Risk prediction models for incident systemic lupus erythematosus among women in the Nurses' health study cohorts using genetics, family history, and lifestyle and environmental factors.

Objective: Systemic lupus erythematosus (SLE) is a severe multisystem autoimmune disease that predominantly affects women. Its etiology is complex and multifactorial, with several known genetic and environmental risk factors, but accurate risk prediction models are still lacking. We developed SLE risk prediction models, incorporating known genetic, lifestyle and environmental risk factors, and family history.

Methods: We performed a nested case-control study within the Nurses' Health Study cohorts (NHS). NHS began in 1976 and enrolled 121,700 registered female nurses ages 30-55 from 11 U.S. states; NHSII began in 1989 and enrolled 116,430 registered female nurses ages 25-42 from 14 U.S. states. Participants were asked about lifestyle, reproductive and environmental exposures, as well as medical information, on biennial questionnaires. Incident SLE cases were self-reported and validated by medical record review (Updated 1997 American College of Rheumatology classification criteria). Those with banked blood samples for genotyping (∼25% of each cohort), were selected and matched by age (± 4 years) and race/ethnicity to women who had donated a blood sample but did not develop SLE. Lifestyle and reproductive variables, including smoking, alcohol use, body mass index, sleep, socioeconomic status, U.S. region, menarche age, oral contraceptive use, menopausal status/postmenopausal hormone use, and family history of SLE or rheumatoid arthritis (RA) were assessed through the questionnaire prior to SLE diagnosis questionnaire cycle (or matched index date). Genome-wide genotyping results were used to calculate a SLE weighted genetic risk score (wGRS) using 86 published single nucleotide polymorphisms (SNPs) and 10 classical HLA alleles associated with SLE. We compared four sequential multivariable logistic regression models of SLE risk prediction, each calculating the area under the receiver operating characteristic curve (AUC): 1) SLE wGRS, 2) SLE/RA family history, 3) lifestyle, environmental and reproductive factors and 4) combining model 1-3 factors. Models were internally validated using a bootstrapped estimate of optimism of the AUC. We also examined similar sequential models to predict anti-dsDNA positive SLE risk.

Results: We identified and matched 138 women who developed incident SLE to 1136 women who did not. Models 1-4 yielded AUCs 0.63 (95%CI 0.58-0.68), 0.64 (95%CI 0.59-0.68), 0.71(95% CI 0.66-0.75), and 0.76 (95% CI 0.72-0.81). Model 4 based on genetics, family history and eight lifestyle and environmental factors had best discrimination, with an optimism-corrected AUC 0.75. AUCs for similar models predicting anti-dsDNA positive SLE risk, were 0.60, 0.63, 0.81 and 0.82, with optimism corrected AUC of 0.79 for model 4.

Conclusion: A final model including SLE weighted genetic risk score, family history and eight lifestyle and environmental SLE risk factors accurately classified future SLE risk with optimism corrected AUC of 0.75. To our knowledge, this is the first SLE prediction model based on known risk factors. It might be feasibly employed in at-risk populations as genetic data are increasingly available and the risk factors easily assessed. The NHS cohorts include few non-White women and mean age at incident SLE was early 50s, calling for further research in younger and more diverse cohorts.