Abstract
OBJECTIVES: To evaluate the diagnostic and prognostic performance of the neutrophil to lymphocyte ratio (NLR) and platelet to lymphocyte ratio (PLR) in SLE and to integrate these biomarkers into an interpretable artificial intelligence (AI) model for clinical decision support. DESIGN: We conducted a two-phase mixed-methods study: (1) a meta-analysis of 50 studies (n=12 850 patients with SLE), compliant with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses, and (2) the development and validation of an XGBoost machine learning model, guided by the Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis-AI, with SHapley Additive exPlanations (SHAP) explainability. SETTING: Our analysis used multicentre data from global SLE registries, including cohorts from Asia, Europe, North America and Africa. PARTICIPANTS: The study included adults (≥18 years) who met the 2019 European League Against Rheumatism/American College of Rheumatology classification criteria for SLE, with NLR and PLR measured via standardised complete blood count. Comparator groups consisted of healthy controls and patients with non-SLE autoimmune diseases. INTERVENTIONS: NLR and PLR were assessed as biomarkers for SLE activity and complications. Our AI model integrated these ratios with standard clinical biomarkers and multi-omics data. PRIMARY AND SECONDARY OUTCOME MEASURES: The primary outcome was diagnostic accuracy (measured by area under the curve (AUC), sensitivity and specificity) for active SLE (defined as Systemic Lupus Erythematosus Disease Activity Index ≥6). Secondary outcomes included prognostic value (HRs for lupus nephritis, cardiovascular events and mortality) and treatment response monitoring. RESULTS: Our analysis demonstrated that NLR has superior diagnostic accuracy for active SLE compared with PLR, with a pooled AUC of 0.85 vs 0.78 (p=0.02). NLR showed pooled sensitivity and specificity of 78% and 82%, respectively, while PLR showed 70% and 75%. Elevated NLR (>3.5) and PLR (>185) predicted higher risks of lupus nephritis (HR=2.1 and 1.8, respectively), cardiovascular events (HR=2.3 and 1.9) and mortality (HR=3.1 and 2.1; all p<0.01). We identified significant ethnic variations, with optimal NLR cut-offs of >3.1 for Asian populations, >2.8 for Caucasian populations and >3.4 for African populations. The AI model achieved an AUC of 0.87 in training and 0.82 in validation, with NLR emerging as the top predictive feature (SHAP score=0.25). CONCLUSION: NLR outperforms PLR in SLE diagnosis and risk stratification, with validated cut-offs that vary significantly by ethnicity. The integration of these biomarkers into AI models enhances predictive accuracy, supporting the use of NLR and PLR as cost-effective tools for SLE management.