# Real-world validation of the SLERPI diagnostic model with concordance and discordance analysis across established SLE classification criteria

## TL;DR

This study evaluates a new diagnostic model for lupus, showing it performs well in identifying early and atypical cases compared to traditional methods.

## Contribution

The study introduces and validates the SLERPI diagnostic model, demonstrating its superior performance in early and atypical SLE detection.

## Key findings

- SLERPI achieved the highest sensitivity (99.2%) and AUC (0.989) in diagnosing SLE.
- SLERPI outperformed traditional criteria in early disease detection (≤ 1 year, sensitivity 98.0%).
- Discordant cases revealed phenotype-specific patterns, with SLERPI showing the lowest discordance.

## Abstract

Systemic lupus erythematosus (SLE) is a clinically heterogeneous disease in which early and atypical presentations frequently challenge existing classification frameworks. The Systemic Lupus Erythematosus Risk Probability Index (SLERPI) was developed as a probabilistic diagnostic aid, but its real-world performance relative to established classification criteria across disease phenotypes remains incompletely characterized.

In this multicenter, cross-sectional study, we evaluated 1,281 participants, including 655 expert-confirmed SLE patients and 626 controls with other rheumatic diseases. Diagnostic performance of SLERPI, ACR-1997, SLICC-2012, and EULAR/ACR-2019 criteria was assessed against expert clinical diagnosis as the reference standard. Subgroup analyses were performed for early disease (≤ 1 year), sex, disease duration, and major organ involvement. Concordance and discordance between criteria were examined using UpSet plots, detailed phenotypic comparisons, and hierarchical cluster analysis of discordant cases. Net reclassification improvement (NRI) was used to quantify incremental diagnostic information.

All four systems demonstrated high diagnostic accuracy, with sensitivities ranging from 95.1–99.2% and specificities from 87.7–90.4%. SLERPI achieved the highest sensitivity (99.2%) and AUC (0.989), particularly excelling in early disease (≤ 1 year, sensitivity 98.0%, AUC 0.987). Net reclassification improvement favored SLERPI over ACR-1997 (+ 2.7%), SLICC-2012 (+ 1.6%), and EULAR/ACR-2019 (+ 4.3%). Concordance across systems was substantial, with 91.6% of patients classified by all four sets. Discordant cases (8.4%) revealed phenotype-specific patterns: ACR-1997 frequently missed immunologically active or hematologic-dominant cases, while EULAR/ACR-2019 underperformed in mucocutaneous-predominant disease. Cluster analysis identified four coherent subgroups, underscoring heterogeneity in missed classifications. SLERPI showed the lowest discordance, with residual misclassifications confined to hematologic-dominant phenotypes.

SLE classification frameworks show substantial overlap in real-world practice, with discordance driven by phenotype-specific prioritization of disease domains rather than random failure. SLERPI complements established classification criteria by supporting identification of early and atypical SLE presentations, while traditional criteria remain essential for research standardization. Integrating probabilistic diagnostic tools with classification frameworks may enhance SLE recognition across diverse clinical contexts.

The online version contains supplementary material available at 10.1186/s13075-026-03749-2.