Approaches to Likert-type Ordered Categorical Data Analysis
(English)Manuscript (preprint) (Other academic)
Purpose: Ordinal variables are common pharmacodynamic responses, but when the number of categories reaches 11, current approaches to fit this type of data may not be suitable. This article aims at exploring modeling approach candidates for 11-point ordered categorical data.
Methods: The study consisted of two sets of 100 stochastic simulations and estimations in NONMEM VI. A first set was generated with a baseline response ordered categorical model, whose parameter values were derived from a real case study with 100 observations on average per individual. A second set included a linear drug effect observed in a 4 parallel dose arm treatment trial. Simulated data were analyzed with an ordered categorical (OC), a truncated generalized Poisson (PO), and a logit-transformed continuous (CO) models.
Results: The dose-response OC model needed 13 parameters, compared to 7 and 6 with the other models. Score distributions were closely mimicked by OC resimulations at all dose levels. The agreement with PO resimulated scores was best at low doses, in contrast to CO showing less discrepancy at high doses.
Conclusions: Truncated count (PO) or logit-transformed continuous (CO) models may be alternatives to ordered categorical models for Likert scores in case of sparse data or in presence of serial correlations.
Ordered categorical, count, Poisson, continuous, Likert, scale, score, NONMEM, power
IdentifiersURN: urn:nbn:se:uu:diva-150922OAI: oai:DiVA.org:uu-150922DiVA: diva2:409327