Background Traditional predictor analysis identifies characteristics that determine binary outcomes in EIA but variability is still observed in disease courses. Growth mixture modeling allows refinement of predictor analysis by identifying an optimal number of clusters of individuals with similar disease trajectories in observed data.
Objectives Determine which baseline characteristics define trajectory clusters, and examine differences in medication use in the clusters.
Methods The cluster-based trajectory modeling strategy was applied to data from the Canadian Early Arthritis Cohort (CATCH) study. Patients were assigned to mutually exclusive trajectories by their DAS28 scores over time, based on a posterior membership probability. To determine variables distinguishing the clusters, cross-tabulations of individual-level cluster assignments with individual-level characteristics were carried out. ANOVA was used to compare differences between clusters, applying the Bonferroni correction for multiple comparisons.
Results Five mutually exclusive clusters (Cl-1 to 5) were identified (2,215 patients, 24 months of follow-up). Cl-1 (n=519, 24.4%) begins in low disease activity and achieves remission; Cl-2 (n=337, 20.3%) begins in high disease activity and achieves remission; Cl-3 (n=759, 27.6%) begins in moderate disease activity and achieves low disease activity; Cl-4 (n=503, 21.3%) begins in high disease activity and achieves moderate disease activity; and Cl-5 (n=97, 6.3%) begins and remains in high disease activity. The clusters varied in the proportion of individuals meeting 2010 RA classification criteria (Cl-1 61%, Cl-5 97%) and being seropositive (Cl-1 56% RF and 49% anti-CCP, Cl-5 72% RF and 63% anti-CCP). Socio-demographics and health status varied in cluster assignment. Patients in Cl-1 were younger (<50 years 54%, 32%>42% in Cl-2-5), with more males (36%, 23-25% in Cl-4, 5). Cl-1 patients had fewer comorbidities (mean 1.9, Cl-5 3.1) and more who never smoked (50%, 29% in Cl-5). Cl-5 had more minority populations (77% Caucasian or European, 93% in Cl-1), lower income (24% <$20,000 per year, 8% in Cl-1), and lower employment rates (39%, 70% in Cl-1). These clusters differed in therapeutic use, with Cl-4, 5 patients using lesser and/or lower doses of methotrexate by 24 months (31.5% and 17.5% respectively, vs 38.7%>43.6% for Cl-1-3, p=0.017). Cl-5 more frequently required steroids at the initial visit (72.8%, 50.1-57.3% in Cl-1-4) and 37.5% were still on steroids by 24 months, p=0.004. Cl-5 patients required early start of biologic therapy (34.0% by month 12, 2.2-14.9% in Cl-1-4, p<0.001).
Conclusions Disease trajectories in EIA are influenced by socio-demographic and health status factors including age, sex, comorbid disease, smoking status, ethnicity, income, and employment. Differing therapeutic requirements are reflected in these trajectories. To achieve further improvements in EIA outcomes, population health strategies addressing social determinants of health and patient-tailored treatment strategies are needed.
Disclosure of Interest None declared