Glucocorticoid exposure and social support deficits as longitudinal predictors of long-term neuropsychiatric damage in SLE: insights from the LUMINA cohort (LXXXIII)

OBJECTIVE: To identify longitudinal predictors of neuropsychiatric damage in SLE and compare predictors of organic versus all neuropsychiatric outcomes. METHODS: We studied 657 patients from the multiethnic Lupus in Minorities: Nature vs Nurture cohort (5944 person-visit observations). Organic neuropsychiatric damage (seizures, cerebrovascular accident, neuropathy, transverse myelitis) and all neuropsychiatric damage (organic plus cognitive impairment/psychosis) were defined using the Systemic Lupus International Collaborating Clinic/American College of Rheumatology Damage Index. Random survival forests with time-varying covariates modelled 173 predictors. Feature importance was assessed by permutation methods and SHapley Additive exPlanations (SHAP). RESULTS: 92 patients (14.0%) had organic and 197 (30.0%) had any neuropsychiatric damage; models were trained on incident events (48 and 90, respectively). Random survival forests achieved C-indices of 0.738 (organic) and 0.775 (all neuropsychiatric damage) outperforming Cox regression. For organic damage, glucocorticoid highest ever daily dose was the strongest predictor, followed by social support deficits and retirement status, exceeding disease activity measures. Tangible material support deficits specifically dominated the social support signal. SHAP dependence analysis suggested a model-derived inflection in risk contribution between approximately 40 and 60 mg/day, with current dose contributing approximately 2.8-fold more to model-predicted risk for all neuropsychiatric damage than for organic damage alone (mean |SHAP| 0.022 vs 0.008). For the secondary outcome, physician global assessment, fatigue and pain ranked highest, suggesting distinct predictor profiles for cognitive versus organic outcomes. Social determinants of health contributed independently of clinical severity. CONCLUSIONS: In this exploratory analysis, glucocorticoid exposure and tangible social support deficits emerged as leading, potentially modifiable predictors of neuropsychiatric damage in SLE, often outranking disease activity measures, with implications for glucocorticoid stewardship, tangible-support screening and outcome-specific management. External validation in an independent contemporary cohort is required prior to clinical translation.