Ten replicates, one run

A bench tech runs the glucose Level 2 control ten times in one run: 98, 99, 100, 101, 102, 103, 99, 101, 100, 102 mg/dL, against an assigned target of 100 mg/dL. Ten replicates in one run do not satisfy the structured multi-day replicate design a laboratory uses to formally verify precision and bias.

The mean is the sum of the ten results divided by ten: 1005 divided by 10 equals 100.5 mg/dL. Each result's deviation from the mean is squared and summed, giving a sum of squared deviations of 22.5. This replicate set is a sample used to estimate a larger variance, and estimating the sample mean uses one degree of freedom; therefore dividing by n − 1, or nine, gives a variance of 2.5. The square root of the variance gives a standard deviation of 1.58 mg/dL.

The coefficient of variation restates that spread as a percentage of the mean: 1.58 divided by 100.5, times 100, equals 1.57%. A %CV this low, for this analyte and this concentration, describes tight clustering; it says nothing on its own about whether the results are close to the target.

Bias answers that second question: the observed mean of 100.5 mg/dL minus the assigned target of 100 mg/dL equals plus 0.5 mg/dL, or plus 0.5%. Comparing that bias against the allowable difference established earlier for this concentration (plus or minus 8 mg/dL) shows the bias is well inside the allowable range. This run supports two separate, defensible conclusions: the method is precise at this concentration in this run, and it shows a small, non-actionable positive bias against the assigned target. It does not, by itself, establish the method's within-laboratory precision or reproducibility, and it does not establish a measurement uncertainty estimate.

Report %CV and bias as two separate numbers from the same data set, and do not let a passing bias hide a failing %CV or the reverse.