Search in the Abstract Database

Abstracts Search 2014

* = Presenting author

P566. A matrix-based prediction model for primary response to infliximab in Crohn's disease patients

T. Billiet1, M. de Bruyn1, K. Claes1, V. Ballet2, X. Liu3, R. Kirkland3, K. Drake3, S. Lockton3, F. Princen3, S. Singh3, M. Ferrante2, G. Van Assche2, P. Rutgeerts2, I. Cleynen1, S. Vermeire2, 1Department of Clinical and Experimental Medicine, KU Leuven, Translational Research in GastroIntestinal Disorders, Leuven, Belgium, 2University Hospitals Leuven, Department of Gastroenterology, Leuven, Belgium, 3Prometheus Laboratories, Department of Research and Development, San Diego, United States

Background

Primary non-response (PNR) to TNF antagonists in IBD still holds a challenge for clinicians. Furthermore, with the advent of anti-integrin molecules, selecting the right therapeutic class for a given patient would be welcomed. We designed a matrix-based prediction model, which may avoid exposure in patients who are unlikely to have benefit of the drug.

Methods

201 anti-TNF naïve Crohn's disease patients started on infliximab (IFX) induction 0–2-6 were used. For each patient, clinical information and biological markers (CRP, albumin, …) were collected prior to IFX start. Baseline serum TNF load and the IBD serology 7 panel (Prometheus Laboratories Inc.) were also available. PNR was defined as no clinical benefit after induction regimen. Univariate and correlation analyses were performed to select possible predictors. A combination of final predictors was selected through multiple regression based on the Bayesian information criterion (BIC). Finally, predicted probabilities were calculated and were organized into a matrix.

Results

The incidence of PNR was 8%. Duration of disease, age at first IFX, BMI, previous surgery, serum albumin, ASCAA, ASCAG and serum TNF load differed substantially between responders and non-responders in univariate analysis (all p < 0.2). The parsimonious BIC in a multiple regression analysis withheld three independent final predictors (p < 0.05): age at first IFX, BMI and previous surgery. The AUC for this model was 0.79 and predictors remained significant after bootstrapping. BMI was categorized into ≤20 (28% of patients), 21–25 (45%) and ≥26 (27%) and age at first IFX was categorized into ≤25 years (27%), 26–64 years (68%) and ≥65 years (5%). Previous surgery was the strongest associated predictor of PNR with an OR of 5.18 [1.51–23.96]. The matrix model using these final predictors showed a good spread of primary response rates for the different categories with a matching color code.

Figure: Matrix presentation of predicted rates of primary response to infliximab for different patient categories based on the final prediction model.

Conclusion

We developed a matrix-based prediction tool to aid physicians in optimizing therapeutic decisions. After stringent selection of predictors only age at first IFX, BMI and previous surgery were withheld. A younger age has been associated with primary response in several independent cohorts. BMI is also known to influence response and this is assigned to the weight based dosing of IFX. Previous surgery has also been associated with PNR and might reflect more refractory disease. The next step is to validate this matrix in an independent cohort and to construct a matrix-prediction tool for secondary loss of response.