人工智能加速 FDA 審查流程

作為一項開創(chuàng)性舉措，FDA 正在利用生成式人工智能加快其審查流程。這一舉措旨在提高效率，縮短評估新醫(yī)療器械和治療方法所需的時間。

一位專業(yè)人士分享了他們的親身經歷：

"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."

他們重點介紹了幾個給他們的工作帶來變革的人工智能使用案例：

1. 實質等同器械搜索：人工智能簡化了對實質等同器械的搜索，使監(jiān)管呈件更加高效。

2. 科學文獻檢索：人工智能擅長篩選大量科學文獻，快速提供相關信息。

3. 510(k) 數據庫分析：定制的人工智能代理可以瀏覽510(k) 數據庫，查找人工查找耗時的特定情況。

4. 高級器械分類：人工智能驅動的工具不僅能對產品代碼進行分類，還能幫助更好地組織和理解器械。

另一位專業(yè)人士強調了將人工智能融入更廣泛的戰(zhàn)略領域：

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

一位評論者指出

"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."

安全風險管理和威脅建模

在醫(yī)療器械開發(fā)過程中，人們越來越多地從威脅建模的角度來看待安全風險管理。AAMI TIR57 安全風險流程與威脅建模宣言中的四個關鍵問題完全一致：

1. 我們在做什么？

- 在安全風險管理計劃中定義系統和范圍。

2. 什么會出錯？

- 推動風險分析和評估。

3. 我們該怎么辦？

- 選擇并實施風險控制措施，然后重新評估殘余風險。

4. 我們做得足夠好嗎？

- 在風險管理報告中記錄，并在生產和后期制作中持續(xù)監(jiān)控。

美國食品和藥物管理局現在要求在提供 ISO 14971 安全文檔的同時提供單獨的安全風險文檔---沒有捷徑可走。

效率和清晰度：清晰的一對一映射可避免重復工作，并為審核人員提供他們所需的準確信息。

作為醫(yī)療器械的 SDK

另一種創(chuàng)新策略是將軟件開發(fā)包(Software Development Kit, SDK)作為醫(yī)療器械部署。制造商不需要獲得整套醫(yī)療器械的許可，而是將核心圖像分析組件作為 SDK 清關，然后其他開發(fā)人員就可以將其集成進來。

優(yōu)點

減輕驗證負擔：使用 SDK 的客戶無需進行大量研究。

靈活性：允許更廣泛地采用并集成到各種應用中。

該領域的專家強調：

"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."

本文由廣州佳譽醫(yī)療器械有限公司/佛山浩揚醫(yī)療器械有限公司聯合編輯