Premium
P4‐372: Comparison of a slopes analysis using a mixed model to a traditional change from baseline analysis in an Alzheimer's disease clinical trial setting
Author(s) -
Horton Scott,
Hendrix Suzanne,
Johnson Troy,
Zavitz Kenton,
Green Robert C.,
Schneider Lon S.,
Swabb Edward
Publication year - 2008
Publication title -
alzheimer's and dementia
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 6.713
H-Index - 118
eISSN - 1552-5279
pISSN - 1552-5260
DOI - 10.1016/j.jalz.2008.05.2442
Subject(s) - missing data , imputation (statistics) , statistics , power analysis , mixed model , pooled analysis , random effects model , meta analysis , medicine , mathematics , econometrics , algorithm , confidence interval , cryptography
Background: Until recently, clinical trials in Alzheimer’s disease have designated change from baseline with LOCF at the end of the study as the primary analysis (traditional analysis). A phase 2 trial of tarenflurbil started in 2003 used a slopes analysis based on a mixed effects model with no data imputation as the primary analysis (slopes analysis). This slopes analysis was designated as the primary analysis for the US phase 3 trial of tarenflurbil. Methods: The phase 3 study of tarenflurbil 800 mg bid is an 18 month study with the ADAS-cog and the ADCS-ADL designated as co-primary efficacy outcomes, and the CDR-sb as a secondary outcome. The primary analysis method uses a slopes analysis. The power of this analysis is compared to a traditional analysis. The estimates obtained from the two models are compared for bias under different assumptions about the pattern of missing data. Results: The slopes analysis provides estimates of rate of change rather than clinical outcome at a single time point, and includes data from all time points in the analysis without the need for imputation using LOCF or ignoring missing data using OC. This method produces unbiased estimates when data is missing at random which has an advantage over an LOCF approach that produces biased estimates in a degenerative disease even when data is missing completely at random. The power of the slopes analysis is greater, requiring about half the number of subjects of a traditional analysis. Conclusions: A slopes analysis is generally preferable to a traditional analysis because it has larger power and less bias in most situations. Reporting outcomes in terms of slopes (rates of decline) may also be more appropriate for emphasizing long-term outcomes in degenerative diseases.