Why Borderline Counts Need Careful Tracking
In a 3+3 dose-escalation design, a count of one dose-limiting toxicity (DLT) does not have a single meaning independent of its denominator. One DLT among the first three participants is an interim result: it calls for more participants at that same dose. One DLT among six participants is a completed evaluation under the usual rule and permits escalation. Two DLTs among six lead to a different action: stop escalation at that dose and consider the dose below it, if one exists.
The distinction matters operationally. Escalating immediately after observing one DLT in the first cohort skips the required expansion; enrolling more participants after the cumulative count reaches two among six continues at a dose where the standard rule says to stop. This tutorial traces each branch using hypothetical participants and dose levels.
The Decision Branches
Assume dose levels are ordered from lower to higher, each participant’s DLT status is assessed over the protocol-defined window, and each participant counted is evaluable under the protocol. Cohorts are assessed before the next dose assignment. The standard rule is applied to the cumulative DLT count at the current dose:
- With 0 DLTs among 3 participants, escalate to the next dose level.
- With 1 DLT among 3 participants, enroll 3 additional participants at the same dose.
- After expansion to 6 participants, escalate if the cumulative count is 0 or 1 DLT among 6.
- After expansion to 6 participants, stop escalation at that dose if the cumulative count is 2 or more DLTs among 6. If a lower dose has been tested and is eligible under the rule, the dose below is the candidate for further consideration.
The branches follow by enumerating the possible second-cohort outcomes after 1 DLT in the first 3. If the next 3 have \(x\) DLTs, where \(x\) can be 0, 1, 2, or 3, then the cumulative count is \(1+x\) among 6. Thus, \(x=0\) gives 1/6 and escalation; \(x=1\), 2, or 3 gives at least 2/6 and stopping escalation at that dose. The decision is based on the cumulative count, not on treating each cohort’s count as a separate final decision.
This is a rule-based decision, not a statistical test of a toxicity probability. Its immediate purpose is to specify the next enrollment and dose transition given the observed counts. Protocol provisions govern details such as evaluability, delayed outcomes, dose-level limits, and what to do if the lowest dose is too toxic.
Worked Example: One DLT Among the First Three
In a hypothetical trial, participants are treated at dose level 2 after dose level 1 has been assessed. The first cohort at dose level 2 has 1 DLT among 3 participants. What is the next action?
The observed count is \(y=1\) and \(n=3\), so the result is 1/3. Under the standard rule, this is not yet a reason to escalate or to stop at dose level 2. It is the branch that requires expansion.
The trial enrolls 3 additional participants at dose level 2. The cohort enrollment at this dose will then be \(3+3=6\). The dose does not change while those additional participants are assessed: the trial remains at dose level 2.
At this point, the total number of DLTs is not yet known. The next decision must wait until the additional participants’ outcomes are assessed. If none of the additional 3 has a DLT, the cumulative result will be 1/6 and the rule permits escalation to dose level 3. If one or more of the additional 3 has a DLT, the cumulative count will be at least 2/6 and escalation at dose level 2 stops. Treating 1/3 as though it already meant “escalate” would bypass the required expansion.
When the Final Count Is One Among Six
One DLT among six is reached through a specific path in the standard design: the first cohort has 1 DLT among 3, so three more participants are treated at that same dose, and none of those additional participants has a DLT. The cumulative count then meets the rule for escalation. The initial 1/3 count and the final 1/6 count are sequential results, not contradictory decisions.
Worked Example: One DLT Among Six and Escalation
In a hypothetical trial, the first cohort at dose level 1 has 1 DLT among 3 participants. The protocol’s assessment window is complete for these participants, and all three are evaluable. Apply the next step and suppose the additional cohort has 0 DLTs.
First, the 1/3 result requires 3 more participants at dose level 1. The second cohort contributes 0 DLTs among 3. Adding the two cohorts gives \(1+0=1\) DLT among \(3+3=6\) participants, or 1/6.
Because the cumulative count is 1 DLT among 6, the standard rule permits escalation to dose level 2. No additional participants are enrolled at dose level 1 under this branch. If dose level 2 is available and escalation is permitted by the protocol, the next cohort is treated at dose level 2.
The rule does not establish that dose level 1 has a true DLT probability of exactly one-sixth. It says that the observed count meets the design’s prespecified escalation criterion. The estimate \(1/6\) is based on only six participants and has substantial sampling uncertainty.
When the Final Count Is Two Among Six
If the first cohort has 1 DLT among 3 and the second cohort has 1 DLT among 3, the cumulative outcome is 2 DLTs among 6. The first cohort’s result still calls for expansion; the second cohort’s result completes the evaluation and changes the action. Once the total reaches 2/6, the standard rule does not call for a third cohort at that dose.
Worked Example: Two DLTs Among Six and De-escalation
In a hypothetical trial, dose level 2 has been reached after dose level 1 had 0 DLTs among 3 participants. The first cohort at dose level 2 has 1 DLT among 3, so 3 additional participants are enrolled at dose level 2. One of those additional participants experiences a DLT.
At dose level 2, the cumulative count is \(1+1=2\) DLTs among \(3+3=6\) participants. The dose-level 2 result is therefore 2/6. Under the standard rule, escalation stops at dose level 2; the trial does not proceed to dose level 3.
Because dose level 1 was previously assessed and had 0 DLTs among 3, the rule points to the dose below dose level 2 as the candidate for further consideration. This does not mean dose level 1 has been proved safe or that its toxicity probability is zero. It means the rule-based transition is downward rather than upward from dose level 2. Whether dose level 1 is selected as the MTD, whether enrollment there is extended, and whether the trial stops altogether depend on the protocol’s prespecified stopping and selection provisions.
A Compact Branch-by-Branch Check
The following table makes explicit both the enrollment at the dose under evaluation and the immediate transition. “Expand” means adding a second cohort at the same dose, not moving to the next dose.
| Observed result at current dose | Enrollment decision | Immediate dose transition under the standard rule |
|---|---|---|
| 0 DLTs among 3 | No expansion at that dose | Escalate to the next dose, if available |
| 1 DLT among 3 | Enroll 3 more at the same dose; total becomes 6 | Wait for the expanded cohort’s assessment |
| 1 DLT among 6 | No further enrollment required by this branch | Escalate to the next dose, if available |
| 2 or more DLTs among 6 | No further enrollment at that dose under this branch | Stop escalation there; consider the dose below if available and permitted |
| 2 or more DLTs among 3 | No expansion at that dose under the standard rule | Stop escalation there; consider the dose below if available and permitted |
The table assumes that outcomes are classified and participants are counted according to the protocol. If a participant has not completed the DLT assessment period, or is unevaluable, the team must follow the protocol’s timing and replacement rules rather than treating an incomplete observation as a confirmed absence of DLT. Similarly, a study’s maximum dose, lowest dose, or other stopping provisions may limit the transitions shown.
Diagnostics, Comparisons, and Limitations
A useful operational diagnostic is to record, for each dose, the first-cohort count, whether expansion was triggered, the second-cohort count if applicable, and the cumulative count before the transition. This makes it easier to detect common bookkeeping errors, such as escalating on 1/3, counting the second cohort without adding the first cohort’s DLTs, or enrolling beyond six participants at the dose without a protocol-specific reason.
The main nearby alternatives have different enrollment or timing behavior. The 3+3+3 design can add a further cohort under its own rules; a Rolling Six design can enroll participants in a way that accommodates overlapping DLT observation; and time-to-event 3+3 rules address toxicity whose assessment is delayed. These designs should not be treated as interchangeable versions of the same decision table. When applying the standard 3+3 branches here, use the trial’s specified design and do not import another design’s enrollment rules.
The simplicity of the 3+3 rule is also a limitation. Small cohorts mean that one additional DLT can change the action, while the observed proportion in six participants is imprecise. The rule does not use all available information to estimate a dose-toxicity curve, and a dose transition is not a guarantee that the next dose is safe. It provides a transparent operational convention; dose selection and safety oversight still depend on the full protocol and clinical review.
Check Your Understanding
For each case, identify the enrollment at the current dose and the next dose transition. Assume the standard 3+3 rule and that all participants are evaluable with completed DLT assessments.
- A first cohort at dose level 3 has 1 DLT among 3. What happens next, before the additional outcomes are known?
- At a dose, the first cohort has 1 DLT among 3 and the second cohort has 0 among 3. State the cumulative result and transition.
- At a dose, the first cohort has 1 DLT among 3 and the second cohort has 1 among 3. State the cumulative result and transition.
- A trial starts at its lowest dose and observes 2 DLTs among the first 3 participants. Can the standard rule move down to a lower tested dose?
Solutions. (1) Enroll 3 additional participants at dose level 3; do not escalate yet. The final decision depends on the cumulative count among 6. (2) The cumulative result is \((1+0)/(3+3)=1/6\), so escalate to the next dose if one is available and permitted. (3) The cumulative result is \((1+1)/(3+3)=2/6\), so stop escalation at that dose and consider the dose below, if available. (4) No. There is no lower dose level in the trial to which the rule can move; follow the protocol’s lowest-dose stopping provision rather than inventing a lower level or escalating.