abstract details

The summaries are free for public use. ARTHROS will continue to add and archive summaries of articles deemed relevant to ARTHROS by our Faculty.

Data-Driven Patient Clustering and Differential Clinical Outcomes in the Brigham and Women

Author

Arthritis Care Res (Hoboken). 2020 Oct 1. doi: 10.1002/acr.24471. Online ahead of print.

Jeffrey R Curtis 1, Michael Weinblatt 2, Kenneth Saag 1, Vivian P Bykerk 3, Daniel E Furst 4 5 6, Stefano Fiore 7, Gregory St John 8, Toshio Kimura 8, Shen Zheng 7, Clifton O Bingham 3rd 9, Grace Wright 10, Martin Bergman 11, Kamala Nola 12, Christina Charles-Schoeman 4, Nancy Shadick 2

Author Information

1  University of Alabama at Birmingham, Birmingham, AL, USA.

2 Brigham and Women's Hospital, Boston, MA, USA.

3 Hospital for Special Surgery, New York, NY, USA.

4 University of California, Los Angeles, CA, USA.

5 University of Washington, Seattle, WA, USA.

6 University of Florence, Florence, Italy.

7 Sanofi, Bridgewater, NJ, USA.

8 Regeneron Pharmaceuticals, Inc, Tarrytown, NY, USA.

9 Johns Hopkins University, Baltimore, MD, USA.

10 Private Practice, New York, NY, USA.

11 Drexel University College of Medicine, Philadelphia, PA, USA.

12 Lipscomb University College of Pharmacy & Health Sciences, Nashville, TN, USA.

Abstract

Objective: To use unbiased, data-driven, principal component (PC) and cluster analysis to identify patient phenotypes of rheumatoid arthritis (RA) that might exhibit distinct trajectories of disease progression, response to treatment, and risk for adverse events.

Methods: Patient demographic, socioeconomic, health, and disease characteristics recorded at entry into a large, single-center, prospective observational registry cohort, the Brigham and Women's Rheumatoid Arthritis Sequential Study (BRASS; NCT01793103), were harmonized using PC analysis to reduce dimensionality and collinearity. The number of PCs was established by eigenvalue >1, cumulative variance, and interpretability. The resulting PCs were used to cluster patients using a k-means approach. Longitudinal clinical outcomes were compared between the clusters over 2 years.

Results: Analysis of 142 variables from 1443 patients identified 41 PCs that accounted for 77% of the cumulative variance in the dataset. Cluster analysis distinguished five patient clusters: (1) less RA disease activity/multimorbidity, shorter RA duration, lower incidence of comorbidities; (2) less RA disease activity/multimorbidity, longer RA duration, more infections, psychiatric comorbidities, healthcare utilization; (3) moderate RA disease activity/multimorbidity, more neurologic comorbidity; (4) more RA disease activity/multimorbidity, shorter RA duration, more metabolic comorbidity, higher BMI; (5) more RA disease activity/multimorbidity, longer RA duration, more hepatic, orthopedic comorbidity and RA-related surgeries. The clusters exhibited differences in clinical outcomes over 2 years of follow-up.

Conclusion: Data-driven analysis of the BRASS registry identified five distinct phenotypes of RA. These results illustrate the potential of data-driven patient profiling as a tool to support personalized medicine in RA. Validation in an independent dataset is ongoing.