# Clinical workflow priorities: Rewire definitions and evidence collection

**Source ID:** `use-case-source-clinical-priorities-2026-09-28`. **Authored:** 28 September 2026. Primary sources were retrieved during the automated exploration on **2026-09-28**; their publication dates are retained below.

This is a **Rewire-authored workflow definition and sourced research brief**. It is not an external professional guideline, an independent benchmark or validated model evidence. Clinical demand selected these questions before catalogue coverage was considered. Definitions can be listed while comparison collection remains planned. The work did not test models, access patient records or obtain human clinical expert approval.

**Review:** automated_source_review by Codex, 2026-09-28T14:52:02.000Z. The review checked consistency with the completed exploration, ten-case synthesis and evidence backlog. Qualified clinical review is a requirement for evaluating future clinical comparisons; it is not a prerequisite for defining the user need.

## C1 — Rare-disease candidate ranking

Diagnostic scientists need to select variants and genes for review during initial ES/GS interpretation. Start with a defined congenital-anomaly or developmental-disorder population, retaining phenotype, pedigree, variant calls and coverage. The output is a reasoned candidate queue, including conflicts and unresolved cases.

The [ACMG pediatric ES/GS guideline](https://www.nature.com/articles/s41436-021-01242-6) (1 July 2021) supports first- or second-tier testing in its specified population. The [NHS test directory](https://www.england.nhs.uk/publication/national-genomic-test-directories/) (page updated 16 July 2026; rare/inherited directory v9) establishes a commissioned service context. [Genomics England’s Exomiser validation](https://pipeline-rd-help.genomicsengland.co.uk/Mira/variant-prioritisation-approaches/exomiser/exomiser-performance/) (page 20 March 2024) documents an operational prioritisation workflow. Its retrieval of known diagnostic variants is not unselected diagnostic yield.

Collect same-input comparisons of conventional phenotype/pedigree/frequency-aware analysis, including a pinned Exomiser configuration, with and without a molecular-effect method. [AlphaGenome](https://www.nature.com/articles/s41586-025-10014-0) (28 January 2026) is a candidate for scoped molecular predictions, not a standalone diagnostic system. Outcomes should include case-level candidate recovery at a fixed review budget, effort, missed classes and abstention. Separate families, centres and time; freeze knowledge; audit label circularity. Independent qualified adjudication and explicit calling-failure denominators are needed before diagnostic claims.

## C2 — Unresolved rare-disease reanalysis

A clinical genomics service needs to identify unresolved cases worth reopening and explain what changed. Preserve the original report, analysis date, calls and filters, then record updated phenotypes, knowledge and analysis methods. A variant reevaluation is narrower than reanalysis of the whole case; the [ACMG statement](https://www.nature.com/articles/s41436-019-0478-1) (24 April 2019) establishes this workflow distinction without supplying a universal interval.

Candidate implementation evidence includes [Exomiser reanalysis](https://www.nature.com/articles/s41525-024-00456-2) and [Talos](https://www.nature.com/articles/s41591-026-04477-5) (24 June 2026). The Talos report distinguishes diagnoses from diagnosed individuals and initial from subsequent reanalysis cycles. Those denominators must remain attached to any later extraction; no yield is promised here.

Collect two comparisons: incremental programme yield against the original unresolved cohort over a stated interval, and added method value against refreshed conventional analysis using identical updated knowledge, phenotypes and calls. Measure confirmed new diagnoses, false alerts, review effort and reasons for resolution. Separate new information from algorithmic benefit, avoid future-knowledge and originating-laboratory label leakage, and retain unresolved cases. Qualified independent review must confirm findings. Controlled cohort access and reusable patient-level data may be unavailable even when implementation papers are public.

## C3 — BRCA1/BRCA2 germline interpretation

Germline variant scientists need auditable evidence and criterion assignments for hereditary-cancer classification. The initial scope is confirmed germline BRCA1/BRCA2 variants under the relevant, pinned ClinGen ENIGMA specification. Inputs include transcript, assay limits, population frequency, segregation, functional evidence and evidence date; outputs retain conflicting and uncertain findings.

[ASCO’s germline-panel guideline](https://ascopubs.org/doi/10.1200/JCO.24.00662) (17 May 2024) documents established testing needs. The [ASCO–SSO breast-cancer guideline](https://ascopubs.org/doi/10.1200/JCO.23.02225) (4 January 2024) states that VUS should not drive management. [ClinGen ENIGMA recommendations](https://pmc.ncbi.nlm.nih.gov/articles/PMC11393667/) (online 13 August 2024) provide gene-specific evidence calibration. Other hereditary-cancer genes require their own specifications.

Compare an evidence-support method with manual criteria-based review using the same evidence cutoff. Collect severe classification disagreements, criterion errors, unresolved-case coverage and review time against blinded independent expert adjudication. Audit functional-training overlap and classifications that already used the predictor. Preserve classification dates and gene/domain separation appropriate to the claim.

Functional activity is supporting evidence, not a complete classification. Tumour-only results do not confirm germline status; germline classification does not estimate individual absolute risk or select somatic treatment. Family-history risk can remain after a negative panel. Qualified hereditary-cancer review is needed to evaluate comparison outcomes.

## C4 — Somatic small-variant oncogenicity

Somatic curators and molecular pathologists need methods that help classify SNVs and small indels while retaining insufficient or contradictory evidence. Oncogene missense and tumour-suppressor loss-of-function mechanisms require separate protocols. Record variant/reference, transcript, tumour, origin confidence, allele fraction, quality, biological evidence and classification criteria.

The [ClinGen/CGC/VICC consensus](https://pmc.ncbi.nlm.nih.gov/articles/PMC9081216/) (online 29 January 2022) addresses oncogenicity interpretation; its small framework-validation set is not an independent large-model benchmark. The [ClinGen oncogenicity SOP](https://clinicalgenome.org/docs/somatic-oncogenicity-sop/) (v1.0 approved 6 May 2021) distinguishes its scope from germline predisposition and therapeutic value. [ClinVar documentation](https://www.ncbi.nlm.nih.gov/clinvar/docs/clinsig/) maintains separate classification types.

Gather dated oncogenicity-specific assertions and orthogonal functional evidence, such as [Ng et al.’s cell-viability assays](https://pmc.ncbi.nlm.nih.gov/articles/PMC5926201/) (12 March 2018), preserving assay and mechanism. Compare methods with explicit criteria-based review; assess precision/recall, calibration, abstention, criteria fidelity and reviewer effort. Audit residues, allelic series, studies, intended unseen genes and training/assertion overlap. Qualified independent somatic adjudication is needed.

Germline-benign assertions are not automatic somatic negative controls. Molecular activity and oncogenicity do not establish treatment response. Fusions, rearrangements and CNVs require separate protocols, and tumour-only sequencing does not confirm somatic origin.

## C5 — EGFR-mutant NSCLC actionability and resistance evidence

Molecular tumour boards need correctly scoped evidence for advanced EGFR-mutant NSCLC, including progression after EGFR-targeted therapy. Inputs must include biomarker, tumour/stage, prior therapies, sampling time, tissue/plasma, co-alterations, jurisdiction and evidence cutoff. Outputs are source-linked sensitivity/resistance assertions with regimen, setting, native evidence tier, contradictions and missing information.

[ESMO’s NGS recommendations](https://pubmed.ncbi.nlm.nih.gov/38834388/) (online 27 May 2024; corrigendum 20 February 2025) establish the relevant workflow. The corrected full guidance needs extraction before detailed grading. [CIViC’s dated releases](https://docs.civicdb.org/en/latest/using/data_releases.html) and [curation documentation](https://docs.civicdb.org/en/latest/curating/evidence.html) support a versioned evidence corpus. The [FLAURA resistance analysis](https://www.nature.com/articles/s41467-023-35961-y) (27 February 2023; correction 1 June 2023) offers primary evidence in a defined therapy context, but patient-level genetic data are unavailable.

Compare retrieval-assisted review with versioned lookup and manual primary-source review at equal effort. Collect relevant-evidence recall, unsupported assertions, context/direction accuracy, citation support, contradiction handling and review time. Freeze dates, hold out studies/profiles and include inapplicable and uncertain examples. Preserve source-specific tiers and qualified molecular-oncology adjudication. Evidence retrieval does not establish treatment benefit or trial eligibility; absence of evidence is not resistance.

[OncoKB’s licensing FAQ](https://faq.oncokb.org/licensing) requires explicit AI-benchmarking permission and prohibits AI training. Keep permissions separate from question selection. No restricted access or permission request occurred in this work; native source terms must be checked at acquisition.

The original exploration recorded intermittent access to some publisher/NHS pages and used primary indexed text where necessary. Detailed protocols, supplements, corrections, contemporary specifications and reuse terms remain collection work. These definitions create no model mappings, performance scores, clinical recommendations or human approvals.