Accurate characterization of which patients bene?t highly from a treatment or program in Alzheimer's or HIV dis- eases are central for knowing which treatments work for which patients, and to plan effectively for the others. A major challenge for this is heterogeneity of these diseases. Until now, treatment studies for Alzheimer's disease with comorbidities have shown little if any ef?cacy. Also, for HIV/AIDS patients in resource -constrained settings, only a small fraction use antiretroviral treatment (ART) or bene?t from a given program to increase ART uptake. Standard methods to characterize which patients bene?t from such treatments/programs, ?rst construct a predic- tor using standard statistical criteria, and then use that predictor to characterize high-bene?t patients. For such methods, therefore, the clinical goal ? to characterize high-bene?t patients ? is considered only at the implemen- tation stage, and is not used for the construction of the method. In earlier work, we have shown that such methods can dramatically misrepresent high-bene?t patients; and we have developed a type of method that directly links the clinical goal (high bene?t) into the construction of the characterization mechanism. We were motivated by: a study to reduce agitation in patients with Alzheimer's disease; and a study to increase ART uptake among HIV patients in Vietnam. We have shown that methods that lack this clinical link can miss and underestimate high bene?t patients by a factor of 2 or more, compared to even simple methods of this new type. In this proposal, we will develop and apply such new clinically-targeted statistical methods for characterizing high-bene?t patients. Such methods will allow physicians and patients to make better choices of best treatments and programs, with potential to bene?t millions of patients. The proposed methods are developed for three aims, and, following the preliminary work, are motivated by and will be applied to Alzheimer's and HIV studies.
Aim 1. Develop methods to characterize patients who highly bene?t from treatment in randomized con- trolled trials. These methods are signi?cant because they can identify high bene?t patients who would be missed when using standard methods.
Aim 2. Develop methods to characterize patients with high bene?t and high risk in randomized trials. Here, we will develop methods to characterize, patients with high bene?t, among those with high risk of an adverse event. These methods can allow patients to better balance risk and bene?t of treatments.
Aim 3. Develop methods to characterize patients who highly bene?t from treatment in observational studies. We will use methods to transform observational studies to a study as close as possible to a randomized one, where we can then extend the methods of aims 1 and 2. These methods are signi?cant where randomized trials are not easy to conduct. They will be tested using the above two studies, and also at a PEPFAR (President's Emergency Program for AIDS Relief) site, to characterize patients who receive low bene?t from the program, in order to provide to them extra support.

Public Health Relevance

. Accurate characterization of which patients benefit highly from treatments for Alzheimer's disease or interventions for HIV programs such as the US-sponsored PEPFAR, is critical for administering effective interventions to the appropriate patients, but is threatened when the clinical goal is not used in the construction of the statistical methods. Evidence shows that dramatically inaccurate characterizations by existing methods can be corrected if new methods use the clinical goal in the core of their construction. This project proposes to develop such new methods to accurately characterize high benefit patients from experimental and nonexperimental studies.

National Institute of Health (NIH)
National Institute of Allergy and Infectious Diseases (NIAID)
Research Project (R01)
Project #
Application #
Study Section
Special Emphasis Panel (ZRG1)
Program Officer
Gezmu, Misrak
Project Start
Project End
Budget Start
Budget End
Support Year
Fiscal Year
Total Cost
Indirect Cost
Johns Hopkins University
Biostatistics & Other Math Sci
Schools of Public Health
United States
Zip Code