Why Accelerate Early Dose Escalation?
In an early-phase dose-finding study, the starting dose may be well below a dose that produces dose-limiting toxicity (DLT). Escalating only in small steps from that point can require many cohorts before participants receive doses that meaningfully inform safety. Accelerated titration addresses this by using larger dose increments early, then switching to a more cautious cohort-based escalation rule when a prespecified toxicity signal is observed.
The switch is central to the design. Accelerated titration is not simply a rapid schedule: the protocol must specify the initial increment, what counts as the trigger, how participants at the triggering dose are handled, and the conventional escalation rule used afterward. These choices define the design and must be set before outcomes are reviewed.
Specify the Rule Before Enrolling Participants
A protocol needs more detail than “escalate rapidly until toxicity.” It should define the ordered dose levels, the size of early increments, the unit of escalation (usually cohorts of one during the accelerated phase), the trigger and its observation window, and the conventional procedure that applies after the trigger. It must also say whether the triggering dose is expanded to a larger cohort, and what happens if the planned maximum dose is reached without a trigger.
One illustrative version uses single-participant cohorts and doubles the dose at each early step, provided no trigger is observed. A trigger might be a specified grade of treatment-related toxicity, whether or not that event meets the protocol’s DLT definition. At the first trigger, escalation pauses and the triggering dose is assessed under a conventional cohort rule. For example, the protocol may complete a three-person cohort at that dose and then continue with conventional 3+3 decisions using smaller dose increments. This is an example of a rule, not a universal definition: accelerated titration designs differ in their trigger thresholds and transition details.
The DLT definition and observation window should be stated before applying either phase. In particular, a participant whose DLT window is incomplete should not be treated as having completed a trigger-free window unless the protocol explicitly defines a different decision rule. The early phase’s larger steps also do not remove the need for safety review, appropriate eligibility criteria, or prespecified stopping procedures.
How the Early Phase Changes Transition Probability
The accelerated phase uses fewer participants at early dose levels than a design that enrolls a cohort at every level. This can move the study through low doses more quickly, but it also means that early decisions are based on very limited information. A simple probability calculation shows how a trigger can end the phase.
Suppose, for illustration, that the probability of a transition-triggering event at early dose level \(j\) is \(q_j\), and that one participant is assessed at each of \(m\) levels. If participant outcomes are independent conditional on their dose levels, the probability of no trigger at any of these levels is the product of the probabilities of no trigger at each level. Therefore:
This calculation is a justification for the expression, not an estimate of safety from a small trial. The \(q_j\) values are generally unknown, and the probability may change with dose. The design’s observed path depends on the actual participants and the protocol rule; it cannot be predicted just by counting dose levels.
Worked Example: Probability of a Trigger Across Early Levels
Consider a hypothetical accelerated phase with three dose levels. For illustration only, suppose the trigger probabilities are \(q_1=0.02\), \(q_2=0.05\), and \(q_3=0.12\), with one participant assessed at each level and independent outcomes conditional on dose. Find the probability of at least one trigger by the third level.
The probability of no trigger is:
Thus the probability of at least one trigger is \(1-0.81928=0.18072\), or about 18.1%. This is a consequence of the hypothetical probabilities, not a safety guarantee or an operating characteristic that can be inferred from three observed participants. If the trigger probabilities are misspecified, or outcomes are dependent in a way the calculation ignores, this numerical result will not describe the trial.
Worked Dose Paths
The following examples use hypothetical protocols and participants. In each, the stated observation window is complete before the next escalation decision. The examples use a common illustrative transition: at the first trigger, complete a three-participant cohort at that dose, then apply conventional 3+3 rules. The actual protocol must specify its own transition and subsequent dose levels.
Worked Example: Switching After a Lower-Grade Toxicity
A hypothetical trial starts at 10 mg. During accelerated titration, it doubles the dose after each fully observed participant with no grade 2 or higher treatment-related toxicity. A grade 2 event is the transition trigger, but only a protocol-defined DLT counts in the 3+3 decision. The dose ladder includes 10, 20, 40, 80, and 100 mg; after transition, increments are smaller.
At 10, 20, and 40 mg, the three participants have no grade 2 or higher event in their completed observation windows. The accelerated rule therefore gives \(10\times2=20\) mg, \(20\times2=40\) mg, and \(40\times2=80\) mg. At 80 mg, the participant has a grade 2 event that is not a DLT. The trigger has nevertheless occurred, so the design must switch; it cannot justify doubling to 160 mg merely because the event was not a DLT.
Under the specified transition rule, two more participants are enrolled at 80 mg to complete the cohort of three. Suppose none has a DLT. The 80 mg cohort then has \(0/3\) DLTs, so the conventional 3+3 rule permits escalation. The next protocol level is 100 mg, a 25% increase because \((100-80)/80=0.25\). The trial now uses the conventional cohort rule rather than the early doubling rule. The grade 2 event remains part of the safety record; it is not erased by the DLT count.
Worked Example: A DLT at the Triggering Dose
A second hypothetical protocol uses singleton cohorts at 15, 30, and 60 mg, with doubling allowed only when the completed participant’s DLT window has no transition trigger. A DLT triggers the switch. After transition, the protocol completes a three-person cohort at the triggering level and follows conventional 3+3 decisions.
Suppose the participant at 15 mg has no trigger, so the next level is \(15\times2=30\) mg. The participant at 30 mg also has no trigger, so the next level is \(30\times2=60\) mg. At 60 mg, a DLT occurs. The design stops accelerated escalation immediately; it does not proceed to \(120\) mg.
Two additional participants are treated at 60 mg to complete three. Suppose neither has a DLT. The cohort result is \(1/3\) with a DLT, so the conventional rule calls for expansion to six. If the next three have no DLTs, the total is \(1/6\), which is at most one DLT in six; the conventional 3+3 rule permits escalation to the next prespecified dose level. This example shows why the transition trigger and the later cohort decision must both be applied: the DLT triggers the switch, while the completed cohort result determines the subsequent conventional decision.
Worked Example: Reaching the Planned Maximum Without a Trigger
A hypothetical protocol permits doubling from 25 mg through 50, 100, and 200 mg, but 200 mg is the planned maximum dose. Each early level has one participant, and escalation requires a fully observed window without the protocol’s trigger.
Suppose all four participants complete their windows without a trigger. The arithmetic path is \(25\times2=50\), \(50\times2=100\), and \(100\times2=200\) mg. At 200 mg the protocol prohibits further escalation. The absence of a trigger in these four participants does not establish that 200 mg is safe, nor does it authorize a dose beyond the planned maximum. The protocol must specify whether to stop, enroll a conventional cohort, or take another prespecified action at the cap. Without such a rule, the decision is not determined by accelerated titration alone.
Applying and Auditing the Transition
A transition decision can be checked as a sequence of explicit questions. This is particularly useful when several toxicity classifications or observation windows are involved.
Verify that the participant received the planned dose level and that no dose assignment or outcome has been counted twice.
Confirm that the protocol-defined observation window is complete, or apply only the explicit rule for incomplete follow-up.
Use the prespecified event type and grade. Do not replace a stated trigger with a different threshold after seeing the result.
If the trigger occurred, stop using accelerated increments and apply the transition rule. If it did not, check the permitted next dose and the maximum dose.
Document the dose, outcome classification, window status, trigger status, and resulting rule-based action.
Warning signs include escalation before a toxicity window is complete, doubling a dose after a trigger because the event was not a DLT, or switching at an event that does not meet the protocol’s stated threshold. A further problem occurs when the transition is described only as “move to 3+3” without stating how many participants are added at the triggering dose or which dose level comes next. These are not minor reporting details: different interpretations can lead to different dose paths.
What the Design Gains—and What It Cannot Establish
Compared with conventional 3+3 escalation from the first dose, accelerated titration can reduce the number of participants treated at low doses and can reach higher dose levels sooner when the starting dose is far below active or toxic doses. The cost is sparse early information: with a singleton cohort, one participant’s outcome determines whether escalation continues. A trigger can also occur at a dose that is not a DLT if the protocol deliberately uses a more sensitive threshold.
Larger increments can pass over doses between tested levels. Therefore, the dose where a trigger appears does not identify the precise dose at which toxicity begins. The design also does not, by itself, estimate a dose-toxicity curve or establish that the selected dose is optimal. Its decisions are consequences of prespecified rules and limited observed outcomes, not proof that all lower levels are acceptably safe or that all higher levels are unsafe.
A conventional 3+3 design from the outset uses cohorts and decision rules at each dose level, generally providing more observations at early levels but potentially requiring more participants there. Accelerated titration trades some of that early information for speed, then uses conventional cohort escalation after the trigger. Neither rule-based approach removes uncertainty, and neither should be described as a guarantee of finding the true maximum tolerated dose.
Check Your Understanding
Answer each question, then compare your reasoning with the solutions below.
- A protocol doubles from 12 mg to 24 mg only if the completed participant has no grade 2 or higher toxicity. At 24 mg, a grade 2 event occurs but is not a DLT. What happens to the escalation rule?
- At three successive levels, hypothetical trigger probabilities are 0.01, 0.04, and 0.10. Under the independent-outcome assumption, calculate the probability of at least one trigger.
- Why is it not valid to treat a participant with an incomplete DLT observation window as having no DLT unless the protocol explicitly permits that decision?
- A protocol reaches its maximum planned dose with no trigger in its singleton cohorts. What can and cannot be concluded from that fact?
Solutions. (1) The grade 2 event meets the stated transition trigger even though it is not a DLT. Stop accelerated increments and apply the protocol’s transition rule at 24 mg. (2) The probability of no trigger is \(0.99\times0.96\times0.90=0.85536\). Thus the probability of at least one trigger is \(1-0.85536=0.14464\), or about 14.5%. (3) The event window is not yet complete, so a later event could still meet the DLT definition. Treating incomplete follow-up as a completed negative observation can lead to escalation without the evidence required by the rule. (4) The trial has reached its prespecified cap without observing a trigger in those participants. This does not prove the dose safe, justify going above the cap, or determine the next action unless the protocol specifies what to do at the maximum dose.