Result Interpretation in Skewed Populations-Are Common Reference Intervals Inadequate?

Reference intervals (RIs) are defined as the central 95th percentile of a healthy population, and many biomarkers exhibit a skewed result distribution. Here we demonstrate interpretive variations for results from a gaussian vs skewed distribution RI.

De-identified patient data were extracted from the laboratory information system for sodium and total testosterone, representing approximately gaussian and skewed reference populations, respectively. RIs and the Box-Cox variables (lambda, mu, and sigma) were obtained using refineR, with individual results and population distributions evaluated using RI midpoint vs median, interquartile range (IQR) values, z scores, abnormality flagging rates, and cumulative percentiles.

For sodium, the RI midpoint, population mean, and median were similar at 138.8 and 139.0 mEq/L, respectively. Total testosterone demonstrated a skewed result distribution with RI midpoint (473 ng/dL; 16.3 nmol/L) closer to the 75th percentile (522 ng/dL; 18.1 nmol/L), impacting interpretation when results fall in the lower half of the RI. By illustration, a testosterone result of 294 ng/dL (10.2 nmol/L) was misinterpreted as "borderline low." Reporting the z score of -0.55 and cumulative percentile of 0.28 may have provided assurance this result is well within a "disease-free" population range, preventing unnecessary investigation. Conversely, for a case of serial sodium results of 143 to 146 mEq/L in a polydipsic patient, presentation of z scores (1.97 to 3.01) and cumulative percentiles 0.97 to 1.00 may have prompted further investigation, preventing delayed disease recognition.

RIs help interpret laboratory results vs a "disease-free" population; however, skewed distributions may affect result interpretation when clinicians lack knowledge of population result distributions. Interpretive aids may improve comprehension.