P117 Machine-learning derived characteristics associated with tapering TNF inhibitors in individuals with rheumatoid arthritis
P117 Machine-learning derived characteristics associated with tapering TNF inhibitors in individuals with rheumatoid arthritis
Background/Aims
Tapering of TNF inhibitor (TNFi) drugs may be considered in some patients to reduce risks and costs. Selecting appropriate patients is not always straightforward and may be influenced by age, sex, comorbidity and disease activity state. We sought to identify predictors for dose tapering in a real-world clinical setting. Algorithmic extraction, selection and analysis of relevant patient sub-cohorts could enable identification of relevant predictors associated with TNFi dose tapering.
Methods
Our institution has a Rheumatology Biologics database running prospectively for over 15 years. Our approach for patients with RA receiving TNFi has been to dose-taper by one third and then 50% if remission achieved (defined as DAS28<2.6 on two occasions more than 6 months apart with no corticosteroid use). Prescribing, disease activity scores and demographics were extracted using SQL along with comorbidity coding, pathology results and anthropometric data. Data were anonymised and analysed in Python 3.8 within our institutions’ Trusted Research Environment (TRE). Pandas, NumPy and StatsModels python packages were used for the analysis in Jupyter notebooks. 49 covariates were considered clinically relevant and included in the regression analysis. Recursive feature elimination (RFE) was performed using logistic regression (LR) with threshold p-value of 0.05. The primary outcome was tapering of TNFi, algorithmically identified by a temporal increase in dosing interval or decrease in dosage. To avoid multiple-drug confounding, only the most recent TNFi data was included for each patient.
Results
663 patients with RA were initiated on TNFi between 5th November 2001 and March 2nd 2020. 491 (74.1%) were female with a mean age of 65.5 (SD ± 14.2) years. 261 (39.4%) received adalimumab, 209 (31.5%) etanercept, 74 (11.2%) infliximab, 82 (12.4%) certolizumab and 37 (5.6%) golimumab. Concurrent methotrexate (MTX) was seen in 34.5% (n = 22), either oral or subcutaneous. There was no change in the likelihood of tapering associated with depression, hypothyroidism, obesity, smoking or seropositivity for RF or anti-CCP. Those taking MTX were more likely to taper their biologics (OR 3.33, 95%CI 1.83-6.09, p = <.000), as were patients who were coded as having type 1 or 2 diabetes (7.4% n = 49, OR 3.23, 95%CI 1.32-8.25, p = 0.011), Higher DAS28 CRP score (OR 0.548, 95%CI 0.38-0.78, p = 0.001), and DAS 28 ESR score (OR 0.71, 95%CI 0.53-0.95, p = 0.021) significantly decreased chance of tapering.
Conclusion
Concurrent methotrexate use increases likelihood of subsequent tapering in patients with RA receiving TNFi. Unexpectedly, patients with diabetes were also more likely to taper, however due to low numbers of patients in this group and the width of confidence intervals this should be interpreted cautiously. As expected, patients with high disease activity scores were less likely to taper. This algorithm driven approach produced results largely in keeping with clinical intuition, however these methods may aid in future selection of tapering cohorts.
Phillips, Thomas
30ef6ddd-1f4a-4791-89e5-37c092fcba51
Bhandari, Megha
3fa5520b-34ee-4294-8fd7-38082ae86bad
Stammers, Matthew
9350205a-3938-4d75-8e86-233a38cdbb0e
Fraser, Simon
135884b6-8737-4e8a-a98c-5d803ac7a2dc
George, Michael
0bcd25e6-33a4-46fe-be5f-6e2096f98a56
Lin, Sharon
41d7f448-17f8-435f-8900-59f42b5dd783
Lwin, May
c9c177cc-2633-4aee-b282-9f767b2b0760
Holroyd, Christopher
38511e1e-7504-45d0-ab00-eacf22108b7a
Edwards, Christopher
dcb27fec-75ea-4575-a844-3588bcf14106
Phillips, Thomas
30ef6ddd-1f4a-4791-89e5-37c092fcba51
Bhandari, Megha
3fa5520b-34ee-4294-8fd7-38082ae86bad
Stammers, Matthew
9350205a-3938-4d75-8e86-233a38cdbb0e
Fraser, Simon
135884b6-8737-4e8a-a98c-5d803ac7a2dc
George, Michael
0bcd25e6-33a4-46fe-be5f-6e2096f98a56
Lin, Sharon
41d7f448-17f8-435f-8900-59f42b5dd783
Lwin, May
c9c177cc-2633-4aee-b282-9f767b2b0760
Holroyd, Christopher
38511e1e-7504-45d0-ab00-eacf22108b7a
Edwards, Christopher
dcb27fec-75ea-4575-a844-3588bcf14106
Phillips, Thomas, Bhandari, Megha, Stammers, Matthew, Fraser, Simon, George, Michael, Lin, Sharon, Lwin, May, Holroyd, Christopher and Edwards, Christopher
(2023)
P117 Machine-learning derived characteristics associated with tapering TNF inhibitors in individuals with rheumatoid arthritis.
Rheumatology, 62, [kead104.158].
(doi:10.1093/rheumatology/kead104.158).
Abstract
Background/Aims
Tapering of TNF inhibitor (TNFi) drugs may be considered in some patients to reduce risks and costs. Selecting appropriate patients is not always straightforward and may be influenced by age, sex, comorbidity and disease activity state. We sought to identify predictors for dose tapering in a real-world clinical setting. Algorithmic extraction, selection and analysis of relevant patient sub-cohorts could enable identification of relevant predictors associated with TNFi dose tapering.
Methods
Our institution has a Rheumatology Biologics database running prospectively for over 15 years. Our approach for patients with RA receiving TNFi has been to dose-taper by one third and then 50% if remission achieved (defined as DAS28<2.6 on two occasions more than 6 months apart with no corticosteroid use). Prescribing, disease activity scores and demographics were extracted using SQL along with comorbidity coding, pathology results and anthropometric data. Data were anonymised and analysed in Python 3.8 within our institutions’ Trusted Research Environment (TRE). Pandas, NumPy and StatsModels python packages were used for the analysis in Jupyter notebooks. 49 covariates were considered clinically relevant and included in the regression analysis. Recursive feature elimination (RFE) was performed using logistic regression (LR) with threshold p-value of 0.05. The primary outcome was tapering of TNFi, algorithmically identified by a temporal increase in dosing interval or decrease in dosage. To avoid multiple-drug confounding, only the most recent TNFi data was included for each patient.
Results
663 patients with RA were initiated on TNFi between 5th November 2001 and March 2nd 2020. 491 (74.1%) were female with a mean age of 65.5 (SD ± 14.2) years. 261 (39.4%) received adalimumab, 209 (31.5%) etanercept, 74 (11.2%) infliximab, 82 (12.4%) certolizumab and 37 (5.6%) golimumab. Concurrent methotrexate (MTX) was seen in 34.5% (n = 22), either oral or subcutaneous. There was no change in the likelihood of tapering associated with depression, hypothyroidism, obesity, smoking or seropositivity for RF or anti-CCP. Those taking MTX were more likely to taper their biologics (OR 3.33, 95%CI 1.83-6.09, p = <.000), as were patients who were coded as having type 1 or 2 diabetes (7.4% n = 49, OR 3.23, 95%CI 1.32-8.25, p = 0.011), Higher DAS28 CRP score (OR 0.548, 95%CI 0.38-0.78, p = 0.001), and DAS 28 ESR score (OR 0.71, 95%CI 0.53-0.95, p = 0.021) significantly decreased chance of tapering.
Conclusion
Concurrent methotrexate use increases likelihood of subsequent tapering in patients with RA receiving TNFi. Unexpectedly, patients with diabetes were also more likely to taper, however due to low numbers of patients in this group and the width of confidence intervals this should be interpreted cautiously. As expected, patients with high disease activity scores were less likely to taper. This algorithm driven approach produced results largely in keeping with clinical intuition, however these methods may aid in future selection of tapering cohorts.
This record has no associated files available for download.
More information
e-pub ahead of print date: 24 April 2023
Identifiers
Local EPrints ID: 477291
URI: http://eprints.soton.ac.uk/id/eprint/477291
ISSN: 1462-0324
PURE UUID: 6d78600e-c495-4971-83f4-07016e42fd2a
Catalogue record
Date deposited: 02 Jun 2023 16:36
Last modified: 11 Jul 2024 01:46
Export record
Altmetrics
Contributors
Author:
Thomas Phillips
Author:
Megha Bhandari
Author:
Matthew Stammers
Author:
Michael George
Author:
Sharon Lin
Author:
May Lwin
Author:
Christopher Holroyd
Download statistics
Downloads from ePrints over the past year. Other digital versions may also be available to download e.g. from the publisher's website.
View more statistics