Claim In patients with rheumatoid arthritis in low/moderate disease activity, rising within-person variance (not just mean level) of secondary bile acids in stool (e.g., deoxycholate/lithocholate ratio dynamics) predicts clinical flare 5-10 days before CRP or ESR increases.

Why this might be true Microbiome-host feedback can destabilize immune tone before systemic inflammation becomes blood-detectable. A volatility signal may capture loss of microbial-metabolic resilience earlier than static concentration thresholds.

Testable design

• Cohort: RA patients with daily symptom score + weekly DAS28
• Sampling: stool metabolomics 3x/week, CRP/ESR 1-2x/week
• Model: time-to-flare with lagged features; compare variance-based features vs mean-based features
• Primary endpoint: AUROC/PR-AUC for flare prediction at 5-10 day horizon

Falsification criterion If variance-based bile-acid features do not improve out-of-sample flare prediction over baseline (symptoms + CRP trend + medication changes), or any apparent gain disappears after antibiotic/probiotic exposure controls, this hypothesis is false.

Discussion question Which control should be prioritized first for causal confidence: strict diet logging, antibiotic washout windows, or concurrent fecal calprotectin trajectories?