Build a defensible top-down estimate

Choose stable periods that represent intended routine operation, including relevant days, operators, reagent lots and recalibrations where the data allow. Plot the sequence. Investigate shifts and exclusions; do not remove inconvenient points merely to reduce the estimate. Control-material behaviour may not fully represent patient specimens.

For a single routine result, an appropriate long-term standard deviation estimates within-laboratory reproducibility. The standard error of the QC mean answers a different question. More QC measurements improve knowledge of the SD; they do not make each patient result arbitrarily more precise.

Estimate SD or CV at relevant concentrations and examine the relationship. Do not root-sum-square SDs from low and high controls as if both independently contributed to one result. Pool only when the same precision model and conditions are defensible.

Recalibration variability may already appear in long-term data. Uncertainty in an assigned calibrator value may not. Establish what each term represents before combining it. Investigate bias and decide whether correction is needed; do not conceal a known problem by widening an interval.