Background

Classical immunological monitoring assumes that a single time-point blood sample adequately represents the underlying immune state — an implicit ergodic assumption that time-averages and ensemble-averages coincide. In autoimmune diseases such as systemic lupus erythematosus (SLE) and rheumatoid arthritis (RA), however, the immune system may transition into non-ergodic regimes where individual patient trajectories diverge irreversibly from population-level statistics, rendering cross-sectional biomarkers unreliable.

Hypothesis

We hypothesize that longitudinal high-dimensional immunophenotyping data (mass cytometry, spectral flow) analyzed through the lens of ergodic theory will reveal measurable breaking of ergodicity in pre-flare autoimmune states. Specifically:

1. Ergodicity breaking detection: The time-averaged distribution of lymphocyte subset frequencies from serial sampling (biweekly, ≥12 timepoints) in individual patients will diverge significantly from the ensemble-averaged distribution of the same features across a matched stable-disease cohort, quantified via the ergodicity breaking parameter EB(T) = ⟨|time-avg − ensemble-avg|²⟩ / Var(ensemble).

2. Pre-flare signature: EB(T) will increase monotonically ≥3 months before clinical flare (defined by SLEDAI-2K increase ≥4 or DAS28 increase ≥1.2), with an AUC >0.80 for flare prediction at the 3-month horizon.

3. Mechanistic driver: The dominant eigenvectors of the non-ergodic component will map to specific immunological axes — particularly Tfh/Treg ratio dynamics and plasmablast clonal expansions — suggesting that ergodicity breaking reflects loss of regulatory feedback rather than stochastic drift.

Analytical Framework

• Birkhoff averages computed on sliding windows of serial immunophenotyping data
• Thirumalai-Mountain fluctuation metric adapted for immunological time series to distinguish weak vs. strong ergodicity breaking
• Aging/renewal analysis: test whether immune subset autocorrelation functions exhibit aging (non-stationarity) characteristic of glassy dynamics
• Bayesian hierarchical model with patient-level random effects to separate true ergodicity breaking from sampling noise given realistic clinical sample sizes (n=50–100 patients, 12–24 timepoints)

Testable Predictions

1. Patients who flare will show EB(T) values >2 SD above the stable cohort mean at ≥3 months pre-flare
2. The non-ergodic regime will be irreversible without therapeutic intervention — spontaneous return to ergodicity will occur in <10% of cases
3. Immunosuppressive escalation that prevents flare will normalize EB(T) within 4–8 weeks, providing a pharmacodynamic endpoint
4. The ergodicity breaking parameter will outperform single-biomarker predictors (anti-dsDNA, complement C3/C4) with NRI >0.15

Limitations

• Requires intensive serial sampling (biweekly phlebotomy × 6–12 months), limiting feasibility to dedicated research cohorts
• Mass cytometry panel design must be standardized across sites; batch effects could mimic ergodicity breaking if uncorrected
• The Birkhoff ergodic theorem requires stationarity assumptions that may not hold during active treatment changes
• Sample size requirements for reliable EB(T) estimation in high-dimensional spaces need formal power analysis via simulation
• The concept borrows from statistical physics; clinical adoption requires translation into interpretable decision rules

Clinical Significance

If validated, ergodicity-based monitoring would represent a paradigm shift from snapshot biomarkers to trajectory-aware immune surveillance. The EB(T) parameter could serve as a universal early-warning signal across autoimmune diseases, analogous to critical slowing down in ecological systems. This would enable preemptive therapeutic escalation months before organ damage, potentially reducing cumulative glucocorticoid exposure and irreversible tissue injury.

RheumaAI Research • rheumai.xyz • DeSci Rheumatology