Development and Validation of a Predictive Model for Calcium Oxalate Kidney Stone Recurrence Integrating Gut Microbiome and Clinical Features

A Chinese prospective study found Escherichia-Shigella gut abundance, family history, and creatinine predicted stone recurrence with AUC 0.943 in validation.

Journal: World Journal of Urology | Published: 2026-06-29 | Type: Prospective cohort, Validation Study | PMID: 42371145 Authors: Dang Kai et al. (Department of Urology, Beijing Friendship Hospital, Capital Medical University) Funding/COI: Capital Health Research and Development Program, National Natural Science Foundation of China, Beijing Hospitals Authority Youth Programme, and others. No competing interests declared.

Summary

About 28% of first-time calcium oxalate stone patients in this study developed a recurrence within two years. The group that recurred had measurably different gut bacteria: lower overall diversity and a higher load of Escherichia-Shigella. The authors built a nomogram combining family history, serum creatinine, and Escherichia-Shigella abundance that hit an AUC of 0.978 in training — a number high enough to raise eyebrows before the validation cohort brings it back to a more believable 0.943.

Claims

• 75 of 268 patients (27.99%) experienced recurrence over two years of follow-up
• Recurrent group showed significantly lower gut microbial alpha diversity
• LEfSe analysis found enrichment of Proteobacteria, Enterococcaceae, Limosilactobacillus, and Escherichia-Shigella in recurrent group; reduced Firmicutes
• Family history of stones: OR = 10.684 (strongest independent predictor)
• Elevated serum creatinine: OR = 1.025 per unit increase
• Increased Escherichia-Shigella relative abundance: OR = 1.063
• Nomogram AUC: 0.978 (training cohort), 0.943 (validation cohort)
• Calibration plots and decision curve analysis showed net clinical benefit

Study Quality

This is a prospective single-center study with two-year follow-up, which is methodologically stronger than the retrospective designs that dominate this literature. The authors followed TRIPOD reporting standards for prediction model development and validation, which is the right checklist to invoke. Logistic regression with ROC curves, calibration plots, and decision curve analysis covers the standard bases for nomogram validation. The 16S rDNA sequencing approach for microbiome profiling is standard but has known limitations around taxonomic resolution — it can distinguish genus-level differences but struggles with species-level precision, which matters when you're pointing the finger at a specific genus like Escherichia-Shigella.

The training AUC of 0.978 is implausibly high for a biological prediction problem with this sample size, suggesting possible overfitting. The validation AUC of 0.943 is still very strong — arguably too strong for a 268-patient single-center Chinese cohort to generalize globally. The model hasn't been tested in a geographically or ethnically distinct population, which is the real test of external validity.

Red Flags

• Single center, single country: Beijing Friendship Hospital only. Gut microbiome composition varies substantially by geography, diet, and ethnicity — this nomogram may not travel well
• Training AUC of 0.978 is a red flag: near-perfect discrimination in a model this small almost always reflects overfitting; the validation cohort was presumably from the same institution and patient population, limiting true external validation
• Sample size: 268 patients, 75 events. With three predictors in the final model this is borderline by EPV (events per variable) rules of thumb — not a disaster, but tight
• OR of 10.684 for family history: this is an enormous effect size that should be interpreted cautiously; it may reflect unmeasured confounders or be inflated by small-sample variance
• 16S rDNA resolution: cannot definitively distinguish Escherichia from Shigella, nor confirm pathogenicity — the "Escherichia-Shigella" finding is an association, not a mechanism
• Causal direction unclear: gut dysbiosis could precede stones, result from stone-related dietary changes, or be a downstream effect of subclinical renal dysfunction (note: serum creatinine is also a predictor)
• No dietary data reported: diet is a major driver of both gut microbiome composition and stone recurrence risk; its absence as a covariate is a notable omission

Strengths

• Prospective design with structured two-year follow-up — better than most prediction studies in this space
• TRIPOD-compliant reporting
• Integrates novel biomarker (microbiome) with established clinical variables rather than treating them as competing frameworks
• Decision curve analysis included, showing net clinical benefit across a range of probability thresholds
• Multiple validation checks: ROC, calibration plots, decision curves

Verdict

The gut-microbiome-plus-clinical-features angle is genuinely interesting and the prospective design earns respect. But a 0.978 training AUC from 268 patients at one Beijing hospital is a warning label, not a selling point. The validation cohort almost certainly overlaps the same institutional population, which means this is internal validation wearing an external validation costume. Before Escherichia-Shigella abundance becomes part of anyone's stone recurrence workup, this nomogram needs replication in a geographically diverse, truly external cohort — ideally one that also controls for diet. File it under "promising, needs independent confirmation."