| name | chart-review-crc-nccn-adherence |
| description | Colorectal cancer (CRC) NCCN guideline-concordance review. Activate when assessing whether a CRC patient's care adhered to NCCN — molecular (MSI/MMR) testing, staging workup, adequate lymph-node harvest, and stage-appropriate adjuvant chemo / neoadjuvant chemoradiation — answering concordance questions per patient from pathology, radiology, oncology, and operative notes plus structured EHR data.
|
CRC chart review — STORE abstraction + NCCN concordance
For each in-scope CRC patient you answer the questions in
references/questions/, in two tiers: confirm eligibility, then abstract the
STORE 2025 data elements (the core chart-review questions). The NCCN
concordance verdicts are then computed deterministically by the rule-engine
from those facts (references/rules/concordance.yaml) — you do not judge
concordance directly.
Procedure
- Eligibility + context (
eligibility.yaml, T0): confirm this is a CRC
case, the subsite (colon vs. rectal), and the AJCC stage group. Answer first —
these gate scope and set the N/A (Excluded) conditions for the
stage-conditional rules.
- Abstraction (
abstraction.yaml, T1) — the core chart-review questions:
abstract the STORE 2025 data elements (primary site, histology, grade, TNM,
LVI, node counts, margins, MSI/MMR, surgery, chemo, radiation) plus the
baseline staging facts (cross-sectional imaging, baseline CEA). Answer
verbatim where possible — these facts feed the concordance rules, so
accuracy here drives the verdicts.
- Concordance is automatic. The rule-engine evaluates the five NCCN rules in
references/rules/concordance.yaml over your eligibility + abstraction
answers — each resolves to Concordant / Non-concordant / Excluded (N/A), and a
non-concordant verdict gets an attribution category
(references/attribution.yaml). You do NOT answer these; if a verdict looks
wrong, correct the underlying abstracted fact.
- Cite evidence for every abstraction answer: a verbatim note quote (path /
radiology / oncology / operative), or a structured row (labs/orders) when
that is the source. Do not infer from parametric knowledge.
Scope (from the CRC EvoSkill design)
- Abstraction (Section 2, Domains 1–4): ~18 STORE data elements across cancer
identification, staging (TNM), surgical/pathologic features, first-course
treatment, and baseline staging workup — the primary chart-review questions
(the only thing the agent answers).
- Concordance (Section 3): five NCCN rules computed by the rule-engine —
MSI/MMR testing, staging-workup completeness, ≥12-node harvest, Stage III
adjuvant chemo, locally-advanced rectal neoadjuvant chemoradiation.
Intentionally out of scope: the study's baselines / Base+Human-Skills /
EvoSkill conditions, difficulty levels, data-prep/linkage, train-val-test, and
evaluation metrics — this package is the task rubric only.
