Pharmaceutical Monitoring Plan Generator（药品药学监护计划生成器）

Overview

Generate two Word documents for any drug–population combination:
1. Comprehensive Monitoring Plan（完整版监护计划）— evidence-based, with priority tiers (P0–P3), drug info, baseline assessment, dynamic monitoring, references
2. Quick-Reference Handbook（临床速查手册）— bedside action card format, 7 modules, printable

Both documents include Vancouver-style numbered references verified against literature search results.

Required Inputs

Before starting, confirm (or ask the user for) these parameters:

Workflow（7-Step SOP）

Execute the following steps in order. Each step builds on the previous.

Step 1: Literature Search（文献检索）

Conduct systematic web searches to gather evidence for the drug–population pair. Search in both Chinese and English. Collect at minimum 10–15 references covering:

• RCTs / landmark trials — the highest-level evidence for this drug in this population
• Guidelines / expert consensus — Chinese (中华医学会/相关学组) and international (WHO, AAP, ESC, NICE, etc.)
• Systematic reviews / meta-analyses — Cochrane, JAMA, BMJ, specialty journals
• PK/PD studies — pharmacokinetics, therapeutic drug monitoring (TDM) data
• Safety / ADR literature — common adverse reactions, organ-specific effects
• Drug interaction studies — CYP metabolism, transporter interactions

Search queries (adapt to the specific drug):
1. "{drug_name_en}" "{population_en}" guidelines systematic review 2023 2024
2. "{drug_name_en}" pharmacokinetics "{population_en}" therapeutic drug monitoring
3. "{drug_name}" "{population}" 指南 专家共识 诊疗建议
4. "{drug_name_en}" adverse effects safety "{population_en}" review
5. "{drug_name}" "{population}" 药物相互作用 CYP

Verification rule: For each reference, verify existence via DOI, PubMed ID, or journal website URL. Never fabricate citations. If a reference cannot be verified, discard it and find a replacement.

Step 2: Evidence Synthesis & Priority Assignment（证据整合与优先级分级）

Organize all collected evidence into a structured knowledge base, then assign monitoring priority tiers using the "Frequency × Severity × Clinical Operability" framework:

Key principles:
- The most common medication error for this drug → P0
- The most frequent ADR with bedside-visible signs → P1
- TDM is P3 unless the drug has a narrow therapeutic index
- Any monitoring already covered by routine ICU/NICU care → do not duplicate; just reference it

Step 3: Generate Comprehensive Monitoring Plan Word Document（生成完整版）

Run scripts/generate_full_plan.py with a JSON data file containing the structured monitoring content. See references/template_data_format.md for the JSON schema.

The full-plan document includes these sections:
1. Priority tier legend (P0–P3)
2. Part 1: Pre-prescription checkpoints (P0)
3. Part 2: Dynamic monitoring — P1 (daily), P2 (periodic), P3 (as-needed)
4. Part 3: Drug information & standard dosing regimen
5. Part 4: Baseline assessment
6. Part 5: Key time-node assessments
7. Part 6: Discontinuation evaluation
8. Part 7: Priority overview table (quick reference card)
9. Part 8: Monitoring record template
9. Part 9: Patient/family education
10. References (Vancouver numbered format)

Step 4: Generate Quick-Reference Handbook Word Document（生成速查手册）

Run scripts/generate_handbook.py with the same JSON data file.

The handbook document includes 7 modules (action-oriented, bedside-ready):
1. Dosing decision quick-reference
2. Pre-prescription checklist (6–8 red-line items)
3. Daily monitoring action card (P1/P2/P3 rows, 5 columns: what to watch / how to check / normal / abnormal → action)
4. Key time nodes (5 decision points)
5. Drug interaction warning card
6. Bedside monitoring record form (single-page printable with signature area)
7. Family education talking points
8. References (abbreviated, same numbering as full plan)

Step 5: Reference Cross-Check（参考文献交叉校验）

After generating both documents, run scripts/verify_refs.py to:
1. Extract all [N] citation numbers from tables and paragraphs in both .docx files
2. Extract all reference entries from the end of both documents
3. Verify: every cited number has a matching reference, and every reference is cited at least once
4. Report any discrepancies and fix them

Step 5b: Package Insert Calibration（说明书校准）⭐ NEW

Purpose: After the monitoring plan is generated and references are verified, calibrate all monitoring content against the official drug package insert (PI / 药品说明书) to ensure there are no factual contradictions with the approved labeling.

Why this step matters:
- Monitoring plans integrate evidence from multiple sources (guidelines, RCTs, expert consensus) that may differ from or extend beyond the approved labeling.
- Any contradiction between the monitoring plan and the official PI must be explicitly labeled as 超说明书用药 with documented evidence and informed consent requirements.
- This protects clinical staff and patients, and ensures the plan is legally and medically defensible.

How to obtain the package insert:
1. NMPA official database (首选): https://www.nmpa.gov.cn/datasearch/home-index.html — search the exact drug name, click "说明书"
2. 药智网: https://db.yaozh.com/ — supports full-text PI search with manufacturer version filtering
3. 丁香园用药助手: https://drugs.dxy.cn/
4. For imported drugs (FDA label): https://labels.fda.gov/

Save the PI as a .txt or .docx file. Then run scripts/calibrate_pi.py:

python scripts/calibrate_pi.py \
    --data <json_path> \
    --pi <package_insert_txt_or_docx_path> \
    --output <report_docx_path>

If no PI file is available, run without --pi; the script will print an online search guide and generate a report with all 7 domains marked as NEEDS REVIEW (human verification required).

7 Calibration Domains:

Verdict interpretation:
- ✅ PASS — No discrepancy detected; proceed.
- ⚠️ WARNING — Difference found; may be evidence-based extension beyond PI. Must be reviewed by pharmacist + physician and labeled accordingly.
- ❌ CONFLICT — Direct contradiction with PI (e.g., contraindicated route used, prohibited population). Must be corrected before clinical use.
- 🔍 NEEDS REVIEW — No PI text provided; human review required.

Action rules after calibration:
- For each ⚠️ WARNING: add a footnote ※超说明书用药，依据：[文献来源] in the monitoring plan.
- For each ❌ CONFLICT: correct the monitoring plan immediately; do not deliver until resolved.
- The calibration report .docx (with sign-off table for pharmacist + physician) must be kept as a documentation artifact.

Important: This step is AI-assisted only. All flagged items must be reviewed and signed off by an experienced clinical pharmacist and attending physician before the plan is used clinically.

Step 6: Deliver & Summarize（交付与总结）

Present both files to the user via open_result_view and deliver_attachments. Provide a concise summary:
- Total references count and evidence level distribution
- Key P0 items (most critical safety checkpoints)
- Key P1 items (most common ADR / core efficacy monitoring)
- Notable clinical insights from the literature search

Script Usage

All scripts require python-docx (pip install python-docx).

generate_full_plan.py

python scripts/generate_full_plan.py --data <path_to_json> --output <output_docx_path>

generate_handbook.py

python scripts/generate_handbook.py --data <path_to_json> --output <output_docx_path>

verify_refs.py

python scripts/verify_refs.py <docx_path_1> <docx_path_2> ...

calibrate_pi.py（说明书校准，Step 5b）

# With PI file (recommended):
python scripts/calibrate_pi.py \
    --data <path_to_json> \
    --pi <package_insert.txt_or_docx> \
    --output <drug_name_校准报告.docx>

# Without PI file (generates NEEDS REVIEW report + prints online search guide):
python scripts/calibrate_pi.py --data <path_to_json>

Output: A Word report with 7-domain calibration table, suggested corrections, human review checklist, and sign-off block for pharmacist + physician.

Data Format

The JSON data file follows this schema (see references/template_data_format.md for full specification):

{
  "drug": {
    "name_cn": "药品中文名",
    "name_en": "English generic name",
    "category": "药理分类",
    "mechanism": "作用机制",
    "indication_primary": "主要适应证",
    "indication_secondary": ["次要适应证1", "次要适应证2"],
    "loading_dose": "负荷剂量描述",
    "maintenance_dose": "维持剂量描述",
    "dose_adjustments": [
      {"scenario": "情境", "adjustment": "调整方案"}
    ],
    "administration": [
      {"item": "给药途径/速度/溶媒等", "requirement": "要求"}
    ],
    "drug_interactions": [
      {"drug": "合并药物", "interaction": "相互作用", "action": "处置建议"}
    ],
    "dose_conversion_note": "剂量换算注意事项（如有）"
  },
  "population": {
    "name": "目标人群描述",
    "key_chars": ["特征1", "特征2"]
  },
  "monitoring": {
    "p0_prescription_checks": [
      {"item": "核查项", "risk": "常见错误/风险", "method": "核查方法", "refs": ["[1]","[2]"]}
    ],
    "p1_daily": [
      {"category": "监护类别", "items": [
        {"content": "具体监护内容", "frequency": "频率", "threshold": "触发阈值/正常标准", "action": "异常处置", "refs": ["[3]"]}
      ], "clinical_context": "简述该类监护的临床背景"}
    ],
    "p2_periodic": [
      {"category": "监护类别", "items": [
        {"content": "具体监护内容", "frequency": "频率", "threshold": "触发阈值", "action": "异常处置", "refs": ["[4]"]}
      ]}
    ],
    "p3_as_needed": [
      {"category": "监护类别", "items": [
        {"content": "具体监护内容", "frequency": "频率/时机", "threshold": "阈值/指征", "action": "处置", "refs": ["[5]"]}
      ], "note": "特别说明（如TDM不推荐常规）"}
    ],
    "time_nodes": [
      {"time": "时间点", "action": "做什么", "criteria": "判断标准", "next_step": "下一步"}
    ],
    "discontinuation": {
      "criteria": ["停药条件1", "停药条件2"],
      "warnings": ["不宜停药信号1"],
      "post_discontinuation": "停药后监护要点",
      "refs": ["[6]","[7]"]
    },
    "baseline_assessment": {
      "patient_info": [
        {"item": "评估项", "content": "内容", "significance": "临床意义"}
      ],
      "organ_function": [
        {"system": "器官系统", "indicators": "关注指标", "significance": "临床意义"}
      ]
    },
    "family_education": [
      {"topic": "沟通要点", "content": "通俗版内容"}
    ]
  },
  "references": [
    {"num": 1, "text": "Author A, Author B. Title. Journal. Year;Vol(Issue):Pages. DOI: xxx."},
    {"num": 2, "text": "中华医学会XX学组. XX共识. 中华XX杂志. Year;Vol(Issue):Pages."}
  ]
}

Quality Checklist

Before delivering, verify:
- [ ] All references are real and verifiable (DOI/PubMed URL confirmed)
- [ ] All [N] in text match reference list (verified by script)
- [ ] P0 items represent the highest-frequency medication errors for this drug
- [ ] P1 items are bedside-feasible with no extra cost
- [ ] TDM priority is justified by therapeutic index width
- [ ] Both documents share the same reference numbering
- [ ] Handbook fits within ~8 A4 pages with narrow margins
- [ ] [Step 5b] calibrate_pi.py has been run; no unresolved ❌ CONFLICT items remain
- [ ] [Step 5b] All ⚠️ WARNING items have been reviewed and labeled (超说明书用药注明依据)
- [ ] [Step 5b] Calibration report .docx generated and ready for pharmacist + physician sign-off
- [ ] [Step 5b] All D6 (special population) issues resolved — neonatal/off-label use explicitly labeled