Abstract
A PREDICTION MODEL FOR SWITCHING OF BIOLOGICS AND TSDMARDS IN RHEUMATOID ARTHRITIS
Full text
L. Cappelli, G. Reed, J. KremerJohns Hopkins School of Medicine, Rheumatology, Baltimore, United States of America
UMass Chan Medical School, Medicine, Worcester, United States of America
Corrona Research Foundation, Rheumatology, Delray Beach, United States of America
Background Patients with rheumatoid arthritis (RA) have many options for biologic therapy both within treatment classes and across differing mechanisms of action, but the lack of predictability of treatment response and failure can be frustrating for patients and physicians. Rheumatologists longitudinally collect patient and physician reported measurements as a part of routine clinical care that can be used for prediction of outcomes . Harnessing data regularly collected to monitor RA to inform treatment decision making could improve patient care.
Objectives To develop a readily usable model with data collected in clinical care at preceding visits to predict the probability of switching biologic at a subsequent clinic visit
Methods Patients were drawn from the CorEvitas (formerly CORRONA) registry and were adults over the age of 18 with a diagnosis of RA. The study matched patients who switched biologics with control patients who had not switched biologics; the cohort was divided into a training and test set for prediction model development and validation. Switchers: RA patients initiating a biologic/tsdmard and switching to another biologic/tsDMARD with at least two prior visits to the switch while on drug. Visits prior to the switch were between 2 and 12 months before. Control visit: Patient visits while still on initiated drug with at least 2 visits prior while on drug; next visit could not be a discontinuation or switch. A control visit was matched to a switch visit within 2 months of time from initiation by drug class and prior biologic experience. Pairs of control-switch visits were divided into a training (60%) and test set (40%). Using the training set, best subset regression, lasso and elastic net methods were used to determine the best potential models. Area under the ROC curve was used for the final selection of best model and estimated coefficients of this model were applied to the test data set to predict switching. Change in disease activity measures had a non-linear association with switching and were fit using linear splines with a knot at zero.
Results A total of 5050 patients were included, 3016 in the training and 2034 in the test dataset. The average age was 59.6, the majority were female (3998, 79.2%), and the average duration of RA at the time of switch or control visit was 12.8 years. The final model included prior CDAI by category, prior patient pain measurement, change in CDAI from baseline, age group, and number of prior biologics which were significantly associated (either positively or negatively) with switching biologics (Table 1). The area under the ROC curve was 0.690 for this model with the training dataset (Figure 1). The model was then applied to the test data with similar performance; the area under the ROC curve was 0.687 (Figure 1).
Conclusion We developed a simple model to determine the probability of switching biologics for RA at the following clinic visit. This could be incorporated into the electronic medical records or used as an application to give clinicians more information about their patient’s trajectory and likelihood of failing a biologic.
References
Novella-Navarro M et al. Predictive model to identify multiple failure to biological therapy in patients with rheumatoid arthritis. Ther Adv Musculoskelet Dis. 2022 Oct 6;14:1759720X221124028.
Image/graph:
Table 1. Prediction model in training data set
Predictor
Odds Ratio
95% CI
coefficient
CDAI prior to switch
Remission (ref)
1
Low
1.75
(1.390 to 2.203)
0.559
Moderate
2.394
(1.845 to 3.107)
0.873
Severe
2.964
(2.141 to 4.103)
1.086
Pt Pain prior to switch (0-100)
1.01
(1.006 to 1.013)
0.010
Age (yrs)
≤40 (ref)
1
41-50
0.958
(0.684 to 1.341)
-0.043
51-60
0.823
(0.603 to 1.123)
-0.195
61-70
0.747
(0.549 to 1.017)
-0.291
>70
0.606
(0.436 to 0.843)
-0.501
Prior no. of biologics/tsDMARDs
0
1
1
0.904
(0.749 to 1.090)
-0.101
2
0.788
(0.630 to 0.984)
-0.239
3+
0.706
(0.558 to 0.893)
-0.348
Change in CDAI at prior to switch (spline at zero)
<0
1.011
(1.004 to 1.018)
0.011
>0
1.044
(1.022 to 1.067)
0.043
CDAI: clinical disease activity index; DMARD: disease modifying anti-rheumatic drugs; ts: targeted synthetic; ref: reference
Acknowledgements: NIL.
Disclosure of Interests Laura Cappelli Grant/research support from: Bristol-Myers Squibb, George Reed Consultant of: CorEvitas, Corrona Research Foundation, Joel Kremer Shareholder of: CorEvitas, Consultant of: Lilly.
Keywords: Prognostic factors, Registries
DOI: 10.1136/annrheumdis-2023-eular.2097Citation: , volume 82, supplement 1, year 2023, page 492Session: Rheumatoid arthritis - prognosis, predictors and outcome
(Poster View)
3 organizations
Organization
Johns Hopkins School of MedicineOrganization
UMass Chan Medical SchoolOrganization
Corrona Research Foundation