Background. Internal quality control (IQC) limits are commonly derived from the standard deviation (SD) of 20 results, following CLSI C24. Limits that are too narrow increase false alarms and consume resources, whereas limits that are too wide fail to detect clinically significant error. At present, no quantitative rule constrains both ends of this interval simultaneously.
Objective. To develop and validate a decision framework for setting IQC limits based on a "floor-ceiling corridor" of imprecision: the ceiling is the maximum imprecision that still maintains sigma ≥ 3 given the total allowable error (TEa) and the measured bias; the floor is the system's long-term imprecision (CV-LT); within this corridor, control limits are preferentially set at the maximum historical CV across quality control lots (CVh).
Materials and methods. We performed a retrospective analysis of the complete IQC dataset from a Roche cobas c501 automated chemistry analyzer, comprising 48 control series (24 analytes × 2 concentration levels), with n = 55-550 results per series, collected continuously on a single lot of control material. Outliers were removed using an iterative ±3SD rule. Desirable imprecision (CV-D) and desirable bias (B-D) were obtained from the EFLM 2021/Ricós biological variation database. Sigma was calculated as σ = (TEa − |Bias|)/CV. Trueness of the mean was verified using two external quality assessment (EQA) rounds from two consecutive cycles (VEQAS, 22 analytes), taking the Roche c501 peer-group mean as the reference value; the EQA bias was compared directly with the bias calculated against the manufacturer's assigned value. The probability of false rejection was modeled for the 1-3s rule with N = 2.
Results. SD estimated from 20 points systematically deviated from the long-term SD (median ratio 1.11; range 0.42-2.20): 25/48 series yielded limits that were too narrow and 11/48 yielded limits that were too wide. With the TEa currently in use and the measured bias, 12/48 series had σ < 3, 12 series had σ = 3-4, and 11 series achieved σ ≥ 6; when the desirable TEa based on biological variation was applied instead, the number of series with σ < 3 rose to 22. Applying the floor-ceiling framework, 28 series had long-term CV below the desirable imprecision (their limits could be widened to CV-D without exceeding the ceiling), 8 series lay within the corridor, and 12 series exceeded the ceiling. The expected number of false alarms fell from 741 to 46 per year across all 48 series (a 94% reduction). Quality goal indices identified imprecision as the limiting factor in 34/48 series. EQA cross-checking showed that the true bias exceeded the bias calculated against the manufacturer in 16/22 analytes, and the direction of the IQC deviation agreed with EQA in only 14/22 analytes. AST showed a constant error of approximately +1.9 U/L (a bias of +27.4% at the low concentration), which reduced the true sigma from 4.17 to −0.53 - an effect that was undetectable when checked against the manufacturer at the mid-level control.
Conclusion. The floor-ceiling corridor allows control limits (SD) to be set quantitatively, avoiding both extremes of being too narrow and too wide, while clearly identifying analytes that cannot be controlled by widening the limits and instead require method improvement. In parallel, EQA cross-checking is a mandatory step before accepting the mean as the chart center, because the IQC mean compared against the manufacturer conceals the true bias in most analytes. Both steps can be performed with existing data, without additional resources.