ACR Meeting Abstracts

ACR Meeting Abstracts

  • Meetings
    • ACR Convergence 2024
    • ACR Convergence 2023
    • 2023 ACR/ARP PRSYM
    • ACR Convergence 2022
    • ACR Convergence 2021
    • ACR Convergence 2020
    • 2020 ACR/ARP PRSYM
    • 2019 ACR/ARP Annual Meeting
    • 2018-2009 Meetings
    • Download Abstracts
  • Keyword Index
  • Advanced Search
  • Your Favorites
    • Favorites
    • Login
    • View and print all favorites
    • Clear all your favorites
  • ACR Meetings

Abstract Number: 1607

Machine Learning Identifies Biomarker of Non-Response to Methotrexate in Rheumatoid Arthritis

Vincent Bouget1, Julien Duquesne1, Paul-Henry Cournède2, Bruno Fautrel3, Francis Guillemin4, Pascal de Jong5, Judith Heutz6, Marloes Verstappen7, Annette van der Helm-van Mil8, Xavier Mariette9 and Samuel Bitoun10, 1Scienta Lab, Paris, France, 2Paris-Saclay University - CentraleSupelec, Paris, France, 3Sorbonne University Paris, France and Pierre Louis Institute of Epidemiology and Public Health, Paris, France, Paris, France, 4Université de Lorraine, APEMAC, Vandoeuvre-les-nancy, France, 5ErasmusMC, Hendrik Ido Ambacht, Netherlands, 6Erasmus Medical Center, Rotterdam, Netherlands, 7Leiden University Medical Center, Leiden, Netherlands, 8Leiden University Medical Center, Erasmus Medical Center, Leiden, Netherlands, 9Paris-Saclay University, Rueil Malmaison, Ile-de-France, France, 10Paris-Saclay University, Paris, France

Meeting: ACR Convergence 2022

Keywords: Bioinformatics, Biomarkers, Disease-Modifying Antirheumatic Drugs (Dmards), rheumatoid arthritis

  • Tweet
  • Click to email a link to a friend (Opens in new window) Email
  • Click to print (Opens in new window) Print
Session Information

Date: Sunday, November 13, 2022

Title: Abstracts: RA – Treatment II: Pre- and Early Disease

Session Type: Abstract Session

Session Time: 4:30PM-6:00PM

Background/Purpose: Methotrexate (MTX) is the first line treatment for rheumatoid arthritis (RA). Unfortunately, 30% to 40% of RA patients do not respond to MTX with a risk of joint erosion. On the other hand, several second-line treatments exist to treat patients with inadequate response. Identifying patients unlikely to respond to MTX before initiating the treatment is therefore a major goal to improve RA management.

Using machine learning models based on simple clinical and biological data, we aimed at unveiling simple clinical markers and/or biomarkers of non-response to MTX.

Methods: We used data from the ESPOIR and Leiden early arthritis cohorts to train a machine learning model (LightGBM) predicting response to MTX. We included patients that fulfilled the 2010 EULAR/ACR criteria and were treated with MTX in monotherapy +/- steroids. The model uses patient’s characteristics at treatment initiation. The predictive variables were selected using backward feature selection. The endpoint was to predict the therapeutic response, defined as the EULAR response 9 months (+/- 6 months) after treatment initiation.

To identify biomarkers, we used Shapley values to analyze how variables are used by the model to predict response. For a given patient and a given variable, Shapley values quantify how much this variable contributed to the predicted therapeutic response. On the training set, we first computed the Shapley values. Based on these Shapley values, patients were then clustered into subgroup of responders/non-responders using the K-means algorithm. We validated the results on the tREACH trial.

Results: We included 433 patients from the ESPOIR cohort, 243 from the Leiden cohort and 138 from the t-REACH trial. Among the 25 available variables, the feature selection automatically selected DAS28, lymphocytes count, white blood cells, creatininemia, AST, ALT, swollen joints count, and corticosteroids co-treatment as predictors of EULAR response. In the replication cohort, the model achieved an AUROC of 0.75 (CI 0.66-0.83) (Table 1).

The strongest predictor of response was DAS28. As expected, high DAS28 was associated to response and low DAS28 was associated with non-response (Figure 1 – left panel: yellow, green, and purple clusters). Interestingly, for patients with moderate activity (DAS28 between 3.2 and 5.1), DAS28 is a weak predictor of response (Figure 1 – left panel: red and cyan clusters are close to x-axis). In this population, which is currently the most frequent population in which DMARDs are introduced, blood lymphocytes count was potent to discriminate responders and inadequate responders (Table 2): around 60% of patients above 2900 cells/mm3 were non-responders vs 30% in the whole population (Figure 1 – right panel: red cluster is strongly correlated to non-response). This simple rule obtained using machine learning can be used in clinical practice without any algorithm to identify potential non-responders.

Conclusion: Using clinical and biological variables, we accurately predicted inadequate response to MTX. Thanks to our machine learning model, we identified high blood lymphocytes count as a simple biomarker of non-response for RA patients with moderate activity. It could have direct impact in clinical practice.

Supporting image 1

Table 1. Machine learning model performances.

Supporting image 2

Figure 1. Shapley values graphs A positive Shapley value represents how much the variable influences the prediction towards response for a given patient. Each patient is represented by a point. Patients are clustered based on Shapley values.
Left panel Shapley values of the model computed on the training set for DAS28.
Right Panel: Shapley values of the model computed on the training set for Lymphocytes count.

Supporting image 3

Table 2. Summary of high-lymphocytes-count subgroup statistics.


Disclosures: V. Bouget, None; J. Duquesne, None; P. Cournède, None; B. Fautrel, Pfizer, Novartis, Roche, Sanofi-Aventis, SOBI, UCB; F. Guillemin, None; P. de Jong, None; J. Heutz, None; M. Verstappen, Bristol-Myers Squibb(BMS); A. van der Helm-van Mil, None; X. Mariette, AstraZeneca, Bristol Myers Squibb, Galapagos, GSK, Novartis, Pfizer; S. Bitoun, None.

To cite this abstract in AMA style:

Bouget V, Duquesne J, Cournède P, Fautrel B, Guillemin F, de Jong P, Heutz J, Verstappen M, van der Helm-van Mil A, Mariette X, Bitoun S. Machine Learning Identifies Biomarker of Non-Response to Methotrexate in Rheumatoid Arthritis [abstract]. Arthritis Rheumatol. 2022; 74 (suppl 9). https://acrabstracts.org/abstract/machine-learning-identifies-biomarker-of-non-response-to-methotrexate-in-rheumatoid-arthritis/. Accessed .
  • Tweet
  • Click to email a link to a friend (Opens in new window) Email
  • Click to print (Opens in new window) Print

« Back to ACR Convergence 2022

ACR Meeting Abstracts - https://acrabstracts.org/abstract/machine-learning-identifies-biomarker-of-non-response-to-methotrexate-in-rheumatoid-arthritis/

Advanced Search

Your Favorites

You can save and print a list of your favorite abstracts during your browser session by clicking the “Favorite” button at the bottom of any abstract. View your favorites »

All abstracts accepted to ACR Convergence are under media embargo once the ACR has notified presenters of their abstract’s acceptance. They may be presented at other meetings or published as manuscripts after this time but should not be discussed in non-scholarly venues or outlets. The following embargo policies are strictly enforced by the ACR.

Accepted abstracts are made available to the public online in advance of the meeting and are published in a special online supplement of our scientific journal, Arthritis & Rheumatology. Information contained in those abstracts may not be released until the abstracts appear online. In an exception to the media embargo, academic institutions, private organizations, and companies with products whose value may be influenced by information contained in an abstract may issue a press release to coincide with the availability of an ACR abstract on the ACR website. However, the ACR continues to require that information that goes beyond that contained in the abstract (e.g., discussion of the abstract done as part of editorial news coverage) is under media embargo until 10:00 AM ET on November 14, 2024. Journalists with access to embargoed information cannot release articles or editorial news coverage before this time. Editorial news coverage is considered original articles/videos developed by employed journalists to report facts, commentary, and subject matter expert quotes in a narrative form using a variety of sources (e.g., research, announcements, press releases, events, etc.).

Violation of this policy may result in the abstract being withdrawn from the meeting and other measures deemed appropriate. Authors are responsible for notifying colleagues, institutions, communications firms, and all other stakeholders related to the development or promotion of the abstract about this policy. If you have questions about the ACR abstract embargo policy, please contact ACR abstracts staff at [email protected].

Wiley

  • Online Journal
  • Privacy Policy
  • Permissions Policies
  • Cookie Preferences

© Copyright 2025 American College of Rheumatology