Information overload remains a significant barrier for patients searching for clinical trials online. We present a framework to dynamically refine clinical trial search results towards a manually reviewable number of studies. At this aim, a set of frequent features are extracted automatically offline from the eligibility criteria of a representative subset of clinical trials. These features are then presented dynamically to the users based on feature co-occurrences and user interactions to reduce the number of resulting documents. Experiments on ClinicalTrials.gov showed that using an average of four features, it is feasible to reduce the results of a simple search from over a thousand to ten.