Background

Clinicians have long observed that autoimmune diseases like RA and SLE exhibit seemingly random fluctuations in disease activity despite stable treatment regimens. Prediction models degrade sharply beyond short time horizons.

Hypothesis

We propose that autoimmune disease activity time series (DAS28, SLEDAI-2K measured monthly) are deterministically chaotic rather than stochastic, characterized by:

1. Positive Lyapunov exponents (λ₁ > 0), indicating sensitive dependence on initial conditions
2. Fractal attractor dimension between 2.5-4.0, suggesting low-dimensional chaos embedded in high-dimensional clinical space
3. Prediction horizon ≈ 1/λ₁ ≈ 8 weeks, beyond which trajectory divergence makes point prediction unreliable

Methodology

• Apply Takens embedding (delay coordinates) to monthly DAS28/SLEDAI time series from BIOBADAMEX/COVAD cohorts
• Estimate maximum Lyapunov exponent using Rosenstein algorithm
• Compute correlation dimension (Grassberger-Procaccia)
• Compare surrogate data testing (Theiler et al.) to distinguish chaos from colored noise

Implications

If autoimmune trajectories are chaotic rather than random:

• Short-term prediction (<8 weeks) is feasible and should be exploited clinically
• Long-term prediction is fundamentally bounded — no amount of data will extend the horizon
• Treatment strategies should shift from target-to-treat to adaptive control (chaos control theory: small perturbations at critical moments can stabilize trajectories)
• Ensemble forecasting (like weather prediction) replaces point prediction for >8 week horizons

This reframes the clinical paradigm: autoimmune disease is not random but deterministic — just inherently unpredictable beyond a finite horizon.

References

• Strogatz SH. Nonlinear Dynamics and Chaos. Westview Press, 2015.
• Rosenstein MT, et al. A practical method for calculating largest Lyapunov exponents. Physica D. 1993.
• Glass L. Dynamical disease: mathematical analysis of human illness. AIP Conf Proc. 2001.