ELISA Validation under ISO 15189:2022 — a practical playbook

ELISA is one of the most widely run assays in clinical and biopharma labs, and one of the most frequently flagged during ISO 15189 inspections. This playbook walks through what an auditor actually looks for in 2026 under ISO 15189:2022 and how to produce a validation package that stands up to scrutiny without slowing down the lab.

What ISO 15189:2022 changed for ELISA

The 2022 revision consolidated ISO 15189 and ISO 15189:2012/Amd.1 and pulled management requirements closer to ISO/IEC 17025. For ELISA validation the practical impact is in three areas:

• Risk-based thinking (clause 5.5): you must justify validation scope against clinical risk. A research-use ELISA for exploratory biomarker work does not need the same protocol as a companion-diagnostic ELISA.
• Metrological traceability (clause 7.3.4): calibrators must chain back to a higher-order reference material, with the chain documented. Where no reference material exists (many cytokines), you must state the method used to assign value and the uncertainty budget.
• Reporting of measurement uncertainty (clause 7.3.4, 7.4.1): the combined uncertainty — not just intra-run CV — must be computable and available on request. Most legacy ELISA validation packages do not do this.

The minimum validation set

For a quantitative ELISA used in clinical decision-making, the following parameters are non-negotiable. Auditors will ask for every one of them.

1. Linearity and working range

Dilute five to seven levels spanning 20% below the lowest clinically useful concentration to 20% above the highest. Fit linear regression and report slope, intercept, r², and residuals. A slope outside 0.9–1.1 or r² below 0.98 fails. Do not confuse the standard curve (4PL typical) with the linearity check — they are different experiments.

2. Precision — repeatability and intermediate precision

Run each level ≥ 5 replicates within one run (repeatability) and across ≥ 20 days with different operators and reagent lots (intermediate precision). Report %CV per level. Typical acceptance: intra-run %CV ≤ 10%, inter-run %CV ≤ 15% at mid-range; relax at LoQ. CLSI EP05-A3 remains the reference design.

3. Accuracy / trueness

Compare to a reference method or certified reference material on ≥ 40 samples spanning the working range. Report Bland–Altman plot, Passing–Bablok regression, and mean bias with 95% CI. If no reference method exists, use spike-recovery: recovery should land in 80–120%.

4. Limit of Detection (LoD) and Limit of Quantitation (LoQ)

Use CLSI EP17-A2. LoD is where detection probability ≥ 95%; LoQ is the lowest concentration meeting a defined total error goal (often 20–25%). Auditors will reject LoD calculated from a single blank SD — use a proper probit or precision-profile approach.

5. Matrix effects and interference

Test hemolysis, lipemia, icterus and any clinically relevant interferents (biotin, heterophile antibodies for sandwich ELISAs). CLSI EP07 is the reference. Document the highest tested concentration that produces ≤ 10% bias.

6. Carry-over, edge-of-plate, and stability

Plate edge-effects are the single most under-documented failure mode. Run a blank pattern and high-concentration pattern on the same plate for at least three lots; report positional bias per well. Stability: reagent, sample, and on-board stability, each with an explicit acceptance criterion.

The documentation package auditors expect

1. Validation plan — approved before experiments start, with pre-specified acceptance criteria. Inspectors check the plan date is before the raw data date.
2. Risk assessment — FMEA covering sample handoff, operator variability, reagent lot change, equipment drift, and software changes. One page, signed.
3. Raw data — plate layouts, OD readings, curve fits, with a clear link from each reported number back to the source file. This is where most labs fail an audit.
4. Statistical report — linearity, precision, accuracy, LoD/LoQ, interference, uncertainty budget. Plots, not just tables.
5. Uncertainty budget — ISO/IEC Guide 98-3 (GUM) approach. Identify sources, quantify, combine. One A4 page is typical.
6. Traceability statement — calibrator → working standard → higher-order reference, with lot numbers and values.
7. Conclusion and approval — signed by lab director and quality manager, with explicit statement of fitness for intended use.

Where labs lose time (and how to avoid it)

From reviewing real validation packages, three patterns account for most of the effort overrun:

• Reconstructing statistics from spreadsheets six months after the lab work was done. Run statistics the same week as the raw data — memory fades, assumptions get lost, and errors compound.
• Rewriting the report for each reviewer. Pre-agree the structure and plot style with QA before you start. Treat the report template as a controlled document.
• Ad-hoc curve fitting. Different software gives different 4PL fits. Lock down the software version, fit method, and weighting scheme in the validation plan and never deviate.

How AiLabrix fits in

AiLabrix automates the statistics-to-report stretch of ELISA validation: drop in your plate-reader CSV, pick the assay template (ELISA, 4PL, linearity, precision, LoD), and the 22-agent pipeline produces a signed PDF with all the elements above — linearity fit with residuals, Bland–Altman vs. reference, precision profile, LoD/LoQ per EP17, uncertainty budget, traceability chain — plus an append-only audit log of every numerical step. The lab keeps full control of the raw data (self-hosted), and the report structure matches what ISO 15189:2022 inspectors actually ask for.

If you want to see it on your own ELISA data, write to [email protected] with two lines about your assay and we will set up a 30-minute session.