Greetings all, We've had some good news recently - our NIH proposal to apply measures of semantic relatedness to the problem of monitoring adverse drug interactions in clinical and pharmaceutical records has been funded. It has been a long and winding road with this proposal, so it's very satisfying to see it end so well. And of course it's not really an ending at all, simply the start of a new and hopefully very long running project. The initial grant period is for three years, but we are optimistic it will go longer (assuming good results, which we are confident we can attain).
This project is being carried out jointly with Serguei Pakhomov and Brian Isetts, both of the College of Pharmacy in the Twin Cities. Below is a short introduction to the project. BTW, we are going to be seeking RAs to work on this project here in Duluth starting Fall 2009, so if you have any friends, relatives, colleagues, etc. who might be interested in coming to Duluth for an MS in Computer Science and to work on a project like this, please feel free to forward them this note and have them contact me at [EMAIL PROTECTED] for additional details...and of course more on the MS in CS in Duluth can be found here ... http://www.d.umn.edu/cs/degr/grad/ Thanks! Ted ======================================= Semantic Relatedness for Active Medication Safety and Outcomes Surveillance Accurate and reliable methods to analyze the ever-increasing volume of medical information are critical to improving patient safety, a major priority in health care in the United States. The NIH Roadmap initiative emphasizes the role of information technology in improving the assessment of clinical outcomes. Post-marketing passive surveillance of outcomes associated with medication use has been recognized as a necessary component in drug safety monitoring to overcome the limitations of pre-marketing clinical trials. A potential adverse drug reaction is identified based on statistical models that rely on counting how many times a group of related outcomes as a whole occurs in a population. Currently, adverse drug reactions are monitored using a mix of voluntary and mandatory reporting mechanisms all of which have limitations including underreporting and poor quality of documentation. The electronic medical record (EMR) and the electronic therapeutic record (ETR) hold promise to overcome these limitations; however, they contain much of the needed information in unstructured text. While the EMR documents patients' overall health states including symptoms, diagnoses, procedures and the plan of treatment. The ETR is created by a pharmaceutical care practitioner specifically to document exposure to medications as well as health problems associated with medication use. Whether written or dictated, text has long been the principal method of recording medical history and physical exams. Even the most modern EMRs and ETRs recognize text fields as essential for providing the clinician flexibility in documenting new or detailed observations. Thus, the analysis of this information for drug safety monitoring requires sophisticated Natural Language Processing (NLP) techniques including methods to discover meaningful groups of related but not necessarily synonymous terms. Creating such groups of related terms is currently a manual process that lacks the flexibility and efficiency required for active drug safety monitoring from large volumes of unstructured text of EMRs and ETRs. Thus, there is a critical need to develop technology capable of automated analysis of large volumes of clinical text to discover groups of related terms as well as meaningful descriptions of the relations within and among the groups. Methods for computing semantic relatedness is a class of computational techniques that can be used to create groups of related terms automatically by using information from large corpora of text and existing ontologies. While these approaches have proven effective in addressing a number of NLP problems, their utility in drug safety monitoring has not yet been demonstrated. Our long term goal is to improve patient safety and treatment effectiveness by providing reliable methods for complete and timely identification of groups of related terms that may be indicative of adverse drug reactions. ======================================= -- Ted Pedersen http://www.d.umn.edu/~tpederse
