chart-review-cancer-diagnosis

name	chart-review-cancer-diagnosis
description	Extract cancer histology type, distant-metastasis status, and local-recurrence status from a patient's clinical notes; disease extent is computed from the latter two. Evidence-cited. Triggers on: cancer type, histology, disease extent, metastatic, local recurrence.

Procedure

This is a notes-only, pan-cancer phenotype task. You answer three leaf fields directly:

• cancer_type — the histology type (enum).
• has_distant_metastasis — yes / no_info (no separate no: "documented negative" and "not addressed" are both no_info).
• has_local_recurrence — yes / no_info (same — only yes or no_info).

disease_extent is a computed field (derived from has_distant_metastasis and has_local_recurrence) — do NOT answer it directly.

1. list_notes to see the chart. Prioritize pathology / surgical-pathology / biopsy / cytology notes first, then treating-oncologist progress notes, then imaging. On a large chart, use search_notes for high-signal terms ("final diagnosis", "carcinoma", "metastatic", "stage IV", "M1", "recurrence") to jump to the relevant spans instead of reading everything. Use read_note/read_notes to read the candidates in full.

2. list_criteria + read_criteria(["cancer_type", "has_distant_metastasis", "has_local_recurrence"]) to get each field's enum of allowed answers and its extraction guidance. Follow the criterion's mappings and no_info rules exactly.

3. For each leaf field, commit one answer via set_field_assessment(field_id, answer, confidence, evidence, rationale). The answer MUST be one of the field's enum values.

   Evidence rules — cite the SMALLEST span that supports the answer:

   • Quote the single sentence/phrase that justifies it (well under ~300 chars). Prefer the pathology "FINAL DIAGNOSIS" line. Use find_quote_offsets to get exact offsets so the faithfulness gate passes.
   • Do NOT cite the whole note. A full-document citation is not acceptable.
   • The cited span must be affirmative — never cite "no evidence of metastatic disease" to support has_distant_metastasis=yes; a negation means the answer is no_info, not yes.
   • For no_info: ALWAYS cite one short span — the section you checked where the info would appear if present (Assessment/Plan, Diagnosis, Staging).

Decision rules (apply to all leaf fields)

• Source priority: surgical pathology > biopsy/cytology pathology > treating-oncologist note > imaging. Read pathology first.
• Affirmative only: extract only what is affirmatively documented for THIS patient. Never infer or guess. Exclude negated findings ("no evidence of…", "rule out…") and family history ("mother with … cancer").
• Distant vs regional: has_distant_metastasis=yes requires distant (M1) spread — a named distant organ or distant nodes, or M1/Stage IV. Regional nodes (N1/N2/N3) are NOT distant → no.
• Recurrence: has_local_recurrence=yes requires an explicit recurrence word ("recurrent/relapse/locally recurrent") after prior definitive treatment; a positive margin alone is not recurrence.
• Conflicts: if notes disagree, take the higher-priority source (and, for histology, the most recent pathology); record the conflict in rationale.
• Confidence: high = explicit pathology final diagnosis or explicit staging (M1/Stage IV); medium = oncologist narrative / inferred mapping; low = imaging-only or ambiguous → prefer no_info over a low-confidence guess.

4. Commit the THREE leaf fields (cancer_type, has_distant_metastasis, has_local_recurrence) via set_field_assessment before finishing — every leaf must have a value (use no_info if absent). Do NOT commit disease_extent — it is derived automatically from the two has_* fields. Do NOT call set_review_status. Once the three leaves are committed, emit a one-line summary and stop.