Clinical chemistry method validation — CLSI EP05, EP09, EP17 done in one pipeline

Clinical chemistry is the high-volume, low-glamour engine of most clinical labs: thousands of results a day on assays like creatinine, ALT, glucose and lipid panels. It is also where method-validation shortcuts compound the fastest — a small systematic bias in creatinine shifts CKD staging for an entire patient population. This piece lays out the CLSI-aligned validation pipeline that makes the numbers defendable.

The validation stack — CLSI references

• CLSI EP05-A3 — precision (repeatability + within-lab).
• CLSI EP06 — linearity.
• CLSI EP07-A3 — interference.
• CLSI EP09-A3 — method comparison and bias estimation.
• CLSI EP17-A2 — LoB / LoD / LoQ.
• CLSI EP28-A3c — reference interval verification or establishment.
• CLSI C24 — quality control and Westgard multi-rule.

The five checkpoints auditors test

1. Precision — not just "the CV"

Two levels of control (near decision point and far) × 20 days × 2 runs/day × 2 replicates per EP05. ANOVA decomposes into repeatability (Sr) and within-lab (SL). Acceptance thresholds come from biological variation databases (EFLM) — creatinine desirable SL < 2.2%, ALT < 5.4%, glucose < 2.2%.

2. Method comparison against the previous / reference method

Minimum 40 samples spanning the analytical measurement range per EP09. Run both Passing-Bablok and Deming regression (they differ when random error is asymmetric), plus Bland-Altman. Report slope, intercept, mean bias, and 95% CI on all three. A slope 0.95–1.05 with 95% CI that contains 1.0 is typically acceptable; specifics depend on analyte.

3. Reference interval verification

Establish new intervals (EP28, ≥ 120 samples from a defined reference population) only if you cannot verify an existing one (EP28 verification, ≥ 20 samples, ≤ 2 outside). Reference intervals are population-specific — pediatric, geriatric, pregnancy, ethnicity — and age/sex-partitioned intervals are the norm for most analytes.

4. Measurement uncertainty

ISO 15189:2022 requires an available-on-request uncertainty per analyte. The top-down approach is practical: combined SL from EP05 + bias uncertainty from EP09 + calibrator uncertainty from the manufacturer. Report U = 2 × combined standard uncertainty at the medical decision point.

5. Interference + commutability

Hemolysis, icterus, lipemia (HIL indices) per EP07 — every analyzer vendor publishes their interference profile but you must verify it locally. Commutability of controls and EQA materials is the under-documented failure mode: if your controls behave differently from patient samples at the decision point, your whole QC program is decoupled from reality.

Ongoing QC that actually catches drift

Westgard multi-rule (1_3s, 2_2s, R_4s, 4_1s, 10x) on two control levels per run, with Levey-Jennings charts. Sigma-metric tracking per analyte (TEa / CV) lets you size the QC effort — a 6-sigma process needs less QC than a 3-sigma process. Monthly review of peer-group comparison data from the QC vendor flags instrument drift before it shows in patient results.

Where labs slip

• Validation done once, then forgotten. Revalidate on reagent lot change, instrument major service, software upgrade.
• Verifying reference intervals by copying the manufacturer's. Legally defensible only if your population matches theirs — often it does not.
• Stale SOPs. The SOP says "compare 40 samples" but the data file has 38. Auditors count.

How AiLabrix fits

Drop the analyzer export (Roche, Abbott, Siemens, Beckman). The pipeline runs EP05 ANOVA, EP09 Passing-Bablok + Bland-Altman + Deming, EP17 LoB/LoD/LoQ, reference interval verification, and builds the ISO 15189 uncertainty budget. The PDF has every plot, every table, and the traceability chain of calibrators. Westgard engine runs daily on your QC files and flags drift before a re-validation becomes mandatory. [email protected].