人工智慧醫療器材的許可與給付之探討：以台灣和美國為例

Abstract

研究背景與目的：各國皆有越來越多的人工智慧（Artificial Intelligence, AI）/機器學習（Machine Learning, ML）醫療器材取得上市許可，鑒於美國為全球醫療器材的重要市場，與其相對成熟豐富的AI/ML新技術醫療器材給付經驗，本研究欲以台灣和美國為例，探討在台灣和美國AI/ML醫療器材之許可現況，以及給付收載情形與特性。 研究方法：由台灣和美國主管機關官方網站之公開資料，包括產品統計清單、工作報告和醫療器材許可證資料庫等，擷取出已許可且符合應用AI/ML技術的醫療器材，蒐集其醫療器材名稱、許可日期、風險等級、治療分類及來源等資訊分析所得結果，以簡單的描述性統計與視覺化圖表呈現已許可AI/ML醫療器材之品項特性。再由台灣和美國主管機關官方網站之公開資料及相關文獻研究，彙整並探討已許可AI/ML醫療器材之給付收載情形。 研究結果：截至2024年3月為止，台灣和美國分別有159項和882項AI/ML醫療器材經主管機關許可上市。台灣所有納入分析的AI/ML醫療器材皆為中風險的第二等級；美國納入分析的AI/ML醫療器材則有877項（99.4%）屬於中風險的第二等級，另有4項（0.5%）為高風險的第三等級和1項（0.1%）未分級。台灣和美國皆以放射學科學用裝置佔多數（台灣75.5%、美國76.1%）。給付收載方面，2023年7月台灣首款AI醫療器材正式納入健保特材給付，為一項心臟血管醫學科學使用的新式進階血液動力學及連續性血壓監測感應器；美國已有至少10項應用在放射學科學、心臟血管醫學科學、胃腸病科學及泌尿科學、眼科學等治療領域的AI/ML醫療器材，透過醫院門診前瞻性支付制度（Hospital Outpatient Prospective Payment System, HOPPS）的新技術門診支付代碼（Ambulatory Payment Codes, APCs）或健康照護常用處置代碼系統（Healthcare Common Procedure Coding System, HCPCS）、住院前瞻性支付制度（Inpatient Prospective Payment System, IPPS）的新技術附加支付（New Technology Add-on Payment, NTAP），或是聯邦醫療保險醫師費用支付標準表（Medicare Physician Fee Schedule, MPFS）被准予納入聯邦醫療保險（Medicare）的給付範圍。 結論：在台灣和美國已許可的AI/ML醫療器材皆以放射學科學的治療分類佔領導地位，風險分級以中風險的第二等級醫療器材為多數，並主要透過第二等級醫療器材查驗登記與510(k)的許可途徑上市。台灣已許可的AI/ML醫療器材中有超過半數同時在美國許可上市。在給付收載方面，台灣有1項、美國有10項AI/ML醫療器材納入保險給付，有無支持性證據及是否為病人帶來效益為關鍵評估因素。然而由於上市審查制度與給付考量面向不同，如有類似品第二等級醫療器材查驗登記與510(k)通常可免除臨床資料，若因此而未規劃執行臨床試驗和/或相關的效益研究，在申請給付時可能面臨相當大的挑戰。

Objective: More and more Artificial Intelligence (AI)/ Machine Learning (ML)-based medical devices are authorized for marketing in many countries. In view that the United States (U.S.) is an important market for medical devices globally and has relatively mature and rich experience in the benefits of medical devices with AI/ML-based new technology, this study intends to take Taiwan and the U.S as examples to explore the current state of AI/ML-based medical devices approved in Taiwan and the U.S. and the conditions and characteristics of benefits coverage. Method: The medical devices were screened from the open data on the Taiwan and U.S. official websites, including product statistics, working papers, license database, etc., and identified those which were approved and utilized AI or machine learning (ML) technology. The name of the medical device, approval date, classification, medical specialty and place of origin were collected for investigation via descriptive statistics and visualized diagrams. Furthermore, based on the open data on the Taiwan and U.S. official websites and related research, the approved AI/ML-based medical devices which have been covered in the healthcare system will be investigated for the benefit characteristics. Results: Until March 2024, there were 159 and 882 medical devices approved by Taiwan and the U.S. competent authorities, respectively. All Taiwan AI/ML-based medical devices investigated were Class II with medium risk. In all U.S. AL/ML-based medical devices investigated, 877 (99.4%) were Class II with medium risk, 4 (0.5%) were Class III with high risk, and the other 1 (0.1%) was unclassified. The majority of medical specialties in Taiwan and U.S. were radiology devices (75.5% in Taiwan; 76.1% in the U.S.). In terms of payment, the 1st AI-based medical device was officially granted the benefit payment in Taiwan in July 2023. The device was a cardiovascular device for a new advanced hemodynamic and continuous blood pressure monitor. In the U.S., there have been at least 10 AI/ML-based medical devices in the treatment areas such as radiology, cardiovascular, gastroenterology and urology, ophthalmology, etc., covered by the Medicare Ambulatory Payment Codes (APCs) or Healthcare Common Procedure Coding System (HCPCS) under the Hospital Outpatient Prospective Payment System (HOPPS), New Technology Add-on Payment (NTAP) under the Inpatient Prospective Payment System (IPPS) or Medicare Physician Fee Schedule (MPFS). Conclusion: Radiology plays a leading position in the medical specialties of the AI/ML-based medical devices approved in Taiwan and the U.S. Majority are classified as Class II medium-risk medical devices, and are mainly granted through Class II medical device registration or 510(k) clearance for marketing. More than half of AI/ML-based medical devices approved in Taiwan are also authorized for marketing in the U.S.. In terms of benefits of AI/ML-based medical devices, one from Taiwan and ten from the U.S. have been granted the coverage of benefits. The key evaluation factors are whether there is supporting evidence and whether it brings benefits to patients. However, due to the different nature of marketing authorization systems comparing to benefit considerations (for example, clinical data is usually exempted for Class II medical device registration and 510(k) review pathway), it may be quite challenging to apply for benefits if there is no plan to conduct clinical trials and/or related effectiveness studies.