Noah Brostoff 1, Ashley Markazi 1, Kathleen Morgan-Kehr 1
1 Simulations Plus, Inc. (Research Triangle Park, United States)
Objectives:
Routine diagnostic assays (e.g., blood counts, complement, autoantibodies) are standard in the clinical evaluation of systemic lupus erythematosus (SLE) but remain underutilized for predicting clinical benefit, contributing to trial heterogeneity and equivocal outcomes. Quantitative systems pharmacology (QSP) modeling provides a mechanistic framework to integrate multimodal biomarkers, characterize disease heterogeneity, and improve prediction of treatment response for both trial design and precision medicine.
We present a framework that links biomarker-defined disease activity and clinical/composite multi-organ metrics within a QSP SimPops (virtual population) model of SLE and cutaneous lupus erythematosus (CLE). The approach enables predictive simulations under alternative trial entry criteria/subgroup definitions. Results are shown for a published therapy to demonstrate proof of concept and illustrate clinical and development implications.
Methods:
A mechanistic QSP model of SLE and CLE was developed to represent key inflammatory and immune pathways driving skin and joint disease, including type I interferon signaling, B-cell/autoantibody dynamics, and complement pathways.
SimPops were generated and optimized using an integrated ordinary differential equation (ODE) and machine learning framework linking biomarker networks to clinical outputs.
SimPops were calibrated and validated against published clinical trials for anifrolumab, belimumab, litifilimab, deucravacitinib, IVIg, lanraplenib, sifalimumab, ustekinumab, daxdilimab, rontalizumab, ianalumab, tabalumab, epratuzumab, and rituximab. Published entry criteria were applied, subsetting the SimPops and generating entry-criteria-matched placebo arms. Following optimization and validation, predictive simulations were conducted.
We evaluated theoretical enrichment strategies using commonly applied subgroup thresholds, including: high anti-dsDNA (≥30,090 pg/mL), low C3 (≤9e8 pg/mL), low/very low C4 (≤1.6e8 / ≤1e8 pg/mL), SLEDAI-2K ≥ 8, CLASI-A ≥ 8, and SJC ≥ 4.
Assessed clinical endpoints included probabilistic membership in CLASI-20/50/70/90, SLEDAI-2K-4, SRI-4, and BICLA response. Results are presented as predicted therapy–placebo deltas and cohort inclusion rates. Anifrolumab simulations were evaluated in the context of TULIP-2 [1] (baseline model criteria: SLEDAI-2K ≥ 6; steroid ≤ 40 mg/day).
Results:
Model calibration was driven by a proprietary objective function incorporating 2,294 constraints, split across fitting and validation, including longitudinal timecourses and discrete endpoints of continuous and binary response analytes, biomarker heuristics, and Poisson-distributed flare rates. As a public-facing metric, the percentage of the mean, median, and binary fitting and validation data points that fall within the model predicted credible intervals is reported. The model achieved 70% of fitting data and 60% of validation data falling within the model’s 90% credible intervals.
Among enrichment strategies, SLEDAI-2K ≥ 8 (vs. baseline ≥ 6) provided the most favorable balance between efficacy amplification and cohort retention. Across endpoints, the composite “sum-of-deltas” increased from 161 to 229 while maintaining 67% inclusion. Absolute drug response rates increased across endpoints, suggesting higher predicted probability of response in patients meeting this criterion.
CLASI ≥ 8 enrichment substantially amplified SRI-4 and BICLA, increasing the sum-of-deltas to 215, but reduced inclusion to 21%. In contrast, SJC ≥ 4 did not enhance deltas, suggesting skin severity preferentially enriches composite response relative to joint severity.
Biomarker-based enrichment (anti-dsDNA, C3, C4) consistently increased absolute CLASI responses and often exceeded original entry-criteria pass rates, indicating strong linkage between biomarker activity and cutaneous improvement. However, deltas were not uniformly enhanced, highlighting that higher absolute drug response does not necessarily translate to improved therapy–placebo separation. These strategies incurred substantial inclusion trade-offs (7–54% inclusion).
Conclusions:
This framework provides multi-level, quantitative support for SLE/CLE drug development, trial design, and clinical practice.
• For developers, it enables prospective optimization of trial enrichment strategies by balancing efficacy signal amplification against cohort retention and estimating probability of success.
• For clinicians, it supports estimation of individual patient response probabilities using routine laboratory and clinical measures.
In the TULIP-2 anifrolumab context, key findings include:
• SLEDAI-2K ≥ 8 enhances all analyzed deltas with sustainable inclusion costs.
• Skin severity preferentially enriches SRI-4 and BICLA deltas at higher inclusion cost.
• Biomarker-defined activity strongly increases absolute CLASI response but does not uniformly increase trial deltas, and often pairs with high inclusion loss.
References:
[1] TULIP-2 (Phase III, IV anifrolumab)
Furie R, et al.
Two-Year, Randomized, Controlled Trial of Anifrolumab in Moderate-to-Severe Systemic Lupus Erythematosus (TULIP-2).
New England Journal of Medicine. 2019;382:211–221.
PMID: 31851731
DOI: 10.1056/NEJMoa1912196
Reference: PAGE 34 (2026) Abstr 12190 [www.page-meeting.org/?abstract=12190]
Poster: Drug/Disease Modelling - Other Topics