2026 INAHTA 國際大會 Day 2：AI 治理與全生命週期醫療科技評估的雙重挑戰

2026 年 INAHTA 大會第二天完美落幕。今天的議程聚焦於兩個顛覆醫療科技評估（HTA）未來的核心議題：「人工智慧的實務與風險挑戰」以及「全生命週期醫療科技評估（Lifecycle HTA）」。大會依舊秉持嚴格的保密與不對外引述原則，讓各國代表能毫無保留地分享最前線的實戰痛點與深刻反思。

AI 應用的務實邊界與風險防範

在 AI 應用方面，各國機構已經跨越了初期的盲目探索，進入了務實的邊界管理階段。大家普遍達成共識：AI 目前最適合處理邊界明確的行政與庶務工作，如文獻初步篩選或撰寫民眾版衛教摘要。

然而，與會代表強調，絕對不能讓 AI 越俎代庖進行最終的價值判斷或獨立分析，以防範 AI 幻覺帶來的嚴重風險。更令人警醒的是，大會中揭露了產業界已經開始提交由 AI 生成的「合成世代（Synthetic cohorts）」作為對照組，甚至包含了完全由 AI 生成的經濟模型。

AI 生成證據面臨的審查困境

面對這些前所未有的數位證據，HTA 機構正面臨缺乏一致審查標準的巨大挑戰。各國代表強烈呼籲應建立跨國共通的評估工具與規範，以免在各國標準不一下被業界各個擊破。

數位時代的智財權與人才危機

此外，數位時代的智財權與人才危機也在今日引發熱烈討論。許多機構發現，其公開發布的 HTA 報告正被外部的網路機器人大量抓取，用以訓練商業 AI 模型。產業界更甚至利用這些模型進行「模擬決策審查」來擬定應對機構的策略。

這促使大家深刻反思資料主權、智財權保護與資安防護的急迫性。

人才養成與批判思維的隱憂

高度依賴 AI 工具可能導致年輕研究員喪失批判性思考與基礎評估能力，這被視為機構面臨的重大隱憂。為了確保核心科學技能的傳承與驗證真實實力，許多機構與學界的培訓及考核方式已經開始從傳統的撰寫書面報告，轉回更重視臨場反應的口頭報告與實體口試。

全生命週期醫療科技評估的機制與困境

下午的議程則轉向了同樣充滿挑戰的全生命週期醫療科技評估。隨著醫療科技與臨床實務發展日新月異，單一時間點的初始評估已不足以應付真實世界證據的動態變化。

再評估的法規基礎與資源現況

雖然已有少數國家將定期再評估納入法規，賦予其改變給付決策的法律基礎，並成功實現了醫療資源的重新投資與撤資；但多數機構坦言，受限於龐大的資源消耗與量能不足，目前仍只能採取依需求觸發的被動模式。

過時科技篩選的共同難題

各國共同面臨最困難且成本最高的環節，在於如何精準選定必須進行再評估的過時科技主題。對此，與會者強烈期盼能建立跨國的情報共享機制，互通各國正在重新檢視的候選主題與排定優先順序的標準，藉此大幅降低獨自摸索的成本。

跨國協作：拒絕重複發明輪子

總結這兩天的國際盛會，「拒絕重複發明輪子」成為全場最具共鳴的核心精神。無論是面對 AI 狂潮的治理規範、統一要求廠商揭露 AI 使用的免責聲明，還是全生命週期評估的資源困境，單打獨鬥已不再可行。

唯有透過國際間的緊密協作與透明共享，我們才能在擁抱科技創新的同時，堅守 HTA 的科學嚴謹與公眾信任。科技的浪潮不會停歇，但人類的智慧與跨國的合作網絡，將是我們最好的導航。

大家在各自的專業領域中，又是如何看待 AI 生成的醫療證據或是專業人才技能退化的挑戰呢？歡迎留言交流分享！

English version below

INAHTA 2026 Day 2: Dual Challenges of AI Governance and Lifecycle Health Technology Assessment

Day 2 of the 2026 INAHTA conference concluded with powerful insights into two transformative issues shaping the future of health technology assessment (HTA): “Practical Applications and Risk Challenges of Artificial Intelligence” and “Lifecycle Health Technology Assessment (Lifecycle HTA)”. The conference maintained its strict confidentiality and non-attribution principles, enabling delegates from participating countries to share frontline challenges and profound reflections without reservation.

Pragmatic Boundaries and Risk Mitigation in AI Applications

In AI applications, national HTA institutions have moved beyond early exploratory phases into practical boundary management. A broad consensus emerged: AI is currently best suited for well-defined administrative and routine tasks, such as initial literature screening or drafting public-facing health education summaries.

However, delegates emphasized that AI must never usurp the final value judgments or independent analyses, protecting against the serious risks posed by AI hallucinations. Notably, the conference revealed that industry has already begun submitting AI-generated “synthetic cohorts” as comparison groups, and even entirely AI-generated economic models.

Scrutiny Challenges and the Need for Unified Standards

Faced with unprecedented digital evidence, HTA institutions confront a critical lack of consistent review standards. Delegates strongly called for establishing transnational common assessment tools and frameworks to prevent industry from exploiting divergent national standards.

Intellectual Property Rights and Talent Crisis in the Digital Age

Another pressing concern emerged: many institutions discovered their publicly released HTA reports being systematically scraped by external web bots to train commercial AI models. Industry actors even leveraged these models to conduct “simulated decision reviews” to strategize their approach to institutions.

This prompted deep reflection on data sovereignty, intellectual property protection, and cybersecurity urgency.

Safeguarding Scientific Capacity Through Education Reform

Over-reliance on AI tools risks eroding young researchers’ critical thinking and foundational assessment capabilities—a major institutional concern. To preserve core scientific skill transmission and validate genuine expertise, many institutions and academic programs have shifted training and assessment methods away from traditional written reports toward oral presentations and in-person examinations emphasizing real-time reasoning.

Mechanisms and Challenges in Lifecycle Health Technology Assessment

The afternoon focused on the equally complex challenge of lifecycle health technology assessment. As medical technology and clinical practice evolve rapidly, single time-point initial assessments no longer suffice to address dynamic real-world evidence.

Regulatory Foundations and Resource Realities

While some countries have incorporated periodic reassessment into regulations, enabling legal grounds to change reimbursement decisions and successfully achieved reinvestment and disinvestment of health resources, most institutions acknowledge that resource constraints and limited capacity restrict them to reactive, needs-triggered reassessment modes.

The Common Dilemma of Identifying Obsolete Technology

All participating countries face the most difficult and resource-intensive challenge: precisely selecting which outdated technology topics warrant reassessment. Delegates strongly advocated establishing transnational intelligence-sharing mechanisms to exchange candidate topics under review and priority-setting standards, thereby dramatically reducing costs of independent exploration.

International Collaboration: Refusing to Reinvent the Wheel

Summarizing the two-day international forum, “refusing to reinvent the wheel” emerged as the most resonant core principle across the conference. Whether addressing AI governance frameworks, unified requirements for vendor AI disclosure disclaimers, or resource constraints in lifecycle assessment, individual efforts no longer suffice.

Only through tight international collaboration and transparent knowledge-sharing can we embrace technological innovation while maintaining HTA’s scientific rigor and public trust. The tide of technology will not cease, but human wisdom and transnational collaborative networks remain our best navigation.

How do you, within your professional domains, perceive AI-generated medical evidence and the challenge of professional skill atrophy? Welcome to share your thoughts in the comments!