12.3 Method Selection, Optimization, and CAP Analytic Validation

12.3 Method Selection, Optimization, and CAP Analytic Validation

Quick Answer: Optimize first (titer, retrieval, detection), freeze the SOP, then validate. CAP 2024 Goldsmith published practice: at least 90% overall concordance; nonpredictive laboratory-developed assays 10 positive plus 10 negative (director may document fewer than 20 total); predictive laboratory-developed assays 20 plus 20; unmodified FDA predictive kits follow the manufacturer, or 20 plus 20 if the insert is silent; HER2 and PD-L1 each scoring combination 20 plus 20; dual predictive and nonpredictive use with the same scoring treated as predictive 20 plus 20. Do not treat doubled 20-plus-20 nonpredictive / 40-plus-40 predictive sample sizes as current 2024 practice.

This OpenExamPrep section is independent teaching on IHC method selection and analytic validation, using published CAP 2024 Goldsmith guideline numbers as source facts. It is not a CAP or ASCP publication and does not claim College or Board approval.

Method selection and optimization come before the concordance study

Selection means choosing a clone, a detection chemistry, a retrieval class, and an intended specimen stream (FFPE in 10% NBF, cytology, decalcified bone) that can answer the clinical question. Optimization means finding working conditions: serial dilutions of a concentrate, retrieval buffer and time, incubation, and detection until expected positives are specific and morphology is intact.

If you skip optimization and jump to a 20-plus-20 study on a random 1:50 dilution, you may "validate" a noisy or insensitive method and then be stuck with it. Optimization slides are not the validation set, but they belong in the file as the reason those conditions were locked. Change one variable at a time during optimization. Then freeze the recipe in the SOP. The validation run uses that frozen method against a comparator. After go-live, informal tweaks are method changes (Section 12.1) and may need 2-plus-2 confirmation, director-determined tissue, or full revalidation depending on what changed.

A titer that looks pretty on a 3+ tumor and dirty on a negative tonsil is not finished. Include a low expressor in optimization if the assay will be scored. Retrieval that shreds fatty breast while rescuing a nuclear marker is not finished. Detection that requires a biotin block on liver is not interchangeable with a biotin-free polymer without a file.

Laboratory-developed versus FDA-cleared or FDA-approved assays

U.S. laboratories distinguish:

• Unmodified FDA-cleared or FDA-approved IHC kits used as the manufacturer specifies, on the intended specimen type: the laboratory verifies that the kit works in this laboratory (personnel, water, instruments, fixative stream).
• Laboratory-developed tests (LDTs) and modified manufacturer assays (different dilution, retrieval, detection, or specimen type than the insert): the laboratory validates the method it actually runs.

CAP 2024 Goldsmith Good Practice Statement 1 says laboratories must analytically validate all laboratory-developed IHC assays and verify all FDA-cleared IHC assays before reporting results on patient tissues. Predictive manufacturer kits that are unmodified follow manufacturer verification instructions (GPS 5); if the insert does not give numbers, the published fallback is a minimum of 20 positive and 20 negative tissues.

A kit used off-label—for example an insert-limited breast kit applied to a different organ scoring scheme without a file, or a ready-to-use antibody diluted further—is not unmodified. Treat it as a method you must validate.

Comparators from statement 1, at a high level

A validation study needs a truth source. Published means include, but are not limited to:

• Expected architectural and subcellular localization of the antigen (membrane versus nucleus versus cytoplasm)
• Prior testing of the same tissues with a validated or verified assay in the same laboratory
• Testing the same tissues in another laboratory that already has a validated or verified assay
• A non-immunohistochemical method (ISH, flow cytometry, molecular)
• Results from a laboratory that tested the same tissues for a clinical trial
• Percent-positive rates documented in published clinical trials (use cautiously; case mix differs)
• IHC on cell lines with known protein amounts
• Previously graded proficiency-testing tissue challenges, if available, compared with the graded responses

You do not need every bullet on every assay. You need a documented comparator that makes sense for the marker. A cytokeratin LDT may lean on expected localization plus a previously verified clone. A predictive HER2 LDT should not rely only on "it looks membranous."

Concordance: 90 percent overall (strong recommendation 2)

For initial analytic validation or verification of every assay used clinically, laboratories should achieve at least 90% overall concordance between the new assay and the comparator or expected results. Overall means the whole set, not a hidden subgroup of easy 3+ tumors.

Worked concordance math:

• Nonpredictive 10 plus 10 (20 tissues): 18 of 20 agree → 90%, meets the published floor if discordances are investigated. 17 of 20 → 85%, does not meet 90%, even if "all the positives worked."
• Predictive 20 plus 20 (40 tissues): 36 of 40 = 90%. 35 of 40 = 87.5% fails the published concordance floor.

Discordant cases are the study, not an annoyance to hide. Localization errors, cutoff disagreements, and preanalytic mismatches belong in the report to the medical director.

Initial sample sizes: 2024 Goldsmith numbers

A common doubled-sample-size story uses 20 positive plus 20 negative tissues for nonpredictive assays and 40 plus 40 for predictive assays. That pair is not CAP 2024 Goldsmith practice, and it was not the 2014 CAP analytic-validation floor either. The 2024 published minima are 10 plus 10 nonpredictive and 20 plus 20 predictive—the same category floors as the 2014 guideline. Directors may still choose more tissue. They must document if they use fewer than the published minimum totals.

Notes on GPS 3–5: the validation set should include high and low expressors for positive cases when appropriate and should span the expected range of clinical results for markers reported with semiquantitative or numerical scoring.

Worked examples

Example A — Nonpredictive LDT (CK7). Optimize titer and retrieval. Freeze the SOP. Select 10 CK7-positive tissues (including weak or focal positives, not only strong lung adenocarcinoma) and 10 expected negatives, all in the laboratory's 10% NBF FFPE stream (GPS 8). Comparator: expected localization plus a previously verified CK7. Need at least 90% overall concordance (strong rec 2). If a rare antigen truly cannot yield 20 tissues, the director documents why fewer than 20 total suffice (GPS 3). That exception is a file, not a habit of testing two leftover tumors and going live.

Example B — Predictive LDT (laboratory-developed ER on FFPE). After optimization, minimum 20 ER-positive and 20 ER-negative tissues, spanning strong, weak, and negative, including cases near the laboratory's reporting cutoff. Concordance at least 90% against a comparator (prior validated assay, another laboratory, or an appropriate non-IHC method). This is GPS 4, not 40 plus 40.

Example C — Unmodified FDA predictive kit. Open the package insert. If it specifies verification materials and counts, follow them (GPS 5). If it is silent, use 20 plus 20, with the same 90% concordance floor.

Example D — Distinct scoring (strong rec 6). The laboratory will report PD-L1 clone X with tumor proportion score for one intended use and combined positive score for another. Those are distinct scoring systems. Validate or verify each combination with 20 plus 20, including challenges that matter for that use (low-percentage tumors; immune-cell-rich cases for CPS). The same logic applies when more than one HER2 scoring scheme is in service: each assay–scoring combination gets its own 20-plus-20 file. Running one 20-plus-20 and declaring every possible PD-L1 score "validated" does not match statement 6.

Example E — Dual use, same scoring (GPS 7). Ki-67 will be used both as a general proliferation description and as a predictive cutoff using the same scoring rules. Treat the assay as predictive: 20 plus 20, not 10 plus 10.

Example F — Cytology alternate fixative (conditional rec 9, GPS 10). ER was validated on NBF tissue. The laboratory now wants ER on alcohol-fixed cytology cell blocks. That is a new fixation method: perform a separate validation. Use 10 positive plus 10 negative if possible; because ER is predictive, the director should consider more. Document if 10 plus 10 is not feasible.

Example G — Decalcified bone (GPS 11). If IHC is regularly done on decalcified tissues, test a sufficient number of such tissues so assays achieve expected results. The director chooses how many positives and negatives and which predictive versus nonpredictive markers. Do not assume the soft-tissue NBF file automatically covers acid-decalcified bone.

Example H — New lot versus clone change. New lot of the same clone: 1 plus 1 (GPS 12). Same clone, new vendor, or a dilution change, or a change in incubation or retrieval time with the same method: 2 plus 2 (GPS 13). New clone: full revalidation (GPS 15)—for a predictive LDT that means another 20-plus-20-class study, not a 1-plus-1 lot check. Changing retrieval method (citrate to EDTA, new pH, new heat platform), detection system, processor, stainer platform, or moving the laboratory: sufficient tissues per director (GPS 14), not a silent swap.

These numbers are published starting minima, not a prohibition on using more tissues when the marker is scored, rare, or noisy. They are also not permission to treat doubled 20-plus-20 / 40-plus-40 counts as if Goldsmith 2024 required them.