2016 Volume 36 Issue 3 Pages 113-122
Physicians decide to order laboratory tests on the basis of differential diagnoses expected from the patients’ symptoms. This requires up-to-date clinical knowledge, but it is difficult for physicians to be aware of the latest advances in all medical fields. Previous studies have proposed systems to determine differential diagnosis by using information regarding the patients’ symptoms, but some problems remain unsolved: the systems do not recommend the exact tests required; physicians require time to enter the query into the system; and manual development and maintenance of the knowledge base are time-consuming and expensive. The purpose of this study is to develop a system for automatically suggesting an individual order on the basis of the previous order; this system is expected to solve the problems faced when using previous systems. We propose two steps of methods to extract pairs of laboratory test items: First, we picked up the diagnoses associated with the test automatically by association rule mining. Second, we created test pairs for the same diagnosis (type 1) or disease group (type 2). To identify the disease group, we used network analysis. Using these methods, we acquired the test pairs that should be ordered on the same day. The test pairs were checked by a medical doctor; the accuracy of the test pairs was found to be 70.0% for type 1 and 89.0% for type 2. The results indicate that our system should be useful for recommending laboratory test items in addition to providing diagnosis or differential diagnosis.
