← 返回 Skills 市场
tlb1201

disease-cea-auto

作者 tlb1201 · GitHub ↗ · v1.0.1 · MIT-0
cross-platform ⚠ suspicious
78
总下载
0
收藏
0
当前安装
2
版本数
在 OpenClaw 中安装
/install disease-cea-auto
功能描述
疾病药物经济学自动评价 Skill — 对任意指定疾病,自动设计适合的 Markov / 决策树模型框架, 联网遴选当前最常用治疗药物,搜索模型参数(有效率、AE率、效用值、费用等), 以中国最新人均 GDP(1倍)为 QALY 支付阈值,计算每种药物的增量成本效果比(ICER)与 货币化净收益(NMB),从大到...
使用说明 (SKILL.md)

\r \r \x3C!-- ============================================================\r SKILL: disease-cea-auto · v1.0 · 2026-04-27\r Disease-Specific Pharmacoeconomic Auto-Evaluation\r 中文 / English Bilingual Skill\r ============================================================ -->\r \r

Disease-Specific Pharmacoeconomic Auto-Evaluation Skill\r

疾病药物经济学自动评价 Skill\r

\r ---\r \r

概述 / Overview\r

\r 本 Skill 帮助你对任意指定疾病完成一次端到端的药物经济学评价:\r 自动确定模型框架 → 遴选主流药物 → 联网搜索参数 → 运行成本效果分析 →\r 以我国最新人均 GDP(1倍)为支付阈值计算货币化净收益(NMB)→ 排序输出报告与 Python 代码。\r \r This Skill performs an end-to-end pharmacoeconomic evaluation for any specified disease:\r auto-design model → select key drugs → web-search parameters → run CEA →\r compute NMB using China's latest GDP per capita (1×) as WTP threshold → rank and report.\r \r ---\r \r

执行流程 / Execution Workflow\r

\r

阶段一:模型框架设计 / Phase 1 — Model Design\r

\r 中文指令:\r

  1. 根据用户输入的疾病名称,判断疾病是慢性进展性(chronic/progressive) 还是\r 急性/治愈性(acute/curative):\r
    • 慢性进展性疾病 → 使用 Markov 模型(状态:通常含疾病缓解期、进展期、重度/终末期、死亡)\r
    • 急性/治愈性疾病 → 使用 决策树模型(分支:治疗成功、治疗失败、不良反应)\r
    • 如同时存在急性发作和长期管理(如哮喘、心血管病) → 混合模型\r
  2. 明确说明模型的健康状态定义、循环周期(cycle length)、时间范围(time horizon)、\r 贴现率(discount rate),并解释设定依据。\r
  3. 以中英文表格列出模型参数清单(见阶段二)。\r \r English Instructions:\r
  4. Based on the disease provided, classify as chronic/progressive or acute/curative:\r
    • Chronic/progressive → Markov model (states: typically remission, mild, moderate, severe, death)\r
    • Acute/curative → Decision tree (branches: success, failure, AE)\r
    • Mixed (acute exacerbations + long-term, e.g., asthma, CVD) → Hybrid model\r
  5. State the model's health states, cycle length, time horizon, discount rate, with justification.\r
  6. List all required parameters in a bilingual table (see Phase 2).\r \r ---\r \r

阶段二:药物遴选 / Phase 2 — Drug Selection\r

\r 中文指令:\r

  1. 使用 web_search 联网搜索该疾病当前国内外最常用/一线/二线治疗药物,\r 参考来源:中国临床指南、国家医保目录、UpToDate、PubMed、药智网等。\r
  2. 遴选标准:优先纳入①中国医保目录内药物;②国内外指南推荐的一线/二线药物;\r ③近5年上市或获批的代表性新药(如有)。\r
  3. 遴选数量:不超过20种代表性药物/方案,确保覆盖不同作用机制和费用区间。\r
  4. 以表格输出:药物名称(中英文)、适应症、作用机制、是否医保、上市年份。\r \r English Instructions:\r
  5. Use web_search to find current first-line/second-line drugs for the disease,\r referencing Chinese clinical guidelines, NRDL, UpToDate, PubMed, etc.\r
  6. Selection criteria: ① NRDL-listed drugs; ② guideline-recommended drugs;\r ③ representative new drugs approved in the last 5 years (if any).\r
  7. Target no more than 20 representative drugs/regimens covering different mechanisms and cost ranges.\r
  8. Output as a bilingual table: drug name (CN/EN), indication, mechanism, NRDL status, approval year.\r \r ---\r \r

阶段三:参数搜索 / Phase 3 — Parameter Search\r

\r 中文指令:\r 对每种遴选药物,使用 web_search 搜索以下参数(每个参数均需注明文献来源):\r \r | 参数类型 | 说明 | 优先来源 |\r |----------|------|----------|\r | 临床疗效 | 有效率、ORR、PFS、OS(适用时)| RCT、Meta分析 |\r | 效用值(utility)| 各健康状态下的 QoL 权重(0-1)| EQ-5D 研究 |\r | 药物费用 | 年均药品费用(元)| 国家医保谈判价、药智网、公立医院价格 |\r | 疾病管理费用 | 门诊/住院/辅助检查费用(元/年)| 国内成本测算研究 |\r | 不良反应率及处理费用 | 3/4级 AE 发生率及对应费用 | RCT、安全性数据 |\r | 转换概率(Markov) | 各状态间年转换概率 | RCT、自然史研究 |\r \r 若某参数无直接文献支撑,优先参考同类药物或同类疾病研究,并标注"外推"。\r \r English Instructions:\r For each selected drug, use web_search to retrieve (cite every source):\r \r | Parameter | Description | Priority Source |\r |-----------|-------------|-----------------|\r | Clinical efficacy | Response rate, ORR, PFS, OS | RCT, meta-analysis |\r | Utility values | QoL weights per health state (0–1) | EQ-5D studies |\r | Drug cost | Annual drug cost (CNY) | NRDL negotiated price |\r | Disease management cost | Outpatient/inpatient/diagnostic (CNY/yr) | Chinese cost studies |\r | AE rate & cost | Grade 3/4 AE rate and management cost | RCT safety data |\r | Transition probabilities | Annual transition probs between states | RCT, natural history |\r \r If a parameter lacks direct evidence, extrapolate from analogous drugs/diseases and label as "extrapolated."\r \r ---\r \r

阶段四:人均 GDP 获取 / Phase 4 — WTP Threshold (GDP per Capita)\r

\r 中文指令:\r

  1. 使用 web_search 搜索"中国最新人均 GDP"(优先查国家统计局最新年度数据,通常在每年1月公布)。\r
  2. 搜索词示例:中国 2024 人均GDP 国家统计局\r
  3. 1倍人均 GDP 作为 QALY 的货币化支付阈值(WTP)。\r
  4. 在报告中明确注明:数据来源、统计年份、具体数值(元/人/年)。\r \r English Instructions:\r
  5. Use web_search to find "China latest GDP per capita" (prefer NBS official annual data).\r
  6. Sample query: China 2024 GDP per capita National Bureau of Statistics\r
  7. Use 1× GDP per capita as the WTP threshold for QALY valuation.\r
  8. Report: data source, year, exact value (CNY/person/year).\r \r ---\r \r

阶段五:成本效果分析与 NMB 计算 / Phase 5 — CEA & NMB Calculation\r

\r 中文指令:\r

  1. 标准治疗或安慰剂作为参照方案(比较组)。\r
  2. 对每种药物计算:\r
    • 增量成本(ΔC) = 干预组总成本 - 对照组总成本\r
    • 增量效益(ΔE) = 干预组总 QALY - 对照组总 QALY\r
    • ICER = ΔC / ΔE(元/QALY)\r
    • 货币化净收益(NMB) = ΔE × WTP - ΔC(元)\r
  3. 按 NMB 从大到小排列所有药物(NMB>0 表示具有成本效果,\x3C0 则不具有)。\r
  4. 同时进行单因素敏感性分析(至少对效用值、药物费用、转换概率各做±20%变动)。\r \r English Instructions:\r
  5. Use standard of care or placebo as the comparator.\r
  6. For each drug, compute:\r
    • Incremental cost (ΔC) = Total cost (intervention) − Total cost (comparator)\r
    • Incremental effectiveness (ΔE) = Total QALY (intervention) − Total QALY (comparator)\r
    • ICER = ΔC / ΔE (CNY/QALY)\r
    • Net Monetary Benefit (NMB) = ΔE × WTP − ΔC (CNY)\r
  7. Rank all drugs by NMB descending (NMB > 0 = cost-effective; \x3C 0 = not cost-effective).\r
  8. Perform one-way sensitivity analysis (±20% on utility values, drug costs, transition probabilities).\r \r ---\r \r

阶段六:Python 代码输出 / Phase 6 — Python Code Output\r

\r 中文指令:\r 根据前述参数,生成完整可运行的 Python 代码,要求:\r

  • 使用 pandasnumpymatplotlib 标准库\r
  • 代码结构:① 参数定义模块;② Markov/决策树计算模块;③ ICER/NMB 计算模块;④ 排序与可视化模块\r
  • 生成两张图:① 成本效果平面散点图(CE plane);② NMB 条形图(按大到小排序)\r
  • 代码中所有变量名和注释使用中英文双语(变量名英文,注释中英文并行)\r
  • 代码末尾调用 print 输出汇总结果表格\r \r English Instructions:\r Generate complete, runnable Python code based on the parameters collected, with:\r
  • Libraries: pandas, numpy, matplotlib\r
  • Structure: ① Parameter definition; ② Markov/decision-tree computation; ③ ICER/NMB calculation; ④ Ranking & visualization\r
  • Two figures: ① CE plane scatter plot; ② NMB bar chart (descending order)\r
  • All variable names in English; comments bilingual (CN+EN parallel)\r
  • Final print output of summary table\r \r ---\r \r

阶段七:报告输出 / Phase 7 — Scientific Report Output\r

\r 中文指令:\r 按以下科学论文格式输出简明结果报告(每段内容均中英文并行):\r \r

# [疾病名称] 多药成本效果分析报告\r
# Cost-Effectiveness Analysis Report for [Disease Name] — Multiple Drugs\r
\r
## 1. 研究背景 / Background\r
## 2. 研究方法 / Methods\r
   2.1 模型结构 / Model Structure\r
   2.2 研究视角与时间范围 / Perspective & Time Horizon\r
   2.3 数据来源 / Data Sources\r
   2.4 支付阈值 / WTP Threshold\r
## 3. 参数汇总表 / Parameter Summary Table(中英文表头)\r
## 4. 结果 / Results\r
   4.1 基础分析 / Base-Case Results(NMB排序表 + ICER表)\r
   4.2 敏感性分析 / Sensitivity Analysis\r
## 5. 结论与政策建议 / Conclusions & Policy Implications\r
## 6. 参考文献 / References\r
```\r
\r
**English Instructions:**\r
Output a concise scientific report in the above structure,\r
with **every section bilingual (CN+EN parallel paragraphs)**.\r
\r
---\r
\r
## 参数默认值 / Default Settings\r
\r
| 参数 / Parameter | 默认值 / Default | 说明 / Note |\r
|------------------|-----------------|-------------|\r
| 贴现率 Discount rate | 5% per year | 中国药物经济学指南推荐 / Chinese guideline |\r
| 循环周期 Cycle length | 1 year (chronic) / per episode (acute) | 依疾病类型调整 |\r
| 时间范围 Time horizon | 10–20 years (chronic) / 1–5 years (acute) | 依疾病类型调整 |\r
| 研究视角 Perspective | 卫生体系视角 / Healthcare system | 含直接医疗费用 |\r
| WTP 阈值 WTP threshold | 1× China GDP per capita (最新值,联网获取) | 依 Phase 4 实时获取 |\r
| 敏感性分析范围 SA range | ±20% on key parameters | 单因素 one-way |\r
\r
---\r
\r
## 质量控制要求 / Quality Control\r
\r
**中文:**\r
- 每个参数来源必须注明(作者、年份、期刊/数据库)\r
- 若参数为外推或假设,必须标注并在敏感性分析中重点测试\r
- NMB 排序表必须包含置信区间或不确定性说明\r
- Python 代码必须可直接运行,不依赖外部私有数据文件\r
\r
**English:**\r
- Every parameter must be cited (author, year, journal/database)\r
- Extrapolated/assumed parameters must be labeled and prioritized in SA\r
- NMB ranking table must include confidence intervals or uncertainty notes\r
- Python code must run standalone without private external data files\r
\r
---\r
\r
## 示例触发 / Example Triggers\r
\r
- "帮我做2型糖尿病的药物经济学评价"\r
- "对比IPF常用药物的成本效果"\r
- "肺癌靶向药的ICER和NMB计算"\r
- "哪种NSCLC一线治疗方案净收益最高"\r
- "Do a multi-drug CEA for COPD"\r
- "Rank asthma biologics by NMB"\r
- "Compare cost-effectiveness of HER2+ breast cancer therapies"\r
\r
---\r
\r
*Skill version: 1.0 | 创建日期 / Created: 2026-04-27 | 作者 / Author: TLB*\r
安全使用建议
This skill appears to do what it says: it auto-builds CEA models, uses web searches to pull literature and price data, and runs calculations using the included Python template. Before installing/using: (1) do not include any patient-identifiable or sensitive health data in prompts — the skill requires live web searches that may transmit prompt text to external services; (2) confirm the web sources the skill cites (NRDL, NBS, RCTs) and validate any extrapolated parameters — automated scraping can be incomplete or pick low-quality sources; (3) review the filled-in parameters in the generated Python code before running it (to ensure correct units, year, discounting, comparator choice); (4) be aware the script requires numpy/pandas/matplotlib to run locally but the skill itself does not declare installs; install those libraries in a trusted environment if you plan to execute code. If you need stricter privacy, run the analysis locally with vetted parameter inputs rather than letting the agent perform web searches.
功能分析
Type: OpenClaw Skill Name: disease-cea-auto Version: 1.0.1 The skill bundle 'disease-cea-auto' automates pharmacoeconomic evaluations by instructing an AI agent to perform extensive web searches for clinical data and drug pricing, and subsequently generate and execute Python code for Markov or decision tree modeling. While the workflow is logically sound and aligned with its stated scientific purpose, the requirement for the agent to perform broad, automated network requests (web_search) and dynamic code generation/execution represents a significant attack surface and high-risk capability set. Per the provided criteria, these risky capabilities are classified as suspicious even when plausibly needed for the stated purpose (SKILL.md and scripts/cea_analysis.py).
能力评估
Purpose & Capability
Name/description match the actual behavior: SKILL.md instructs automated model design, web-based parameter retrieval, CEA calculations and output of Python code/report. The included Python template (cea_analysis.py) implements Markov/decision-tree calculations consistent with that purpose. No unrelated credentials, binaries, or system access are requested.
Instruction Scope
Runtime instructions explicitly require using web_search to find drugs, RCTs, utilities, costs and the latest China GDP per capita and to cite sources. The instructions do not ask the agent to read system files or environment variables. Caution: because the skill performs live web searches, any user-provided sensitive or patient-identifying information included in prompts could be sent to external sites via the search tool — the SKILL.md does not admonish against including PHI.
Install Mechanism
No install spec is present; this is instruction-plus-template code only. The supplied Python script is local and does not perform downloads or execute remote code. This lowers installation risk.
Credentials
The skill declares no required environment variables, credentials, or config paths. The code does not read environment variables or external secrets. Requested resources (web searches for literature, GDP, drug prices) are proportionate to the pharmacoeconomics use case.
Persistence & Privilege
always:false and normal model invocation are set. The skill does not request elevated persistence or modify other skills or system-wide settings.
如何使用
  1. 确保已安装 OpenClaw(本地或 Docker 部署)
  2. 在对话框中输入安装命令:/install disease-cea-auto
  3. 安装完成后,直接呼叫该 Skill 的名称或使用 /disease-cea-auto 触发
  4. 根据 Skill 的参数说明提供必要输入,即可获得结构化输出
版本历史
v1.0.1
- Increased the maximum number of representative drugs/regimens to 20 in the drug selection phase, allowing broader inclusion. - Updated instructions to reflect the new limit (“不超过20种”/“no more than 20”) in both Chinese and English guidance. - No other functional or descriptive changes were made.
v1.0.0
disease-cea-auto v1.0.0 - Initial release. - Automatically designs a Markov or decision tree pharmacoeconomic model for any specified disease. - Identifies and selects representative treatment drugs via web search, prioritizing those in Chinese guidelines and NRDL. - Searches and sources key model parameters (efficacy, utility values, costs, AE rates, etc.) with cited references. - Uses China’s latest per capita GDP (1×) as the willingness-to-pay (WTP) threshold for QALY. - Calculates and ranks incremental cost-effectiveness ratio (ICER) and net monetary benefit (NMB) for each drug. - Outputs both full Python code (with bilingual comments) and a scientific paper–style bilingual report.
元数据
Slug disease-cea-auto
版本 1.0.1
许可证 MIT-0
累计安装 0
当前安装数 0
历史版本数 2
常见问题

disease-cea-auto 是什么?

疾病药物经济学自动评价 Skill — 对任意指定疾病,自动设计适合的 Markov / 决策树模型框架, 联网遴选当前最常用治疗药物,搜索模型参数(有效率、AE率、效用值、费用等), 以中国最新人均 GDP(1倍)为 QALY 支付阈值,计算每种药物的增量成本效果比(ICER)与 货币化净收益(NMB),从大到... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 78 次。

如何安装 disease-cea-auto?

在 OpenClaw 或 Claude Code 对话框中运行命令「/install disease-cea-auto」即可一键安装,无需额外配置。

disease-cea-auto 是免费的吗?

是的,disease-cea-auto 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。

disease-cea-auto 支持哪些平台?

disease-cea-auto 跨平台运行,可在任意部署了 OpenClaw / Claude Code 的环境中使用(cross-platform)。

谁开发了 disease-cea-auto?

由 tlb1201(@tlb1201)开发并维护,当前版本 v1.0.1。

💬 留言讨论