Assessing Treatment Effect Heterogeneity in the Citalopram for Agitation in Alzheimer’s Disease Clinical Trial: A Subgroup Analysis to Guide Personalized Treatment Selections.

subgroup analyses;clincal trials;biostatistics;treatment effect heterogeneity;Alzheimer"s Disease;Public Health Studies

Rein, Lisa EgnerFrangakis, Constantine E. ;

Johns Hopkins University

关键词: subgroup analyses;    clincal trials;    biostatistics;    treatment effect heterogeneity;    Alzheimer";    s Disease;    Public Health Studies;   

Others  :  https://jscholarship.library.jhu.edu/bitstream/handle/1774.2/37263/REIN-THESIS-2014.pdf?sequence=1&isAllowed=y

瑞士|英语

来源: JOHNS HOPKINS DSpace Repository

【 摘 要 】

Alzheimer;;s disease (AD) is a devastating neurodegenerative condition with symptoms of cognitive decline, behavioral disturbances, and ultimately mortality. As there is currently no cure, improvements for management of AD symptoms are desperately needed. The Citalopram for Agitation in Alzheimer;;s Disease (CitAD) study examined off-label use of citalopram, a selective serotonin reuptake inhibitor, for management of agitation symptoms. The primary analysis showed a greater average decrease in agitation symptoms and an increase in a potentially serious adverse event with citalopram compared to placebo. Physicians want to know if the treatment effect is heterogeneous, and if so, which patients have the greatest potential to benefit; given the risks, it may be unethical to prescribe the drug to patients with little chance of benefit. Subgroup analyses are employed to assess heterogeneity of effect across subgroups defined by categorical baseline covariates. This is typically done by calculating subgroup treatment effects in a stratified dataset and testing for the interaction between treatment and the baseline covariate. This approach is not very comprehensive, as it only examines one covariate at a time while patients hold multiple characteristics simultaneously. Another limitation of this approach is the use of parametric models which carry the assumption of correct mean model specification. For our subgroup analysis, we employed the two-stage estimation method developed by Cai et al. In the first stage, parametric working models are used to calculate the approximate treatment effect, the difference in potential outcomes under citalopram versus placebo, for each participant based on multiple baseline covariates. This predicted treatment effect is called the index score; patients with the same combination of baseline covariates have the same index score and are considered a subgroup. In the second stage, patients are grouped by index score allowing non-parametric estimation of subgroup treatment effects using observed data. Using this approach, we found evidence for treatment effect heterogeneity. CitAD participants with the largest predicted treatment effects were more likely to be living outside long-term care facilities, within the middle age range (ages 76-82), with minimal cognitive impairment (MMSE 21-30), within the middle baseline agitation range (NBRS-A 6-8), and not taking lorazepam.