01
Predicting SLE Flare Risk After Steroid Withdrawal

Domain: Clinical Research | Rheumatology | Predictive Modeling

Timeline: Fall 2024 – Spring 2025

Tools: R, GLMM, Nonlinear Splines, Longitudinal Analysis

Abstract

This study identifies risk factors for Systemic Lupus Erythematosus (SLE) flare recurrence following glucocorticoid withdrawal (<7.5mg/day) and evaluates the utility of nonlinear predictive modeling. Leveraging longitudinal data from the FORWARD Lupus Registry (1999– 2024), we analyzed 1085 participants, narrowing it to 95 patients post-withdrawal. A Generalized Linear Mixed Model (GLMM) with natural splines for continuous predictors (age, BMI, PHQ-8 depression scores, global severity) was compared to a linear baseline model. The nonlinear GLMM significantly outperformed the baseline (p = 0.02), achieving an AUC of 0.913, and revealed U-shaped age trends (higher risk <50 and >70 years), an inverse BMI relationship (lower BMI = higher risk), and nonlinear depressive symptom effects (peak risk at moderate PHQ-8 scores). Global severity exhibited a near-linear risk increase, while Medicare coverage and higher BMI emerged as protective. These findings underscore the value of flexible modeling in capturing complex risk interactions, advocating for personalized monitoring of high-risk subgroups (e.g., younger patients, lower BMI), and further research into socioeconomic and metabolic mediators.

Introduction

Systemic Lupus Erythematosus (SLE) is a chronic autoimmune disease marked by unpredictable flares that contribute to irreversible organ damage and reduced quality of life. Glucocorticoids are a mainstay of SLE management, but their long-term use necessitates careful withdrawal strategies to balance efficacy and safety. Despite clinical guidance, post-withdrawal flare risk remains poorly characterized, particularly in patients tapering to low doses (<7.5mg/day). This study leverages longitudinal data from the FORWARD Lupus Registry (1999–2024) to identify risk factors and evaluate a flexible predictive model incorporating nonlinear relationships between covariates and flare probability.

Data & Methods

The study cohort included 1,085 participants from the FORWARD Lupus Registry who completed ≥3 consecutive 6-month questionnaires. After applying the inclusion criteria (glucocorticoid withdrawal to <7.5mg/day), 209 patients were identified, with 95 retained after excluding missing data. Flare status was defined via patient-reported outcomes. Predictors included age, BMI, PHQ-8 depression scores, global disease severity, and comorbidities (diabetes, renal disease, allergies).

Two Generalized Linear Mixed Models (GLMMs) were compared: a baseline model with linear terms and a nonlinear model using natural splines for continuous variables (age, BMI, PHQ-8, global severity). Model performance was evaluated via likelihood ratio testing, the area under the curve (AUC), and calibration plots. Analyses accounted for repeated measures using patient-level random intercepts.

Results & Discussion

The nonlinear GLMM significantly outperformed the baseline model (*p = 0.02*), achieving an AUC of 0.913 (vs. 0.909). Age exhibited a U-shaped relationship with flare risk, peaking in patients <50 and >70 years. Lower BMI was strongly associated with increased risk, plateauing at higher BMI levels. Depressive symptoms (PHQ-8) showed a nonlinear trend, with risk peaking at moderate scores. Global severity demonstrated a near-linear increase, emerging as the strongest predictor.

Flares were linked to higher global severity, depressive symptoms, renal disease, and allergies. Protective effects were observed with Medicare coverage and higher BMI. These findings highlight the limitations of linear models in capturing complex risk interactions. For instance, the nonlinear age trend suggests distinct biological or socioeconomic vulnerabilities in younger and older patients, while Medicare’s protective role may reflect healthcare access disparities. The inverse BMI relationship challenges conventional assumptions, warranting investigation into metabolic or inflammatory mediators.

Conclusion

This study underscores the value of flexible modeling in SLE flare prediction, revealing nonlinear risk patterns that linear approaches overlook. Clinically, younger patients, those with lower BMI, and individuals with moderate depressive symptoms may benefit from intensified post-withdrawal monitoring. The protective roles of Medicare and higher BMI suggest socioeconomic and metabolic factors as modifiable targets. Future work should validate these findings in broader cohorts and explore mechanistic pathways, such as BMI’s dual role in inflammation and metabolic health. These insights align with Alira Health’s mission to advance personalized care in autoimmune diseases.