentration emerged as independent risk factors. The Random Forest model achieved AUCs of 0.758 and 0.714 in the training and validation sets, respectively. Calibration curves demonstrated strong concordance between predicted probabilities and observed outcomes. Furthermore, DCA confirmed a favorable net clinical benefit within a 20%–60% risk threshold range. A predictive model integrating gut microbiota and neuroimaging features was successfully developed and validated through internal validation for assessing AD severity. The model suggests robust discrimination, calibration, and clinical utility, suggesting its potential as an objective and reliable quantitative tool to support early identification and individualized intervention in AD, pending further external validation.