Document resource
Introduction Systemic sclerosis (SSc) is a clinically heterogeneous disease with variable organ involvement and outcomes. Reliable biomarkers are needed to improve risk management. Persistent systemic inflammation, captured by longitudinal C-reactive protein (CRP) measurements, may help predict disease course. Integrating biomarker-defined phenotypes with machine learning could enhance clinical stratification. We aim to examine the relationship between CRP-defined inflammatory phenotypes and machine learning–derived clusters, and to evaluate associations with organ involvement, lung function decline, and survival.Material and Methods Patients from the EUSTAR registry with at least 3 visits including CRP were analysed. Inflammatory phenotype was classified over 24 ± 6 months as: non-inflammatory (<5 mg/L), low-grade (5–9.9 mg/L), intermediate-grade (10–14.9 mg/L), or high-grade (15 mg/L or more). Baseline characteristics were compared using Chi 2/ANOVA. Survival was assessed with Kaplan–Meier and multivariable Cox models adjusting for demographics, disease duration, cutaneous subset, and organ involvement. Longitudinal %pFVC and %pDLCO were modelled with mixed-effects models. Unsupervised k-means clustering (k=4) on baseline %pFVC, %pDLCO, modified Rodnan skin score, and organ involvement identified data-driven groups.Results Among 4,474 patients, 3,019 (67.5%) were non-inflammatory, 377 (8.4%) low-grade, 369 (8.2%) intermediate, and 709 (15.8%) high-grade. Ten-year survival was 82.4% in non-inflammatory, 71.1% in low-grade, 69.5% in intermediate, and 63.8% in high-grade groups (p<0.001). In adjusted models, intermediate/high grades were associated with increased risk of mRSS progression (HR 1.9, 95% CI 1.4–2.5), digital ulcers (HR 2.3, 1.7–3.1), right bundle branch block (HR 1.8, 1.2–2.7), auricular arrhythmias (HR 1.7, 1.2–2.5), ILD (HR 1.6, 1.2–2.1), %pFVC decline of at least 10 (HR 1.5, 1.1–2.0), and %pDLCO decline of at least 15 (HR 1.7, 1.3–2.2) at 12 months. Intermediate/high grades were also linked to composite endpoints of any heart (IRR 1.9, 1.5–2.5) and any organ involvement (IRR 2.0, 1.6–2.6). K-means clustering identified four groups: C0 (preserved lung, low mRSS, minimal organ involvement, enriched for non-inflammatory); C1 (mild lung decline, moderate mRSS, more GI involvement, enriched for intermediate); C2 (intermediate measures, balanced phenotypes); and C3 (markedly reduced %pFVC and %pDLCO, a high prevalence of ILD, and greater gastrointestinal involvement and was enriched for high-grade).Conclusions Persistently elevated CRP in SSc, particularly at moderate/high levels, defines a phenotype with greater multi-organ involvement and worse survival. CRP-based phenotyping provides a pragmatic prognostic tool that complements machine learning clustering. These findings support closer monitoring and tailored management of inflammatory phenotypes and may inform future trial design and stratification.Abstract OC.39 Figure 1Kaplan-Meier survival curve according to inflammatory phenotypeAbstract OC.39 Figure 2Forest plot of adjusted hazard ratios for organ involvement and persistent intermediate-high inflammatory phenotypeAbstract OC.39 Figure 3Cluster profile heatmap for k-means-derived subgroups in systemic sclerosis (k-4)