Medical Research — Evidence-Based Writing Protocol

Mandatory when writing medical articles for the OB/GYN RND system.

---

1. Evidence Grading — Oxford CEBM

Every recommendation/protocol in articles MUST include Level of Evidence:

Level	Evidence Type	In-Article Notation
I	Systematic Review / Meta-analysis / Large RCT	[LoE: I]
II	Small RCT / High-quality Cohort study	[LoE: II]
III	Case-control study / Retrospective Cohort	[LoE: III]
IV	Case series / Cross-sectional	[LoE: IV]
V	Expert opinion / Consensus	[LoE: V]

Rules:

• Each pathology article (Tier 2) must have at least 3 recommendations with LoE
• Each technique article (Tier 3) must have at least 2 recommendations with LoE
• Each foundation article (Tier 1) must have at least 1 LoE
• Include LoE inline right after the recommendation, e.g.: Aspirin 150mg from week 12 for high-risk group [LoE: I — ASPRE trial 2017]

---

2. Citation Standard — Mandatory References

Accepted Sources (in priority order)

Priority	Source	Citation Example
P0	National MoH Guidelines	[MoH 2023 — Diagnosis and Treatment Guidelines]
P1	ACOG Practice Bulletin	[ACOG PB #234, 2024]
P1	ASRM Committee Opinion	[ASRM CO 2023]
P1	ESHRE Guideline	[ESHRE Guideline 2024]
P2	WHO Recommendation	[WHO 2023]
P2	Cochrane Systematic Review	[Cochrane 2023]
P3	NICE / RCOG	[NICE NG000, 2024]
P3	UpToDate / Standard Textbook	[Williams Obstetrics 26th Ed]

Citation Rules

• Each article MUST have a ## REFERENCES section at the end
• Each pathology article (T2) must cite ≥ 3 different guidelines
• Each technique article (T3) must cite ≥ 2 guidelines
• Each foundation article (T1) must cite ≥ 2 guidelines
• Cite inline within text, e.g.: According to ACOG PB #234 (2024), IVF success rate in women > 40 is 15-20%
• DO NOT accept vague citations: "According to research", "Experts suggest"

---

3. Anti-Hallucination Rules — No Fabrication

🔴 CRITICAL: The most important rules for medical content.

Rule	Description	Violation Example
No fabricated statistics	Do not invent percentages or statistical figures	❌ "Success rate 87.3%" (no source)
No fabricated drugs	Only list real medications with correct dosages	❌ Inventing non-existent brand names
No fabricated guidelines	Only cite real, existing guidelines	❌ "ACOG PB #999" (doesn't exist)
No fabricated studies	Do not create fake author names/years	❌ "According to Smith et al. (2024)..." (fabricated)
When uncertain → state clearly	Write "Data varies..." or use ranges	✅ "Rate ~10-20% (varies by study)"

Safe Phrases When Exact Data Is Missing

✅ "Rate ranges from X-Y% depending on the study"
✅ "Based on available literature, estimated at approximately..."
✅ "Local data is limited; international reports indicate..."
✅ "Further research needed to confirm"
❌ "Exactly XX.X%" (when no specific source exists)

---

4. Clinical Safety Markers — Medical Markup

Each pathology article (T2) MUST include all 4 markers:

Marker	Purpose	Format
> 💊 CLINICAL NOTE	Most important practice consideration	Blockquote
> ⚠️ RED FLAG	When urgent intervention / referral is needed	Blockquote
> 📚 REFERENCE	Primary guideline for this topic	Blockquote
> ⚕️ DISCLAIMER	Specialist consultation required	Blockquote at end

Mandatory Disclaimer (end of each article)

> ⚕️ **DISCLAIMER:** This article is for medical reference purposes only.
> All diagnostic and treatment decisions **require specialist physician guidance**.
> Do not self-apply protocols without specific clinical evaluation.

---

5. Cross-Reference Protocol

Inter-Article Links

• When mentioning a pathology/technique with its own article → MUST include article code
• Format: → See details: [VSN-04] Endometriosis
• Section ## CROSS-LINKS must list all related articles

ICD-10

• Pathology articles (T2) with ICD-10 codes → MUST include in header
• Format: > **ICD-10:** N97.0 | **Group:** VSN

---

6. Medical Writing Audit — 8-Dimension Evaluation

When reviewing medical articles, score across 8 dimensions (1-10):

Dimension	Question	Passing Standard
1. Evidence Quality	Are LoE markers on key recommendations?	≥ 3 LoE markers
2. Citation Depth	Are specific guidelines cited?	≥ 3 different sources
3. Clinical Accuracy	Are drug doses, metrics correct?	No medical errors
4. Structure	Correct template, all sections present?	100% sections with content
5. ICD-10 Compliance	Correct ICD-10 code present?	Header has ICD-10
6. Safety Markers	Red Flag, Disclaimer present?	4/4 markers
7. Cross-References	Links to related articles?	≥ 2 cross-links
8. Word Count	Meets minimum length?	T1≥1500, T2≥2000, T3≥1500

Scoring:

• 72-80: Excellent — publish ready
• 56-71: Good — minor revisions needed
• 40-55: Average — needs more evidence
• < 40: Fail — rewrite

---

7. Template Enhancement Rules

When creating new articles, inject these additional requirements into the prompt:

For Pathology Articles (T2)

EVIDENCE-BASED REQUIREMENTS:
1. Each treatment recommendation includes [LoE: I-V] + source
2. "DIAGNOSIS" section must have a test table with "Source Guideline" column
3. "TREATMENT" section must cite specific protocols (ACOG/MoH/ASRM)
4. End of article has "## REFERENCES" listing ≥ 3 guidelines
5. End of article has medical DISCLAIMER
6. DO NOT fabricate statistics — use estimated ranges when exact data is unavailable

For Technique Articles (T3)

EVIDENCE-BASED REQUIREMENTS:
1. Each procedure step cites source protocol (ASRM/ESHRE/MoH)
2. Drug tables must have a "Guideline" column
3. Success rates must cite source + confidence interval
4. End of article has "## REFERENCES" ≥ 2 guidelines
5. End of article has medical DISCLAIMER

---

Verification Script

Run after article generation to check quality:

python3 ~/.gemini/antigravity/skills/medical-research/scripts/evidence_checker.py <output_dir>

Script checks: References, LoE markers, ICD-10, Disclaimer, word count, cross-links.

---

Quick Diagnostic

Question	If NO	Action
Has REFERENCES section?	Article lacks sources	Add ≥ 3 guidelines
Has LoE for recommendations?	Unclear reliability	Assign LoE I-V
Has ICD-10 in header?	Missing coding	Add ICD-10
Has RED FLAG?	Missing clinical warning	Add ⚠️ marker
Has DISCLAIMER?	Legal risk	Add disclaimer at end
Do statistics have sources?	Suspected hallucination	Convert to estimated ranges