A prospective longitudinal investigation tracking 990 individuals diagnosed with early rheumatoid arthritis across fifteen years has demonstrated that long-term structural joint damage follows three distinct trajectory pathways rather than a uniform progression pattern. The multicenter investigation, conducted through the Better Anti Rheumatic FarmacO Therapy cohort known as the BARFOT study across six outpatient rheumatology centers in southern Sweden, monitored structural joint degradation in the hands and feet from clinical disease onset. Applying unsupervised machine learning via K-means clustering to standardized fifteen-year changes in cortical bone erosions and joint space narrowing revealed three clear progression profiles. The largest subgroup of participants, accounting for more than two-thirds of the cohort, exhibited minimal structural joint degradation over the decade and a half of follow-up, while a distinct intermediate subgroup developed moderate damage and a compact high-risk subset experienced aggressive, unremitting structural destruction despite active pharmacological intervention.
The study cohort was drawn from an initial population of 1,634 participants enrolled between 1992 and 2006 who fulfilled the 1987 American College of Rheumatology classification criteria with symptom duration of no more than twelve months at inclusion. Among those evaluated at the fifteen-year assessment milestone, 990 patients possessed complete sets of serial radiographs at baseline and fifteen years to allow full structural analysis. Within this analyzed population, female patients predominated at 71 percent, the median patient age stood at 54 years, median symptom duration prior to enrollment was six months, and 71 percent tested positive for serum autoantibodies. Total recruitment spanned two distinct management eras, with 451 participants entered during the 1990s and 539 enrolled during the 2000s, reflecting clinical practice patterns across periods when biological therapeutic options were progressively integrated into standard rheumatology practice across southern Sweden.
Structural articular assessment relied on standardized posterior-anterior radiographs of the hands and feet obtained at baseline and scheduled intervals of one, two, five, eight, and fifteen years. Films were systematically quantified using the van der Heijde modification of the Sharp scoring methodology, evaluating 32 joint spaces across both hands and 12 joints across the feet to derive composite totals. The system assigns a total Sharp score ranging from 0 to 448 units, encompassing a discrete erosion score scale of 0 to 280 units and a joint space narrowing range of 0 to 168 units. Erosions were defined morphologically by observable cortical breaks, and baseline erosive status required confirmed cortical disruption in the hands, wrists, or feet. General progression was recognized when annual Sharp scores advanced by at least 3 units, while rapid radiographic progression was marked by an annual escalation of 5 units or more, evaluated by two independent readers who demonstrated high agreement with intraclass correlation coefficients spanning 0.940 to 0.998.
To classify longitudinal structural evolution without arbitrary clinical cutoffs, investigators implemented K-means cluster analysis utilizing standardized z-scores derived from the fifteen-year changes in erosion and joint space narrowing metrics. Iterative mathematical assignment evaluated cluster models ranging from two to six groups, ultimately identifying a three-cluster solution as optimal based on cluster separation, internal variance minimization, and clinical interpretability. Cluster 1 incorporated 54 individuals characterized by severe, unyielding joint destruction. Cluster 2 encompassed 258 patients who exhibited moderate long-term structural progression. Cluster 3 formed the primary cohort majority with 678 individuals demonstrating notably modest structural progression over the entire fifteen-year timeframe. Radiographic protocol compliance proved consistent across all subsets, with 68 percent of participants in clusters 1 and 2 and 69 percent in cluster 3 completing radiographs at all six evaluation intervals, while more than 95 percent of each cluster completed at least four imaging assessments.
Divergence in radiographic outcomes between the three trajectory groups emerged immediately from early follow-up and widened progressively over time. Patients allocated to cluster 1 started with a baseline median total Sharp score of 4.0 units and escalated to a median score of 108 units by year fifteen, with every patient surpassing the standard threshold of 3 units per year and 93 percent fulfilling criteria for rapid radiographic progression. In cluster 2, median baseline Sharp scores began at 3.0 units and rose to 55 units at study completion. Conversely, cluster 3 presented with a median baseline Sharp score of zero and reached a median of only 11 units after fifteen years. Despite persistent disease activity throughout the general cohort, 29 percent of all 990 patients displayed no detectable bone erosions on radiographs at fifteen years, and 68 percent of that non-erosive subgroup had maintained completely erosion-free scans across every scheduled clinical follow-up visit throughout the observation period.
Longitudinal clinical disease activity mirrored these structural divisions, measured systematically through the composite Disease Activity Score across 28 joints alongside swollen joint counts, tender joint counts, visual analogue scales for pain and global health, and erythrocyte sedimentation rates. At baseline, cluster 2 exhibited higher disease activity, elevated inflammatory sedimentation rates, and greater tender joint involvement than cluster 1, while cluster 3 presented with younger median age, elevated baseline tender joint counts, and higher composite disease activity scores despite lower radiographic scores. Throughout subsequent follow-up assessments at two, five, eight, and fifteen years, clusters 1 and 2 maintained significantly worse disease activity, swollen joint counts, and sedimentation rates relative to cluster 3. Sustained low disease activity, defined by composite disease scores at or below 3.2 from year two onward, occurred in only 13 percent of cluster 1 and 23 percent of cluster 2, compared with 28 percent among patients in cluster 3.
Physical functional capacity, captured using the validated Swedish version of the Stanford Health Assessment Questionnaire scored from zero to three, demonstrated enduring real-world impacts from cumulative joint damage and active synovitis. While baseline physical dysfunction scores did not separate the groups definitively, functional divergence became pronounced during long-term monitoring.
Treatment strategies recorded throughout the fifteen-year period adhered to Swedish nationwide therapeutic recommendations, relying on conventional synthetic disease-modifying agents alongside selective biologic therapies. Methotrexate served as the predominant conventional agent across all cohorts, while tumor necrosis factor inhibitors constituted the majority of biological prescriptions. Initial management showed high baseline treatment rates without statistically significant disparities between groups, as glucocorticoids or conventional disease-modifying antirheumatic drugs were initiated in 88 percent of cluster 1, 93 percent of cluster 2, and 95 percent of cluster 3. Conversely, biologic utilization climbed steadily over time across all clusters, particularly after five years in cluster 1, where biologic uptake reached 17 percent and coincided with a deceleration in erosion rate, although aggressive joint destruction remained unhalted.
Multivariate backward stepwise logistic regression modeling, utilizing the low-progression cluster 3 as the reference baseline, uncovered distinct baseline predictors driving group assignment. Seropositivity emerged as the sole common predictive denominator for both progressive groups, independently raising odds for cluster 1 and cluster 2. For cluster 1, belonging was predicted specifically by autoantibody seropositivity and having fewer tender joints at initial diagnosis, with a 50-year-old seropositive patient possessing zero tender joints yielding a 19 percent probability of rapid destruction compared to 6 percent when ten tender joints were present. For cluster 2, independent predictors included seropositivity, older patient age, female sex, and baseline radiographic erosions. A 70-year-old seropositive woman presenting with an initial baseline erosion score of 10 units carried an estimated 65 percent probability of entering cluster 2, whereas a seronegative 50-year-old male with zero erosions carried only an 11 percent probability.
The study authors contextualized these long-term findings within evolving rheumatology paradigms, observing that modern ultrasound and magnetic resonance imaging excel at detecting soft-tissue synovitis and tenosynovitis before bone disruption occurs, yet conventional radiography remains a reliable, standardized approach for tracking structural disease progression over decades. Methodological strengths highlighted by the investigators include the large patient sample, prospective follow-up across fifteen years, standardized reading protocols, and inclusion within routine outpatient clinical care across six centers. Acknowledged limitations involved the smaller size of cluster 1 and historical cohort composition, as 59 percent of cluster 1 participants were enrolled during the 1990s when early therapeutic strategies were less intensive. Although statistical adjustments accounted for enrollment period and treatment variables, investigators emphasized that distinct progression trajectories illustrate heterogeneous therapeutic needs requiring tailored early treatment interventions.
Subject of Research: Fifteen-year prospective cluster analysis of radiographic joint damage progression and predictive clinical factors in early rheumatoid arthritis
Article Title: Radiographic outcome over 15 years in patients with early rheumatoid arthritis – a BARFOT-study
Article References: Andersson, M. L. E., Hafström, I., & Forslind, K. (2026). Radiographic outcome over 15 years in patients with early rheumatoid arthritis – a BARFOT-study. BMC Rheumatology, 10(1), Article 46. https://doi.org/10.1186/s41927-026-00660-w
Image Credits: AI Generated
DOI: 10.1186/s41927-026-00660-w
Keywords: rheumatoid arthritis, radiographic progression, BARFOT study, Sharp van der Heijde score, joint space narrowing, bone erosion, K-means cluster analysis, disease activity score, erythrocyte sedimentation rate, health assessment questionnaire, seropositivity, rheumatoid factor, anti-citrullinated protein antibodies, disease-modifying antirheumatic drugs, tumor necrosis factor inhibitors, prospective cohort study, long-term joint damage, tender joint count
News Source: Ophelia Keating. (November 11, 2026). Long-Term Early Rheumatoid Arthritis Study Identifies Three Distinct Joint Damage Trajectories. Scienmag.



