Premium
PII‐61
Author(s) -
Dunne A.
Publication year - 2006
Publication title -
clinical pharmacology and therapeutics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.941
H-Index - 188
eISSN - 1532-6535
pISSN - 0009-9236
DOI - 10.1016/j.clpt.2005.12.186
Subject(s) - goodness of fit , categorical variable , statistics , likelihood function , likelihood ratio test , medicine , mathematics , score test , placebo , maximum likelihood , alternative medicine , pathology
BACKGROUND Modelling a categorical response involves the use of a link function which links the response and the independent variables. This project examines a profile likelihood based method for choosing the most appropriate link function and compares it with the standard goodness‐of‐link test. METHODS The data are a subset of those collected during a randomised, double‐blind, placebo controlled, forced dose escalating, clinical trial of oral oxybutynin tablets for the treatment of urinary incontinence and kindly supplied by Alza Corp. At the end of each week patients recorded the severity of dry mouth during the week. The severity of dry mouth on an ordinal scale was used to represent the adverse effect. The present study used the adverse effect data from 32 patients. RESULTS The goodness‐of‐link test underestimated the value of the parameter in the link function and had a larger standard error than the likelihood based estimate. The profile likelihood gives the maximum likelihood estimate without any approximation. CONCLUSIONS The profile likelihood approach to deciding on the link function is superior to the goodness‐of‐link test because it provides the actual maximum likelihood estimate of the link function parameter. In addition it gives a visual indication of how sensitive the likelihood is to changes in this parameter. Clinical Pharmacology & Therapeutics (2005) 79 , P52–P52; doi: 10.1016/j.clpt.2005.12.186