BIOP02-140 v1: 폐 공변량 기준선 + KRAS/EGFR 증분가치 — kras_g12c 🔴 강등 권고 - #136
Merged
Conversation
purity(ABSOLUTE)/stage/site/cohort 공변량-only vs 공변량+H&E 결합모델, holdout 5-fold in-holdout CV(H&E patient_proba가 holdout에만 존재해 방법 정정, 결과 계산 전 발견). histology_lusc 양성대조 통과(Δ+0.345). kras_g12c: 공변량-only(0.802)가 H&E-only(0.681)를 능가, 결합 시 증분 거의 0(Δ=-0.0004) → 🔴 결정지도 강등 권고. egfr_activating: 같은 방향 이나 n_pos=15<25로 판정 보류. 순도 층화(참고용, 표본 작음) 포함. 독립 리뷰 요청 대상: 지용기(braveji), DoD 명시.
5 tasks
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
BIOP02-140 v1 — 폐 공변량 기준선 · 증분가치 (KRAS/EGFR/histology)
claim_level: hypothesis_only·critic_status: pendingv1 범위
티켓 본문의 완화 옵션대로 폐(LUNG_NSCLC) 한 암종, KRAS·EGFR·histology_lusc 세 엔드포인트만.
대장·위·두경부는 후속 카드로 분리 제안(#6 BIOP02-139가 HPV 등을 스킵한 것과 같은 판단).
방법
COVARIATE_BASELINE_PREREGISTRATION.md로 커밋(사전등록).fetch_covariates_lung.py로그가 근거).mil_cost_results.json)이 holdout에만 존재함을 발견 — 이 시점까지 수치 미계산 상태였음 — 방법을 holdout 5-fold in-holdout CV로정정(문서에 정정 내역 명시, 검정력 손실 없음).
run_mil_cost.py의bootstrap_auc()재사용(n=1000, seed=42), 재학습 없음, CPU only.결과
티켓 인용 수치 검증: "조직형만으로 예측하면 0.793" → cohort 단일공변량 재계산 시 0.7795(5-fold OOF), 방향·결론 일치.
핵심 발견
KRAS는 티켓의 우려보다 더 나쁩니다 — purity·stage·site·cohort를 다 합친 공변량-only 모델(0.802)이 H&E(0.681)를 확실히 앞서고, H&E를 얹어도 증분이 없습니다.
결정지도에서 kras_g12c를 🔴로 명시(H&E가 아니라 공변량/조직형이 예측)하는 걸 제안합니다. egfr_activating은 같은 방향이지만 n_pos=15로 확증 부족이라 판정
보류로 남겼습니다. 순도 층화는 참고용(층당 n_pos 5~10)으로 포함, 판정에는 안 씀.
DoD
결정지도(
CROSS_CANCER_DECISION_MAP.md) 반영은 리뷰 통과 후 별도로 진행합니다(critic_status pending 상태로 셀프 반영 안 함).관련: Jira BIOP02-140, BIOP02-139(#6, 같은 묶음)