Our primary goal was to create an outcome change score index similar to a standard rheumatoid arthritis (RA) model utilizing real-world data in systemic lupus erythematosus (SLE) patients that occurred during their phase 3 trials with a Food and Drug Administration-approved drug.

We utilized raw data from trials of belimumab for the treatment of SLE. Data were split 80/20 into training/validation sets. Index variables present in a majority of patients and with face validity were selected. Variables were scored for each patient as percentage improvement from baseline after one year. The percentage of placebo- and drug-treated patients considered improved after the application of various criteria was ascertained. Logistic regression was employed to determine the ability of the new index to predict treatment assignment.

A total of 1693 subjects had data for analyses. Eight variables were chosen: arthritis, rash, physician global assessment, fatigue, anti-double stranded DNA antibodies, C3, C4 and C-reactive protein. In the training dataset, ≥20% improvement in ≥4 of eight variables produced the largest difference between placebo- and drug-treated patients (22.1%) with an acceptable rate of improved placebo-treated patients (25%). This resulted in an odds ratio for belimumab (10 mg/kg) vs placebo of 2.7 (95% CI: 2.0–3.6; p < 0.001). However, in the validate dataset the odds ratio was not significant at 1.3 (95% CI: 0.8–2.2; p = 0.863).

The index created from training data did not achieve statistical significance when tested in the validation set. We have speculated why this happened. Is the lack of success of therapeutics for SLE caused by ineffective medications, study design and outcome instruments that fail to inform us, or is the heterogeneity of the disease too daunting? The lessons learned here can help direct future endeavors intended to improve SLE outcome instruments.