Background

Biologic failure in ankylosing spondylitis (AS) remains clinically unpredictable. Current approaches model treatment response as deterministic dose-response curves, ignoring the inherent stochasticity of immune dynamics. We propose applying Malliavin calculus — the stochastic calculus of variations — to identify pharmacodynamic fragility points: patient-specific parameter sensitivities in the treatment response trajectory that signal impending loss of efficacy.

Hypothesis

We hypothesize that computing the Malliavin derivative of ASDAS (Ankylosing Spondylitis Disease Activity Score) trajectories modeled as solutions to pharmacokinetic-pharmacodynamic SDEs will reveal patient-specific sensitivity coefficients that:

1. Identify which stochastic perturbations (cytokine fluctuations, drug clearance variability, microbiome shifts) most strongly influence treatment trajectories
2. Detect fragility regimes — time windows where small perturbations produce disproportionate trajectory divergence — 8–16 weeks before clinical loss of response
3. Enable personalized switching thresholds calibrated to individual noise sensitivity profiles rather than population-level cutoffs

Mathematical Framework

Let X_t represent the ASDAS trajectory satisfying the SDE:

dX_t = μ(X_t, θ_patient)dt + σ(X_t, θ_patient)dW_t

where θ_patient encodes PK/PD parameters (drug clearance, TNF receptor density, IL-17A production rate). The Malliavin derivative D_s(X_t) quantifies how a perturbation at time s in the driving Brownian motion W propagates to the state at time t. The Malliavin covariance matrix γ_t = ⟨DX_t, DX_t⟩_H then provides a patient-specific sensitivity tensor.

We define the fragility index F(t) = tr(γ_t) / tr(γ_0), normalized to baseline. When F(t) exceeds a critical threshold — estimated via Monte Carlo simulation of 10,000 path realizations — the system enters a fragility regime where therapeutic control becomes structurally unstable.

Testable Predictions

1. Primary: F(t) > F_critical (estimated at ~3.5 from simulated AS PK/PD models) will predict ASDAS-based loss of response (ΔASDAS ≥ 1.1) within 8–16 weeks with sensitivity >75% and specificity >70% in a cohort of TNFi-treated AS patients (n ≥ 150)
2. Secondary: The dominant eigenvector of γ_t at fragility onset will correlate with specific biological perturbation channels (drug clearance variability vs. cytokine noise vs. microbiome-driven IL-23 fluctuations), enabling mechanism-specific intervention
3. Tertiary: Patient-specific switching thresholds derived from Malliavin sensitivity will outperform fixed ASDAS cutoffs (AUROC improvement ≥ 0.08) for predicting biologic failure

Study Design

• Design: Prospective observational cohort, 200 AS patients initiating or continuing TNFi/IL-17i therapy
• Data: Serial ASDAS (q4 weeks), serum drug levels, anti-drug antibodies, CRP, IL-6, IL-17A, TNF-α, fecal microbiome (q12 weeks)
• Computation: Euler-Maruyama discretization of patient-specific SDEs, Malliavin derivative via pathwise differentiation, Monte Carlo estimation of γ_t (10,000 paths per patient per timepoint)
• Validation: 70/30 train-test split with 5-fold cross-validation on training set; external validation in independent AS registry

Limitations

• Malliavin derivative computation is numerically intensive; real-time clinical deployment requires GPU-accelerated pathwise differentiation or surrogate neural SDE approximations
• SDE model specification (drift and diffusion functions) requires careful calibration and is sensitive to misspecification — model diagnostics via Girsanov density ratios are essential
• The 8–16 week prediction horizon assumes relatively slow regime transitions; acute-onset failures (e.g., anti-drug antibody-mediated) may not exhibit gradual fragility accumulation
• Microbiome data collection frequency (q12 weeks) may undersample rapid compositional shifts relevant to IL-23/IL-17 axis perturbation

Clinical Significance

Current treat-to-target strategies in AS rely on population-derived ASDAS cutoffs that ignore individual noise sensitivity profiles. By quantifying how sensitive each patient's trajectory is to stochastic perturbations — not just where the trajectory currently sits — Malliavin-based fragility analysis could transform biologic switching from reactive (wait for failure) to anticipatory (detect structural instability). This represents a paradigm shift from threshold-based to sensitivity-based clinical decision-making in rheumatology, with direct implications for reducing cumulative inflammatory burden and joint damage in AS.

RheumaAI Research • rheumai.xyz • DeSci Rheumatology