Clinical Trial Design

Evaluating Contribution of Effect in Novel Cancer Drug Combinations

A sponsor-side framework for selecting evidence and trial designs to show what each component contributes, while separately evaluating the safety and benefit-risk of the combination.

1. The Co-Development Challenge in Modern Oncology

The landscape of oncology treatment has radically shifted over the past decade. Monotherapies are increasingly being replaced by rational drug combinations, particularly involving immune checkpoint inhibitors (IO) paired with targeted biologics, chemotherapies, or other novel IO agents. The goal is to achieve synergistic efficacy, overcome acquired resistance mechanisms, and elicit durable responses in refractory tumors.

However, combining multiple agents introduces regulatory and clinical complexities, including the need to understand each component's contribution to benefit relative to added toxicity. Under 21 CFR 300.50, a fixed-combination prescription drug must include each component in a quantity that contributes to the claimed effects, with exceptions described in the regulation. FDA's recommendations for demonstrating contribution of effect depend on the development scenario and available evidence; the rule should not be read as requiring one identical trial design for every oncology combination.

2. Trial Design Strategies: Add-on vs. Factorial Designs

The appropriate design depends on the drugs' development history, indication, prior evidence, and whether the question is the contribution of each component or the benefit of adding a new drug to established standard of care. An "add-on" design (A+B vs. A alone) can test whether adding investigational drug B improves outcomes over A, but it does not, by itself, isolate the contribution of A when both A and B are new investigational drugs. The FDA's 2025 oncology-specific draft guidance discusses multiple development scenarios and potential use of external data; it is a nonbinding draft, not final guidance.

It is important to distinguish an add-on efficacy question from a component-contribution question. If A is an established standard of care for the same indication and the trial asks whether adding B improves outcomes (A+B vs. A), that is an add-on-to-standard-of-care question. The FDA's July 2025 oncology draft guidance on contribution of effect explicitly excludes this setting. By contrast, when A and B are being developed as a novel combination and the evidence must establish that each contributes to the regimen, the study may need component arms or a scientifically justified alternative.

DesignMain questionKey limitation
A+B vs. A (established same-indication SOC)Does adding B improve the established regimen?Does not establish the contribution of A when both A and B are new investigational components.
A+B vs. A vs. B (± SOC)Does each component contribute, and how does the combination compare with SOC?May be infeasible or unethical if monotherapy arms expose participants to inadequate treatment; requires adequate power and prespecified comparisons.
Modified three-arm designCan the contribution of a component with limited monotherapy activity be isolated?The arm structure and estimand must fit the known activity of each component; it is not a universal substitute for a full factorial design.
A+B vs. A with fit-for-purpose external evidenceCan reliable component-level data reduce the need for a prospective monotherapy arm?Cross-study differences and unmeasured confounding can undermine causal interpretation.

When both investigational components have meaningful single-agent activity and can ethically be administered separately, a randomized factorial design is generally the clearest way to estimate each component's contribution. A conventional multi-arm structure may include A+B, A, B, and a standard-of-care arm. If one component has limited monotherapy activity, a modified design may be more appropriate; the 2013 codevelopment guidance discusses scenario-specific options, including a comparison of the combination with the more active drug when Phase 2 evidence already suggests the combination is superior. The 2025 oncology draft also discusses adaptive factorial designs that may drop futile arms, provided adaptations, type I error, and power are addressed prospectively.

Do not infer that every oncology combination requires all four arms. Feasibility, equipoise, disease severity, available treatment, the activity of each component, and the evidence already available should determine the design. The scientific question and primary estimand should be agreed before choosing the number of arms.

3. Pharmacological Rationale and Preclinical Synergy

A scientifically credible rationale should support evaluating a combination, drawing on mechanistic understanding, nonclinical findings, and early clinical evidence as appropriate. However, formal demonstration of synergy is not a universal prerequisite for every combination. Preclinical models and biomarkers can inform the rationale, but results from these sources do not by themselves establish clinical benefit or prove that each drug contributes to the observed effect.

Combination development should also evaluate pharmacokinetic and pharmacodynamic interactions, overlapping toxicities, dose selection, and schedule as appropriate to the agents. For example, overlapping hepatotoxicity or effects on the same metabolic pathway may warrant additional studies or dose adjustments. Contribution-of-effect evidence is one part of the overall benefit-risk assessment; it does not replace the need to characterize safety, dose, and interaction risks.

4. When Can External Data Support Component Contribution?

The July 2025 FDA draft guidance recognizes that a complete randomized factorial design may not always be feasible. External data may be considered when the combination has a strong biological rationale, disease natural history is sufficiently predictable, the combination is expected to produce a large effect, or a component has already shown limited activity alone. This is an evidentiary alternative to be justified—not permission to compare unrelated response rates from different studies.

For a comparative analysis, patient-level data are substantially more informative than published summary results. The FDA draft identifies comparability of population and treatment setting, clinically relevant covariates, data provenance and auditability, missingness, exposure and outcome ascertainment, and a prespecified statistical analysis plan as important considerations. Summary-level results from published studies should generally be treated as hypothesis-generating rather than as a reliable substitute for a prospective component arm.

External-comparator fitness-for-purpose check

  • Population: Are disease subtype, line of therapy, eligibility criteria, baseline prognosis, and prior treatment sufficiently comparable?
  • Time: Are treatment eras, index dates, follow-up windows, and subsequent therapies aligned?
  • Variables: Are key prognostic and predictive covariates measured consistently, with missingness characterized?
  • Endpoints: Are response criteria, assessment schedules, censoring rules, and independent-review methods comparable?
  • Data integrity: Can source data, transformations, exposure, outcomes, and analysis populations be traced and audited?
  • Analysis: Is the protocol and statistical analysis plan prespecified, including methods for measured confounding and sensitivity analyses?

Propensity-score methods, outcome regression, weighting, or Bayesian approaches may help address measured differences, but none can reliably remove bias from unmeasured confounding or repair fundamentally non-comparable data. FDA's draft also highlights time-related bias for time-to-event outcomes, incomplete death capture in some real-world sources, and differences in tumor assessment. A large observed treatment effect may strengthen the case, but does not automatically make a weak comparator fit for regulatory inference.

5. Match the Endpoint to the Contribution Question

A response endpoint such as ORR, supported by duration-of-response information, or a persuasive pharmacodynamic/response biomarker may provide an earlier direct signal of antitumor activity. However, an early response difference does not necessarily show that every component contributes to longer-term outcomes such as progression-free survival or overall survival. A component could increase initial tumor shrinkage without improving durability, or could add toxicity that offsets an early efficacy gain.

For external comparisons, endpoint definitions and assessment schedules must be aligned. Overall survival is objective, but real-world death data may be incomplete and subsequent anticancer treatments can confound interpretation. Time-to-event analyses also require comparable index dates, follow-up, and censoring. The endpoint strategy should therefore be tied to the clinical question, disease context, and evidence available—not selected only because it produces an earlier readout.

6. Contribution of Effect Is Not the Same as Combination Safety

Evidence that each component contributes to efficacy does not establish that the regimen has an acceptable benefit-risk profile. The 2025 oncology draft guidance explicitly does not address safety or dose-selection considerations. Sponsors still need to characterize pharmacokinetic and pharmacodynamic interactions, overlapping or synergistic toxicities, dose and schedule, exposure-response relationships, and clinically relevant concomitant medications as appropriate to the agents.

For example, a combination may improve response while increasing clinically important hepatotoxicity, myelosuppression, QT prolongation, or infection risk. Dose optimization may change the balance between contribution and toxicity; a component's contribution at one dose cannot automatically be assumed at another. Safety monitoring and dose-modification rules should be developed as a parallel workstream rather than treated as a by-product of an efficacy comparison.

7. Development-Stage Decision Framework

  1. Define the regulatory scenario: two or more investigational drugs; an investigational drug plus a drug approved for another indication; or drugs approved for other indications. Confirm whether the question is instead a conventional add-on trial to same-indication standard of care, which is outside the scope of the 2025 draft.
  2. State the component-level hypothesis: specify what each drug is expected to contribute and what evidence would distinguish contribution from biological plausibility alone.
  3. Map existing evidence: summarize monotherapy activity, early combination data, nonclinical rationale, prior safety experience, and the relevance of available external datasets.
  4. Select the strongest feasible design: favor randomized component arms when feasible; justify modified factorial or external-data approaches when a full factorial design is not appropriate.
  5. Prespecify endpoint and analysis: define estimands, multiplicity/type I error, power, handling of missing data, external comparator construction, and sensitivity analyses before examining comparative outcomes.
  6. Discuss the strategy with FDA early: seek feedback on the evidence package and design, particularly when external data or nonstandard comparisons are central to the argument.
  7. Run a separate benefit-risk assessment: integrate efficacy contribution with dose, schedule, interaction, and toxicity evidence before advancing or filing the regimen.

Scope note: FDA's July 2025 document remains a draft and is not for implementation as of November 2026. It describes the Agency's current thinking for comment purposes, not a binding requirement. The 2013 codevelopment guidance remains the final guidance for its narrower scope of two or more new investigational drugs. Confirm current FDA materials and obtain program-specific regulatory advice before relying on a particular design.

Toolkit Tip: For a simple, prespecified comparison of categorical response proportions between two independent arms, our Chi-Square Calculator can calculate a chi-square P-value and Cramér's V. Check expected cell counts and consider an exact method when counts are sparse. This calculator does not adjust for stratification, covariates, multiplicity, time-to-event outcomes, or the causal limitations of external comparisons; it is not a substitute for a trial-specific statistical analysis plan.