From Phase II Evidence to a Confirmatory Candidate
A phase II dose-ranging study informs a decision; it does not, by itself, settle every question that a confirmatory study must answer. The practical task is to decide which dose or doses have a credible prospect of providing clinically relevant benefit, an acceptable safety profile, and a regimen that can be tested and delivered as planned. A dose can have the largest estimated efficacy and still be a poor candidate if its harms, uncertainty, or operational demands are unacceptable.
Advancement is therefore a structured judgment across evidence dimensions, not simply a rule to choose the dose with the smallest p-value or largest observed effect. The decision should use prespecified criteria where possible, describe the uncertainty that remains, and distinguish evidence supporting a dose from questions the confirmatory trial is intended to resolve. The earlier material on Combining Dose-Response Evidence Across Trials can inform the evidence base; the decision here is how that evidence supports a next-stage candidate.
The decision should begin with the intended confirmatory question and population. A result from a short study in a narrowly selected group may be less informative for a longer trial in a broader population. The endpoint, treatment duration, background therapy, and dose regimen should also be relevant to the confirmatory plan. If those features differ substantially, the phase II evidence may still be useful, but its limitations should be explicit rather than hidden inside a single summary estimate.
Define the Decision Before Comparing Doses
A useful decision framework states what counts as a worthwhile efficacy signal, what safety findings are unacceptable or require mitigation, and how uncertainty will affect advancement. The exact thresholds depend on clinical context and should not be chosen after seeing which dose happens to pass. They are decision criteria, not universal statistical constants.
These screens are not a substitute for judgment. An estimate just inside a threshold with wide uncertainty is not equivalent to a precise estimate comfortably inside it. Likewise, an estimate outside a threshold may warrant more study if the uncertainty is large and the potential benefit is important, but that rationale must be distinguished from claiming the threshold was met. Confidence intervals and other uncertainty summaries describe how compatible the evidence is with a range of values; they do not guarantee that the true effect lies within a particular interval.
A brief utility argument explains why the most efficacious estimate need not determine the choice. Suppose a decision-maker summarizes benefits by \(B(E)\) and harms by a nonnegative weight \(\lambda\), defining \(U(d)=B(E(d))-\lambda R(d)\) among doses that meet essential safety and feasibility constraints. For two eligible doses \(d_1\) and \(d_2\), the second has greater utility exactly when:
This is a justification for making trade-offs explicit, not a prescription to assign an arbitrary numerical value to benefit or harm. If stakeholders cannot defend a common utility scale, keep the dimensions visible and explain the decision qualitatively. A hard safety constraint should not be overridden merely because a weighted average looks favorable.
Worked Example: A Stronger Efficacy Estimate Does Not Automatically Win
Worked Example: Comparing Efficacy and Safety Against Prespecified Criteria
Hypothetical data. In a randomized phase II trial, the efficacy endpoint is a response rate, and higher values are better. The response rates are 12% on placebo, 18% on 40 mg, 25% on 80 mg, and 29% on 160 mg. The estimated treatment-placebo efficacy contrasts are therefore:
For illustration, the team has set a clinically relevant point-estimate threshold of 8 percentage points and a safety boundary of 5 percentage points for excess discontinuations due to adverse events versus placebo. The observed discontinuation rates are 7% on placebo, 8% on 40 mg, 10% on 80 mg, and 16% on 160 mg. The estimated excesses are:
The illustrative efficacy screen rules out 40 mg as the lead candidate because its 6-point estimate is below 8 points. Both 80 mg and 160 mg exceed the efficacy threshold. However, the estimated excess discontinuation rate for 160 mg is 9 percentage points, which exceeds the 5-point safety boundary. The 80-mg estimate is 3 points, below that boundary. On these point-estimate criteria, 80 mg is the clearest candidate to advance.
The 160-mg discontinuation excess is 6 percentage points higher than the 80-mg excess: \(9-3=6\). That difference is a comparison between doses, not the quantity assessed against the specified safety boundary. The boundary concerns each dose’s excess discontinuation rate versus placebo; it is the 160-mg excess of 9 points that exceeds 5 points.
The decision is not a declaration that 80 mg is risk-free or proven optimal. Suppose the 95% confidence interval for its efficacy contrast is 8 to 18 percentage points and the interval for its excess discontinuation rate is −1 to 7 points. The efficacy interval is compatible with effects at or above the illustrative efficacy threshold, while the safety interval includes values above the 5-point boundary. The dose may be advanced provisionally, but the safety uncertainty should inform monitoring, sample size, and the confirmatory benefit-risk assessment. The interval does not establish that the true excess risk is below 5 points.
Why Retain More Than One Dose?
A single dose is not always the most informative confirmatory choice. Retaining two doses can be justified when both appear viable and the remaining uncertainty is consequential: for example, a lower dose may offer a favorable safety or tolerability profile while a higher dose may provide additional benefit. Testing both can help determine whether the increment in efficacy is clinically meaningful and whether the lower dose delivers enough benefit with fewer harms. This is strongest when the confirmatory objectives, dose comparison, and analysis plan are specified in advance.
Retaining several doses is not a way to avoid making a decision. Each additional arm uses participants and resources, and additional comparisons can complicate interpretation. The planned design must account for the fact that more than one dose is being tested, including prespecified comparisons and appropriate control of the overall error rate where relevant. It should also ensure that each dose has a meaningful role in answering the confirmatory question.
Worked Example: Keeping Two Plausible Doses
Hypothetical data. In another phase II program, placebo-adjusted efficacy gains are estimated as 9 percentage points for 40 mg and 14 points for 80 mg. Their respective 95% confidence intervals are 5 to 13 and 10 to 18 points. Excess discontinuation estimates versus placebo are 1 point for 40 mg and 3 points for 80 mg, with intervals of −2 to 4 and 0 to 5 points. The prespecified efficacy point-estimate threshold is 8 points and the safety boundary is 5 points.
Both doses meet the point-estimate screens: \(9\geq8\) and \(14\geq8\) for efficacy, while \(1\leq5\) and \(3\leq5\) for excess discontinuations. The intervals indicate that uncertainty remains; in particular, the lower confidence limit for the 40-mg efficacy estimate is below the 8-point threshold. The data therefore do not show that both doses will certainly exceed that threshold.
The estimated incremental efficacy of 80 mg over 40 mg is \(14-9=5\) percentage points, and the incremental excess discontinuation estimate is \(3-1=2\) points. Those contrasts help frame the trade-off. They are not substitutes for comparing each dose with placebo against the stated absolute screening thresholds. Retaining both doses may be reasonable if the confirmatory question includes whether the additional 5-point efficacy estimate is worth the additional 2-point estimated discontinuation risk. A design that tests only 80 mg would not directly answer whether 40 mg offers a preferable benefit-risk balance.
The rationale for two doses is strongest when the phase II evidence leaves a real, decision-relevant distinction unresolved. If the lower dose has little chance of adequate benefit, or if the higher dose is plainly unacceptable on safety grounds, carrying both forward may add complexity without useful information.
Check Relevance, Uncertainty, and Feasibility
Before committing to a candidate, examine whether the evidence is sufficiently reliable and applicable. Useful checks include the amount of information at each dose, missing outcome data, adherence and dose interruptions, the consistency of efficacy and safety patterns across important groups, and whether the endpoint and follow-up duration match the confirmatory purpose. Model-based estimates should be checked against observed data and plausible alternative assumptions. These are decision diagnostics: a dose recommendation that changes sharply under reasonable choices deserves a more qualified rationale.
Feasibility is part of the decision, but it is a different type of evidence. Consider dosing burden, formulation, supply, visit schedule, recruitment, and whether participants can follow the intended regimen. A dose that is difficult to produce or administer may not be a practical confirmatory candidate even if its phase II efficacy estimate is attractive. Conversely, operational convenience is not evidence that the dose is effective or safe.
Worked Example: A Feasibility Screen Changes the Candidate Set
Hypothetical data. A program is considering two doses for a 52-week confirmatory trial. A feasibility rule, specified before reviewing the phase II completion data, requires at least 85% of participants in a dose arm to complete the planned 12-week phase II regimen. Of 100 participants assigned 80 mg, 88 complete; of 100 assigned 160 mg, 74 complete. The completion proportions are:
Against the illustrative 85% rule, 80 mg passes and 160 mg does not. The observed difference in completion is \(88\%-74\%=14\) percentage points. That difference is descriptive; it does not by itself show why participants did not complete, nor does it prove that the same difference would occur in a longer or broader trial. The completion data should be reviewed alongside reasons for discontinuation, exposure, and safety. If the doses are otherwise plausible, the 160-mg regimen may require further feasibility work or modification before confirmatory testing rather than being assumed viable from its efficacy estimate alone.
A common failure is to treat one favorable summary as a complete decision: “the highest dose had the largest response, so advance it.” This ignores safety, uncertainty, and the possibility that benefit has begun to plateau. The opposite error is to treat an uncertain result as if it were evidence of no benefit, or to label a dose safe because its estimated risk is below a boundary. The correction is to state what each estimate says, what it does not say, and how the remaining uncertainty affects the decision.
Other warning signs include selecting thresholds after inspecting the results, comparing an absolute safety boundary with the difference between two active doses, and advancing a regimen that does not match the intended confirmatory treatment strategy. A decision record should identify the evidence considered, the criteria applied, the dose or doses selected, the main uncertainties, and the reason alternatives were not selected.
Choosing an Advancement Strategy
The principal options are to advance one dose, advance more than one dose, or defer selection while obtaining additional evidence. A single-dose strategy can be efficient when one candidate has a favorable and adequately supported profile and testing another dose would not resolve an important question. A multi-dose strategy can preserve a meaningful benefit-risk comparison, at the cost of design and interpretive complexity. Deferral can be appropriate when evidence is too uncertain, safety concerns remain unresolved, or the planned regimen is not yet feasible.
| Strategy | When it may fit | Main limitation to address |
|---|---|---|
| Advance one dose | A candidate has a credible efficacy signal, acceptable safety evidence, and a feasible regimen. | Confirm that an alternative dose would not answer an important unresolved benefit-risk question. |
| Advance multiple doses | Several doses remain plausible and their comparison could change treatment decisions. | Plan the comparisons and account for the added sample-size and interpretive demands. |
| Defer selection | Key safety, efficacy, or feasibility uncertainty is too consequential for the planned confirmatory trial. | Specify what additional evidence would resolve the decision rather than repeating dose-ranging without a clear purpose. |
No strategy eliminates uncertainty. The aim is to choose a confirmatory test that can answer a clinically useful question with an interpretable dose rationale. Phase II evidence supports that choice; it should not be presented as a guarantee of confirmatory success.
Check Your Understanding
Answer each question, then compare your reasoning with the solutions.
- A dose has an estimated efficacy gain of 11 percentage points against a prespecified 8-point threshold, but its 95% confidence interval is 3 to 19 points. What does the interval add to the advancement decision?
- The safety boundary is a 5-point excess discontinuation rate versus placebo. The estimated excess is 8 points at 120 mg and 2 points at 60 mg. What is the correct boundary comparison?
- Two doses meet the efficacy and safety point-estimate screens, and the higher dose has a larger efficacy estimate. Give one reason to retain both for confirmatory testing.
- In a hypothetical trial, 81 of 90 participants complete a regimen. Calculate the completion proportion. Does it meet a prespecified minimum of 85%?
- Why is a dose with the largest phase II efficacy estimate not automatically the best dose to advance?
Solutions. (1) The estimate exceeds the threshold, but the interval includes effects below it; the estimate is promising, not precise evidence that the threshold is exceeded. (2) Compare each dose’s excess versus placebo with 5 points: 120 mg has an 8-point excess and is above the boundary; 60 mg has a 2-point excess and is below it. Do not compare the 6-point difference between doses with the boundary. (3) The lower dose may provide adequate benefit with less safety burden; retaining both can let the confirmatory study assess whether the higher dose’s additional benefit justifies its additional risk. (4) \(81/90=0.90=90\%\), so it meets the 85% minimum. (5) The largest efficacy estimate may come with greater harm, substantial uncertainty, or an infeasible regimen; advancement considers the full evidence and the confirmatory question.