醫療科技的明日浪潮：數位健康與 AI 智慧醫療的「實證、價值與可近性」國際研討會全紀錄

在這個生成式人工智慧以前所未有速度顛覆人類社會的時代，醫療科技正迎來繼農業、電力與網際網路之後的「第四次技術革命」。當演算法開始執行臨床診斷、遠距醫療應用程式成為日常處方，醫療決策者、監管機構與病人團體，究竟該如何重新定義醫療科技評估與健保給付框架？

2026年8月31日，由財團法人醫藥品查驗中心（CDE）主辦的「數位健康與人工智慧智慧醫療科技之實證、價值與可近性國際研討會」在台北張榮發基金會國際會議中心盛大舉行。本次大會匯聚了來自荷蘭、德國、澳洲及台灣的頂尖政策制定者與方法學專家，共同探討在「實證極度不確定」的科技海嘯下，如何建構一個兼顧創新、公平、病人主體性與醫療體系永續性的給付制度。

一、變革中的起手式：CDE 擘劃「及時與調適性」的評估未來

大會由 CDE 林首愈副執行長揭開序幕。林副執行長在開幕致詞中指出，數位健康與人工智慧科技的爆發式成長，為現代醫療體系帶來了顛覆性的臨床與經濟機遇，但同時也為傳統的醫療科技評估帶來巨大挑戰。

新一代的醫療科技通常具有「迭代快速、演算法黑盒子、數據依賴度高」等特性，使得產品在市場進入時的實證往往存在極高的不確定性。林副執行長強調，未來的醫療科技評估機制必須朝向「更具及時性」與「具調適性」的方向演進。決策者在衡量臨床實證、經濟效益、病人可近性與社會價值時，必須在「支持醫療科技創新」與「確保醫療體系（如健保制度）的財務永續性」之間取得完美的動態平衡。這不僅是台灣的當務之急，更是全球醫療科技評估決策社群的共同命題。

二、實證方法學的跨國躍升：歐盟 HTx 與 SUSTAIN-HTA 的「方法落地與沙盒探索」

HTx 專案：實證融合與人工智慧經濟評估指引

荷蘭Utrecht University藥物醫療科技評估教授、荷蘭國家醫療保險研究所（ZIN）特約顧問 Wim Goettsch 教授登場，以歐盟近年兩大指標性專案 HTx 與 SUSTAIN-HTA 為核心，分享了歐洲如何處理複雜醫療科技的實證與決策不確定性。

Goettsch 教授首先介紹了執行期長達五年的 HTx 專案（2019-2024）。該專案旨在開發新方法，以提供更客製化的醫療有效性與成本效益資訊。其核心成果包含：

跨數據源融合：面對隨機對照試驗數據不足時，HTx 研究如何有效結合真實世界數據進行因果推論數據合成，並在 CRAN 發布了開源 R 套件 crossnma。

個人化預測模型：開發預測模型，協助醫師根據患者個人的復發風險高低，選擇最合適且最具成本效益的藥物療程，案例涵蓋糖尿病與多發性硬化症。

首創 CHEERS-AI 指引：與國際藥物經濟及療效研究學會合作，針對包含人工智慧介入之醫療科技的經濟評估，量身打造了 CHEERS-AI（人工智慧介入健康經濟評估報告標準），填補了全球在人工智慧經濟學評估報告品質指引上的空白。

SUSTAIN-HTA 專案：推廣落地的「六大支柱」與「沙盒」

相較於 HTx 側重於方法學開發，SUSTAIN-HTA 則是一項「協調支持行動」，核心在於方法學的推廣、轉譯與實務落地。該專案建構了六大互聯的建築模塊：

方法觀測台（Methods Observatory, SUSTAIN-OBS）：一個雙向、開源的平台，將新興醫療科技評估方法進行系統化分類，方便各國評估機構檢索並貢獻新方法。

評估沙盒環境（Sandbox, SUSTAIN-SB）：這是一項極具創意的機制。它模擬金融或監管沙盒的概念，為醫療科技評估機構建立一個「安全、無真實決策壓力」的模擬空間。SUSTAIN 專案在此沙盒中針對醫療科技評估機構最棘手的「替代指標」進行案例試辦，讓各國評估代表能在不影響真實健保給付的前提下，共同測試、反思新評估工具的適用邊界，進而促進跨國共識。

Wim Goettsch 教授的核心觀點

「評估的方法學其實早就準備好了，真正的瓶頸在於推廣與落地。」

「數據品質才是決定性的限制，而非統計學。如果原始數據很差，再先進的統計新方法也救不回來。」

三、數位給付的先鋒與陣痛：德國 DiGA 快速通道六年「政策壓力測試」

德國在 2020 年率先推出的「數位健康應用（DiGA）快速通道」，被全球視為數位醫療健保給付的「大膽實驗」。德國醫療品質與效率研究所（IQWiG）國際事務部主任 Dr. med. Alric Rüther，以極為坦率且實證導向的視角，剖析了這項制度運作六年來的真實數據與政策痛點。

DiGA 快速通道的「暫時給付與實證引導給付」機制

在德國由支付方與提供方共同治理的健保制度下，DiGA 提供了一條快速通往健保列名的綠色通道：

審查時效：DiGA 須先取得 CE 認證（風險等級 I、IIa 或 IIb），並向聯邦藥品和醫療器械研究所（BfArM）遞交申請，BfArM 必須在3個月內做出決定。

分流機制：

  • 正式列入給付：若送件時已有充足的比較性研究，證實具備「積極醫療照護效益」（包含醫療效益或病人相關之結構/流程改善效益）。
  • 暫時性列入：若實證仍未臻成熟但「具備潛力」，可先列入給付12至24個月。在此期間，廠商必須一邊接受健保給付，一邊執行「實證引導之給付」臨床試驗，以累積實證數據。

定價機制：

  • 第一年自由定價：由廠商自主訂定每季的處方價格（雖設有價格上限）。
  • 第二年起協商定價：第一年過後，健保署會嚴格依據廠商「實際累積的實證等級」，與廠商談判調降後續的給付價格。

現實的骨感：DiGA 制度面臨的四大挑戰

根據德國截至 2026 年中最新的官方統計數據，DiGA 快速通道實施成效如下：

實證品質堪憂：在總計 251 件申請案中，僅有 60 件成功獲得「正式列入」，29 件為「暫時列入」，其餘多遭駁回或由廠商撤回。Rüther 醫師直指，許多 DiGA 的對照組僅為單純的「等待清單」，而非面談心理治療等活性常規治療，且研究多為未設盲、短期（3個月）、流失率高，實證強度依然偏低。

使用高度集中：DiGA 的實際使用極為不均，其中肥胖症控制應用程式佔了高達 44.1% 的處方量，背痛（5.9%）與耳鳴（4.5%）次之。

醫病互動脫鉤與高流失率：統計顯示，有高達20%的患者在拿到應用程式的短短數天內便完全放棄使用。這與臨床上缺乏醫師的介入、監管與持續引導密切相關。

價格與價值嚴重失衡：在第一年自由定價下，DiGA 的平均季給付價格高達 500 歐元（約合台幣 1.7 萬元），部分甚至高達 2,077 歐元。相較之下，德國醫師執行一次繁複的專業遠距監測服務，僅能向健保申領 126 歐元，但一個功能類似的 DiGA 應用程式卻能拿走 605 歐元，兩者價值評估顯然嚴重脫鉤。

Alric Rüther 醫師的總結

德國經驗是一場震撼的政策壓力測試——「我們究竟該拿健保資金去資助多大程度的不確定性？資助多久？又該以何種價格買單？」早期可近性的確達成了，但臨床價值的落實與合理的定價，至今仍是巨大的考驗。

四、誰的價值說了算？人工智慧時代下「病人自主權」與真實溫度的維護

「當我們在討論人工智慧與數位科技的醫療價值時，是學術界、科學家、還是開發商說了算？還是由真正與疾病共處的病人說了算？」

國際醫療科技評估組織網絡會長、澳洲「Patient Voice Initiative, PVI」執行長 Ann Single 女士，以其深耕病人組織數十年的經驗，為科技導向的研討會注入了深刻的人文反思。

數位健康對病人的隱性威脅與數位落差

Ann Single 指出，數位科技的介入雖然能打破地理疆界，卻也可能讓病人陷入更深的不平等：

被動接受者的危機：許多人工智慧或智慧醫材的設計，往往將病人預設為「數據提供者」與「被動接收指令者」，這無形中剝奪了病人的「主體能動性」。

不對等與排他性：演算法偏見、數據二次利用隱私疑慮、黑盒子演算法的不可解釋性，以及數位素養、寬頻網絡可近性造成的數位落差，都可能讓弱勢病人更加邊緣化。

缺乏病人參與的開發：研究顯示，在澳洲現有的數位健康產品中，高達四分之三（75%）在研發階段完全沒有任何病人參與。

澳洲 Patient Input Buddy 試辦計畫：用人工智慧協助病人發聲

為了打破一般民眾面對繁複官僚醫療科技評估申請表格時的困惑與障礙，PVI 與藥廠 Boehringer Ingelheim 及澳洲四大病人組織（肺臟、硬皮病、罕見癌症基金會）深度共同開發了 Patient Input Buddy 平台。

對話式引導：病人可透過手機或電腦，用最自然、口語的打字或語音，回答人工智慧機器人關於確診經歷與日常生活受影響的提問。

結構化重組與保留「病人語彙庫」：人工智慧會將病人零碎的回答，自動套入給付評估委員會規定的報告格式中。最關鍵的原則是，人工智慧在進行重組與摘要時，絕對不能去修改、轉譯或「過度科學化」病人原始的用語與口吻，必須完整保留病人真實的聲音與靈魂。

隱私與自主提交：平台採取「無痕運行」設計，完全不擷取、不儲存病人的個人隱私數據。草稿整理完成後由病人審查並自行透過個人電郵下載寄送。

五、顛覆性浪潮下的道德警鐘：醫療人工智慧的倫理、公平與素養

醫學期刊《IJTAHC》總編輯、皇家紐澳外科醫師學會（RACS）研究稽核處長 Wendy Babidge 博士，則將視角拉升至宏觀的全球人工智慧治理與倫理層面。

歷史鏡鑑：失控演算法的真實傷害

English version below

Digital Health and AI Medical Technology: A Comprehensive Record of the International Conference on Evidence, Value, and Accessibility

In an era where generative AI is disrupting human society at an unprecedented pace, medical technology is experiencing the “fourth technological revolution” following agriculture, electricity, and the internet. As algorithms begin executing clinical diagnoses and telemedicine apps become routine prescriptions, medical decision-makers, regulatory bodies, and patient groups face a critical question: how should we redefine Health Technology Assessment (HTA) and reimbursement frameworks?

On August 31, 2026, the Center for Drug Evaluation (CDE) organized the “International Conference on Digital Health and AI Medical Technology: Evidence, Value, and Accessibility” at the Chang Jung-Sen Foundation International Conference Center in Taipei. The conference brought together leading policymakers and methodological experts from the Netherlands, Germany, Australia, and Taiwan to explore how to build a reimbursement system that balances innovation, equity, patient autonomy, and healthcare system sustainability amid the “technological tsunami of extreme evidence uncertainty.” This article presents the most cutting-edge and in-depth insights from the entire conference.

The Opening Move: CDE’s Vision for “Timely and Adaptive” Assessment

The conference opened with remarks from Ms. Shou-Yu Lin, Vice Executive Director of the CDE. Lin emphasized that the explosive growth of digital health and AI technology presents both revolutionary clinical and economic opportunities, while simultaneously creating massive challenges for traditional HTA evaluation.

New-generation medical technologies typically possess characteristics of “rapid iteration, algorithmic black boxes, and high data dependency,” resulting in extremely high evidence uncertainty at market entry. Lin stressed that future HTA mechanisms must evolve toward being “more timely” and “adaptive.” Decision-makers must achieve a perfect dynamic balance between “supporting medical technology innovation” and “ensuring the financial sustainability of healthcare systems (such as national health insurance).” This is not only Taiwan’s urgent task but also a shared challenge for the global HTA decision-making community.

Cross-National Advancement in Evidence Methodology: The EU’s HTx and SUSTAIN-HTA Projects

HTx Project: Evidence Integration and AI Economic Evaluation Guidelines

Professor Wim Goettsch from Utrecht University, a drug HTA specialist and consultant to the Dutch National Health Insurance Research Institute (ZIN), shared how Europe is addressing evidence and decision uncertainty in complex medical technologies, focusing on two flagship EU projects: HTx and SUSTAIN-HTA.

Goettsch first introduced the HTx project (2019-2024), a five-year initiative designed to develop new methods for providing more personalized medical effectiveness and cost-benefit information. Its core achievements include:

Multi-source data fusion (RWD and RCT network meta-analysis): When faced with insufficient clinical trial (RCT) data, HTx research demonstrates how to effectively combine real-world data (RWD/RWE) with causally-informed data synthesis, releasing the open-source R package “crossnma” on CRAN.

Personalized predictive models for diabetes and multiple sclerosis (MS): Developed predictive models to help physicians select the most appropriate and cost-effective medication regimens based on individual patient relapse risk.

Pioneering CHEERS-AI guidelines: In collaboration with the International Society for Pharmacoeconomics and Outcomes Research (ISPOR), HTx created CHEERS-AI (Consolidated Health Economic Evaluation Reporting Standards for AI interventions), filling a global gap in quality guidelines for AI economic evaluation reports.

SUSTAIN-HTA Project: Six Pillars and the “Sandbox” for Implementation

In contrast to HTx’s focus on methodological development, SUSTAIN-HTA is a “coordinated support action” centered on the dissemination, translation, and practical implementation of methods. The project constructed six interconnected architectural modules:

Methods Observatory (SUSTAIN-OBS): A bidirectional, open-source platform that systematically categorizes emerging HTA assessment methods, enabling various national evaluation institutions to search and contribute new methods.

Assessment Sandbox (SUSTAIN-SB): An innovative mechanism modeled after financial and regulatory sandboxes, creating a “safe, pressure-free simulation space” for HTA institutions. The SUSTAIN project has piloted case studies on the most challenging HTA issue—surrogate outcomes—within this sandbox, allowing international assessment representatives to jointly test and reflect on new evaluation tools’ applicable boundaries without affecting real health insurance coverage, thereby promoting cross-national consensus.

Professor Wim Goettsch’s Insightful Observations:

“The methodological foundations for assessment are already in place. The real bottleneck is uptake and implementation.”

“Data quality is the decisive limitation, not statistics. If the raw data is crap, no advanced statistical methods can save it.”

Digital Reimbursement Pioneers and Growing Pains: Germany’s DiGA Fast-Track After Six Years

Germany’s 2020 introduction of the “Digital Health Application (DiGA) fast-track,” has been viewed globally as a “bold experiment” in digital healthcare reimbursement. Dr. med. Alric Rüther, Head of International Affairs at the German Institute for Quality and Efficiency in Health Care (IQWiG), provided a candid and evidence-driven analysis of this system’s six years of actual data and policy challenges.

The DiGA Fast-Track’s “Provisional Reimbursement and CED” Mechanism

Within Germany’s Bismarck health insurance system, jointly governed by payers (sickness funds) and providers (physicians/hospitals), DiGA offers a fast-track green channel to health insurance listing:

Review timeline: DiGA must first obtain CE certification (risk levels I, IIa, or IIb) and submit an application to the Federal Institute for Drugs and Medical Devices (BfArM), which must make a decision within 3 months.

Triage mechanisms:

  • Final listing: If the submission already includes sufficient comparative research demonstrating “positive healthcare effects” (including medical benefits or patient-relevant improvements in structure/processes).

  • Provisional listing: If evidence is still immature but “shows promise,” the product can be provisionally listed for 12 to 24 months. During this period, manufacturers must conduct “coverage with evidence development” (CED) clinical trials while receiving health insurance reimbursement.

Pricing mechanism:

  • First-year free pricing: Manufacturers independently set quarterly prescription prices (though price ceilings apply).

  • Negotiated pricing from year two onward: After the first year, the health insurance fund strictly negotiates price reductions based on the manufacturer’s “actual accumulated evidence level.”

The Harsh Reality: Four Major Challenges Facing DiGA

According to Germany’s official statistics as of mid-2026, the DiGA fast-track has faced significant challenges:

Questionable evidence quality: Of 251 total applications, only 60 achieved “final listing,” 29 received “provisional listing,” and the remainder were rejected or withdrawn by manufacturers. Dr. Rüther noted that many DiGA studies used only “care as usual” control groups rather than active comparative treatments like face-to-face psychotherapy, with research often lacking blinding, being short-term (3 months), and showing high dropout rates.

Highly concentrated usage: DiGA use is extremely uneven, with obesity management apps accounting for 44.1% of all prescriptions, followed by back pain (5.9%) and tinnitus (4.5%).

Decoupled physician-patient interaction and high dropout rates: Statistics show that 20% of patients abandoned the app within days of receiving it. This is closely related to the lack of physician involvement, oversight, and continuous guidance in clinical practice.

Price-value mismatch: Under first-year free pricing, the average quarterly DiGA reimbursement reached 500 euros (approximately 17,000 TWD), with some exceeding 2,077 euros. In comparison, a German physician performing complex professional telemonitoring services receives only 126 euros from health insurance, while a functionally similar DiGA app commands 605 euros—a stark disconnect in value assessment.

Dr. Alric Rüther’s Summary:

Germany’s experience represents a shocking policy stress test—”How much uncertainty should we fund with health insurance resources? For how long? At what price?” Early accessibility was indeed achieved, but the realization of clinical value and reasonable pricing remain enormous challenges.

Whose Value Counts? Patient Autonomy and Human Touch in the AI Era

“When we discuss the medical value of AI and digital technology, who determines it—academics, scientists, developers, or the patients who actually live with disease?”

Ann Single, President of the Health Technology Assessment International (HTAi) and Executive Director of Australia’s Patient Voice Initiative (PVI), brought profound humanistic reflection to the technology-focused conference, drawing on decades of patient advocacy experience.

Hidden Threats of Digital Health and the Digital Divide

Single highlighted how digital technology, while breaking geographic barriers, may push patients into deeper inequality:

The crisis of passive recipients: Many AI and smart medical device designs preset patients as “data providers” and “passive recipients of instructions,” implicitly stripping patients of “agency.”

Inequality and exclusivity: Algorithm bias, privacy concerns over secondary data use, the inexplicability of black-box algorithms, and the digital divide caused by digital literacy and broadband accessibility gaps may further marginalize vulnerable patients.

Lack of patient involvement in development: Research shows that 75% of existing digital health products in Australia had zero patient involvement during development.

Australia’s Patient Input Buddy Initiative: Using AI to Amplify Patient Voices

To break through the confusion and barriers patients face with complex HTA application forms, PVI partnered with pharmaceutical company Boehringer Ingelheim and four major Australian patient organizations (lung, scleroderma, and rare cancer foundations) to develop the Patient Input Buddy platform.

Conversational guidance: Patients can answer an AI chatbot’s questions about their diagnosis experience and daily life impacts using natural, conversational typing or voice on mobile or desktop.

Structured reorganization while preserving “patient lexicon”: The AI automatically fits patients’ fragmented responses into the Pharmaceutical Benefits Advisory Committee (PBAC) required report format. The critical principle is that AI must never modify, translate, or “over-scientize” patients’ original language and tone during reorganization and summarization—the patient’s authentic voice and spirit must be completely preserved.

Privacy and autonomous submission: The platform uses a “traceless” design that completely avoids capturing or storing patient personal data. After organizing the draft, patients review and independently download and submit it via personal email.

Disruption’s Moral Alarm Bell: Medical AI Ethics, Equity, and Literacy

Dr. Wendy Babidge, Editor-in-Chief of medical journal IJTAHC and Head of Research Audit at the Royal Australasian College of Surgeons (RACS), elevated the perspective to a macro level of global AI governance and ethics.

Historical Lessons: Real Harms from Runaway Algorithms

The conference’s final segment emphasized the critical importance of ethical oversight in AI-driven medical technology deployment.