drug-drug

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Drug-Drug Interaction (DDI) Evidence-Based Assessment

药物相互作用（DDI）循证评估

Overview

概述

This skill implements a Retrieval-Augmented Generation (RAG) approach to drug-drug interaction assessment, modeled after the Micromedex Drug-Reax System methodology. Rather than relying on LLM training data alone -- which may be outdated, incomplete, or hallucinated -- every claim in the output is grounded in real-time literature retrieval from peer-reviewed sources with full citation traceability. The structured report mirrors Micromedex Drug-Reax classification grades and is designed to complement (not replace) certified drug information systems and clinical pharmacist expertise.

本技能采用检索增强生成（RAG）方法进行药物相互作用评估，参考Micromedex Drug-Reax System方法论构建。它不依赖大语言模型（LLM）的训练数据——这些数据可能过时、不完整或存在幻觉——输出中的每一项结论都基于从同行评审来源实时检索的文献，且具备完整的引用可追溯性。结构化报告与Micromedex Drug-Reax的分类等级一致，旨在补充（而非替代）认证药物信息系统和临床药师的专业知识。

Trigger

触发条件

```
/drug-drug DrugA DrugB
```
Any query mentioning two drug names and asking about interaction, safety, or concomitant use
Keywords: DDI, interaction, contraindication, concomitant, 交互作用, 併用

```
/drug-drug DrugA DrugB
```
任何提及两种药物名称并询问相互作用、安全性或联合使用的查询
关键词：DDI、interaction、contraindication、concomitant、交互作用、併用

Workflow (execute strictly in order)

工作流程（严格按顺序执行）

Step 0: Drug Name Resolution

步骤0：药物名称解析

Extract two drug names from user input. If the user provides brand names, resolve to INN/generic names. If only one drug is provided or names are ambiguous, ask the user to clarify.

从用户输入中提取两种药物名称。若用户提供品牌名，需解析为国际非专利名称（INN）/通用名。若仅提供一种药物或名称模糊，需请求用户澄清。

Step 1: Literature Search

步骤1：文献检索

Execute at least 5 distinct searches to ensure adequate coverage:

PubMed Search (via web_search)

site:pubmed.ncbi.nlm.nih.gov "DrugA" "DrugB" interaction

site:pubmed.ncbi.nlm.nih.gov "DrugA" "DrugB" pharmacokinetic

site:pubmed.ncbi.nlm.nih.gov "DrugA" "DrugB" CYP450

CrossRef Search (via web_fetch)

https://api.crossref.org/works?query=DrugA+DrugB+drug+interaction&rows=10&sort=relevance

General WebSearch

DrugA DrugB drug interaction clinical significance

DrugA DrugB interaction mechanism CYP enzyme

DrugA DrugB interaction case report adverse event

FDA Label / DailyMed

site:dailymed.nlm.nih.gov DrugA interaction

DrugA DrugB FDA drug interaction warning

PubMed MCP Tool (if PubMed MCP server is connected)

Use

PubMed:search_articles

with

"DrugA" AND "DrugB" AND "drug interaction"

Use
```
PubMed:get_article_metadata
```
for key articles

执行至少5次独立检索以确保覆盖全面：

PubMed检索（通过web_search）

site:pubmed.ncbi.nlm.nih.gov "DrugA" "DrugB" interaction

site:pubmed.ncbi.nlm.nih.gov "DrugA" "DrugB" pharmacokinetic

site:pubmed.ncbi.nlm.nih.gov "DrugA" "DrugB" CYP450

CrossRef检索（通过web_fetch）

https://api.crossref.org/works?query=DrugA+DrugB+drug+interaction&rows=10&sort=relevance

通用网络检索

DrugA DrugB drug interaction clinical significance

DrugA DrugB interaction mechanism CYP enzyme

DrugA DrugB interaction case report adverse event

FDA标签/DailyMed检索

site:dailymed.nlm.nih.gov DrugA interaction

DrugA DrugB FDA drug interaction warning

PubMed MCP工具（若已连接PubMed MCP服务器）
- 使用
```
PubMed:search_articles
```
  ，检索词为
```
"DrugA" AND "DrugB" AND "drug interaction"
```
- 使用
```
PubMed:get_article_metadata
```
  获取关键文献的元数据

Step 2: Evidence Appraisal

步骤2：证据评估

Read

references/evidence-grading.md

for complete grading criteria.

For each retrieved article, extract and document:

Study design (RCT / cohort / case-control / case report / in vitro / review)
Sample size
Key findings (AUC fold-change, Cmax change, clinical events)
Interaction mechanism described
Credibility assessment

Traceability requirement: Every factual claim in the final report must be attributable to a specific source with PMID or DOI. If a claim is inferred from pharmacologic reasoning rather than direct evidence, it must be explicitly labeled as such (e.g., "based on known CYP3A4 inhibition profile" rather than stated as established fact).

Uncertainty signaling: When evidence is limited or conflicting, the report must explicitly state the level of uncertainty. Use phrases such as "limited evidence suggests," "based on case reports only," or "no direct clinical studies available; assessment based on pharmacologic reasoning." Never present uncertain inferences with the same confidence as RCT-level evidence.

阅读

references/evidence-grading.md

获取完整的分级标准。

针对每篇检索到的文献，提取并记录：

研究设计（随机对照试验RCT / 队列研究 / 病例对照研究 / 病例报告 / 体外研究 / 综述）
样本量
关键发现（AUC倍数变化、Cmax变化、临床事件）
描述的相互作用机制
可信度评估

可追溯性要求：最终报告中的每一项事实性结论都必须可归因于特定来源（含PMID或DOI）。若结论是基于药理学推理而非直接证据得出，必须明确标注（例如：“基于已知CYP3A4抑制特性”，而非作为既定事实陈述）。

不确定性提示：当证据有限或存在冲突时，报告必须明确说明不确定性程度。使用诸如“有限证据表明”“仅基于病例报告”或“无直接临床研究可用；基于药理学推理进行评估”等表述。绝不能将不确定的推断与RCT级证据以相同的可信度呈现。

Step 3: Structured Classification

步骤3：结构化分类

Classify using the Micromedex Drug-Reax framework. See

references/evidence-grading.md

for detailed decision trees.

参考Micromedex Drug-Reax框架进行分类。详细决策树见

references/evidence-grading.md

。

3a. Severity

3a. 严重程度

Grade	Definition
Contraindicated	Drugs are contraindicated for concurrent use
Major	Interaction may be life-threatening and/or require medical intervention to minimize or prevent serious adverse effects
Moderate	Interaction may result in exacerbation of the patient's condition and/or require a change in therapy
Minor	Interaction would have limited clinical effects; may augment side effects but generally does not require a change in therapy

等级	定义
Contraindicated（禁忌）	药物禁止联合使用
Major（严重）	相互作用可能危及生命，且/或需要医疗干预以减少或预防严重不良反应
Moderate（中等）	相互作用可能导致患者病情恶化，且/或需要调整治疗方案
Minor（轻微）	相互作用的临床影响有限；可能加重副作用，但通常无需调整治疗方案

3b. Documentation

3b. 证据等级

Grade	Definition
Excellent	Controlled studies have clearly established the existence of the interaction
Good	Documentation strongly suggests the interaction exists, but well-controlled studies are lacking
Fair	Available documentation is poor, but pharmacologic considerations lead clinicians to suspect the interaction exists; or documentation is good for a pharmacologically similar drug
Poor	Documentation is very limited, e.g., only isolated case reports or theoretical rationale
Unlikely	No reasonable pharmacologic basis for the interaction

等级	定义
Excellent（优秀）	对照研究已明确证实相互作用存在
Good（良好）	文献强烈提示相互作用存在，但缺乏充分的对照研究
Fair（一般）	现有文献质量较差，但药理学因素使临床医生怀疑相互作用存在；或针对药理学相似药物的文献质量良好
Poor（较差）	文献非常有限，例如仅孤立病例报告或理论依据
Unlikely（不可能）	无合理的药理学依据支持相互作用存在

3c. Onset

3c. 发作时间

Grade	Definition
Rapid	Clinical effects of the interaction occur within 24 hours
Delayed	Clinical effects of the interaction occur after 24 hours
Not specified	Onset not clearly documented in the literature

等级	定义
Rapid（快速）	相互作用的临床效应在24小时内出现
Delayed（延迟）	相互作用的临床效应在24小时后出现
Not specified（未明确）	文献中未明确记录发作时间

3d. Mechanism

3d. 作用机制

Classify the interaction mechanism into:

Pharmacokinetic (PK):

CYP450 enzyme inhibition (specify isoform: CYP3A4, CYP2D6, CYP2C19, CYP2C9, CYP1A2, etc.)
CYP450 enzyme induction
P-glycoprotein (P-gp) / transporter-related
Protein binding displacement
Renal tubular secretion competition
Absorption-level (pH alteration, chelation, etc.)

Pharmacodynamic (PD):

Additive effect (e.g., QTc prolongation, bleeding risk, CNS depression)
Synergistic effect
Antagonistic effect

将相互作用机制分为以下类别：

药代动力学（PK）：

CYP450酶抑制（注明亚型：CYP3A4、CYP2D6、CYP2C19、CYP2C9、CYP1A2等）
CYP450酶诱导
P-糖蛋白（P-gp）/转运体相关
蛋白结合置换
肾小管分泌竞争
吸收层面（pH改变、螯合作用等）

药效动力学（PD）：

叠加效应（例如QTc间期延长、出血风险、中枢神经系统抑制）
协同效应
拮抗效应

Step 4: Report Generation

步骤4：报告生成

Produce the final report in the following format:

markdown

undefined

按照以下格式生成最终报告：

markdown

undefined

Drug-Drug Interaction Assessment Report

药物相互作用评估报告

Drug Pair

药物组合

Drug A: [Generic Name] ([Brand Names])
Drug B: [Generic Name] ([Brand Names])

药物A： [通用名]（[品牌名]）
药物B： [通用名]（[品牌名]）

Structured Classification

结构化分类

Parameter	Grade	Note
Severity	[Grade]	[note]
Onset	[Grade]	[note]
Documentation	[Grade]	[note]

参数	等级	备注
严重程度	[等级]	[备注]
发作时间	[等级]	[备注]
证据等级	[等级]	[备注]

Interaction Effect

相互作用效应

[Description of the clinical effect of the interaction — what happens when these drugs are used together]

[描述相互作用的临床效应——联合使用这些药物会产生什么结果]

Clinical Management

临床管理建议

[Specific clinical management recommendations:]

Whether to avoid concomitant use
Dose adjustments required
Monitoring parameters
Alternative drug suggestions

[具体临床管理建议：]

是否避免联合使用
是否需要调整剂量
监测参数
替代药物建议

Probable Mechanism

可能的作用机制

[Pharmacological mechanism — PK and/or PD, including specific enzymes, transporters, or receptors involved. Include PK data (AUC/Cmax changes) if available from studies.]

[药理学机制——PK和/或PD，包括涉及的特定酶、转运体或受体。若有研究提供的PK数据（AUC/Cmax变化），请一并包含。]

Evidence Sources

证据来源

[Key references: Author, Journal, Year, PMID/DOI]

[关键参考文献：作者、期刊、年份、PMID/DOI]

Disclaimer

免责声明

This report is generated by AI using a retrieval-augmented generation (RAG) approach and is intended as a clinical decision support aid only. It does not constitute medical advice and cannot replace certified drug information systems (Micromedex, Lexicomp, Clinical Pharmacology) or the judgment of qualified healthcare professionals. AI-generated content carries inherent limitations including potential for incomplete literature retrieval, misinterpretation of source data, and inability to account for individual patient factors. All clinical decisions must be made by licensed practitioners with access to complete patient information, institutional formulary policies, and current prescribing guidelines. When in doubt, consult a clinical pharmacist or drug information specialist.

---

本报告由AI通过检索增强生成（RAG）方法生成，仅作为临床决策支持工具。 不构成医疗建议，不能替代认证药物信息系统（Micromedex、Lexicomp、Clinical Pharmacology）或合格医疗专业人员的判断。AI生成内容存在固有局限性，包括可能的文献检索不完整、源数据解读错误，以及无法考虑个体患者因素。所有临床决策必须由获得执业许可的从业者做出，且需获取完整的患者信息、机构处方政策和当前处方指南。如有疑问，请咨询临床药师或药物信息专家。

---

Key Principles

核心原则

Retrieval before generation: Never rely on LLM training data alone. Always perform real-time literature search before generating any assessment. Training data may be outdated, incomplete, or reflect unverified sources (social media, blogs). Real-time retrieval from peer-reviewed databases ensures currency and reliability.
Multi-source cross-validation: Use at least PubMed + CrossRef + WebSearch (three sources). Cross-validate findings across sources to reduce the risk of hallucination or bias from any single retrieval.
Full citation traceability: Every factual claim must cite its source (PMID, DOI, or URL). This enables clinicians to fact-check AI-generated answers against the original sources -- a core requirement for explainable AI (XAI) in clinical settings.
Conservative grading: When evidence is insufficient, err on the side of higher severity and lower documentation grade (err on the side of caution). Patient safety takes precedence over precision.
Mechanism is king: Even without clinical studies, if the pharmacologic mechanism is clear (e.g., known CYP inhibitor + known CYP substrate), document it and assign at least Fair documentation.
Transparency of uncertainty: The report must clearly distinguish between evidence-supported facts, pharmacologic inferences, and areas of genuine uncertainty. Never present inferred information with the same confidence as controlled-study evidence.
Context matters: Note dose-dependent or population-specific differences when evidence supports them (e.g., high-dose vs. low-dose regimens may carry different risk profiles).
Complement, not replace: This tool supports clinical decision-making but does not substitute for professional judgment. The output is designed to be reviewed by a clinician, not acted upon autonomously.
Language: The report body defaults to English. If the user writes in another language, respond in that language but keep drug names, grade labels, and pharmacological terms in English.

先检索再生成：绝不单独依赖LLM训练数据。在生成任何评估前，必须执行实时文献检索。训练数据可能过时、不完整或来自未经验证的来源（社交媒体、博客）。从同行评审数据库实时检索可确保内容的时效性和可靠性。
多源交叉验证：至少使用PubMed + CrossRef + WebSearch三个来源。跨来源验证研究结果，以降低单一检索导致的幻觉或偏见风险。
完整引用可追溯性：每一项事实性结论都必须引用其来源（PMID、DOI或URL）。这使临床医生能够对照原始来源核实AI生成的答案——这是临床环境中可解释AI（XAI）的核心要求。
保守分级：当证据不足时，倾向于更高的严重程度和更低的证据等级（优先考虑谨慎性）。患者安全优先于准确性。
机制优先：即使没有临床研究，若药理学机制明确（例如已知CYP抑制剂 + 已知CYP底物），也需记录该机制并至少评定为Fair证据等级。
不确定性透明化：报告必须明确区分证据支持的事实、药理学推断和真正存在不确定性的领域。绝不能将推断信息与对照研究证据以相同的可信度呈现。
情境相关性：当证据支持时，需注明剂量依赖性或人群特异性差异（例如高剂量与低剂量方案可能具有不同的风险特征）。
补充而非替代：本工具支持临床决策，但不能替代专业判断。输出内容仅供临床医生参考，不能直接用于自主决策。
语言适配：报告主体默认使用英文。若用户使用其他语言提问，则以该语言回复，但药物名称、等级标签和药理学术语保留英文。

Reference Files

参考文件

```
references/evidence-grading.md
```
: Complete evidence grading criteria, decision trees, CYP450 quick reference, and common PD interaction patterns

```
references/evidence-grading.md
```
：完整的证据分级标准、决策树、CYP450速查指南和常见PD相互作用模式