Eight Days of a Creatinine Drift and 260 Reported Results

Convert to SD units (SD = 0.15, mean = 4.20):

• Preceding five days: -0.3, +0.1, -0.1, +0.2, -0.2 SD
• Day 1: -0.1 | Day 2: +0.1 | Day 3: +0.8 | Day 4: +1.2 | Day 5: +1.6 | Day 6: +2.7 | Day 7: +1.0 | Day 8: +0.7

Which rules fire:

• 10x (ten consecutive points on one side of the mean). Days 1 through 5 are all above the mean, and they are preceded by four days above the mean as well, giving nine consecutive positive values by Day 5. Under a 10x rule the run is out of control at Day 5 - one day before anyone acted - and the pattern is a systematic shift, not random scatter.
• 1-3s on Day 6. 4.61 is +2.7 SD, which trips a 1-3s rejection. That is an immediate rejection of the run.

Was the Day 6 repeat defensible? No, on two grounds. First, a repeat on the same aliquot is not an independent repeat; it re-measures the same cup and cannot distinguish a real shift from a pipetting or bubble problem. Second, and more important, a single in-control repeat does not override a documented multi-day shift. Once 10x has fired, the correct response is investigation and corrective action, not re-running until a value looks acceptable. Releasing 38 results on that basis converted a QC event into a patient-result event.

The bench tech is four months qualified and unsupervised; that is a training and escalation failure at least as much as an individual one.

Characterize the error first. Level 1 was stable while level 2 drifted, so the bias is roughly proportional - small at low creatinine, larger at high creatinine. At a mean of 4.20 the Day 5 shift is about +0.25 mg/dL (+6%) and the Day 6 value about +0.4 mg/dL (+10%). At creatinine 1.0 mg/dL, a 6-10% proportional bias is 0.06-0.10 mg/dL, well within clinical noise.

Who is most affected:

• Patients near a decision threshold at high creatinine - for example, a value of 3.9 reported as 4.2, or a 4.4 reported as 4.7. Dosing of renally cleared drugs, contrast decisions, and dialysis timing all sit on these boundaries.
• eGFR is auto-reported from creatinine, so the bias propagates into a calculated value that clinicians treat as a discrete category (CKD stage boundaries at 60, 45, 30, 15 mL/min/1.73 m2). A CKD 3b patient could be reclassified as stage 4.

Look-back design:

1. Extract all creatinine results from Day 1 through the day the problem is corrected - approximately 260.
2. Rank by creatinine value and flag those close to a decision threshold, then flag those with a same-patient prior result suggesting an unexplained step-up.
3. Re-test a representative sample of retained specimens to estimate the actual bias at each level rather than assuming the QC drift equals the patient bias.
4. Notify the medical director and the ordering clinicians for the flagged subset; do not blanket-amend 260 results without evidence of clinical significance.

Document the analysis and its limits - you cannot retest everything, and saying so is part of a defensible look-back.

Corrective action, in order:

1. Stop reporting. Close the analyte in the LIS if the middleware allows it, and notify the laboratory medical director and the on-call supervisor - 'the supervisor is on leave' is a reason to escalate, not a reason to wait.
2. Investigate the reagent change first. A new lot on Day 1 with a proportional upward drift is the leading hypothesis. Check lot-to-lot comparison data, reconstitution and storage, expiration, and whether onboard stability was exceeded. Repeat the calibration with fresh, properly handled reagent.
3. Recalibrate and re-verify. Calibration verification should be repeated, and the level count and acceptance criteria must match your accreditor's checklist requirement rather than the two levels used. Verify across the reportable range, not just at two points.
4. Confirm with a comparison study. Run 20 patient samples by the old and new conditions and evaluate bias against a defined acceptance limit tied to the clinical decision points, not a convenient percentage.
5. Reinstate QC and document the return to control before any patient result is released, with the technologist, date, and time recorded.

Documentation: the QC charts, the rule violations and dates, the investigation notes, the corrective action, the look-back analysis, and the medical director's sign-off. Retain per your retention policy (CLIA requires at least two years for most records).

Escalation and people: brief the bench tech without blame in writing - the system failed to teach escalation. Add a rule that any 10x or 2-2s event requires supervisor notification before release, and audit compliance monthly for the next quarter.