ABSTRACT
Clinical trials are often conducted under idealized and rigorously controlled conditions to ensure internal
validity, but such conditions, paradoxically, compromise trials external validity (i.e., generalizability to the target
population). Low trial generalizability has long been a concern and widely documented across different clinical
areas. For instance, participants of Alzheimer's disease (AD) clinical trials are systematically younger than AD
patients in the general population. Overly restrictive eligibility criteria are arguably the biggest yet modifiable
barriers causing low generalizability. The FDA has launched numerous initiatives, primarily through
broadening eligibility criteria, to promote enrollment practices so that trial participants can better reflect the
population who would most likely use the treatment if approved. Nevertheless, trial sponsors and investigators
are reluctant to broaden eligibility criteria due to concerns over potential increases in risk of serious adverse
events (SAEs) and its negative impact on the investigational drug’s safety and effectiveness profile. As a
result, many elderly patients are excluded from AD trials either explicitly through an age restriction or implicitly
through excluding clinical characteristics more prevalent in the elderly. There is a gap between the need to
broaden trial criteria and ways available to fulfill the need in practice. Previous studies, including ours, have
validated and used the Generalizability Index of Study Traits (GIST), the best available quantitative, eligibility-
driven, a priori generalizability measure, in a number of disease domains. GIST scores can potentially be used
to guide adjustments to criteria towards better population representativeness. However, there are key barriers
for its adoption in practice, especially in AD trials: (1) the lack of a standardized, computable eligibility criteria
(CEC) framework to translate criteria to data queries – a necessary step to define the populations for
generalizability assessment, and (2) the need to map the mathematical relationships between eligibility criteria
and GIST as well as patient outcomes (i.e. SAE), which answers the critical question how broadened criteria
will affect AD trial’s generalizability and patient outcomes simultaneously. To remove these barriers, we
propose to systematically analyze existing AD trials in ClinicalTrials.gov to create a standardized library of CEC
for AD trials and develop statistical models on how adjustments to eligibility criteria, especially age, would
affect (1) trial generalizability measured by GIST, and (2) outcomes (i.e., SAEs) of the target population,
approximated using real-world data (RWD) from the OneFlorida network. OneFlorida contains linked
electronic health record (EHR), claims, and cancer registries data for ~15 million Floridians. This study will
provide the necessary data to support future development of a trial eligibility criteria design tool that can
optimize trial generalizability while balancing potential increases in risk of SAEs in the target population.