2nd/10_Wiki/Topics/Domain_General/From_Topics_Root/FDA Clearance (Medical Device Approval).md

id, title, category, status, canonical_id, aliases, duplicate_of, source_trust_level, confidence_score, verification_status, tags, raw_sources, last_reinforced, github_commit, tech_stack

id	title	category	status	canonical_id	aliases	duplicate_of	source_trust_level	confidence_score	verification_status	tags	raw_sources	last_reinforced	github_commit	tech_stack
wiki-2026-0508-fda-clearance-medical-device-app	FDA Clearance (Medical Device Approval)	10_Wiki/Topics	verified	self	510k fda-510k premarket-notification	none	A	0.9	applied	regulatory medical-device fda compliance		2026-05-10	pending	language framework english regulatory

FDA Clearance (Medical Device Approval)

매 한 줄

"매 device 가 predicate 에 substantially equivalent 인가의 증명". FDA 의 medical device 시장 진입 경로 — 매 510(k) clearance / De Novo / PMA 의 3 trail. 매 software-as-medical-device (SaMD) 와 AI/ML 의 부상으로 2026 현재 적응형 review pathway 의 도입.

매 핵심

매 Class

• Class I (low risk): 매 general controls. 대부분 exempt.
• Class II (moderate): 매 510(k) submission 필요.
• Class III (high risk, life-supporting): 매 PMA — full clinical trial.

매 경로

• 510(k): predicate device 와 의 substantial equivalence — 매 fastest (3-6 months).
• De Novo: novel low/moderate risk — predicate 의 부재 시.
• PMA (Premarket Approval): Class III — 매 most rigorous, 1-3 year.
• Breakthrough Designation: priority review for unmet need.

매 응용

1. AI 의료기기 — IDx-DR (diabetic retinopathy), Aidoc (radiology triage).
2. Surgical robot — da Vinci, Intuitive.
3. Continuous glucose monitor — Dexcom G7.
4. SaMD — Apple Watch ECG (De Novo), Cardiologs.

💻 패턴

Predicate Search

import requests

def search_510k(device_name: str, limit: int = 50):
    """openFDA 의 510k database 의 predicate 검색."""
    url = "https://api.fda.gov/device/510k.json"
    params = {"search": f'device_name:"{device_name}"', "limit": limit}
    r = requests.get(url, params=params, timeout=30)
    r.raise_for_status()
    return r.json().get("results", [])

Substantial Equivalence Comparison

def compare_devices(subject: dict, predicate: dict):
    """매 indications / technology / performance 의 비교 표 의 생성."""
    rows = []
    for field in ["indications_for_use", "technological_characteristics", "performance"]:
        rows.append({
            "field": field,
            "subject": subject.get(field),
            "predicate": predicate.get(field),
            "different": subject.get(field) != predicate.get(field),
        })
    return rows

Adverse Event Lookup (MAUDE)

def maude_events(device_name: str, since: str = "2024-01-01"):
    url = "https://api.fda.gov/device/event.json"
    params = {
        "search": f'device.generic_name:"{device_name}" AND date_received:[{since} TO now]',
        "limit": 100,
    }
    return requests.get(url, params=params).json().get("results", [])

SaMD Risk Categorization (IMDRF)

def samd_category(intended_use: str, healthcare_situation: str) -> str:
    """IMDRF SaMD: I-IV — 매 information vs treat/diagnose × non-serious/serious/critical."""
    matrix = {
        ("inform", "non-serious"): "I",
        ("inform", "serious"): "II",
        ("inform", "critical"): "II",
        ("drive", "non-serious"): "II",
        ("drive", "serious"): "III",
        ("drive", "critical"): "III",
        ("treat-diagnose", "non-serious"): "II",
        ("treat-diagnose", "serious"): "III",
        ("treat-diagnose", "critical"): "IV",
    }
    return matrix.get((intended_use, healthcare_situation), "unknown")

PCCP (Predetermined Change Control Plan) for AI

pccp:
  modifications:
    - type: retraining
      trigger: quarterly with new data
      validation: hold-out test set AUC > 0.9
    - type: input expansion
      trigger: new sensor model
      validation: equivalence study
  monitoring:
    metrics: [sensitivity, specificity, demographic parity]
    threshold: 5% degradation
    action: rollback + FDA notification

매 결정 기준

상황	Approach
Predicate 존재	510(k)
Novel low-risk	De Novo
Life-supporting	PMA
AI software	510(k) + PCCP
Unmet medical need	Breakthrough

기본값: predicate search 후 510(k) — 매 most devices 의 default.

🔗 Graph

• 변형: 510k

🤖 LLM 활용

언제: predicate search / SE comparison drafting / adverse event summary. 언제 X: 매 final regulatory submission — 매 RA professional review 의 필수.

❌ 안티패턴

• Predicate cherry-picking: 매 weakest predicate 의 선택 — FDA 의 reject.
• Algorithm change without PCCP: 매 retrain 후 silent deploy — adulteration.
• 510(k) for novel device: 매 De Novo 가 필요한 경우 의 wrong path.

🧪 검증 / 중복

• Verified (FDA CDRH guidance, 21 CFR 807, IMDRF SaMD framework).
• 신뢰도 A.

🕓 Changelog

날짜	변경
2026-05-08	Phase 1
2026-05-10	Manual cleanup — FDA pathways + SaMD/PCCP 패턴