A New Creatinine Reagent Lot, a Five-Day QC Shift, and Eighteen Dialysis Patients
On Monday a 600-bed hospital laboratory loads a new lot of creatinine reagent on its chemistry analyzer. Calibration with the calibrator lot already in use is accepted, with a slope within 5% of the previous calibration. Two levels of quality control are run once each morning. The manufacturer's values are level 1 mean 0.80 mg/dL (SD 0.04) and level 2 mean 4.00 mg/dL (SD 0.08).
Level 1, Monday to Friday: 0.79, 0.81, 0.80, 0.78, 0.82. Level 2, Monday to Friday: 4.12, 4.20, 4.22, 4.18, 4.25.
On Friday the renal unit sends a list: 18 stable hemodialysis patients whose creatinine rose by a mean of 0.22 mg/dL (range 0.14–0.31) compared with last month, with no change in dialysis prescription, weight or Kt/V. The laboratory assistant says: "No single point is outside ±3 SD, so we can keep reporting." A second technologist suggests simply repeating the QC.
1. Apply the Westgard rules to these data and decide whether each day's run should have been accepted.
Read the worked answer
Convert level 2 to z-scores: +1.5, +2.5, +2.75, +2.25, +3.13. Level 1 z-scores are −0.25, +0.25, 0, −0.50, +0.50 - unremarkable.
- 1-2s: day 2 onwards exceeds 2 SD; this is a warning rule, not by itself a rejection criterion, but it must trigger inspection of the other rules.
- 2-2s: days 2 and 3 both exceed +2 SD on the same side, so the run is rejected from day 3, and again on days 3–4.
- 4-1s: days 1–4 are all more than 1 SD above the mean on the same side, so the run is rejected on day 4.
- 1-3s: day 5 (+3.13 SD) is a rejection on its own.
- 10x: not yet reached with five points, but the pattern is clearly a shift.
The assistant's statement is wrong: the rules are designed so that a sustained small bias is caught by 2-2s, 4-1s and 10x even when no single point exceeds 3 SD. The QC should have been rejected from day 3 at the latest, and arguably the day-2 result should have prompted a warning investigation.
2. What is the most likely cause, and what investigation would distinguish a calibration problem from a reagent-lot problem?
Read the worked answer
The bias is proportional rather than constant: level 2 is running about 5–6% high, while level 1 is essentially at its mean. A constant additive error would shift both levels by the same absolute amount; a proportional error points to a calibration slope error or a difference between reagent lots.
Investigation, in order:
- Repeat level 2 with a fresh vial and with a newly reconstituted control to exclude a deteriorated or misreconstituted QC material.
- Recalibrate using a fresh aliquot of calibrator, ideally a different calibrator vial, and re-run both levels. If the bias disappears, the earlier calibration was at fault.
- If the bias persists after recalibration, run a split-sample comparison: about 20 patient samples spanning 0.5–8 mg/dL on the old and new reagent lots. A consistent proportional difference confirms a lot-related bias; a random scatter suggests imprecision or an instrument problem.
- Check reagent on-board age and storage, probe and pipetting performance, photometer lamp hours, and whether the lot requires a lot-specific calibrator.
- Check linearity and review the calibration curve and residuals.
A key point is that the calibration slope was "within 5%" of the previous one - the same magnitude as the bias being investigated. That acceptance criterion was too loose for a 5% proportional error to be detected, and it should be reviewed.
3. What should be done about the patient results already reported Monday to Friday?
Read the worked answer
The 18 patients' mean rise of 0.22 mg/dL is the same order as the QC bias, so the results cannot be assumed valid. Do not apply a mathematical correction factor, which would require its own validation.
Practical steps: identify all creatinine results reported from that analyzer since the reagent lot was loaded; retrieve and re-run available samples, or request fresh samples from a subset of patients, on a verified method; if a proportional bias is confirmed, issue corrected reports after medical review and notify the renal unit with an explanation. Record the event as a nonconformance with the root cause, the affected period, the number of patients and the corrective action.
Note also that a delta check would probably not have caught this: the reference change value for creatinine is around 14%, and a 5% shift falls inside it. This is exactly why the QC rules, not the delta check, are the primary defence. Repeating the same QC vial without investigating, as the second technologist suggested, would only have shown that the same biased material gives the same biased result.