The Rolling Six Escalation Design

Why Roll Cohorts?

In a conventional 3+3 design, a dose decision generally waits until the assigned participants have completed the dose-limiting toxicity (DLT) assessment window. When that window is long, waiting can slow a trial. The rolling six escalation design allows enrollment to continue at a dose while some participants are still being assessed, subject to a limit on how many may be treated there. It changes the timing and flow of enrollment; it does not make an unresolved toxicity outcome known.

The key operational distinction is between participants who have been treated at a dose and participants whose DLT assessment is complete. A participant still inside the assessment window occupies an enrollment place at that dose, but cannot be counted as having had no DLT. The rules below describe a common rolling-six implementation. A trial protocol must specify its own assessment window, decision timing, and handling of participants whose outcomes are pending.

The Core Enrollment and Decision Rules

The familiar 3+3 dose decision is the starting point: among three evaluable participants, zero DLTs permits escalation, one DLT calls for expansion, and two or more DLTs calls for de-escalation or stopping at that dose. If a dose is expanded to six evaluable participants, escalation is permitted with no more than one DLT among the six; two or more DLTs lead to de-escalation or stopping. Rolling six adds flexibility in when additional participants can be enrolled, not permission to disregard these toxicity thresholds.

For the examples here, use the following explicit operational version. Before a dose decision is available, the team may enroll additional participants at that dose, up to six total, while outcomes are pending, provided no stopping signal has occurred. Once three participants are evaluable, apply the 3+3 decision. If there are zero DLTs, escalation is allowed under this rule set; if there is one, complete the expansion to six evaluable participants before deciding; if there are at least two, do not escalate. After expansion, at most one DLT among six permits escalation. A pending participant is never counted as evaluable or as DLT-free.

This description makes the decision rule explicit, but a protocol may impose stricter constraints—for example, requiring resolution of all participants already assigned to a dose before opening the next dose. Such constraints must be followed. Enrollment capacity is not itself a dose-escalation criterion.

Counting Capacity While Outcomes Are Pending

Keep separate counts for (1) total enrolled at the dose, (2) evaluable participants, (3) pending participants, and (4) observed DLTs. If \(E\) is the total enrolled and \(P\) is pending, then the number with completed outcomes is \(E-P\), assuming every participant not pending has a final DLT classification. Under the cap, the maximum number of additional enrollments is:

This small calculation prevents a frequent error: adding six more participants because six is the design's maximum. Six is the maximum total at that dose, not the maximum additional number after an initial group. It also prevents a subtler error: using the number enrolled as if it were the number evaluable. For example, three enrolled participants with two outcomes pending are not equivalent to three completed DLT assessments.

How Pending Outcomes Affect Dose Decisions

Enrollment can proceed during an assessment window, but the team must keep reviewing accumulating information. A newly reported DLT can change the decision immediately. For example, if two DLTs have occurred among participants at a dose, the dose is not made acceptable by the fact that other participants remain pending. Under the rule set above, at least two DLTs means no escalation from that dose.

Likewise, “no DLT observed so far” is not equivalent to “zero DLTs among evaluable participants.” The first statement describes incomplete information; the second requires completed assessments. When deciding whether to enroll more people, use the protocol's safety and capacity rules. When deciding whether to escalate, use the required number of evaluable outcomes and the prespecified DLT threshold.

A Short Justification for the Six-Person Cap

The six-person cap bounds the number exposed at one dose before a decision, while allowing more than three participants to be treated without waiting for every assessment to finish before enrollment continues. The capacity calculation follows directly from the definition of the cap. If \(E\) participants have already been assigned, then any additional number \(A\) must satisfy:

Subtracting \(E\) from both sides gives \(A\leq 6-E\). Since enrollment cannot be negative, the number of additional participants is at most \(\max(0,6-E)\). This derivation concerns enrollment capacity only. It does not show that enrolling the maximum is appropriate, nor does it establish that escalation is safe; those judgments depend on completed outcomes and protocol stopping rules.

Monitoring, Alternatives, and Limitations

Rolling enrollment requires reliable, current tracking. For each dose, the study team should be able to identify the total enrolled, the number evaluable, the number pending, the DLT count, and any participants who discontinued or otherwise have a nonstandard outcome. The protocol must explain how such outcomes are classified; they should not be informally treated as completed without DLT. Clear assignment records are also essential when participants at different doses are being assessed at the same time.

A practical review before each enrollment or dose decision is:

Compared with a conventional 3+3 implementation that waits for all assigned participants to finish assessment before enrolling the next group, rolling six can reduce delays by allowing enrollment to continue within the six-person limit. The trade-off is more complex real-time coordination: several outcomes may be pending at once, and dose assignments may overlap in time. A strict wait-for-assessment approach is easier to administer but may lengthen trial conduct. Neither approach turns a small rule-based escalation study into a precise estimate of a dose's toxicity rate; the design remains dependent on its prespecified thresholds and limited sample sizes.

Common Mistakes

• Counting pending participants as DLT-free. A participant without a reported DLT may still be within the assessment window. Keep that participant pending until the protocol-defined outcome is available.
• Applying the cap to new enrollments rather than total enrollments. If five participants are already assigned, the remaining capacity is one, not six.
• Escalating based on the enrolled count. The decision thresholds refer to evaluable DLT outcomes. Three enrolled with two pending do not provide three completed assessments.
• Assuming the cap authorizes enrollment. Six is a maximum, not a target. Safety pauses, stopping rules, or protocol restrictions can require fewer enrollments.
• Treating every rolling-six protocol as identical. The precise timing of escalation and management of pending participants must be stated in the protocol. Follow that rule rather than assuming the cap answers every decision.