人工智能加速 FDA 审查流程

 

作为一项开创性举措，FDA 正在利用生成式人工智能加快其审查流程。这一举措旨在提高效率，缩短评估新医疗器械和治疗方法所需的时间。

 

一位专业人士分享了他们的亲身经历：

 

"I understand why. It has been an absolute game changer. This technology lets me complete tasks in minutes that used to take days. I have spent more time than I would like to admit on developing my custom-built AI agents and tools to suit my workflow."

 

他们重点介绍了几个给他们的工作带来变革的人工智能使用案例：

 

1. 实质等同器械搜索：人工智能简化了对实质等同器械的搜索，使监管呈件更加高效。

 

2. 科学文献检索：人工智能擅长筛选大量科学文献，快速提供相关信息。

 

3. 510(k) 数据库分析：定制的人工智能代理可以浏览510(k) 数据库，查找人工查找耗时的特定情况。

 

4. 高级器械分类：人工智能驱动的工具不仅能对产品代码进行分类，还能帮助更好地组织和理解器械。

 

另一位专业人士强调了将人工智能融入更广泛的战略领域：

 

"Linking it with market access planning and reimbursement and revenue pathways. Regulatory crosswalks, optimal PCCP approaches to name a few."

 

一位评论者指出

 

"Wow, this is pretty cool! Last I heard, FDA reviewers were not allowed to use ChatGPT over fears of IP leakage. This makes sense, but why not for literature review? This is a use case I have found to be an extremely valuable use case for AI without disclosing confidential information to OpenAI and others."

 

"It's great to see people considering practical, non-sensitive applications of AI like this. Literature reviews could definitely benefit from AI's ability to quickly sift through vast amounts of data."

 

安全风险管理和威胁建模

 

在医疗器械开发过程中，人们越来越多地从威胁建模的角度来看待安全风险管理。AAMI TIR57 安全风险流程与威胁建模宣言中的四个关键问题完全一致：

 

1. 我们在做什么？

 

- 在安全风险管理计划中定义系统和范围。

 

2. 什么会出错？

 

- 推动风险分析和评估。

 

3. 我们该怎么办？

 

- 选择并实施风险控制措施，然后重新评估残余风险。

 

4. 我们做得足够好吗？

 

- 在风险管理报告中记录，并在生产和后期制作中持续监控。

 

美国食品和药物管理局现在要求在提供 ISO 14971 安全文档的同时提供单独的安全风险文档---没有捷径可走。

 

效率和清晰度：清晰的一对一映射可避免重复工作，并为审核人员提供他们所需的准确信息。

 

作为医疗器械的 SDK

 

另一种创新策略是将软件开发包(Software Development Kit, SDK)作为医疗器械部署。制造商不需要获得整套医疗器械的许可，而是将核心图像分析组件作为 SDK 清关，然后其他开发人员就可以将其集成进来。

 

优点

 

减轻验证负担：使用 SDK 的客户无需进行大量研究。

 

灵活性：允许更广泛地采用并集成到各种应用中。

 

该领域的专家强调：

 

"Rather than getting a completed medical device cleared, this manufacturer decided to just clear the core image analysis component instead. It would decrease the validation burden on the customer of this SDK because they would not need to run another 527 participant study."

 

"Actually, this is a market of products that are SDKs used by third-party apps like telemedicine. The question goes if the telemedicine app can be marketed without being cleared also itself. How the final interface is presented to the user is a factor that impacts overall performance of the combination."

 

"Indeed a very elegant way. Only nitpick is that I wouldn't call it SOUP since its provenance is very well known."

 

来源：MDR小能手

关键词： 人工智能 医疗器械