Arthritis & Rheumatism, Volume 60,
October 2009 Abstract Supplement

The 2009 ACR/ARHP Annual Scientific Meeting
Philadelphia October 16-21, 2009.


Rapid Prediction of Low Disease Activity at 1 Year Among RA Patients Treated with Certolizumab Pegol

Curtis1,  Jeffrey R., Luijtens2,  Kristel

UAB, Birmingham, AL,
UCB, Brussels, Belgium

Purpose:

Rapidly predicting which RA pts will respond to biologics may optimize outcomes.The objective of our analysis was to determine the prognostic significance of data collected early after starting certolizumab pegol (CZP) to predict low disease activity (LDA) at 1 year.

Methods:

Wks 4 and 12 data from CZP-treated pts in RAPID 1 were used as variables to predict LDA (DAS28 (ESR) <=3.2) at Wk 52. Classification and Regression Tree (CART) software (Salford Systems) identified variables, built a prediction model, and tested its performance using a split sample (ie, separate training/testing datasets). The ability of this model (Model 1) to discriminate between pts who did/did not achieve LDA at Wk 52 was assessed using area under the receiver-operator curve (AUC). For pts for whom the prediction of LDA at Wk52 using Wk 4/12 data was suboptimal, Wk 28 data were added. A second prediction model (Model 2) was constructed in which more complex variables (ie, DAS) were replaced with ones more easily measured in clinical practice (ie, Clinical Disease Activity Index).

Results:

Of 783 pts randomized to CZP (RAPID 1 ITT pop), 703 were included in this analysis (80 pts who withdrew for safety were excluded); 83% female, mean age 51.7 yrs, disease duration 6.1 yrs, concomitant MTX 13.6 mg/wk, baseline DAS28 6.9. Using Wk 4/12 data, LDA at Wk 52 was predicted with ~71–87% accuracy for 80% of pts. For the remaining 20% of pts, adding Wk 28 data resulted in similar accuracy (Figure below). Model 2 had comparable discrimination and accuracy. Overall results from the testing dataset are shown below; results from the training dataset were somewhat better (not shown).

Figure. Model 1, any variable, Weeks 4 and 12

Table. Model discrimination and calibration

Data used from these WksAUC% Correctly Classified
Model 1, Wks 4/120.7674%
Model 1, Wks 4/12/280.8384%
Model 2, Wks 4/120.7176%
Model 2, Wks 4/12/280.8182%

Conclusion:

Using CART, we accurately classified ~75% of RA pts as early responders/non-responders within 12 wks of initiating CZP. Wk 28 data improved predictability by an additional 10%. Classification trees may be useful to prospectively guide the management of individual RA pts treated with CZP and perhaps other biologic therapies.

To cite this abstract, please use the following information:
Curtis, Jeffrey R., Luijtens, Kristel; Rapid Prediction of Low Disease Activity at 1 Year Among RA Patients Treated with Certolizumab Pegol [abstract]. Arthritis Rheum 2009;60 Suppl 10 :1672
DOI: 10.1002/art.26746

Abstract Supplement

Meeting Menu

2009 ACR/ARHP