A simple method to identify real-world clinical decision intervals of laboratory tests from clinical data

• Main Authors: Yuki Hyohdoh, Yutaka Hatakeyama, Yoshiyasu Okuhara
• Format: Article
• Language: English
• Published: Elsevier 2021-01-01
• Series: Informatics in Medicine Unlocked
• Subjects: Reference intervals; Laboratory tests; Ordering patterns; Unsupervised learning; Medical database
• Online Access: http://www.sciencedirect.com/science/article/pii/S2352914821000022

Background: Reference intervals (RIs) and clinical decision limits (CDLs) are important in evaluating laboratory test results. However, physicians not only the simple RIs and CDLs in clinical decision-making, but also flexibly change criteria according to the background characteristics of patients. Objectives: The purpose of this study was to propose a simple method to determine intervals of laboratory results that physicians consider to be acceptable, i.e., the real-world clinical decision intervals (RCDIs), from clinical data. Methods: Blood test data (35 items) from outpatients aged 20 years or older at Kochi Medical School Hospital were included in the study. Data were resampled using the test interval until the next laboratory test order as a weight, and a set of clinically acceptable laboratory test values was identified. Further, the relationship between iron prescription and laboratory test values was evaluated to verify the validity of estimated RCDIs for hemoglobin. Results: Many of the estimated RCDIs were similar to RIs and intervals calculated from clinical reports. Further, estimated age-specific RCDIs of hemoglobin and the trend for iron prescription behavior were consistent, and the predictive model with RCDIs as a threshold identified iron prescriptions better. Conclusion: The present approach visualizes the evidence of physicians’ decisions based on laboratory test orders and shows the possibility that it can be used for phenotyping in specific diseases using electronic medical records, with re-evaluation of conventional RIs and CDLs.