Background

Lupus nephritis (LN) is classified by ISN/RPS biopsy class (I-VI), but class alone poorly predicts long-term renal outcomes. Class IV patients have widely variable trajectories — some achieve complete remission, others progress to ESRD. The missing variable is the dynamics of class transitions.

Hypothesis

We model LN progression as a continuous-time Markov chain (CTMC) with 6 states:

1. Minimal (Class I/II) — mesangial involvement only
2. Focal Proliferative (Class III) — <50% glomeruli
3. Diffuse Proliferative (Class IV) — ≥50% glomeruli
4. Membranous (Class V) — subepithelial deposits
5. Complete Remission — proteinuria <0.5g/24h, normal creatinine
6. ESRD — eGFR <15 or dialysis (absorbing state)

The generator matrix Q contains transition rates q_ij estimated from serial biopsy + labs data. Key predictions:

Transition Rate Hypotheses

1. q(III→IV) > q(IV→III): progression to diffuse is faster than regression (asymmetric rates)
2. q(IV→Remission) with cyclophosphamide > 3× q(IV→Remission) with MMF alone (drug-dependent rates)
3. q(IV→ESRD) is modulated by chronicity index: patients with CI>4 have q(IV→ESRD) > 0.15/year vs <0.03/year for CI≤4
4. Mean sojourn time in Class IV before transition = 18 months (exponential distribution)
5. 10-year dialysis-free survival predicted by CTMC Q-matrix outperforms ISN/RPS class alone (C-index >0.75 vs 0.62)

Mathematical Framework

• Generator matrix Q: 6×6, off-diagonal q_ij = instantaneous transition rate from i→j, diagonal q_ii = -Σ_{j≠i} q_ij
• Transition probability: P(t) = exp(Qt) — matrix exponential gives exact finite-time transition probabilities
• Estimation: Maximum likelihood from interval-censored data (biopsies at irregular times) via msm R package or equivalent
• Covariates: Cox-type regression on transition rates: q_ij(z) = q_ij^0 × exp(β_ij × z) where z = {chronicity index, anti-dsDNA, complement, treatment}

Clinical Application

Given current state + covariate profile, the CTMC computes:

• P(ESRD within 5 years)
• P(Remission within 1 year)
• Expected time in each state
• Optimal biopsy timing (when transition probability is highest)

This transforms LN management from class-based to trajectory-based.

References

• Jackson CH. Multi-state models for panel data: the msm package for R. J Stat Softw. 2011.
• Weening JJ, et al. ISN/RPS classification of lupus nephritis. J Am Soc Nephrol. 2004.
• Austin PC, et al. Practical recommendations for reporting multi-state models. Stat Med. 2020.
• Tektonidou MG, et al. Risk of ESRD in SLE. Arthritis Rheum. 2004.