Background

Systemic Lupus Erythematosus (SLE) flares are notoriously unpredictable. Current models treat flares as independent events with constant hazard rates. However, clinical observation suggests temporal clustering and environmental triggers.

Hypothesis

We hypothesize that SLE flare dynamics can be modeled as a semi-Markov process where:

1. States: Remission → Low Activity → Moderate Flare → Severe Flare
2. Transition probabilities are not constant but depend on:
   • Time since last flare (sojourn time distribution)
   • UV index and temperature (environmental covariates)
   • Medication adherence (binary covariate)
   • Anti-dsDNA and complement levels (biomarker covariates)

Rationale

Traditional survival analysis (Cox PH) assumes proportional hazards — flare risk is constant relative to covariates. But lupus patients exhibit refractory periods post-flare (reduced risk for 2-4 weeks) followed by vulnerability windows (elevated risk at 6-12 weeks). This non-monotonic hazard is characteristic of semi-Markov rather than Markov processes.

Testable Predictions

• Flare interval distributions will be better fit by Weibull or log-logistic than exponential
• UV index >6 will increase the transition probability from Remission to Low Activity by >30%
• The model will outperform standard Poisson regression in predicting next-flare timing (concordance >0.70)

Data Requirements

• Longitudinal cohort with ≥200 SLE patients, ≥2 years follow-up
• Serial SLEDAI-2K measurements (monthly)
• Geocoded weather data (UV, temperature, humidity)
• Medication adherence logs

References

• Petri M, et al. Derivation and validation of SELENA-SLEDAI. Lupus. 2005.
• Golder V, et al. Lupus flare prediction. Lupus Sci Med. 2022.
• Foucher Y, et al. Semi-Markov models in clinical research. Stat Methods Med Res. 2010.