Marion Bouillon-Pichault (1),, Bruno Boulanger (1), Astrid Jullion (1), Fabrice Nollevaux (1)
(1) Arlenda, Belgium
Objectives: In clinical pain studies, the primary endpoint is frequently reported using an 11-point Numeric Rating Scale, ranging from 0, no pain, to 10, worst pain imagined. These data are thus ordinal, categorical and bounded at both sides. Data from a double blind, placebo controlled, and flexible titration study on Drug A were modelled. To account for characteristics of the distribution of the pain scores and the design of the study two different approaches were investigated: a longitudinal and an ordered categorical approach.
Methods: In the longitudinal approach, the pain score was calculated as mean change from baseline over a week in order to be able to treat the dependent variable as normally distributed. A time dependent Emax model was fitted with the baseline as a covariate on the Emax parameter. Dose could not successfully be included in the model. This was probably due to the flexible titration design of the study leading to confounding between subject sensitivity, time and dose received. Thus, the population was divided into sub-populations according to highest received dose or to drug sensitivity level. The Emax model was separately applied to each of these sub-populations in order to estimate the maximal pain reduction per highest dose or per drug sensitivity level.
In the ordered categorical approach, the individual twice daily pain scores were fitted using a model including dose, baseline, period of the day of pain assessment (morning or evening) and time (day) as covariates. This approach allows direct modelling of the cumulative probabilities to reach a pain score.
Results: Both approaches adequately described the data. With the longitudinal approach, the maximal effects of Drug A per highest dose seemed to reach a plateau at the higher doses whereas the maximal effects of Drug A per drug sensitivity level seemed to be stable among the dose range. The ordered categorical approach showed increased cumulative probabilities of lower pain scores when the dose increases.
Conclusions: The longitudinal approach did not succeed in establishing a relationship between dose and efficacy while the ordered categorical approach succeeded in integrating both time and dose in the modelling of pain scores observed in a flexible titration design. However, this approach requires extensive computation time and is less straightforward to apply and to understand than the longitudinal approach.
Reference: PAGE 21 () Abstr 2348 [www.page-meeting.org/?abstract=2348]
Poster: New Modelling Approaches