Predictive Model Links Gut Microbiota and Neuroimaging to Alzheimer Severity

A predictive model combining gut microbiota metrics with neuroimaging features has successfully classified Alzheimer’s disease severity in a cross-sectional study of 340 patients, according to newly reported research published in BMC Neurology. Led by Xiaoyu Zhang, Yan Yu, Weiyong Yu, Fan Yang, and Yan Li, the study addressed long-standing clinical challenges in evaluating disease stages, noting that conventional assessment methods for Alzheimer’s disease are frequently subjective and that single-domain biomarkers inadequately capture the multi-faceted pathophysiology of the progressive neurodegenerative disorder.

The investigative team aimed to develop and validate an objective quantitative instrument for evaluating disease progression risk and informing individualized clinical strategies by exploring the gut-brain axis. Emerging evidence surrounding this axis indicates that integrating multidimensional biological indicators can enhance diagnostic and prognostic precision.

To evaluate this approach, the researchers recruited 340 individuals diagnosed with Alzheimer’s disease, randomly assigning them to either a primary training cohort of 238 participants or an independent internal validation cohort of 102 participants following a predetermined 7:3 distribution ratio.

For each participant, the researchers gathered baseline demographic data alongside a comprehensive panel of biological and radiological measurements. These assessments encompassed profiling of gut microbiota metrics, quantitative neuroimaging evaluations, and targeted fluid biomarkers.

Disease severity among the cohorts was formally stratified using the Clinical Dementia Rating scale, establishing a clear threshold dividing mild impairment, defined by scores of 0 to 0.5, from moderate-to-severe disease manifestations, defined by scores of 1.0 or higher. To isolate the most informative indicators from the collected multi-domain features, the authors executed predictor selection through univariate analysis paired with Least Absolute Shrinkage and Selection Operator regression.

Univariate screening confirmed statistically significant intergroup differences across multiple biological categories, specifically highlighting gut microbiota alpha-diversity, Faecalibacterium relative abundance, Escherichia-Shigella relative abundance, plasma phosphorylated tau 181 concentration, plasma neurofilament light chain concentration, hippocampal volume, and mean whole-brain cortical thickness, each demonstrating significant variance between severity groups at a significance threshold of P < 0.05.

Subsequent multivariate logistic regression modeling delineated distinct independent protective and risk factors associated with Alzheimer’s disease severity. Gut microbiota alpha-diversity, higher relative abundance of Faecalibacterium, preserved hippocampal volume, and greater mean whole-brain cortical thickness were established as independent protective factors against elevated severity.

Conversely, elevated Escherichia-Shigella relative abundance, increased plasma phosphorylated tau 181 concentrations, and higher plasma neurofilament light chain concentrations emerged as statistically independent risk factors for advanced impairment. When assessing discrimination capacity, the predictive model demonstrated robust performance across both cohorts.

The Random Forest model achieved an area under the receiver operating characteristic curve of 0.758 within the training dataset and maintained a comparable area under the curve of 0.714 within the internal validation dataset. Calibration curve assessments indicated strong agreement between model-predicted probabilities and the actual clinical outcomes observed in the study population.

In addition to traditional statistical performance indices, the researchers conducted decision curve analysis to appraise the practical clinical utility of the multi-domain model. This decision curve analysis established a favorable net clinical benefit across an operational risk threshold range spanning 20 percent to 60 percent.

The study authors concluded that the integrated predictive model demonstrated strong internal calibration, discrimination, and clinical utility as a quantitative framework to support early identification and targeted interventions. However, the researchers emphasized that the cross-sectional evaluation remains internally validated, meaning further testing across broader external cohorts is necessary to substantiate its predictive reliability before potential implementation in wider clinical practice.

Subject of Research: Multimodal prediction model for Alzheimer's disease severity using gut microbiota and neuroimaging Article Title: A prediction model for alzheimer’s disease severity using gut microbiota and neuroimaging Article References: Zhang, X., Yu, Y., Yu, W., Yang, F., & Li, Y. (2026).

A prediction model for alzheimer’s disease severity using gut microbiota and neuroimaging. BMC Neurology . https://doi.org/10.1186/s12883-026-05397-0 Image Credits: AI Generated DOI: 10.1186/s12883-026-05397-0 Keywords: Alzheimer's disease, gut microbiota, neuroimaging, gut-brain axis, predictive model, disease severity, Clinical Dementia Rating, Faecalibacterium, Escherichia-Shigella, plasma p-tau181, plasma neurofilament light chain, hippocampal volume, cortical thickness, Random Forest, LASSO regression, biomarkers, decision curve analysis