To identify the risk factors for infection and to develop a model for the prediction of infection in hospitalized patients with active systemic lupus erythematosus (SLE), we carried out a retrospective cohort study with clinical data collected from 121 consecutive patients with active SLE. Seventeen patients had infection within 6 months of beginning steroid therapy for active SLE. Independent multivariate predictors of infection were a decrease in the serum albumin value, an increase in the serum creatinine value, and prednisolone use in a dose of > or = 60 mg/day without methylprednisolone pulse therapy. The error rate of the model by 10-fold cross-validation method was 12%, sensitivity was 65%, specificity was 91%, and positive predictive value was 55%. Four nonsurvivors were correctly discriminated. Use of this model could contribute to earlier diagnosis of infection and may assist decisions regarding empiric antimicrobial administration in patients with SLE.