Modelling Ordered Categorical Allergic Rhinitis Scores in an Environmental Exposure Unit Study
Rik Schoemaker(1), Jakob Ribbing (2)
(1)Exprimo NV, Mechelen, Belgium and (2)Pfizer AB, Sollentuna, Sweden
Introduction: An Environmental Exposure Unit (EEU) study was performed in allergic rhinitis patients to determine the effects of placebo (n=90), fexofenadine (n=225) and cetirizine (n=225) on self-assessed rhinoconjunctivitis symptoms. Patients were exposed to ragweed pollen and treated on two consecutive days. Day 1 consisted of seven hours of ragweed pollen exposure where treatment was started two hours after initiating allergen exposure and Day 2 consisted of six hours of ragweed pollen exposure where treatment was started three hours after initiating allergen exposure.
Methods: Model development was performed using non-linear mixed effects modelling with a proportional odds model. A new modelling approach was obtained by describing both the increase in score due to allergen exposure and the subsequent decrease in score due to treatment where the treatment profile for both active and placebo treatments was described using K-PD-type methodology[1]. Parameters were estimated using the SAEM algorithm in both Monolix and NONMEM VII. Novel graphics were developed to generate visual predictive checks for time profiles of ordered categorical data.
Results: Sensitivity to treatment was quantified using an EDF50-parameter where cetirizine and fexofenadine were shown to induce a significant decrease in the estimated value relative to placebo. Duration of action over both days of treatment was quantified using a half-life parameter where placebo was shown to be associated with a shorter effect half-life. Various predictive checks illustrated the adequacy of the model both for predicting symptom-score time profiles and for predicting derived statistics, such as average score 2-5 hours post-treatment and time to onset of effect. Monolix seemed less critical in the choice of initial estimates than NONMEM and was therefore used initially.
Conclusion: Excellent descriptions were obtained of individual score profiles both for the five scores on the 0-3 point scale and the Total Symptom Severity Complex (TSSC), an aggregated score on the 0-12 point scale. Implementation of a proportional odds model allowed a proper description of the specific ordered categorical nature of the data with model predictions corresponding to the observed data range. The underlying K-PD model allowed both a description of the treatment profile in the absence of concentration measurements and an adequate description of the placebo treatment effect.
Reference:
[1] Jacqmin P, Snoeck E, van Schaick EA, Gieschke R, Pillai P, Steimer JL and Girard P. Modelling Response Time Profiles in the Absence of Drug Concentrations: Definition and Performance Evaluation of the K-PD Model. J Pharmacokinet Pharmacodyn, 34(1), 2007, 57-85.
