Objectives Through the integration of machine-learning approaches, we aim to assess the effect of ANA positivity and a triple-positive serological profile (RF + /ACPA + /ANA +) on disease activity and achieving sustained remission in a Colombian cohort of rheumatoid arthritis (RA) patients. Method This retrospective cohort study included adult RA patients and used clinical and serological data collected from an outpatient follow-up program. Machine learning models, including Support Vector Machines, Random Forest, Extreme Gradient Boosting (XGBoost), Adaptive Boosting (AdaBoost), and K-Nearest Neighbors, were employed to predict sustained remission based on DAS28-ESR, DAS28-CRP, CDAI, and SDAI. Feature selection was performed using Shapley Additive Explanations values; cross-validation and area under the ROC curves (AUC) were used to assess model performance. Results Triple positivity was consistently associated with higher disease activity at baseline and 12 months of follow-up and lower rates of sustained remission compared to RF + /ACPA + /ANA-. This trend was also true when comparing ANA + vs. ANA- patients. The Random Forest model best predicted sustained remission for the different disease activity indices, with an AUC ranging from 0.815 to 0.839. Important features and their SHAP values for DAS28-ESR-based sustained remission included baseline DAS28-ESR (0.252), HAQ (0.147), TJC (0.119), BMI (0.115), age (0.108), SJC (0.095), and RF + /ACPA + /ANA + versus RF + /ACPA + /ANA- (0.021). Conclusions ANA positivity and a triple-positive serological profile are key factors in increased disease activity and poorer outcomes in RA. Furthermore, our predictive modeling results support integrating clinical data and machine learning methods to enhance individualized care and advance precision medicine in treating RA. <Table Float="No" ID="Taba"> <tgroup cols="2"> <colspec align="left" colname="c1" colnum="1" /> <colspec align="left" colname="c2" colnum="2" /> <tbody> <row> <entry nameend="c2" namest="c1"> Key Points • Triple seropositivity (RF +/ACPA +/ANA +) predicts worse disease activity and lower sustained remission in RA, highlighting its potential as a marker of more severe disease phenotypes. • ANA + patients exhibited delayed improvement and persistently higher disease activity, even when not triple-positive, suggesting a distinct role of ANA in RA pathogenesis and outcomes that warrants further attention in clinical evaluation. • Machine learning accurately predicted sustained remission using routine clinical and serological data. The Random Forest model demonstrated the best performance (AUC 0.815–0.839), identifying key predictors such as baseline DAS28-ESR, HAQ, and triple seropositivity, and underscoring the potential of AI in clinical decision support. • CDAI-based analyses revealed clearer distinctions between serological subgroups and more accurately reflected disease trajectories, reinforcing its value as a sensitive, accessible, and patient-centered index, especially in resource-limited settings. </entry> </row> </tbody> </tgroup> </Table>

ANAs and triple positivity effect on disease activity and sustained remission in rheumatoid arthritis: a retrospective real-world machine learning approach

期刊论文
发表日期：

2025年10月24日

查看全文

Kevin Maldonado-Cañón,
Paola Coral-Alvarado,
Paul Méndez-Patarroyo,
Wilson Bautista-Molano,
Gerardo Quintana-López

摘要

Objectives

Through the integration of machine-learning approaches, we aim to assess the effect of ANA positivity and a triple-positive serological profile (RF + /ACPA + /ANA +) on disease activity and achieving sustained remission in a Colombian cohort of rheumatoid arthritis (RA) patients.

Method

This retrospective cohort study included adult RA patients and used clinical and serological data collected from an outpatient follow-up program. Machine learning models, including Support Vector Machines, Random Forest, Extreme Gradient Boosting (XGBoost), Adaptive Boosting (AdaBoost), and K-Nearest Neighbors, were employed to predict sustained remission based on DAS28-ESR, DAS28-CRP, CDAI, and SDAI. Feature selection was performed using Shapley Additive Explanations values; cross-validation and area under the ROC curves (AUC) were used to assess model performance.

Results

Triple positivity was consistently associated with higher disease activity at baseline and 12 months of follow-up and lower rates of sustained remission compared to RF + /ACPA + /ANA-. This trend was also true when comparing ANA + vs. ANA- patients. The Random Forest model best predicted sustained remission for the different disease activity indices, with an AUC ranging from 0.815 to 0.839. Important features and their SHAP values for DAS28-ESR-based sustained remission included baseline DAS28-ESR (0.252), HAQ (0.147), TJC (0.119), BMI (0.115), age (0.108), SJC (0.095), and RF + /ACPA + /ANA + versus RF + /ACPA + /ANA- (0.021).

Conclusions

ANA positivity and a triple-positive serological profile are key factors in increased disease activity and poorer outcomes in RA. Furthermore, our predictive modeling results support integrating clinical data and machine learning methods to enhance individualized care and advance precision medicine in treating RA.

Key Points

• Triple seropositivity (RF +/ACPA +/ANA +) predicts worse disease activity and lower sustained remission in RA, highlighting its potential as a marker of more severe disease phenotypes.

• ANA + patients exhibited delayed improvement and persistently higher disease activity, even when not triple-positive, suggesting a distinct role of ANA in RA pathogenesis and outcomes that warrants further attention in clinical evaluation.

• Machine learning accurately predicted sustained remission using routine clinical and serological data. The Random Forest model demonstrated the best performance (AUC 0.815–0.839), identifying key predictors such as baseline DAS28-ESR, HAQ, and triple seropositivity, and underscoring the potential of AI in clinical decision support.

• CDAI-based analyses revealed clearer distinctions between serological subgroups and more accurately reflected disease trajectories, reinforcing its value as a sensitive, accessible, and patient-centered index, especially in resource-limited settings.