Last Observation Carried Forward: A Crystal Ball?

In spite of widespread criticism and notwithstanding the availability of flexible, broadly valid, and conveniently implemented methods like direct-likelihood and direct-Bayesian analyses, last observation carried forward (LOCF), baseline observation carried forward (BOCF), and complete case analysis...

Full description

Saved in:
Bibliographic Details
Published in:Journal of biopharmaceutical statistics 2009-09, Vol.19 (5), p.872-888
Main Authors: Kenward, Michael G., Molenberghs, Geert
Format: Article
Language:English
Subjects:
Baseline observation carried forward
Bayes Theorem
BOCF
Clinical Trials as Topic - statistics & numerical data
Complete case analysis
Data Interpretation, Statistical
Direct likelihood
Fluvoxamine - therapeutic use
Humans
Likelihood Functions
LOCF
Missing at random
Missing not at random
Models, Statistical
Multicenter Studies as Topic - statistics & numerical data
Patient Dropouts - statistics & numerical data
Pattern-mixture model
Reproducibility of Results
Selection model
Serotonin Uptake Inhibitors - therapeutic use
Treatment Outcome
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:In spite of widespread criticism and notwithstanding the availability of flexible, broadly valid, and conveniently implemented methods like direct-likelihood and direct-Bayesian analyses, last observation carried forward (LOCF), baseline observation carried forward (BOCF), and complete case analysis remain popular for handling incomplete clinical trial data. To the well-known criticisms toward LOCF and BOCF, namely, that they can be either conservative or liberal depending on the context, that they can violate the intention-to-treat principle, and so on, we add that, with the exception of certain degenerate and unrealistic special cases, these methods are incompatible with the sensible requirement that the mechanism governing dropout be independent of future, unobserved measurements, given covariates, past observed measurements, and the current possibly missing measurement.
ISSN:1054-3406
1520-5711
DOI:10.1080/10543400903105406