Evaluation of a system to identify relevant patient information and its impact on clinical information retrieval

Concept-oriented views of electronic medical records are desirable, yet difficult to create. We have developed a system that creates concept-oriented views by identifying relevant patient information, however, previous such systems have received little evaluation. We present here an evaluation of our system's ability to identify relevant patient data and generate concept-oriented views, along with the clinical impact of the generated views. The evaluation was carried out in three parts: First, using physicians and medical literature as gold standards, the system's sensitivity and specificity in identifying relevant information were measured. In some areas, the system demonstrated sensitivity comparable to that of physicians. Second, concept-oriented views were compared with original records and shown to contain significantly less non-specific information. Third, physician volunteers, when answering questions about patient cases using the concept-oriented views and traditional source-oriented views generated by the system, showed a significantly greater accuracy in information retrieval using concept-oriented views.