Institute for Medical Informatics, Statistics and Documentation, Medical University of Graz

Multiple comparisons for non-Gaussian distributed endpoints - using R

Speaker: Prof. Dr. Ludwig A. Hothorn,
Institute of Biostatistics, Leibniz University Hannover, Germany
Date, Time: October 29th, 2013, 3:30 p.m.
Location: IMI Meeting Room (S-05-170), Auenbruggerplatz 2/5, 8036 Graz
Presented By: Institute for Medical Informatics, Statistics and Documentation together with BSSK
Abstract: A discrepancy between the MCP-methods assuming Gaussian distribution and homogeneous variances in the literature (& common software), and the practically occurring different types of endpoints in RCT and toxicology exists, namely: i) proportions (e.g. tumor rates), ii) skewed distributed endpoints (e.g. the ASAT enzyme), iii) survival functions, iv) mortality-adjusted tumor rates (poly-3 estimates without cause-of-death information), v) counts with between-subject-variability (overdispersion) (e.g. number of micronuclei), vi) ordered categorical data (e.g. graded histopathological findings).

Based on the asymptotic approach in general parametric models (Hothorn et al. 2008) and the R packages multcomp, mratios, MCPAN and SimpComp, by means of case studies the estimation of related simultaneous endpoints for different contrast matrices are demonstrated, such as Dunnett-type, Williams-type and Grand-Mean-type. Moreover, the usefulness of a non-parametric version for relative effects (Konietschke et al. 2012) is demonstrated using the R package nparcomp and ratio-to-control tests are explained using the R package, particularly in the case of variance heterogeneity. Finally, user-defined contrast tests, controlling a claim-wise error rate (instead of a family-wise), will be discussed.

References:
  • Hothorn, T; Bretz, F.; Westfall, P: Simultaneous inference in general parametric models.
    Biometrical Journal, 2008; 50(3): 346-363.
  • Konietschke, F; Hothorn, L.A.; Brunner, E: Rank-based multiple test procedures and simultaneous confidence intervals.
    Electronic Journal of Statistics, 2012; 6: 737-758.
 
 Deutsch  :  Contact  :  Imprint  :  Search   » MedUni Graz 
Webmaster, 2013-10-04