Item Response Theory Model as Support for Decision-Making: Simulation Example for Inclusion Criteria in Alzheimer's Trial
Sebastian Ueckert, Andrew C. Hooker, Mats O. Karlsson, Elodie L. Plan
Pharmacometrics Research Group, Dept. Pharmaceutical Biosciences, Uppsala University, Sweden
Objectives: Item Response Theory (IRT) has been introduced in pharmacometrics for the characterization of the Alzheimer's Disease (AD) Assessment Scale - cognitive subscale (ADAS-cog) test [1] and used as well in models describing the Expanded Disability Status Score in multiple sclerosis patients [2] and the Positive And Negative Syndrome Scale in schizophrenic subjects [3]. Benefits in terms of increased power of drug effect detection, enhanced simulation properties and quantification of item information content were highlighted. This study aims to demonstrate a practical application of IRT’s distinct capabilities in the context of drug development (DD) decision-making. The fictitious example was an investigation of the DD question: which Mini-Mental-State Examination (MMSE) [4] inclusion range would deliver the highest probability to detect a hypothetical drug effect in a change from baseline (CFB) analysis of ADAS-cog as primary clinical endpoint for a coming AD trial?
Methods: The process followed an automated stochastic simulation and estimation approach repeated 500 times for different patient populations. The model used for simulation consisted of i) a previously published ADAS-cog IRT model [1], ii) an extension [5] linking the MMSE items to a common cognitive disability hidden variable, based on baseline records from the ADNI [6] database, and iii) a disease modifying drug effect of 30%. The simulations generated replicates of an 18-month placebo controlled trial with 600 subjects selected according to their sampled MMSE values. The estimations accounted for the mean CFB at the last visit using a repeated measures marginal means model. Constant progression rate, baseline correlation, and drug effect over cognitive disability were assumed in this study.
Results: The inherent properties of the simulation model captured several characteristics of the trial data, e.g., increasing skewness for lower MMSE ranges and ceiling/flooring effects of the ADAS-cog score. The power to detect the hypothetical drug effect for Alzheimer’s patients having an MMSE score between 5-10, 10-15, 15-20, and 20-25, was 58%, 92%, 91%, and 74%, respectively.
Conclusions: The IRT pharmacometric approach allowed simulation of realistic clinical data and aided in answering the DD question even though a statistical analysis was intended for the fictitious trial. This example highlights the utility of complex IRT models for DD beyond the analysis of data.
References:
[1] Ueckert S, Plan EL, Ito K, Karlsson MO, Corrigan B, Hooker AC. Improved Utilization of ADAS-Cog Assessment Data Through Item Response Theory Based Pharmacometric Modeling. Pharm Res, 2014 March.
[2] Kalezic A, Savic R, Munafo A, Ueckert S, Karlsson MO. Application of Item Response Theory to EDSS Modeling in Multiple Sclerosis. PAGE, 2013.
[3] Krekels EHJ, Friberg LE, Vermeulen AM, Karlsson MO. Item response theory for the analysis of the placebo effect in Phase 3 studies of schizophrenia. PAGE, 2013.
[4] Folstein MF, Folstein SE, McHugh PR. "Mini-mental state" A practical method for grading the cognitive state of patients for the clinician. J Psychiatr Res, 1975 Nov;12(3):189-98.
[5] Ueckert S, Plan EL, Ito K, Karlsson MO, Corrigan B, Hooker AC. Predicting Baseline ADAS-cog Scores from Screening Information using Item Response Theory and Full Random Effect Covariate Modeling. ACoP, 2013.
[6] ADNI (Alzheimer’s Disease Neuroimaging Initiative). http://www.adni-info.org/
Acknowledgement: This work was supported by the DDMoRe (www.ddmore.eu) project.
