Related Experiment Video
Updated: Jan 16, 2026
Analyses of Proteinuria, Renal Infiltration of Leukocytes, and Renal Deposition of Proteins in Lupus-prone MRL/lpr Mice
Published on: June 8, 2022
Systemic lupus erythematosus damage risk index (SLE-DRI): a simple machine learning-based tool for identifying
Panagiotis Garantziotis1,2, Dionysis Nikolopoulos3, Spyridon Katechis4
1Deutsches Zentrum für Immuntherapie (DZI), Friedrich-Alexander-University Erlangen-Nürnberg and Universitätsklinikum Erlangen, Erlangen, Germany.
Objective:
Organ damage is a key determinant of poor prognosis and increased mortality in systemic lupus erythematosus (SLE). However, no validated clinical tools for predicting damage accumulation currently exist. We sought to develop a machine learning-based model to predict early organ damage in patients with SLE.
Methods:
Classification criteria (American College of Rheumatology (ACR)-1997, Systemic Lupus International Collaborating Clinics (SLICC)-2012, European League Against Rheumatism (EULAR)/ACR-2019) and non-criteria features of a cohort of 914 patients with SLE were analysed to predict damage (defined as SLICC/ACR Damage Index (SDI)) within 5 years since diagnosis. Feature selection and model construction were performed using the least absolute shrinkage and selection operator-logistic regression (LASSO-LR). The best model in 10-fold cross-validation was tested in an external cohort (n=50).
Results:
A LASSO-LR model incorporating 16 criteria and non-criteria features predicted early organ damage with area under the receiver operating characteristic curve (AUC) values of 0.80 (95% CI 0.74 to 0.87) and 0.86 (95% CI 0.76 to 0.97) in the derivation and external validation cohorts, respectively, outperforming the ACR-1997 (AUC 0.70; 95% CI 0.62 to 0.78), SLICC-2012 (AUC 0.74; 95% CI 0.66 to 0.81) and EULAR/ACR-2019 (AUC 0.70; 95% CI 0.63 to 0.78) classification systems. Features most strongly associated with damage included the SLICC-2012 neurological disorder, EULAR/ACR-2019 class III/IV lupus nephritis and among non-criteria manifestations, myocarditis and interstitial lung disease. Operating the model as a binary classifier (early damage versus no damage), it demonstrated high specificity (0.90, 95% CI 0.78 to 0.95). The model can be converted to a simplified scoring system, with a threshold of ≥3 achieving an AUC of 0.86 (95% CI 0.75 to 0.96).
Conclusion:
We developed and validated a clinician-friendly model for early organ damage prediction in SLE, facilitating risk stratification.
