Project Optimus: From MTD to Dose Optimization
Project Optimus：從最大耐受劑量到劑量最佳化
English
For decades, the logic of oncology drug development was borrowed wholesale from the era of cytotoxic chemotherapy: push the dose as high as the patient can tolerate, call that the maximum tolerated dose (MTD), and bring it forward into registration trials. The rationale was intuitive — more drug kills more cancer cells, so the ceiling of tolerability is presumably also the floor of efficacy. But this logic, while adequate for non-specific poisons that kill everything that divides fast, was increasingly ill-fitting for the new generation of molecularly targeted therapies, antibody-drug conjugates (ADCs), immune-activating bispecifics, and checkpoint modulators that entered clinical development in the 2000s and 2010s. These drugs often hit a pharmacodynamic ceiling — a plateau of target engagement — well below their toxicity ceiling. Going higher didn’t mean going better; it just meant going sicker.
FDA’s Oncology Center of Excellence launched Project Optimus precisely to confront this mismatch. The initiative, which has been building momentum since the early 2020s and culminated in a formal August 2024 guidance document (“Optimizing the Dosage of Human Prescription Drugs and Biological Products for the Treatment of Oncologic Diseases”), does not say “lower every dose.” It says: prove that the dose you’ve chosen is justified by a totality of evidence — efficacy signal, pharmacokinetics, pharmacodynamics, safety, tolerability, and patient burden — before you walk into a registration trial. If you cannot show that the dose is better than a lower alternative across this multi-attribute landscape, you haven’t done your homework.
The sotorasib case, discussed in a 2024-2025 JCO paper titled “The Retrofit,” illustrates exactly what happens when this homework is skipped. Sotorasib (Lumakras), a KRAS G12C inhibitor, was approved at 960 mg daily. Post-approval data suggested that 240 mg daily — a quarter of the approved dose — achieved comparable efficacy with less toxicity. By the time this evidence emerged, the label had already been written, formularies had locked in the higher dose, and a postmarketing trial was needed to clean up the mess. A 2024 Nature Reviews Clinical Oncology commentary calculated the incremental cost ineffectiveness: patients, payers, oncologists, and the system all bore costs that proper early dose exploration might have avoided. The JCO editorial framing this as “the retrofit” captures the spirit of what Project Optimus is trying to prevent — it is far cheaper and kinder to patients to optimize dosing during development than to retrofit it after approval.
What does dose optimization actually require in practice? A 2024 CPT: Pharmacometrics & Systems Pharmacology review emphasized that the challenge is not primarily statistical — it is organizational and cultural. Dose optimization has to be baked into the FIH (first-in-human) protocol from the start, not added as an afterthought after phase 2 data are in. This means designing early trials to accumulate data at multiple dose levels, not just racing to MTD; it means using model-informed drug development (MIDD) to connect PK exposure, pharmacodynamic signals, and clinical outcomes; it means defining, in advance, what a “better” dose looks like across both efficacy and tolerability dimensions.
The core framework Project Optimus asks investigators to use is a multi-attribute benefit-risk assessment. For each candidate dose, the question is not “did anyone experience a dose-limiting toxicity (DLT)?” but rather: did this dose bring meaningfully more response? Did it bring meaningfully more toxicity or treatment burden? For targeted therapies taken daily for months or years, chronic low-grade toxicities — diarrhea, fatigue, rash, peripheral neuropathy, oral mucositis, ocular effects — often matter more to patients than single-cycle DLTs. These rarely appear in traditional DLT windows, so they require deliberate endpoint design. The recommended phase 2 dose (RP2D) should integrate all of this, not simply default to the maximum safely administered dose.
By 2024-2025, Project Optimus was beginning to visibly reshape real-world phase 1 protocol design. A 2025 JCO Oncology Advances study audited 367 industry-sponsored phase 1 oncology protocols initiated between 2021 and 2024 at Sarah Cannon Research Institute sites. Bayesian methodology adoption rose from 48% in 2021 to 75% in 2024, and dose optimization plans appeared in 30% of protocols. But the same study revealed a persistent gap: patient-reported outcomes (PROs) were present in only 12% of protocols, and patient advocate engagement showed no meaningful upward trend. This is the tension at the heart of Project Optimus in 2025-2026: the statistical machinery is improving, but the patient voice — the dimension that actually answers whether a dose is livable — is still underpowered in most trial designs.
中文
幾十年來，腫瘤藥物開發的劑量邏輯幾乎全盤繼承自細胞毒殺性化療時代：一路把劑量推到病人能承受的最大值，稱之為最大耐受劑量（MTD），然後帶著它進入註冊試驗。這個邏輯在直覺上說得通——藥給多一點、腫瘤細胞死更多，耐受性天花板大概也是療效地板。但這套思路雖然適用於非特異性的細胞毒殺藥，對 2000 至 2010 年代新興的分子標靶藥、抗體藥物複合體（ADC）、免疫活化雙特異性抗體和檢查點調節劑來說，卻越來越格格不入。這些藥物常常在達到毒性天花板之前，就已到達藥效天花板——也就是說，劑量更高不代表療效更好，只代表毒性更重。
正是為了應對這個錯位，FDA Oncology Center of Excellence 啟動了 Project Optimus。這個計畫自 2020 年代初逐步積累動能，並於 2024 年 8 月正式發布指引《Optimizing the Dosage of Human Prescription Drugs and Biological Products for the Treatment of Oncologic Diseases》。Project Optimus 要說的不是「把所有劑量都降下來」，而是：在走進註冊試驗之前，你必須用一整套證據——療效訊號、藥物動力學、藥效學、安全性、耐受性、病人負擔——來證明你選的劑量是合理的。如果你無法說明這個劑量在這些多維度面向優於較低的替代選擇，那就是功課沒做完。
sotorasib 的案例，在 2024-2025 年 JCO 一篇名為「The Retrofit」的文章中被詳細解剖，完美示範了跳過這份功課的後果。這個 KRAS G12C 抑制劑以每天 960 mg 的劑量獲批，但上市後資料顯示 240 mg——四分之一劑量——可達到相近療效，毒性更少。等到這些證據出來，標籤已寫定，給付已鎖定，只能靠上市後試驗來補課。2024 年《Nature Reviews Clinical Oncology》計算了這種「增量成本無效性」的代價：病人、保險、腫瘤醫師和整個系統都在承擔本可在早期開發就避免的成本。「The Retrofit」這個標題精準捕捉了 Project Optimus 的精神——在開發期間做好劑量最佳化，遠比核准後補救來得便宜，也對病人更仁慈。
Project Optimus 在實務上要求什麼？2024 年 CPT: Pharmacometrics & Systems Pharmacology 的一篇回顧強調，挑戰主要不是統計方法，而是組織文化。劑量最佳化必須從 FIH 方案設計時就嵌入，而不是等到第二期結束後再補。這意味著早期試驗要設計成能在多個劑量層級累積資料，而非一路衝向 MTD；要用模型輔助藥物開發（MIDD）把 PK 暴露、藥效訊號和臨床結果連起來；要事先定義在療效和耐受性兩個維度上，什麼叫做「更好」的劑量。
Project Optimus 要求研究者採用的核心框架，是多屬性效益-風險評估。對每個候選劑量，問的不只是「有沒有人出現劑量限制毒性（DLT）？」，而是：這個劑量是否帶來更有意義的反應？是否也帶來更多不必要的毒性或治療負擔？對每天需要長期服用的標靶藥，慢性低階毒性——腹瀉、疲倦、皮疹、周邊神經病變、口腔炎、眼毒性——對病人的重要性往往遠超過單療程的 DLT。這些在傳統 DLT 視窗裡幾乎看不到，必須靠刻意設計的終點來捕捉。建議第二期劑量（RP2D）應整合所有這些資訊，而不是自動落回「最高安全給藥劑量」。
到 2024-2025 年，Project Optimus 開始在真實世界的一期試驗方案設計上留下明顯痕跡。2025 年 JCO Oncology Advances 的研究審視了 Sarah Cannon Research Institute 在 2021 至 2024 年啟動的 367 份業界贊助第一期腫瘤試驗方案。貝氏方法的採用率從 2021 年的 48% 上升到 2024 年的 75%，劑量最佳化計畫出現在 30% 的方案中。但同一篇研究也揭示一個持續存在的缺口：病人回報結果（PRO）只出現在約 12% 的方案中，病人倡議者參與也沒有明顯的逐年上升趨勢。這就是 2025-2026 年 Project Optimus 的核心張力：統計工具在進步，但病人聲音——真正能回答「這個劑量能不能讓人活下去」的那個維度——在大多數試驗設計中仍然不足。
Key Concepts | 核心概念
- MTD vs. Optimal Dose | MTD vs. 最佳劑量: MTD marks the toxicity ceiling; the optimal dose sits where meaningful efficacy meets acceptable long-term tolerability — these are rarely the same point. MTD 是毒性天花板；最佳劑量是療效與長期耐受性的交叉點，兩者很少重合。
- Multi-attribute benefit-risk | 多屬性效益-風險: Dose decisions must weigh response depth, grade 3+ toxicity, dose interruptions/reductions, discontinuation, and patient-reported tolerability simultaneously. 劑量決策必須同時衡量反應深度、3 級以上毒性、劑量暫停/減量、停藥率與病人回報的耐受性。
- The Retrofit Problem | 補課問題: When dose optimization is skipped during development, postmarketing trials and regulatory actions attempt to correct it — at much higher cost and patient burden. 在開發期間跳過劑量最佳化，上市後試驗與監管行動會嘗試補救——但代價更高、病人負擔更重。
- MIDD in dose optimization | 模型輔助劑量最佳化: Model-informed drug development connects PK exposure, tumor dynamics, and clinical outcomes to support dose decisions across development stages. 模型輔助藥物開發將 PK 暴露、腫瘤動態和臨床結果連結起來，支持跨開發階段的劑量決策。
- PRO gap | 病人回報缺口: Even as statistical designs improve (Bayesian adoption 48%→75%), patient-reported outcomes remain present in only ~12% of phase 1 protocols — the biggest remaining gap in Project Optimus implementation. 統計設計持續進步（貝氏採用 48%→75%），但病人回報結果仍只出現在約 12% 的一期方案中，是 Project Optimus 落地最大的缺口。
Related Pages | 相關頁面
- fda-2024-dose-optimization-guidance — The formal regulatory framework
- fih-starting-dose-selection — How the dose journey begins
- fda-aacr-2025-dose-optimization-trilogy — The three-paper educational series
- benefit-risk-dose-decision-framework — Practical tools for dose decisions
- real-world-protocol-impact — How protocols are actually changing
- postmarketing-dose-optimization-risk-signals — What happens when optimization is skipped