FIH Starting Dose Selection: From Safety Floor to Biological Hypothesis

FIH 起始劑量選擇：從安全底線到生物學假說

English

The first dose given to a human in a novel oncology drug trial is not the output of a formula — it is the output of a set of risk decisions made under uncertainty. For most of the history of cytotoxic chemotherapy, this decision was relatively tractable: you took the highest dose tested in the most sensitive animal species that did not cause severe toxicity (the highest non-severely toxic dose, HNSTD, or the no-observed-adverse-effect level, NOAEL), divided it by a safety factor (often 1/10th for humans relative to rodents, or 1/6th for non-rodent species), and arrived at a starting dose designed to prevent catastrophic first-dose harm. The logic was conservative by design, and for drugs whose toxicity tracks their anti-tumor activity, it worked reasonably well.

But the modern oncology drug landscape has fractured this single-algorithm approach. A KRAS inhibitor, an anti-CD3 bispecific T-cell engager, a TROP-2-directed ADC, and a PSMA-targeting radiopharmaceutical each carry completely different risk profiles at their first dose. The anti-CD3 bispecific may trigger a cytokine storm — a rapid, systemic immune activation — that kills a patient before pharmacokinetics matter. The ADC has toxicity arising from payload release that may behave quite differently from the intact antibody component alone. The radiopharmaceutical has dosimetry — the pattern of radiation deposition — as a primary determinant of organ toxicity. For all of these, dividing an NOAEL by a safety factor is a starting point at best, and a misleading one at worst.

The FDA-AACR 2025 series in Clinical Cancer Research gave this more nuanced picture systematic form. For immune-activating biologics, bispecifics, and cell therapies, the relevant concept is MABEL — the minimal anticipated biological effect level. MABEL is not “the lowest safe dose”; it is the dose at which you first expect to see biological activity, estimated by integrating nonclinical pharmacology data (in vitro potency, receptor occupancy in animal models, species differences in target expression), existing PK/PD models, and any available human receptor expression data. The goal is to choose a first dose that is biologically meaningful — enough to verify that the drug is doing something — but below the threshold where the biology becomes dangerous. For highly potent immune activators, this window can be very narrow, which is why step-up dosing (starting at a low priming dose, then escalating over subsequent cycles) was invented: not as a regulatory formality, but as a genuine risk-control mechanism.

For small molecule targeted therapies, a complementary concept is the pharmacologically active dose (PAD) — the dose at which meaningful target inhibition is achieved in humans, estimated from preclinical exposure-response relationships and first-in-human PK projections. PAD matters because purely toxicity-driven starting dose calculations can place patients in a zone of no pharmacological effect — essentially giving them sub-therapeutic drug for the first several cohorts of a trial, with no benefit and still the inconvenience of trial participation. The tension between MABEL (minimize dangerous biology) and PAD (don’t waste time with pharmacologically inert doses) is a genuine design challenge, especially for drugs where the therapeutic window is wide in preclinical models but poorly characterized in humans.

What does a good FIH starting dose document actually say? According to the 2025 FDA-AACR series, it should answer four questions: First, what is the basis — HNSTD/NOAEL, MABEL, PAD, or an integrated PK/PD model? Second, what safety factor was applied, and why? Third, what are the critical assumptions underlying the calculation, and what happens if they are wrong? Fourth, what is the plausible worst-case toxicity profile at this dose, and what monitoring or risk-mitigation measures are in place to catch it early? For biotherapeutics, the document should also explicitly characterize the anticipated immune biology — cytokine profile, receptor occupancy at first dose, expected PD changes — and explain why sentinel dosing, split dosing, or premedication were or were not required. This is not bureaucratic box-checking; it is the foundation that makes subsequent dose escalation intellectually coherent.

The molecular-class framework proposed in a 2025 Clinical Pharmacology & Therapeutics review provides a practical way to think about which starting-dose questions dominate for which drug type. Class 1 (small molecule targeted therapies and ADCs) asks primarily about target inhibition saturation and payload-related cumulative toxicity. Class 2 (large molecule antagonists) asks about receptor occupancy and the consequences of complete pathway blockade. Class 3 (cancer immunotherapy agonists, including T-cell engagers and checkpoint agonists) asks about first-dose cytokine release risk, the steepness of the dose-immune activation curve, and whether the therapeutic window allows dose escalation at all. Class 4 (molecules with limited single-agent activity) asks a more fundamental question: what biological read-out will even tell you that escalation is doing something? These classes demand different conversations, different preclinical packages, and different risk-mitigation frameworks — and a physician reviewing an FIH protocol should know which class they are dealing with before reading the starting dose rationale.

中文

新型腫瘤藥物臨床試驗中給予人體的第一個劑量，並不是公式計算的產物——它是在不確定性下做出的一組風險決策。在細胞毒殺性化療時代的大部分歷史中，這個決策相對直接：取在最敏感動物物種中測試的最高不致嚴重毒性劑量（HNSTD，即最高非嚴重毒性劑量，或無可觀察不良效應劑量 NOAEL），除以一個安全係數（人類相對囓齒類通常取十分之一，相對非囓齒類取六分之一），得到一個設計用來防止第一劑災難性傷害的起始劑量。這個邏輯天生保守，對於毒性與抗腫瘤活性平行的藥物，運作相當合理。

但現代腫瘤藥物的格局已打破這種單一演算法。KRAS 抑制劑、抗 CD3 雙特異性 T 細胞接合器、TROP-2 靶向 ADC 和 PSMA 靶向放射性藥物，各自的第一劑風險完全不同。抗 CD3 雙特異性抗體可能在藥物動力學尚未發揮作用之前，就引發細胞激素風暴——一種快速、全身性的免疫活化——奪走病人性命。ADC 的毒性源自 payload 釋放，行為可能與完整抗體成分大相逕庭。放射性藥物有劑量測定——輻射沉積的模式——作為器官毒性的主要決定因素。對所有這些藥物，用 NOAEL 除以安全係數頂多是起點，最壞是誤導。

2025 年 FDA-AACR 在《Clinical Cancer Research》的系列文章，把這幅更複雜的圖景整理成系統性框架。對免疫活化生物製劑、雙特異性抗體和細胞療法，相關概念是 MABEL——預期最低生物效應劑量。MABEL 不是「最低安全劑量」；它是你第一次預期看到生物活性的劑量，通過整合非臨床藥理資料（體外效價、動物模型中的受體佔有率、物種間目標表現差異）、現有 PK/PD 模型以及任何可用的人類受體表現資料來估計。目標是選擇一個生物學上有意義的第一劑量——足以驗證藥物在發揮作用——但低於生物學變得危險的門檻。對高效能免疫活化劑，這個窗口可能非常窄，這就是為什麼發明了步進給藥（step-up dosing，從低劑量引導劑量開始，在後續療程中逐步升量）：這不是監管形式，而是真正的風險控制機制。

對小分子標靶藥，一個互補的概念是藥理活性劑量（PAD）——在人體達到有意義目標抑制的劑量，通過臨床前暴露-反應關係和首次人體 PK 預測來估計。PAD 很重要，因為純粹以毒性驅動的起始劑量計算，可能將病人置於無藥理效果的區間——本質上是在試驗的前幾個 cohort 讓他們接受次治療劑量的藥物，沒有獲益卻仍要承受試驗參與的不便。MABEL（最小化危險的生物效應）和 PAD（不要把時間浪費在藥理學上無活性的劑量）之間的張力，是真實的設計挑戰，尤其是對那些在臨床前模型中治療窗口很寬但在人體中特性未知的藥物。

一份好的 FIH 起始劑量文件實際上應該說什麼？根據 2025 年 FDA-AACR 系列，它應該回答四個問題：第一，依據是什麼——HNSTD/NOAEL、MABEL、PAD，還是整合性 PK/PD 模型？第二，採用了什麼安全係數，為什麼？第三，計算背後的關鍵假設是什麼，如果這些假設錯了會發生什麼？第四，在這個劑量下最可能的最壞毒性是什麼，有哪些監測或風險緩解措施可以早期捕捉？對生物製劑，文件還應明確描述預期的免疫生物學——細胞激素譜、第一劑的受體佔有率、預期的 PD 變化——並解釋為何要求或不要求哨兵給藥、分次給藥或預防藥物。這不是官僚式打勾；它是使後續劑量遞增在知識上自洽的基礎。

2025 年《Clinical Pharmacology & Therapeutics》回顧提出的分子類別框架，提供了一種實用方式來思考哪類起始劑量問題在哪類藥物中佔主導。第一類（小分子標靶藥和 ADC）主要問目標抑制飽和度和 payload 相關累積毒性。第二類（大分子拮抗劑）問受體佔有率和完全途徑阻斷的後果。第三類（癌症免疫治療促效劑，包括 T 細胞接合器和檢查點促效劑）問第一劑細胞激素釋放風險、劑量-免疫活化曲線的陡峭程度，以及治療窗口是否允許劑量遞增。第四類（單藥活性有限或缺乏的分子）問一個更根本的問題：什麼生物讀出值能告訴你遞增有任何意義？這些分類需要不同的對話、不同的臨床前資料包和不同的風險緩解框架——審查 FIH 方案的醫師在閱讀起始劑量理由之前，應先知道面對的是哪一類。

Key Concepts | 核心概念

• HNSTD / NOAEL + safety factor | 安全係數法: The classical approach — divide the highest non-severely toxic animal dose by a safety factor. Appropriate for cytotoxics; often insufficient alone for modern targeted agents. 傳統方法——將動物最高非嚴重毒性劑量除以安全係數。適用於細胞毒殺藥；對現代標靶藥通常單獨使用不夠。
• MABEL | 最低預期生物效應劑量: Minimum anticipated biological effect level — calibrated to the pharmacology of immune-activating drugs where dangerous activity can occur before classical toxicity. 預期最低生物效應劑量——針對免疫活化藥物的藥理特性校準，此類藥物危險活性可在經典毒性之前出現。
• PAD | 藥理活性劑量: Pharmacologically active dose — the floor of meaningful target engagement; avoids placing patients in a pharmacologically silent zone. 藥理活性劑量——有意義目標接合的底線；避免讓病人處於藥理學無活性區間。
• Step-up dosing | 步進給藥: A risk-control protocol design where patients receive a low priming dose before escalating — not a formality, but a genuine mechanism to prevent CRS and other early immune toxicities. 病人在遞增前接受低劑量引導——不是形式，而是防止 CRS 和其他早期免疫毒性的真實機制。
• Molecular-class framework | 分子類別框架: Different drug classes (small molecule, antagonist antibody, immune agonist, low single-agent activity) require entirely different starting-dose questions and risk frameworks. 不同藥物類別需要完全不同的起始劑量問題和風險框架。

Related Pages | 相關頁面

• project-optimus-dose-optimization — Why starting dose is just the beginning of optimization
• fda-aacr-2025-dose-optimization-trilogy — The three-paper series contextualizing FIH dose selection
• benefit-risk-dose-decision-framework — How to evaluate each dose level during escalation
• dose-escalation-designs-boin-crm — Statistical designs for FIH escalation