醫療科技評估的核心挑戰：從臨床試驗解讀談起

醫療科技評估人才培訓課程第二天（05/29），進一步深入臨床試驗解讀與存活曲線外推的專業領域。這些內容對於評估藥廠提交的實證資料，以及建立科學性的批判思考能力，都相當有幫助。

比較品選擇與試驗族群偏差的評估困境

臺大臨床藥學研究所的關鍵提醒

臺大臨床藥學研究所蕭斐元教授首先點出重點。她提醒我們，在評估新藥是否納入健保給付時，最核心的問題之一，在於「比較品」的選擇，以及臨床試驗收案族群與實際目標族群之間的差異。

許多創新療法的早期臨床試驗，缺乏與臺灣現行標準治療進行頭對頭直接比較；此外，試驗族群也可能未充分納入亞洲或臺灣病人。這些限制都會增加相對療效評估的不確定性，並進一步影響後續健保給付條件的訂定與成本效益估算。

次族群分析：統計顯著≠臨床結論

國家衛生研究院的統計學警示

國家衛生研究院蕭金福教授針對次族群分析，提出重要的統計學提醒。他強調，次族群分析的主要價值，在於檢視各次族群間的結果是否一致，以及產生新的研究假設，不能直接將其視為臨床試驗的最終結論。

ISIS-2試驗的經典警示案例

他以著名的 ISIS-2 試驗為例。該試驗曾出現「雙子座與天秤座病人的治療效果較差」這類明顯缺乏臨床合理性的分析結果，藉此提醒大家，不能因整體臨床試驗未達預期結果，便在事後進行大量資料探勘（data dredging），再將偶然出現統計顯著差異的特定次族群，包裝成足以支持決策的試驗結論。

隨機分派平衡破裂的後續影響

現場也有學員敏銳地提問，若不同治療暴露組別中的次族群人口學特徵分布不一致，應如何處理？這正呼應教授所提醒的重點：一旦原有的隨機分派平衡遭到破壞，分析結果通常只能用於觀察趨勢，難以據此作出具有充分統計依據的結論。

存活曲線外推：從有限期間到長期成本效益

外推評估的核心原則

在評估長期成本效益時，我們經常需要將有限期間內觀察到的存活資料向後外推。針對這項議題，中國醫藥大學譚家惠助理教授提供了相當具有實務價值的存活外推評估原則。

她提醒評估研究人員，不應直接採信未經驗證的外推結果，也不能僅依賴 AIC 或 BIC 等統計指標選擇模型，還必須同時檢視視覺配適情形，並評估結果是否具有臨床合理性。

外部驗證與台灣本地化檢查

更重要的是，應透過外部驗證進行合理性檢查，將外推結果與臺灣一般人口生命表相互比對，確認模型預測的病人存活率，不會出現明顯高於一般人口的不合理情況。

療效衰退假設的必要性

此外，也不能預設新藥療效會終生維持，而應要求廠商評估療效衰退的可能性。

創新療法的不確定性管理

「一次治癒」宣稱的評估困局

面對學員提問，對於宣稱「僅治療一次即可終生治癒」、但長期效果仍具有高度不確定性的創新療法，應如何評估其外推合理性？專家建議，除了可以考慮採用混合治癒模型外，也應設定不同療效維持年限與療效衰退情境，進行敏感度分析，以釐清各項不確定性對財務影響與健康效益的實際影響。

結論：科學證據與資源配置的平衡

唯有具備扎實的證據評讀與品質評估能力，我們才能在創新醫療科技與有限健保資源之間，找到兼顧科學證據、臨床價值與資源配置的適當決策平衡點。

English version below

Core Challenges in Health Technology Assessment: Clinical Trial Interpretation

The second day of the HTA capacity building training program (May 29) provided an in-depth exploration of clinical trial interpretation and survival curve extrapolation—fields essential for evaluating pharmaceutical evidence submissions and developing critical evidence appraisal skills.

The Assessment Dilemma: Comparator Selection and Trial Population Bias

Key Insights from National Taiwan University

Professor Hsiao Fei-Yuan from the National Taiwan University Department of Clinical Pharmacy identified a critical starting point. She emphasized that when assessing whether new drugs should be included in national health insurance coverage, one of the core questions concerns the selection of comparators and discrepancies between trial enrollment populations and real-world target populations.

Many innovative therapies lack head-to-head direct comparisons with Taiwan’s current standard treatments in early-phase clinical trials; moreover, trial populations often insufficiently represent Asian or Taiwanese patients. These limitations increase uncertainty in relative efficacy assessment and subsequently affect reimbursement conditions and cost-effectiveness calculations.

Subgroup Analysis: Statistical Significance ≠ Clinical Conclusion

Statistical Warning from the National Health Research Institutes

Professor Hsiao Chin-Fu from the National Health Research Institutes highlighted critical concerns regarding subgroup analysis. He emphasized that the primary value of such analyses lies in examining consistency across subgroups and generating new research hypotheses—not in serving as definitive trial conclusions.

The Cautionary Tale of ISIS-2

He illustrated this with the famous ISIS-2 trial, which reported that patients born under Gemini and Libra had poorer treatment outcomes—a finding lacking any clinical rationale. This serves as a stark reminder: when overall trial results disappoint, researchers cannot resort to extensive data dredging, then package statistically significant differences in isolated subgroups as evidence sufficient for policy decisions.

Compromised Randomization Balance and Its Consequences

A perceptive participant asked how to handle cases where demographic distributions of subgroups differ across treatment groups. This question reinforced the professor’s warning: once the balance achieved through randomization is disrupted, analysis results can only suggest trends—they cannot support robust statistical conclusions.

Survival Curve Extrapolation: From Observed Data to Long-term Cost-Effectiveness

Core Principles for Extrapolation Assessment

When assessing long-term cost-effectiveness, we frequently must extrapolate observed survival data beyond the trial observation period. On this topic, Assistant Professor Tan Chia-Hui from China Medical University offered highly practical guidance.

She cautioned evaluators against directly accepting unvalidated extrapolation results, and against relying solely on statistical indices like AIC or BIC for model selection. Instead, extrapolation assessment must include visual fit inspection and clinical plausibility evaluation.

External Validation and Taiwan-Specific Reality Checks

Crucially, external validation should be performed by comparing extrapolation results against Taiwan’s general population life tables, ensuring that predicted patient survival rates do not implausibly exceed general population figures.

The Necessity of Treatment Effect Waning Assumptions

Furthermore, one cannot assume new drug efficacy will persist indefinitely; manufacturers must be required to assess potential treatment effect waning.

Managing Uncertainty in Innovative Therapies

Evaluating “One-Time Cure” Claims

When asked how to assess extrapolation reasonableness for innovative therapies claiming “single-administration lifelong cure” despite high long-term uncertainty, experts recommended considering cure-fraction models and conducting sensitivity analyses across scenarios with varying efficacy duration and treatment effect waning patterns. This approach helps clarify how different sources of uncertainty impact financial impact and health benefits.

Conclusion: Balancing Scientific Evidence and Resource Allocation

Only with solid evidence appraisal and quality assessment capabilities can we navigate the space between innovative health technologies and finite health system resources, achieving decisions that respect scientific evidence, clinical value, and judicious resource stewardship.