Item response theory modelling of motor scores to investigate feasibility of reducing proof-of-concept trial for Parkinson’s disease
Yucheng Sheng, Shuying Yang, Peiming Ma, Chao Chen
Clinical Pharmacology Modeling and Simulation, GSK
Objectives: Parkinson’s disease (PD) is a progressive condition. The total score of Part III (Motor Examination), a composite of 33 categorical items of MDS-UPDRS (Movement Disorder Society United Parkinson’s Disease Rating Scale), is a commonly used efficacy endpoint in clinical trials of anti-PD drugs. It is conceivable that among the 33 items some are more informative than others. In this work, we seek the most informative items in MDS-UPDRS Part III, using Item Response Theory (IRT) modeling analysis, and compare the power for a proof-of-concept study between IRT and conventional methods by simulation.
Methods: We developed a longitudinal IRT model to describe motor examinations of MDS-UPDRS using the Parkinson’s Progression Markers Initiative (PPMI) data set which contains 5 years observations. All motor items from MDS-UPDRS were linked to one latent variable representing "Severity" via probability functions. The score change over time for each item then reflected the longitudinal change in this variable. We assessed the relative informativeness of individual items. Following IRT model validation, trial simulation was conducted to assess PD progression, when using the sum of all 33 item scores or of the most informative ones. Item scores were back-transformed from the IRT model with different, presumed treatment effects. Probability of success (or Assurance) was estimated from drug effects which that were hypothesized to follow a uniform distribution of between 0.1 and 0.5. The ability to detect treatment effects was compared for the sum of item scores and for the “Severity” variable.
Results: Two-parameter logit item response model was used to link the probability of each item score to the latent "Severity". "Severity" change over time was described by a linear model, and "Severity" at baseline was the only influential covariate on the slope. Inter-occasion variability was also added to "Severity". Longitudinal “Severity” change reflect well the item score change over time. Visual predictive check for total scores, which were summed up from back-transformed item score, also suggested the final IRT longitudinal model sufficiently captured the density and changes of total scores. Seven items from left body side – hand movement, finger tapping, pronation-supination, toe tapping, leg agility and rigidity of lower and upper extremity – were identified as the most informative items from all 33 items. The IRT analysis demonstrated a higher power to detect drug effects than the total-score-based analysis, either with all items or with the selected items. Assurance analysis suggested that 300 subjects per arm in a 2-year trial reached more than 70% probability of success from the IRT method with the selected items, whereas that from the conventional method of total score was less than 65%.
Conclusions: The IRT modelling analysis provides higher power to detect drug effects and results in smaller sample size for a proof-of-concept study. For the IRT method, the power was greater when all items were included, while for the conventional total score method the power was greater when only the most informative items were included. Simulations from this IRT model showed the “probability of success” to evaluate design options and promised for more efficient proof-of-concept studies in PD patients.
References:
[1] Goetz CG, et al. Movement Disorder Society-sponsored revision of the Unified Parkinson’s Disease Rating Scale (MDS-UPDRS): Process, format, and clinimetric testing plan. Mov Disord. 2007;22(1):41-47
[2] Ueckert S, et al. Improved utilization of ADAS-cog assessment data through item response theory based pharmacometric modeling. Pharm Res. 2014;31(8):2152-2165.
[3] Novakovic AM, et al. Application of Item Response Theory to Modeling of Expanded Disability Status Scale in Multiple Sclerosis. AAPS J. 2017;19(1):172-179
[4] Lei P-W, et al. Effects of Vertical Scaling Methods on Linear Growth Estimation. Appl Psychol Meas. 2012;36(1):21-39.
[5] Vu TC, et al. Progression of motor and nonmotor features of Parkinson’s disease and their response to treatment. Br J Clin Pharmacol. 2012;74(2):267-283.
[6] O’Hagan A, et al. Assurance in clinical trial design. Pharm Stat. 2005;4(3):187-201.