Europeanhhm

Patient groups in Rheumatoid arthritis identified by deep learning respond differently to biologic or targeted synthetic DMARDs

Maria Kalweit, Andrea M. Burden, Joschka Boedecker, Thomas Hügle, Theresa Burkard

Abstract

Cycling of biologic or targeted synthetic disease modifying antirheumatic drugs (b/tsDMARDs) in rheumatoid arthritis (RA) patients due to non-response is a problem preventing and delaying disease control. We aimed to assess and validate treatment response of b/tsDMARDs among clusters of RA patients identified by deep learning. We clustered RA patients clusters at first-time b/tsDMARD (cohort entry) in the Swiss Clinical Quality Management in Rheumatic Diseases registry (SCQM) [1999–2018]. We performed comparative effectiveness analyses of b/tsDMARDs (ref. adalimumab) using Cox proportional hazard regression.

Introduction

Rheumatoid arthritis (RA) is an heterogenic inflammatory disorder, often presenting a fluctuant disease activity over the disease course [1]. Despite the advent of validated treatment recommendations, it remains challenging in clinical practice to avoid disease flares or overmedication [2]. Moreover, specific biomarkers to guide optimal biologic or targeted synthetic disease modifying antirheumatic drug (b/tsDMARD) use are not available [3].

Material and methods

We clustered patients at cohort entry (i.e., start of b/tsDMARD) using deep embedded clustering (DEC) [17] in combination with the AnyNets-Autoencoder, an adaptive deep adaptive neural network, which is especially designed to work with medical data with missing values [18]. The adaptive architecture outperformed classical feed-forward neural networks in combination with imputation methods (a naive approach to deal with missing values by replacing them with defaults such as e.g. zeros) [19]. DEC simultaneously learns feature representations and cluster assignments in a lower-dimensional space (latent space) using the adaptive deep autoencoder in order to reconstruct the input and iteratively optimizes a clustering objective [20].

Results
We identified 3516 patients with a first-time b/tsDMARD in SCQM between 1999 and 2018 (flow chart in S8 Fig). Selected patient characteristics of the overall population can be seen in Table 1. Patients had a mean age of 55.4 years and 76.2% of patients were women. The median RA duration of patients was 6.2 years, 69.3% were rheumatoid factor (RF) positive, the mean DAS28-esr was 4.3, and 66.6% of patients also used methotrexate at their first b/tsDMARD use.

Discussion
Epidemiologic studies to date have identified predictors of treatment response (or non-response) to certain b/tsDMARDs [24–28]. We compared our results with those because there are no studies that identified response to b/tsDMARDs among RA phenotypes that consist of more than one trait. Observed non-response to golimumab among seronegative women without use of prednisone as well as seropositive patients with higher disease burden and duration may suggest to avoid this treatment in these patients. However, no other significant findings were identified in these strata to help with treatment choice. Thus, adalimumab, the reference treatment may be a suggestion; especially since a study which assessed patient characteristics associated with treatment response to adalimumab suggested also that concomitant csDMARD use as well as a high disease burden was predictive of treatment response [24].

Conclusion
This study used deep learning to suggest RA patient groups which responded differently to certain b/tsDMARDs. Results were validated through stratified analyses according to few most distinct patient characteristics for application in clinical practice. Our results suggests optimal first-line b/tsDMARD use in certain patient groups which is a step forward towards personalizing treatment in RA patients. However, further research in other cohorts is needed to verify our results.

Acknowledgments

We thank all patients and rheumatologists contributing to the SCQM registry, as well as the entire SCQM staff. A list of rheumatology offices and hospitals which contribute to the SCQM registry and a list of supporters of SCQM can be found at http://www.scqm.ch.The professorship of Andrea M Burden is partially supported by PharmaSuisse and the ETH Foundation.

Citation: Kalweit M, Burden AM, Boedecker J, Hügle T, Burkard T (2023) Patient groups in Rheumatoid arthritis identified by deep learning respond differently to biologic or targeted synthetic DMARDs. PLoS Comput Biol 19(6): e1011073. https://doi.org/10.1371/journal.pcbi.1011073

Editor: Jennifer Wilson, Stanford University, UNITED STATES

Received: November 22, 2021; Accepted: April 4, 2023; Published: June 2, 2023

Copyright: © 2023 Kalweit et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Data Availability: Data cannot be shared publicly because the data belongs to a third party (Swiss Clinical Quality Management in Rheumatic Diseases registry, SCQM) and are only available through a contract with them. Furthermore, access to SCQM data requires a collaboration with a certified rheumatologist practicing in Switzerland who also contributes data to SCQM. Finally, data access further requires ethical approval or a waiver of such issued by a local ethics committee. Researchers interested in SCQM data and willing to comply with aforementioned restrictions will be able to obtain access to the data in the same manner as the authors. Contact information of SCQM can be found at their web page at https://www.scqm.ch/. The SAS code written by TB and the Python code written by MK for this project is available from https://github.com/tiozab/comparative-effectivness-in-RA-patient-clusters.

Funding: The author(s) received no specific funding for this work.

Competing interests: I have read the journal’s policy and the authors of this manuscript have the following competing interests: TH has received research grants and honorariums from Pfizer, AbbVie, Novartis, and Janssen, and does consultancy for Eli Lilly, Janssen, and Menarini; no other relationships or activities that could appear to have influenced the submitted work.

 

https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1011073#sec002

       

Equinix accelerated medical research...The Healthcare Patient Experience & Engagement Summit 2024Healthcare Innovation & Transformation SummitHealthcare CNO SummitHealthcare CMO Summit