Background: Cognitive Impairment (CI) is one of the most common manifestations of neuropsychiatric SLE with a prevalence of 38%. Establishing the different clinical phenotypes of CI, monitoring CI progression over time, identifying patients at high risk of developing CI, along with predicting which patients are at risk of persistent CI over time is essential to prevent the accrual of damage and disability. Machine learning (ML) analysis and augmented intelligence, and longitudinal modeling over time are promising methods by which to develop predictive models for better patient stratification and to identify different patients’ phenotypes. Our project aims to understand the heterogeneity of SLE CI and the progression of CI over time. Objectives: To identify clinical phenotypes in CI among SLE patients and use predictive models to pinpoint patients at risk of CI. Aim 1: Identify clinical phenotypes of CI among SLE patients that take into account clinical history, sociodemographic factors and comorbidities, disease burden, antibody and cytokines profile, interferon signature and data from the American College of Rheumatology neuropsychological battery (ACR-NB). Aim 2: Predict which patients are at risk for CI and CI trajectory. Aim 2.1: Identify factors that increase risk of persistent CI, including both modifiable (e.g. medication) and non-modifiable factors (e.g., ethnicity, sex, etc.). Aim 2.2: Predict the trajectory of CI over time. Research Strategy: The study population includes 293 SLE patients followed at the Toronto Lupus Clinic–Cognitive Study. Patients have completed a comprehensive baseline visit (T0) and 173 patients completed 2 follow-up assessments 6 months apart (T1 and T2). Demographic and clinical data, patient-reported outcomes, antibody/cytokines profiles, interferon signature data, along with data from ACR-NB from these 3 visits, will be used to conduct ML analyses; data on 2 future annual visits (T3 and T4) will be collected. Aim 1: Using an established ML alongside augmented intelligence approach for data integration called Similarity Network Fusion (SNF) to identify clinical phenotypes in SLE patients’ CI. Data from T0, T1 and T2 data will be analyzed. Aim 2: Build predictive models of CI at T3 and T4 with generalized linear mixed models using data from T0, T1 and T2. Analyses will incorporate variable selection by L1-penalized estimation. We will build and assess these models in a stratified sample of 80% of our patients and evaluate our models on the remaining 20%. Impact: By identifying CI subtypes, SLE patients at high risk of developing CI and the trajectory of CI, our findings will facilitate the diagnosis of CI and identification of high risk patients. Longer term, results will facilitate the development of treatments or interventions, and will reduce the rate of co-morbidities and lost productivity related to CI in SLE.