Background For patients (pts) with rheumatoid arthritis (RA), early attainment of remission or low disease activity (LDA) predicts better long-term outcomes. Predicting non-response to therapy early in the disease course may avoid additional exposure to ineffective treatments and reduce a patient's potential risk of adverse events.
Objectives To develop an instrument for predicting clinical outcomes for individual RA pts at 6 months based on observed 3-month responses.
Methods A 3-step method was used to predict DAS28-CRP (28-joint disease activity score based on C-reactive protein) <2.6 and LDA (<3.2) at 6 months, using data from the OPTIMA trial in early RA pts: First, baseline (BL) and post-BL candidate predictors including demographics, laboratory assessments and genetic assays, were screened using the Least Absolute Shrinkage and Selection Operator (LASSO) method (1). To obtain an individual prediction score of achieving Week (wk) 26 DAS28-CRP <2.6 or LDA, logistic regression was performed using predictors chosen based on LASSO results and clinical feasibility. Lastly, cutoff values were established by applying the Recursive Partitioning and Regression Tree (2) (RPART) method on the individual predictive score. Prediction rules were recommended based on satisfactory positive and negative predictive values (PPV/NPV), and validated using data from the DE013, DE019, MUSICA and CONCERTO trials in patients with early or established RA. Responders and non-responders (NR) are defined based on attainment of DAS28-CRP <2.6 or LDA at 6 months.
Results Wk 12 DAS28-CRP, physician global assessment of arthritis, and health-assessment questionnaire-disability index (HAQ-DI) were selected from the final list of predictors to calculate the individual prediction score (Table footnote). Individual prediction scores were calculated for 866 pts in OPTIMA, who had both Wk 26 DAS28-CRP <2.6 status and Wk 12 predictors. The prediction rule was applied to 541/866 (62%) pts based on their Month 3 prediction score. Out of 427 predicted NR at Wk 26, 401 predictions were correct (NPV of 94.61%) (Table). Out of 114 predicted responders at Wk 26, 97 predictions were correct (PPV of 85.19%). For the 38% of pts in the hard-to-classify category, the treatment outcome at Wk 26 was unclear due to lack of proper early response indicators; however, a higher prediction score corresponded to a higher rate of DAS28-CRP <2.6. Similarly, for DAS28-CRP LDA, the NPV was 90.80% and the PPV was 90.15%. Prediction accuracy in the validation datasets was consistent with that in OPTIMA; for MTX NR, accuracy was consistently >90% in all the trials.
Conclusions Using this instrument derived from clinical study data, DAS28-CRP <2.6 or LDA at 6 months could be individually predicted for about 60% of RA pts based on clinical outcomes at 3 months. This instrument-based predictive capability may provide quantified guidance for switching non-responders to more effective therapy as early as 3 months.
Tibshirani R. 1996. J of the Royal and Stat Society;58:267–88
Breiman L et al. 1984. Classification and Regression trees (Wadsworth)
Acknowledgement AbbVie: study sponsor, contributed to design, data collection, analysis, and interpretation; writing, reviewing, and approval of final version. Medical writing support: Naina Barretto of AbbVie.
Disclosure of Interest J. Smolen Grant/research support from: AbbVie, Consultant for: AbbVie, X. Wang Shareholder of: AbbVie, Employee of: AbbVie, I. Sainsbury Shareholder of: AbbVie, Employee of: AbbVie, A. Kavanaugh Grant/research support from: AbbVie, Amgen, Astra-Zeneca, BMS, Celgene, Centocor-Janssen, Pfizer, Roche, and UCB, Consultant for: AbbVie, Amgen, Astra-Zeneca, BMS, Celgene, Centocor-Janssen, Pfizer, Roche, and UCB