Combined total error in calculated clinical chemistry tests: component analytical error is magnified and decision threshold gray zones widen.

INTRODUCTION

Calculated tests propagate component analytical errors, yet laboratories monitor only the measured components, rarely the calculated quantity.

METHODS

Using analytical performance specifications (APSs) from the European Federation of Clinical Chemistry and Laboratory Medicine Biological Variation Database, we propagated bias and imprecision through 5 calculated tests (low-density lipoprotein cholesterol [LDL-C], anion gap, albumin-corrected calcium, the Homeostatic Model Assessment of Insulin Resistance [HOMA-IR], and the De Ritis ratio), representing difference, linear correction, product, and ratio structures. Difference-based and linear-based tests were assessed across clinical scenarios; the ratio was validated by Monte Carlo simulation.

RESULTS

At desirable component APSs, the combined total error was 2.1% to 2.8% for albumin-corrected calcium, 25.9% for the De Ritis ratio, and 34% for the HOMA-IR. At typical values, combined errors for LDL-C and the anion gap were 18.5% and 20.8%, respectively, exceeding their respective allowable total error targets (12.0% and 11.3%). Error was concentration dependent: the anion-gap error decreased to 7.4% at 28 mmol/L but increased to 43.0% at 6 mmol/L; the LDL-C error reached 28.6% at 60 mg/dL. The allowable total error-based estimate closely matched the quadrature-bias estimate but underestimated worst-case error.

DISCUSSION: Acceptable component performance does not ensure acceptable performance of calculated tests. Derived quantities require specific APSs, and results near clinical decision limits should be interpreted within concentration-dependent analytical gray zones.