Background

RA disease activity is measured using composite indices (DAS28, SDAI, CDAI) that represent a snapshot — a single point in a continuous trajectory. Treat-to-target strategies assume discrete categories (remission, low, moderate, high activity), but these are imposed thresholds on continuous scores, not biologically distinct states.

Hypothesis

We propose that longitudinal DAS28 trajectories in RA follow a Hidden Markov Model (HMM) with 4 latent states that do not correspond to conventional DAS28 cutoffs:

1. Deep Remission — stable low inflammation with self-correcting dynamics (absorbing tendency)
2. Fragile Remission — low scores but high transition probability to flare (pre-flare state)
3. Oscillating Activity — cycling between moderate and high with period ~3-6 months
4. Progressive Erosive — persistently elevated activity driving structural damage

Key Insight

The critical distinction is between states 1 and 2: both have similar DAS28 values (<2.6), but state 2 has 10× higher transition probability to flare within 3 months. Current clinical practice cannot distinguish them. The HMM can.

Mathematical Framework

• Observation model: DAS28(t) ~ N(μ_s, σ²_s) for latent state s
• Transition matrix: A[i,j] = P(state_j at t+1 | state_i at t), estimated via Baum-Welch
• Covariates: time-varying medication, anti-CCP titer, shared epitope status modify transition probabilities via input-output HMM
• Viterbi decoding: most likely state sequence for each patient

Testable Predictions

1. BIC selects 4-state HMM over 3-state and 5-state models
2. Patients in Fragile Remission (state 2) have >50% probability of flare within 3 months vs <10% for Deep Remission (state 1)
3. HMM-predicted state at month 6 predicts Sharp/van der Heijde score change at year 2 better than DAS28 alone (ΔR² > 0.10)
4. The Progressive Erosive state correlates with radiographic progression rate >5 Sharp units/year
5. Shared epitope positivity increases transition probability from Fragile Remission to Oscillating Activity by >2×

Data Requirements

• Longitudinal RA cohort: ≥300 patients, ≥2 years, DAS28 measured ≥quarterly
• Serial radiographs scored by modified Sharp/van der Heijde
• Genotyping for HLA-DRB1 shared epitope
• Medication changes logged

Clinical Implications

If validated, clinicians could identify patients in Fragile Remission and preemptively escalate rather than wait for the flare that DAS28 cannot predict. This shifts RA management from reactive to predictive.

References

• Rabiner LR. A tutorial on HMMs. Proc IEEE. 1989;77:257-86.
• Siemons L, et al. Trajectories of DAS28 in RA. PLoS One. 2014.
• van der Heijde D. Radiographic progression in RA. J Rheumatol. 1999.
• Smolen JS, et al. Treat-to-target in RA. Ann Rheum Dis. 2010.