A balanced random forest model identified adults at heightened risk of rheumatoid arthritis up to 6 to 18 months before diagnosis, according to researchers reporting in RMD Open.
Among 1,544 first-degree relatives followed for a mean of 7.1 years, 1.7% of participants (n=27) developed rheumatoid arthritis (RA) and 8.2% (n=126) developed seropositive inflammatory arthritis (IA). The model, optimized for high sensitivity, correctly identified most impending RA cases 6 to 18 months before diagnosis, achieving sensitivity of 0.82 with specificity of 0.43; at 18 to 36 months, sensitivity was 0.74 with specificity of 0.48. Negative predictive values approached 0.99 across all models. Seropositive IA predictions demonstrated similarly high sensitivity, ranging from 0.88 to 0.91, respectively.
Romain Aymon, MD, PhD of the Division of Rheumatology, Geneva University Hospitals, Geneve, Switzerland and colleagues reported that rheumatoid factor was the strongest predictor of future RA, followed by clinically suspected arthralgia and having multiple family members with autoimmune disease.
The findings also revealed clinically relevant gaps in traditional risk stratification: 70% of RA converters were ACPA-negative at the time of diagnosis, and 41% remained seronegative—substantially higher than typical estimates. Established risk factors including ACPA, shared epitope, and smoking contributed little to prediction, likely reflecting the low number of converters and the predominance of seronegative disease at onset.
The researchers noted that approximately 30% of IA cases were defined based solely on self-reported swelling, which may have inflated IA counts because mechanical joint disease can mimic inflammatory symptoms. Age, BMI, and alcohol use also appeared influential in IA prediction, suggesting metabolic or degenerative contributors. The authors emphasized that although the model was designed to capture nearly all high-risk individuals, specificity remained limited, underscoring the need for external validation and incorporation of additional biomarkers before clinical implementation.
Funding for the study was provided by the Swiss National Science Foundation, the Jean & Linette Warnery Foundation, and the Rheumasearch Foundation; several authors reported industry relationships outside the submitted work in the study.
Source: RMD Open
