ACR Meeting Abstracts

ACR Meeting Abstracts

Abstract Number: 2211

Identifying Clinical Predictors of Difficult-to-Treat Rheumatoid Arthritis Using Machine Learning-based Techniques

Jiri Baloun1, Lucie Andrés Cerezo2, Tereza Kropáčková3, Aneta Prokopcova4, Kristýna Brábníková Marešová1, Herman Mann5, Jiří Vencovský3, Karel Pavelka6 and Ladislav Šenolt3, 1Institute of Rheumatology, Prague, Czech Republic, 2Institute of Rheumatology and Department of Rheumatology, First Faculty of Medicine, Charles University, Prague, Czech Republic, Prague, Czech Republic, 3Institute of Rheumatology and Department of Rheumatology, First Faculty of Medicine, Charles University, Prague, Czech Republic, 4First faculty of medicine and Institute of Rheumatology, Charles University, Prague, Czech Republic, 5Revmatologický ústav, Praha, Czech Republic, 6Institute of Rheumatology and Charles University, Praha, Czech Republic

Meeting: ACR Convergence 2024

Keywords: ,

Session Information

Date: Monday, November 18, 2024

Title: RA – Diagnosis, Manifestations, & Outcomes Poster III

Session Type: Poster Session C

Session Time: 10:30AM-12:30PM

Background/Purpose: Despite advances in biologic (b-) and targeted synthetic (ts-) disease-modifying anti-rheumatic drugs (DMARDs), a significant subset of rheumatoid arthritis (RA) patients remain symptomatic, meeting the definition of “difficult-to-treat” (D2T) RA1.

This study aimed to identify crucial clinical indicators that can differentiate patients at risk of developing D2T RA at enrollment into the registry (baseline) or one to two years prior to meeting the definition of D2T RA.

Methods: We retrospectively analysed 8,543 RA patients from the Czech Republic biologics registry ATTRA2 (2002 – 2023) who commenced treatment with b/tsDMARDs. D2T RA was defined according to EULAR criteria 1. For comparison, we identified patients in sustained clinical remission as those with a Simple Disease Activity Index (SDAI) < 3.3 and Swollen Joint Counts (SJC) ≤ 1 over two consecutive follow-ups 12 weeks apart. All patients initiated b/tsDMARDs treatment, and their characteristics are listed in Table 1.

Results: Altogether, 641 patients with D2T RA (mean age 50.6 years at baseline; 84% female) were compared with 641 RA patients in sustained remission, matched by age, gender, disease duration, and b/tsDMARD treatment at each time point.

The machine learning model demonstrated accuracy and area under the receiver operating characteristics curve (AUC) ranges of 0.606–0.747 and 0.656–0.832, respectively, for predicting D2T RA (Figure 1). The SHAP analysis identified key predictors of D2T RA, such as clinical disease activity measures, CRP, and duration of b/tsDMARD treatment (Figure 2). The best performance of these predictors was one year before the patients met the D2T RA.

Conclusion: Our study identified clinical features that, at baseline and even up to one year before meeting the D2T RA definition, predict its development with greater likelihood. These findings provide valuable insights that could enhance the early identification and management of patients at risk for D2T RA, potentially improving treatment strategies and outcomes. Future research should focus on optimising and validating these predictive models.

Acknowledgements: Supported by SVV 260 638, MHCR 023728, NU23-10-00434.

References:

1.       1.   Nagy, G. et al. EULAR definition of difficult-to-treat rheumatoid arthritis. Ann Rheum Dis 80, 31–35 (2021).

2.       2.   Skácelová, M. et al. The beneficial effect of csDMARDs co-medication on drug persistence of first-line TNF inhibitor in rheumatoid arthritis patients: data from Czech ATTRA registry. Rheumatol Int 42, 803–814 (2022).

Supporting image 1

Table 1 – Demographic and clinical characteristics. Demographic and clinical characteristics of patients with difficult-to-treat (D2T) rheumatoid arthritis (RA) and patients in sustained clinical remission at baseline and two and one years before D2T classification. Data are presented as median (IQR) or count (%). The subjects of the remission group matched an equal number of patients with D2T disease based on age, gender, duration of disease, and duration of biological treatment.
Abbreviations: ESR – erythrocyte sedimentation rate; CRP – C-reactive protein; SJC28 – Swollen 28-Joint Count; TJC28 – Tender 28-Joint Count; DAS28-ESR – Disease Activity Score 28-Joint Count with Erythrocyte Sedimentation Rate; CDAI – Clinical Disease Activity Index; csDMARDs – Conventional Synthetic Disease-Modifying Anti-Rheumatic Drugs; anti-CCP – Antibodies against cyclic citrullinated peptide.

Supporting image 2

Figure 1 – The performance of machine learning models. Receiver operating characteristics (ROC) curve analyses visualise the performance of all machine learning models in prediction difficult-to-treat (D2T) classification at baseline and two and one years in advance. The table below shows the performance of all machine learning methods at baseline and two and one years before the classification.

Supporting image 3

Figure 2 – Shapley plots for difficult-to-treat (D2T) prediction at each time point. Shapley plots show Shapley additive explanation (SHAP) values in the order of the important variables contributing to D2T disease. Each variable’s feature values change from light (low) to dark (high). For Disease Activity Score 28-Joint Count with Erythrocyte Sedimentation Rate (DAS28-ESR), the lighter colours represent lower DAS28-ESR, whereas the darker colours represent higher DAS28-ESR values. For the one year before D2T figure, increasing (positive SHAP values with darker colours indicate that higher DAS28-ESR are strongly related to D2T, whereas decreasing (negative) SHAP values with lighter colours indicate that lower DAS28-ESR are strongly related to remission. In binary variables, such as glucocorticoids, the darker colours represent ‘yes’, and the lighter colours represent ‘no’. Therefore, linear changes in the feature values are expressed in colour, and the SHAP values (impact on model output) indicate closer to remission or D2T classification based on the feature values.

Disclosures: J. Baloun: None; L. Andrés Cerezo: None; T. Kropáčková: None; A. Prokopcova: None; K. Brábníková Marešová: None; H. Mann: AbbVie/Abbott, 6, Eli Lilly, 6, Janssen, 6, Novartis, 6, Pfizer, 6, SOBI, 6; J. Vencovský: AbbVie/Abbott, 6, Argenx, 2, Biogen, 6, Eli Lilly, 2, Fresenius, 6, Galapagos, 2, Horizon, 2, Merck/MSD, 6, Octapharma, 6, Pfizer, 6, Sobi, 2, Takeda, 6, UCB, 1, 2, 6; K. Pavelka: AbbVie/Abbott, 6, Bristol-Myers Squibb(BMS), 6, Eli Lilly, 6, Merck/MSD, 6, Novartis, 6, Pfizer, 6, UCB, 6; L. Šenolt: AbbVie/Abbott, 1, 6, Eli Lilly, 1, 6, GlaxoSmithKlein(GSK), 1, 6, Janssen, 1, 6, Novartis, 1, 6, Pfizer, 1, 6, UCB, 1, 6.

To cite this abstract in AMA style:

Baloun J, Andrés Cerezo L, Kropáčková T, Prokopcova A, Brábníková Marešová K, Mann H, Vencovský J, Pavelka K, Šenolt L. Identifying Clinical Predictors of Difficult-to-Treat Rheumatoid Arthritis Using Machine Learning-based Techniques [abstract]. Arthritis Rheumatol. 2024; 76 (suppl 9). https://acrabstracts.org/abstract/identifying-clinical-predictors-of-difficult-to-treat-rheumatoid-arthritis-using-machine-learning-based-techniques/. Accessed .

« Back to ACR Convergence 2024

ACR Meeting Abstracts - https://acrabstracts.org/abstract/identifying-clinical-predictors-of-difficult-to-treat-rheumatoid-arthritis-using-machine-learning-based-techniques/