Skip to main content

Appendix D — Reference card

[S] = verified in a primary public source (2026-10-16); [Y] = synthesis. “NV” = not verified: paywalled.

Document map (CLSI editions verified on clsi.org)

Question Document
Precision EP05 Ed4 (2025-12); A3 (2014): 20 days × 2 runs × 2 reps
Verify precision/bias EP15-A3 (2014): ≤5 working days; bias ≥25 results over ≥5 days
LoB/LoD/LoQ EP17-A2 (2012)
Linearity EP06 Ed2 (2020)
Method comparison EP09c Ed3 (2018)
Interference EP07 Ed3 (2018); EP37 tables
Stability / lots EP25 Ed2 (2023); EP26 Ed2 (2022)
Binary tests EP12 Ed3 (2023)
ROC EP24-A2 (2011)
Reference intervals EP28-A3c (2010)
Framework EP19 Ed3 (2022); specimen types EP35 Ed2 (2025)

Design counts for most EP documents: NV.

Formulas

Topic Formula
Within-lab SD √(σ²day + σ²run + σ²rep); CV% = 100·SD/mean [Y]
Satterthwaite df (Σ cᵢMSᵢ)² / Σ (cᵢMSᵢ)²/dfᵢ; 20×2×2 ≈ 70 df
UVL claim·√(χ²₀.₉₅,ν/ν) [Y]
LoB / LoD mean_b + 1.645 SD_b; LoB + 1.645 SD_low
Probit LoD95 10^((1.645 − b₀)/b₁)
Bias at Xc a + (b − 1)·Xc
Total error |bias| + 1.96 SD
PPV Se·p / (Se·p + (1−Sp)(1−p))
LR+ / LR− Se/(1−Sp); (1−Se)/Sp
Wilson CI (p̂ + z²/2n ± z√(p̂q̂/n + z²/4n²)) / (1 + z²/n)
n cases (Se ± d) z²p(1−p)/d²; total = cases/prevalence
AUC P(case > control); DeLong var from structural components
RCV exp(± 1.96 √2 SD_w) − 1
Deming slope (s_yy − δs_xx + √((s_yy − δs_xx)² + 4δs_xy²)) / 2s_xy

Regulatory anchors

Item Fact
IVDR Art 47 / 48 classification / conformity assessment
Art 56 / 57 / 58 performance evaluation / study requirements / additional (58(2) CDx)
Annex I 9.1(a)(b) analytical / clinical parameters
Annex VIII Rule 3(f): CDx = class C
Art 48(7)(8) CDx: consult competent authority (2001/83/EC) or EMA
Annex XIII Part A PEP, PER, CPSP 2.3.2; Part B PMPF
Art 2(7)(38)(40)(41) CDx; scientific validity of analyte; analytical; clinical performance
Transition 2024/1860 legacy to 31 Dec 2027; D 2027, C 2028, B 2029
MDCG 2022-2 clinical evidence for IVDs (Jan 2022)
ISO 20916:2019 clinical performance studies, good study practice
510(k) / De Novo / PMA cleared / granted / approved [Y]
FDA 2007 PPA/NPA, not Se/Sp, vs non-reference
FDA CDx 2014 final; 2016 codevelopment draft; 2020 “Specific Group of Oncology…” final
STARD / QUADAS-2 30 items / 4 domains
ICH E9, E9(R1) 1998, 2019 (by analogy)
CDISC SDTMIG-MD v1.1, ADaMIG-MD v1.0; no IVD standard found

Elecsys pTau217 (FDA K261686, decision 2026-08-19)

Item Fact
Cut-offs pg/mL neg ≤ 0.223; int > 0.223–≤ 0.378; pos > 0.378
Pivotal N 958; PET+ 378 (39.5%); 6-reader visual read
3×2 (PET+/PET−) pos 285/37; int 72/119; neg 21/424
Headline PV pos 88.5%; NPV 95.3%; intermediate 19.9%; PPA 75.4%; NPA 73.1%
Not in summary eGFR analysis; monitoring exclusion (Roche release only)
Papers prototype assay; Hibar 0.189 single cut-off; Hortsch 5.78% indeterminate

Rules of thumb [Y]

  • Se/Sp need a reference standard; otherwise PPA/NPA.
  • Lock cut-offs before validation; fix the ≤ / < convention.
  • Intermediates: report fraction, per-category PV, PPA/NPA with intermediates as misses.
  • Missing bridging samples: ask why; MI needs arm × outcome term; tipping-point.
  • Non-significant ≠ equivalent: use CI inside the limit.
  • Treat NV items as unverified; do not state them as fact.