Item response theory for the analysis of the placebo effect in Phase 3 studies of schizophrenia
Elke H.J. Krekels (1), Lena E. Friberg (1), An M. Vermeulen (2), Mats O. Karlsson (1)
(1) Department of Pharmaceutical Biosciences, Uppsala University, Uppsala, Sweden, (2) Janssen Research & Development, a division of Janssen Pharmaceutica NV, Beerse, Belgium
Objectives: Disease severity in schizophrenia is quantified using the PANSS scale, a composite psychological and functional scale divided into 7 positive, 7 negative, and 16 general items, all scored from 1 to 7. We investigated a new approach based on item response theory (IRT) to simultaneously analyze the scores of all individual items, instead of analyzing the single composite score [1].
Methods: 102,481 records of item-level data were available from 3 Phase 3 studies. Data of each item were modeled as ordered categorical data. The parameters describing the probability curves of each score of each item were estimated as fixed effect. The underlying disease state of individuals at baseline was modeled as a random effect with a fixed variance of 1. Baseline data were used to establish a reference for the probability curves, which were fixed when modeling the longitudinal placebo data. Linear, power, asymptotic and Weibull functions were tested to describe the changes in the disease state as a function of time.
Results: At baseline, the correlations between the disease states on the positive, negative
and general subscales were low; -0.119 for pos-neg, 0.368 for pos-gen, and 0.125
for neg-gen. On all three subscales there was a relatively fast initial
improvement rate in disease state for placebo treated patients that slowed down
later. This was best described by asymptotic functions, with half-lives of
14.4, 15.4, and 11.1 days for the positive, negative and general subscale and disease
states asymptoting to mean variance values of 0.839, 0.419, and 1.49
respectively. This time-course means that at the end of the study (42 days), 64%
of the patients on placebo treatment had a disease state on the positive
subscale that was better than the disease state of the typical individual at
baseline. For the negative and general subscales this was 59% and 71%
respectively.
Conclusions: IRT modeling allows for the analysis of PANSS scores on both the individual item level as well as on the composite scores. The low correlations between the disease states estimated for each individual at baseline on the positive, negative and general subscale suggest that schizophrenia influences the three subscales differently for individual patients. The time-course of the placebo effect does however appear to be rather similar on all three subscales.
Acknowledgement: This work was supported by the DDMoRe (http://www.ddmore.eu/) project.
References:
[1] PAGE 21 (2012) Abstr 2318 [www.page-meeting.org/?abstract=2318].
