Tuesday, September 1, 2009
Comparative Effectiveness Research and Intelligent Search
Comparative effectiveness research (CER) and evidence-based medicine have been one of the boogeymen of the healthcare debates. Fear that, as a result of CER findings, your doctor will not be able to make his or her own decisions about what's best for you but will be forced through some chart of acceptable protocols. None of the physicians or researchers I know are that wimpy.
We don't know what the outcome of the healthcare debate will be but we subscribe to the belief that knowledge is power. The more you know, the better your decisions will be. CER has the potential to deliver for healthcare what nutritional labels did for groceries - information.
To that end, Coronado has been working on a novel intelligent search tool for answering clinical questions. At a high level this is what it does: When faced with a complex patient problem or when you are looking for information to address a research issue, you construct a single query by combining all of the text based information you have about the problem and using that as the search source. You copy and paste all the text-based information you have - clinical reports, image interpretation text, previous findings - and copy it into the search box. When you've entered all the text that you have you can add your own notes, words you are looking for, hunches, terminology that relates to the subject matter in your domain. The search box can handle gigabytes of data. Once that's done, you hit enter and the intelligent search tool evaluates all of the concepts in the search source and then finds and ranks results based on how conceptually and semantically close the documents are to your search source.
Our initial implementation uses the content of 13 of the top diagnostic imaging journals and RadLex, the radiology and diagnostic imaging ontology, to create the concept space and the searchable document set. We started here because two of the top radiology informatics experts from the University of Maryland School of Medicine have been gracious enough to provide us with guidance so we can make sure it works. It also offers us a platform where we can address some pressing search and retrieval issues - convergence. In physician speak this area is dealing with multidisciplinary/multimodality diagnosis, treatment, and monitoring. In regular guy speak what this means is that to answer a question it's not unusual for s clinician or researcher to need to do cross-domain searches covering radiology, chemist, molecular biology, pharmacology, chemotherapeutics and contract agents to get a comprehensive answer to a complex problem. It would be nice to be able to execute a search that gets it rather than having to keep searching and searching and searching.
We have been using this technology successfully in the patent and intellectual property space. Our work on the patent front resulted in our having a deep understanding of the nature and content of scientific, technical and medical documents and understanding how to authors writes these kind of documents, where the good information is buried. As a result of this work our concept space for scientific materials is excellent allowing IP searches that dig deep into the content of the patents and supporting documents. Since the patent space is so wide and the classifications so deep, we have refined the complex subject matter there with some of our own special sauce to help expedite complex searchers. We expect that we can leverage this framework to support a highly focused implementation for medical research. Especially with help from our colleagues.
This tool offers a lot of value in the comparative effectiveness arena. First, you can pose a complex search that will return the documents that are the closest to your topic. This lends itself to helping medical librarians and investigation collaborators on evidence-based medical evaluations assemble the information they need to do their work. At the risk of being corny, less time finding, more time working.
It also supports deep dive information harvesting. While a certain protocol or finding may be a very small part of one investigation it may be extremely critical to current searchers. Conventional search techniques, especially those on the internet, make finding these documents hard to do. Because the intelligent search tool searches on the conceptual and semantic content, it finds all the documents that embody a particular idea even if it's not the most important concept. It makes a whole new universe of information accessible to searchers. You can search the widest possible universe of documents and still get highly refined results.
One of the other areas we are exploring is the ability to do longitudinal searches. Or, as consultants like to call them, retrospective longitudinal searches. We do this in the patent world now. Basically what this means is that you can look back in time to see what evidence is available. Because the technology understands how the words used to describe something have changed over time as well as their context, this may enable searchers looking for evidence to look back in time. This has the potential to accelerate some of the work to establish meaningful comparative effectiveness work without having to wait for new evidence. It has its limitations, it's only as good as the old documents, but it might prove helpful as investigators look across older medical evidence or access deep repositories of reports. We're looking at all that "stuff" stored in the archival PACS systems.
And a final note. It's easier to search. You don't have to distill your search requirements into ands, ors, or nots. You simply write up what you are looking for, stream of consciousness works, or assemble the best scientific and medical documents that describe what it, or assemble a text document where you keep your research notes and use that for the search. It's more natural. It lends itself to serendipitous search - navigating through the information to see what's there, to see what you find. It lets you search the way you think...really.
We published a White Paper called Intelligent Search for Answering Clinical Questions. You can download it from our website. The image if our version of finding the needle in the haystack..just find all the needles.