Background/Purpose: Neuropsychiatric systemic lupus erythematosus (NPSLE) is linked to increased morbidity, mortality, and adverse health-related quality of life. Early disease, a history of NPSLE, aPL positivity, and high disease activity are considered risk factors for NPSLE. However, there is currently no reliable clinical tool to predict neuropsychiatric flares. Recent advancements in machine learning (ML) have demonstrated great potential in aiding clinical decision-making across various medical disciplines. Therefore, we aimed to assess the reliability and effectiveness of ML applications in predicting NPSLE flares within a large cohort of patients with active SLE, yet no ongoing active severe NPSLE.

Methods: We analysed data from five phase III trials (BLISS-52, BLISS-76, BLISS-NEA, BLISS-SC, EMBRACE) after exclusion of patients with baseline neuropsychiatric BILAG score A (N=3638). Neuropsychiatric flares were defined as a transition from BILAG score C, D, or E to score A or B, or from score B to score A in the neuropsychiatric domain of the classic BILAG index throughout a 52-week long follow-up. After constructing panels of variables based on expert knowledge, we employed ML methodology to develop predictive models utilising the least absolute shrinkage and selection operator (LASSO) as well as multivariable logistic regression analysis. A stratified split was applied to the data to partition the study population into a training (70%; N=2547), and a test set (30%; N=1091). The training set was used in model development while the internal validation was developed by a 10 times 10-fold cross validation. The test set was used for validation of the built model, and the performance of the two models was demonstrated using area under the curve (AUC) of the receiver operating curves (ROC), accuracy with a 95% confidence interval (CI), sensitivity, and specificity metrics.

Results: A total of 105 SLE patients (2.89%) experienced a neuropsychiatric flare during follow-up. Knowledge-driven feature selection included a history of NPSLE, disease duration, aCL positivity, clinical SLEDAI-2K, sex, age, and the use of antimalarials. The LASSO and multivariable logistic regression models demonstrated comparable performance, with an AUC of 0.80 and 0.80, sensitivity of 0.61 and 0.61, and specificity of 0.83 and 0.82, respectively. Moreover, both algorithms exhibited appropriate calibration on the test dataset.

Conclusion: The integration of traditional risk factors for NPSLE into ML-based models can predict neuropsychiatric involvement in SLE with high specificity and modest sensitivity. We herein propose a pragmatic, robust, and highly accurate prediction tool forecasting neuropsychiatric flares in patients with SLE. The utilisation of this ML-based tool holds promising prospects for improving patient care and outcomes in real-world settings.

« Back to ACR Convergence 2023

ACR Meeting Abstracts - https://acrabstracts.org/abstract/harnessing-machine-learning-to-predict-neuropsychiatric-events-in-systemic-lupus-erythematosus/