Choosing by Operational Fit
A rule-based dose-escalation design is an algorithm: it specifies how many participants may be treated at a dose, what toxicity information must be available, and how observed outcomes determine the next action. Choosing among designs is therefore not simply a choice between “slow” and “fast.” The relevant question is whether the design’s rules fit the enrollment pace, toxicity follow-up, and safety oversight that a particular trial can support.
The comparison here concerns conventional 3+3, accelerated titration, 3+3+3, and rolling six. They differ in cohort size, the timing of decisions, and how they respond to observed toxicities. A design’s name alone is not a complete protocol: the eligibility and toxicity rules, observation requirements, dose increments, and stopping instructions must be specified in advance.
All four approaches require a prespecified definition of an evaluable participant, a toxicity assessment period, and the outcomes that trigger a decision. The rules also depend on reliable, timely toxicity assessment. If outcomes are pending or participants are not evaluable, the protocol must say what actions are permitted; a team should not treat an unknown outcome as though it were a completed observation.
What the Four Designs Ask the Trial Team to Do
| Design | Typical operational feature | Constraint it may suit | Important qualification |
|---|---|---|---|
| Conventional 3+3 | Begins with cohorts of three; some outcomes require expanding a dose-level cohort to six before deciding. | A straightforward, familiar process when sequential review is practical. | Waiting for required outcomes can slow escalation; rules do not guarantee accurate identification of a dose. |
| Accelerated titration | A common variant uses single participants during initial escalation, then switches to cohort-based escalation after a prespecified toxicity trigger. | Situations where a protocol permits faster early progression and toxicity can be assessed promptly. | Triggers and the switch-back rules vary; one participant’s experience cannot characterize a dose well. |
| 3+3+3 | A protocol-defined staged approach can allow up to three groups of three at a dose before a decision. | When a team wants a specified opportunity for additional dose-level information and can enroll and monitor more participants. | The exact decision thresholds vary. The name alone does not identify a complete algorithm. |
| Rolling six | Allows staggered enrollment, typically with up to six participants at a dose, using both ongoing and completed toxicity information according to protocol rules. | When enrollment is steady and participants can be monitored while others are entering. | Concurrent enrollment does not remove the need to follow dose-assignment and escalation rules. |
In conventional 3+3, the usual rule is to escalate after zero dose-limiting toxicities (DLTs) among three evaluable participants; after one among three, treat three more at that dose; and do not escalate if at least two of six have DLTs. The exact protocol must define the remaining actions as well, including stopping and what to do when outcomes are pending.
Accelerated titration is a family of approaches, not one fixed rule set. For example, a protocol might enroll one participant at a time until a specified toxicity grade or event occurs, then switch to conventional 3+3 cohorts. The trigger, dose increments, and handling of the participant who triggered the switch must all be written down. Rapid early escalation is not appropriate just because few events have yet been observed.
There is likewise no single universal 3+3+3 algorithm. For a concrete illustration below, the term refers to a protocol that starts with three participants, expands to six after one DLT among three, and adds a third group of three if there are two DLTs among six. That protocol must separately specify how the results among nine determine the next action. A different protocol using the label may set different rules; teams must compare actual decision rules, not names.
Rolling six permits staggered enrollment under prespecified limits and uses information about both participants who have completed the assessment period and those still being followed. The protocol’s particular rules determine when another participant may enroll, when escalation is permitted, and how many participants can be at a dose at once. It is not simply “six at a time,” nor permission to escalate on the basis of incomplete evidence.
A Practical Selection Process
Specify the expected enrollment pace, the toxicity assessment period, whether enrollment may overlap during follow-up, and the monitoring capacity available.
For each design, identify how many participants are enrolled at a dose, which outcomes must be available, and what triggers expansion, escalation, or stopping.
Reject or revise a candidate if its required waiting, concurrent monitoring, or added enrollment cannot be supported. Do not assume that a faster design can safely bypass required assessment.
Explain why the selected rule set fits the scenario, what it costs in time or participants, and which operational details must be prespecified.
This process is an operational comparison, not a claim that one design is generally superior. A design can be easy to administer but provide limited information at each dose. A more flexible enrollment scheme can reduce idle time but requires careful tracking of participants whose outcomes are still pending. No choice removes the need to define toxicity outcomes and follow-up precisely.
Worked Example: Steady Enrollment and a Long Observation Period
Suppose a hypothetical trial expects two eligible participants per week. The protocol requires six weeks of toxicity follow-up before an outcome is complete. The site can monitor several participants concurrently, and the protocol’s rolling six rules permit staggered enrollment at the current dose while outcomes accrue. Which design best addresses the stated enrollment constraint?
With conventional 3+3, suppose the first three participants are enrolled as they become available. At two per week, the time to enroll three is \(3 \div 2 = 1.5\) weeks. If the protocol requires the full six-week assessment period for the decision, that cohort cannot support a decision until the last participant completes follow-up: approximately \(1.5 + 6 = 7.5\) weeks after enrollment begins.
Under the stated rolling six rules, six participants could be enrolled at the same dose as they become available, if the rules permit and monitoring capacity is sufficient. Enrolling six at two per week takes \(6 \div 2 = 3\) weeks. The last of those six would complete the six-week period about \(3 + 6 = 9\) weeks after enrollment begins. The advantage is not that all six outcomes are known sooner: it is that enrollment can continue during follow-up instead of requiring the team to wait for a completed cohort before enrolling anyone else at that dose.
Choice: Rolling six is a plausible operational fit because the site can monitor concurrent participants and the specified rules allow staggered enrollment. This choice is conditional on following the protocol’s limits and decision rules; it does not authorize escalation before the required evidence is available. If the site could not monitor overlapping follow-up reliably, the same enrollment pressure would not justify rolling enrollment.
Worked Example: A Protocol-Defined Accelerated Phase
Consider a hypothetical protocol that starts with one participant per dose at 10, 20, and 40 mg. It specifies that a grade 2 or higher toxicity ends single-participant escalation. At 40 mg, the participant experiences a grade 2 event. The protocol then requires a cohort of three at 40 mg before any further escalation. Assume the participant is evaluable and the protocol permits that participant to count toward the cohort.
The first participant at 40 mg counts as one of the required three, so the additional enrollment is \(3 - 1 = 2\) participants. Suppose neither additional participant has a DLT and the first participant’s event is not a protocol-defined DLT. The cohort has \(0/3\) DLTs, so the protocol’s subsequent rule—not the grade 2 trigger by itself—determines whether escalation is allowed. If the protocol switches to conventional 3+3 at that point, its zero-of-three rule permits escalation to the next dose.
The example shows why the trigger must be explicit. A grade 2 event might trigger a change in escalation mode even if it is not a DLT. Conversely, a protocol could choose a different trigger or require a different action. The accelerated phase is a fit only if the follow-up schedule and toxicity review are adequate for the decisions being made.
Worked Example: When More Participants at a Dose Are Deliberate
Suppose a hypothetical team can enroll and monitor nine participants at a dose if needed. It is considering the 3+3+3 variant defined above because it wants an additional stage of observation when the first six participants produce two DLTs. At a particular dose, the first three have one DLT, so the cohort expands. Among the next three, one has a DLT. The total is \(1 + 1 = 2\) DLTs among \(3 + 3 = 6\) participants.
Under conventional 3+3, two DLTs among six means the trial does not escalate from that dose under the usual rule. Under this example’s 3+3+3 rule, the two-of-six result triggers three additional participants at the same dose. Suppose none of those three has a DLT. Then the observed total is \(2 + 0 = 2\) DLTs among \(6 + 3 = 9\) participants.
The protocol must now state exactly what two DLTs among nine imply. For example, it could prescribe stopping escalation at that dose; it could instead specify another action, but that action must be justified and stated before the outcomes are seen. The arithmetic does not determine the decision by itself. The 3+3+3 option is reasonable only if the added enrollment and its prespecified decision rule serve a clear purpose; it should not be selected merely because nine observations sound more informative.
Diagnostics, Limitations, and a Common Mistake
A design choice is going wrong when the operations no longer match the rule set. Warning signs include enrolling participants before required eligibility or follow-up checks are complete, making escalation decisions from pending outcomes as if they were final, inconsistent handling of evaluability, or using different thresholds at different doses without protocol authorization. A clear audit trail should show each participant’s dose, assessment status, observed outcome, and the rule that led to the next action.
The central limitation is that these designs use relatively simple decision rules. A count such as one DLT among three is not a precise estimate of toxicity at that dose, and no rule-based design automatically identifies the dose with the most favorable benefit and risk. The earlier material on operating characteristics emphasizes why a design should be evaluated as a whole rather than judged from a single appealing scenario.
Likewise, accelerated titration is not automatically the right choice for a short trial, and a 3+3+3 design is not automatically safer because it may enroll more participants. Compare the actual rule sets, the outcomes each decision requires, and the burden each imposes. The final protocol should unambiguously specify dose assignments, cohort sizes, assessment timing, escalation and stopping criteria, handling of pending or unevaluable outcomes, and any safety pauses.
Check Your Understanding
Answer each question using the operational rules and constraints in the scenario.
- A site cannot reliably monitor more than three participants at once, and toxicity follow-up takes four weeks. What feature makes rolling six a poor fit unless the monitoring plan changes?
- A protocol calls for one participant per dose until a specified toxicity trigger, then cohort-based escalation. What additional information is needed before calling this an accelerated titration protocol?
- Why is “3+3+3” insufficient as a complete protocol description?
- A team argues that a design allowing up to nine participants at a dose must be safer than conventional 3+3. What is missing from that argument?
Solutions. (1) Rolling six relies on staggered enrollment and monitoring of participants whose outcomes may still be pending; a three-person monitoring limit may prevent the permitted overlap. (2) The exact trigger, dose increments, switch-back rules, cohort requirements, and handling of the triggering participant must be specified. (3) The label does not supply universal expansion thresholds or the decision after nine participants; those rules must be stated. (4) More participants do not establish safety by themselves: the toxicity thresholds, decision rule, follow-up, and monitoring plan determine what the design does with the observations.