Detection and Assessment of Date-Like Anomalies in Clinical Laboratory Reference Ranges

The interpretation of clinical laboratory data is greatly influenced by reference ranges. But these ranges may result in date-like expressions which can block direct numeric processing and obscure the original range limits. In this study, this is explored with the Multi-Center Clinical Laboratory Dataset (MCLD) by integration of Rule Based Detection (RBD), Numeric Parsing Validation (NPV) and Rule Based Candidate Reconstruction (RBCR). The expression for dates is discovered by RBD, the range for interpretation is checked by the NPV, and potential numeric intervals are created by the RBCR. The candidates generated are evaluated for consistency and ambiguity within the candidates. In 145,288 laboratory records examined, 5,054 (3.48%) had date-like reference-range expressions, representing 18 different combinations of test name, expression, and measurement units. 18 cases were coded as numeric candidates. These candidates matched the Normal and Abnormal classifications on 5054 affected records. An ‎additional analysis found compatible alternative intervals in 14 of the 18 cases within a limited ‎search grid of integer boundaries. The results demonstrate that there is not always a single reference interval associated with a particular classification. The framework enables identification of anomalies that fall outside of date ranges, some possible numeric corrections, and the assessment of internal consistency and ambiguity of those corrections.