Simulations of delta check rule performance to detect specimen mislabeling using historical laboratory data

Despite their widespread use, the performance of delta check rules is rarely evaluated because errors are rare and lack a gold standard for detection. In this study we used a simulation-based approach to compare strategies for empirically defining criteria for univariate delta checks, and assessed the performance of these rules for detecting mislabeled specimens in 2 inpatient populations.

We performed simulations using historical laboratory test results by randomly sampling pairs of specimens successively drawn from the same patient or two different patients. We evaluated the performance of delta check rules using a variety of thresholds, including those currently in use in our laboratory.

Mean corpuscular volume had the highest positive predictive value for specimen mislabeling, and produced the fewest false positives. Conversely, rules using other laboratory tests had considerably poorer performance. Several of the "best guess" thresholds historically used in our laboratory, notably those for potassium and anion gap, were predicted to have extremely low yields. In addition, rule performance was not consistent between the two patient populations.

The low yield of delta checks based on any single analyte should prompt careful evaluation of their practical utility. Furthermore, our results indicate that it may not be possible to generalize delta rules across institutions.