|
Activity Number:
|
317
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Tuesday, August 2, 2016 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Biopharmaceutical Section
|
|
Abstract #319864
|
|
|
Title:
|
Adjustment for Categorization in Predictor Variables
|
|
Author(s):
|
Saptarshi Chatterjee* and Sanjib Basu
|
|
Companies:
|
Northern Illinois University and Northern Illinois University
|
|
Keywords:
|
multiplicity ;
cutpoint ;
permutation test ;
minimum p-value
|
|
Abstract:
|
Cutoff detection in prognostic variables is an important area of research in clinical data analysis. Medical practitioners often need to categorize predictor variables in order to interpret the association with the outcome in a more meaningful way. Maximally selected chi-square statistic is often used for categorizing predictor variables. However, it has the disadvantage of possibly in?ating the type I error rate by a signi?cant margin. Several adjustments to correct the p-value of this statistic have been proposed in literature, but remain less used in practice. We propose a permutation based method to overcome the issue of multiple testing that has general applicability. We validate our ?ndings in the setting of both continuous and binary outcomes through extensive simulation studies and show that the proposed method maintains appropriate type-I error while providing substantially improved power than existing comparator methods.
|
Authors who are presenting talks have a * after their name.
