OBJECTIVES : Systemic lupus erythematosus (SLE) is associated with increased morbidity and mortality. In this study, we sought (1) to examine the mortality rate difference between elderly SLE patients versus controls, and (2) to identify characteristics associated with mortality in SLE using machine learning (ML).

METHODS : Medicare Part A/B claims (5% sample) from 2010-2018 were used to identify SLE cases in patients ≥66 years old with continuous enrollment during the 12-month period prior to first SLE diagnosis date (index date). Controls had no evidence of SLE and were identified through exact matching on age group, gender, race and the year of a randomly selected index encounter date in 1:1 ratio. Patients were followed until death, end of enrollment or end of data. Mortality incidence rate ratio (IRR) between cohorts was calculated. To identify characteristics of SLE mortality, LASSO models were built on a high dimensional set of demographics, baseline comorbidities, procedures and service location variables, and evaluated using the area under the ROC curve (AUC). Model performance was reported on unseen data.

RESULTS : The study population included 13,770 SLE cases and controls, respectively (mean age=74.5 years; females=75.2%). The median follow-up was 3.6 years [IQR=1.6-5.8] for controls and 3.4 years [IQR=1.5-5.6] for cases. Mortality was recorded for 21.3% of controls and 24.7% of cases. The IRR was 1.20 [95% CI= 1.14-1.26]. Characteristics more likely associated with deceased SLE cases compared to controls identified through the LASSO model (AUC=83.1%; recall=73.5%; precision=49.0%) were rheumatoid arthritis, nephritis/nephrosis/renal sclerosis, coagulation/hemorrhagic disorders, thyroid disorders, viral infections, gout and coronary atherosclerosis/other heart disease.

CONCLUSIONS : Our results indicate an increased mortality rate in elderly SLE cases compared with non-SLE controls, although lower than previously reported results for all age groups. Our ML model suggests particular risk associated with certain comorbidities and confirms the need for strategies to better manage SLE and its comorbidities.