A perioperative medicine clinical decision support system: foundation, design, development, evaluation, and the standards

Abstract

Abstract 1:Introduction and Background: Decisions regarding medication management during the perioperative period are often made based on clinical anecdotes and vary by provider. Clinical decision support (CDS) tools aid physicians with decision making tasks at the point of care. We have developed a set of perioperative medication management recommendation decision heuristics based on evidence-base literature, clinical notes, and expert opinions. These heuristics will serve as the foundation for a subsequent CDS tool in perioperative medicine. Methods: In this descriptive study, we manually extracted key demographic and medication-related data from the records of 100 randomly-selected patients at the Minneapolis VA's preoperative medicine clinic. We then searched PubMed for studies in perioperative medication management as well as other web sources for expert opinions in the field. Relevant studies, clinical notes, and expert opinions were distilled into an XML-based set of heuristics "rules" file.Results: We have developed medication management recommendation heuristics for the entire VA's formulary of drugs based on evidence-base literature, actual clinical notes, and expert opinions. Discussion and Conclusion: This work shows a proof of concept for the full-scale system development of similar decision support systems---------------------------------------------------------------------------Abstract 2:Background: A clinical decision support tool to manage medications can help perioperative medicine clinicians avoid spending valuable time looking for drug management information during a pre-op physical exam evaluation. Our objective was to develop and validate a clinical decision support (CDS) tool for managing medications perioperatively.Methods: We developed a CDS tool based on a set of heuristics classifiers developed in a previous study, and tested the tool using medication data extracted from the electronic records of 100 randomly selected perioperative medicine patients including medications in use. For each medication, the tool-generated recommendation was compared with actual recommendations in the EHR by experienced preoperative medicine providers.Results: A total of 879 medications were used by the sample population. We extracted 378 "actionable" drugs from the EHR Notes section, compared to 479 identified by the tool, while 334 were identified in both. The total number of "non-actionable" drugs in the EHR notes was 132 compared to 18 flagged by the tool, while 369 were identified by both. In the initial testing phase the tool generated provider-matched recommendations 76% of the time. After correcting for errors and adjudicating the differences by a perioperative medicine domain expert, the tool's matching performance increased to 95%. These results are encouraging.Conclusion: The CDS tool compared favorably with other similar tools and thus can be used as a support tool at the point of care.--------------------Abstract3:Background: Perioperative medicine data is mostly in non-standard, unstructured free text, making the measurement and assessment of clinical outcomes challenging using electronic medical record data. Perioperative medication management, managing the patient's medications during the perioperative period, is a complex clinical problem. SNOMED CT (Systematized Nomenclature of Medicine Clinical Terms) is a comprehensive clinical terminology which provides a consistent way to index, store, retrieve, and aggregate clinical data. Our objective was to validate the use of Medication Therapy Management (MTM) concepts within SNOMED to express perioperative medication management recommendations. Methods: Perioperative medication management recommendations of 100 randomly selected patients were extracted from their electronic medical records. Keyword searches of MTM concepts were performed on the March 2013 of SNOMED CT and candidate concepts were manually extracted and verified for relevance by domain experts. Two domain experts rated the cross-mappings as a "match" or "non-match". Results: A total of 11 unique recommendations were aggregated from the sample population. A search of SNOMED CT yielded 47 concepts. The inter-rater agreement statistic between the two experts was 0.77 (substantial). Conclusion: MTM concepts in SNOMED CT can be used reliably to code perioperative medication management recommendations with sufficient clarity.