Ask Dory – an Intelligent Clinical Trials Decision Engineby Applied Informatics
Ask Dory is an award-winning system that is designed to support the needs of novice, intermediate and advanced users (the gradation based on the knowledge a user has about the clinical trials sought).
Ask Dory supports both recognition and recall based information seeking behavior. Psychological studies have shown that recognition is an easier cognitive task than recall, particularly for novice users who may not have enough domain knowledge to correctly articulate their needs into system queries. For complete novice users, Ask Dory provides an intelligent question and answer based discovery process that uses an innovative decision tree algorithm (an artificial intelligence method that allows data to be partitioned into sets based on the most discriminating variable at each partition iteration).
The algorithm partitions the available clinical trials for a given condition based on several parameters (such as users location, gender, age, treatment status) and dynamically decides what question to ask the patient so that the system can narrow down the list of matching trials as fast as possible. This is a completely new paradigm compared to existing clinical trials search tools that may require users to fill long static questionnaires.
The system is designed to reduce the cognitive burden of patients in trying to identify a suitable trials based on their exclusion/inclusion criteria, which can be rather technical and complicated in the domain of cancer.
The system has a semi-automated process that involves natural language processing and expert reviews to code cancer trials for specific criteria. For example, we have coded all myeloma clinical trials by their treatment status, treatments offered, transplant versus non-transplant and a variety of metrics using this semi-automated process. Ask Dory then uses the decision algorithm to shift through the available trials based on user’s inputs to consumer-friendly questions (such as asking patients about their treatment status or treatment preferences) till the result set is narrowed to 5 or less trials. For expert users it support a traditional advanced search function (providing ability to recall information). It also includes a browse/exploratory mode to supplement the search and the guided approaches.