Stratified testing for treatment effects with missing data.

W J Shih; H Quan

Home/Archive/PubMed

PubMed · 9660633

Stratified testing for treatment effects with missing data.

W J Shih·H Quan

View original source ↗Export metadata ↓

Abstract

Dawson and Lagakos (1993, Biometrics 49,1022-1032) proposed a stratified test for repeated measures data that contain missing observations. They recommended stratification based on missing data patterns and considered sufficient conditions under which the size of the test is properly retained. In this paper, we point out some practical problems with these conditions and illustrate them with their CD4 count example as well as a stimulation study. We give a less stringent condition and delineate its merit. We also discuss what to do when none of the conditions are met.

Explore related subjects

Biometry CD4 Lymphocyte Count Clinical Trials as Topic Computer Simulation Humans Least-Squares Analysis Models, Statistical Patient Dropouts Journal Article

Keep this discovery

Explore connections, maps & timelines

BibTeX RIS

W J Shih, H Quan. 1998. Stratified testing for treatment effects with missing data.. https://pubmed.ncbi.nlm.nih.gov/9660633/

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related citations

PubMed2001

Towards a curriculum for the consultant biostatistician: identification of central disciplines.

The variation in the background of biostatistical consultants is huge. This varying background in formal education complicates communication. It may obscure clients' expectations as well as any discussion on qualifications and registration. To tackle these disadvantages as well as to facilitate the training of a new colleague, we need to find common ground. The aim of this paper is to identify the central disciplines which constitute the foundation of biostatistical consultancy. Any task-aimed curriculum should be based on these disciplines. Curriculum design theory was used to guarantee a logical and rational process. The result is the identification of a sextet of disciplines: applied statistics; methodology; epidemiology; communication; computational science, and personal effectiveness.

Biometry↗

PubMed2001

Synthetic estimators with moderating influence: the carry-over in cross-over trials revisited.

A general approach to combining alternative estimators of a parameter is described and applied to the analysis of 2x2 cross-over trials in which the treatment effect is of principal interest and carry-over is a nuisance feature. The approach requires prior information about the carry-over. Although modest information is sufficient, more information is rewarded by greater efficiency. The combination, or synthetic, estimator exhibits a moderating influence: its mean squared error is a flat function of the carry-over effect in the range of its feasible values, and for the largest feasible carry-over effect it is superior to either of the originally considered estimators. An extension of the approach is outlined and two other applications in medical research discussed. Although the method can be regarded as Bayesian in spirit, we focus on its resampling (frequentist) properties.

Biometry↗

PubMed2001

Planning a reproducibility study: how many subjects and how many replicates per subject for an expected width of the 95 per cent confidence interval of the intraclass correlation coefficient.

Reproducibility of a quantitative outcome is usually assessed by means of the intraclass correlation coefficient (ICC). When we are interested in assessing reproducibility from only one sample, we suggest that the study be planned with regards to the expected width of the 95 per cent confidence interval of the ICC. An approximation of this latter width is derived, which allows appraisal of the influence of n the number of subjects and p the number of replicates. Through simulation studies, the approximation is shown to be of good accuracy and can therefore be used reliably. Optimal designs are also discussed such as the optimal distribution between the number of subjects and the number of replicates per subject for a fixed total number of measures.

Biometry↗