Novel Drug Combination Trial Design

Definition

"A novel combination is a combination of two or more drugs being evaluated in a clinical trial where the combination itself has not been previously approved for the indication under study."

The FDA July 2025 draft guidance on Development of Cancer Drugs for Use in Novel Combinations addresses the fundamental challenge: when multiple drugs are combined, sponsors must characterize each drug's individual contribution to the combination's safety and efficacy profile.

Three combination types covered by the July 2025 draft:

1. Two or more unapproved investigational drugs combined — both drugs lack any regulatory approval
2. Investigational drug + drug approved for a different indication — one drug approved but not for the cancer indication under study
3. Two or more drugs each approved for different indications — both approved elsewhere, but not for the target indication or not in combination

Scope exclusion: The guidance does NOT cover combinations where one drug is already approved for the same indication being studied. Adding a new drug to an existing approved backbone (e.g., adding a new agent to R-CHOP in DLBCL) is covered by different regulatory considerations.

Core problem: Without demonstrating each component's contribution, a sponsor cannot establish that all components of a combination are necessary — leading to patients receiving extra toxicity from components that provide no benefit.

Status: FDA Novel Combinations July 2025 draft = Draft (comment period closed September 15, 2025)

---

Regulatory Position

Historical Context (2013 Codevelopment Guidance)

The 2013 FDA guidance on co-development of two investigational drugs established the framework for factorial designs in dual-investigational combinations. It was limited to two investigational drugs and did not formally address external data alternatives.

July 2025 Draft — Key Advances Over 2013

Aspect	2013 Guidance	2025 Draft
Scope	Two investigational drugs only	Three combination types; includes approved drugs in new indications; extends to 3+ drugs
Factorial design	Preferred approach	Still preferred but acknowledged as often infeasible in oncology
External data	Not formally addressed	Explicitly accepted as alternative to factorial arms with suitability criteria
Biologic rationale	Implied	Explicitly required; nonclinical characterization mandated before Phase 3 initiation
Toxicity justification	Limited	Added toxicity must be explicitly justified by expected contribution to benefit
Contribution definition	Vague	Clearly defined: each drug must show individual contribution beyond other components

FDA's Core Requirement: Contribution of Effect

Every drug in a combination must have its contribution characterized. "Contribution of effect" means: demonstrating that each drug individually contributes to the combination's efficacy beyond what the other drug(s) provide, and that the added toxicity of each drug is justified by its contribution to benefit.

What "contribution of effect" evidence must show:

1. Drug A alone has activity — mechanistic plausibility or early clinical evidence (Phase 1/2 data, PD biomarkers)
2. Drug B alone has activity — independent evidence of mechanism and clinical effect

3. Combination A+B shows greater efficacy than either drug alone — demonstrated via:

   • Factorial design (direct comparison within trial), OR
   • External data (Phase 3 evidence for Drug A alone + Phase 3 evidence for Drug B alone compared to the combination trial results)

4. Added toxicity of each drug is outweighed by efficacy contribution — explicit risk-benefit calculation per component

FDA 2025 explicit statement: "The FDA cannot conclude that all components of a combination are necessary to produce the claimed effect unless the applicant demonstrates the contribution of each drug to the combination's safety and efficacy profile." (July 2025 draft)

---

Factorial Design (Preferred Method)

Structure and Statistical Analysis

2×2 Factorial design: A randomization producing four arms:

Arm 1: Drug A + Drug B (combination) — n₁
Arm 2: Drug A alone — n₂
Arm 3: Drug B alone — n₃
Arm 4: Control/placebo or standard of care — n₄

Primary analysis:

• Main effect of Drug A: (Arm 1 + Arm 2) vs. (Arm 3 + Arm 4)
• Estimates Drug A's effect independent of Drug B

• Sample size: (n₁ + n₂) vs. (n₃ + n₄)

• Main effect of Drug B: (Arm 1 + Arm 3) vs. (Arm 2 + Arm 4)

• Estimates Drug B's effect independent of Drug A

• Sample size: (n₁ + n₃) vs. (n₂ + n₄)

• Interaction (A × B): Synergy or antagonism

• Test: (Arm 1 - Arm 2) - (Arm 3 - Arm 4) — does combination exceed additive effect?

Statistical testing hierarchy (FDA-recommended):

1. Test for interaction first: If A×B interaction p-value > 0.10, conclude no significant interaction and interpret main effects as marginal effects
2. If interaction not significant: Main effects are interpretable; each drug's contribution can be claimed
3. If interaction is significant: Do not interpret main effects in isolation; evaluate each arm separately and determine whether synergy justifies additional toxicity

Alpha allocation — requires pre-specification:

• Option A (Bonferroni): Split α across A effect, B effect, interaction (e.g., α_A = 0.0167, α_B = 0.0167, α_AB = 0.0167)
• Option B (Fixed-sequence): Test interaction first at α=0.10 (exploratory), then main effects at full α=0.05 if no interaction detected
• Option C (Closed testing): Use closed testing principle to maintain FWER while allowing flexible inference

Why Factorial Designs Are Often Infeasible in Oncology

1. Overlapping dose-limiting toxicities (DLTs):

   • Drug A causes dose-limiting neutropenia; Drug B also causes neutropenia
   • Combination requires dose reduction of both drugs (e.g., 75% + 75% instead of 100% + 100%)
   • Single-agent arms (at full 100% dose) become clinically irrelevant for understanding the actual combination being studied
   • Result: Factorial design fails to characterize the approved combination

2. Small biomarker-defined populations:

   • Rare mutations (BRCA-mutant, TMB-high, MSI-H) cannot support four arms with adequate statistical power
   • Accrual timelines become prohibitive (4-8 years instead of 2-3)

3. Ethical concerns:

   • Withholding one active drug from a patient who could receive both violates clinical equipoise in serious cancer settings
   • Single-agent arms may be considered unethical if both drugs are expected to be synergistic

4. Mechanism dependency:

   • Some combinations are only active together (Drug B activates Drug A's mechanism; Drug B alone has no activity)
   • Single-agent arm 3 would be expected to fail regardless, making the arm uninterpretable

5. Competitive/commercial pressure:

   • Sponsors prefer to avoid trial arms that might show a competitor's single agent is superior
   • This is not a scientific reason but influences design practice

---

External Data as Alternative to Factorial Arms

July 2025 draft formally accepts external data when factorial design is infeasible. External data can substitute for one or more single-agent arms, allowing a simpler design:

Criteria for External Data Acceptability

1. Contemporaneous: Data from the same calendar period as the combination trial

   • Recommendation: not >5 years apart (avoids Will Rogers effect from evolving standard of care)
   • Example: 2024 combination trial can use 2020-2024 single-agent data; 2019 data is borderline acceptable with justification

2. Similar population: Matching on key prognostic/predictive factors

   • Same line of therapy (1L vs. 2L vs. maintenance)
   • Same biomarker status (PD-L1+, MSI-H, BRCA mutation, etc.)
   • Same ECOG performance status distribution
   • Same age, ethnicity, geographic region if relevant
   • Avoid using 1L data as comparator for 2L combination trial

3. Endpoint concordance: External data must use identical primary endpoint

   • If combination trial uses PFS, external single-agent data must report PFS (not OS, not ORR)
   • If different endpoint definitions (e.g., RECIST 1.1 vs. Lugano), comparability must be justified

4. Suitability assessment required: Sponsors must submit formal analysis before Study initiation

   • Propensity score matching methodology (if PSM is used)
   • Baseline covariate comparability metrics (SMD < 0.1 preferred for all key covariates)
   • Sensitivity analyses (crude vs. PSM-adjusted; different caliper widths)

Acceptable External Data Sources

• Randomized controlled trials (RCT) — contemporaneous Phase 3 in similar population (gold standard)

• Example: using KEYNOTE-042 (pembrolizumab monotherapy 1L NSCLC) data as external control for pembrolizumab + PARP inhibitor combination in NSCLC

• Propensity-score-matched real-world data (RWD) — with limitations acknowledged in CSR

• Example: using IQVIA/Flatiron EHR data matched to trial population for prior single-agent efficacy

• Individual patient-level data (IPD) meta-analyses — pooling contemporaneous RCTs of the same single agent

• Historical single-arm data — acceptable only when RCT is impossible (e.g., orphan population, unethical control)

Unacceptable External Data

• Data >10 years old without documented population/treatment landscape stability
• Data from different treatment settings (e.g., using 2L PFS data as external control for 1L combination trial)
• Data from populations with substantially different biomarker profiles
• Different primary endpoint definitions without transparent justification
• Data from single-arm studies when RCTs exist

Statistical Approach with External Data

Example: Combination vs. SOC + External Single-Agent Data

Trial design:

- Arm A: Drug X + Drug Y + SOC (n=400)
- Arm B: SOC alone (n=400)
- Primary endpoint: PFS

Contribution of effect (external data):

- Drug X activity: Use Phase 3 RCT data (Drug X + SOC vs. SOC) 
  → HR_X = 0.65 (95% CI 0.52-0.81) from prior trial

- Drug Y activity: Use Phase 3 RCT data (Drug Y + SOC vs. SOC)
  → HR_Y = 0.58 (95% CI 0.45-0.74) from prior trial

- Combination expected effect: 
  → If additive: HR_combo ≈ HR_X × HR_Y = 0.65 × 0.58 = 0.377
  → If multiplicative on log scale: HR_combo ≈ exp[log(0.65) + log(0.58)] = 0.377

Analysis in current trial:

- If HR_current (Drug X + Drug Y + SOC vs. SOC) = 0.40, this is consistent 
  with or slightly better than additive effect

- Each drug's contribution supported by external data
- Combination effect explained by additive contributions of both drugs

---

Design Considerations

Biologic Rationale Requirement (Pre-Phase 3)

The July 2025 draft requires sponsors to provide strong biologic rationale BEFORE Phase 3 initiation:

1. Nonclinical characterization:

   • Mechanism-of-action (MOA) data for Drug A: target identification, cellular pathway, relevant cancer models
   • Mechanism-of-action (MOA) data for Drug B: independent target, pathway, models
   • Nonclinical combination data: In vitro synergy studies (isobolograms, combination index), in vivo tumor model data showing combination superiority over either agent alone
   • Pharmacology interaction: Do the drugs compete for the same pathway or act complementarily?

2. Preclinical combination activity:

   • In vitro synergy: Reduced cell viability at sub-optimal concentrations of both drugs vs. either drug alone
   • In vivo tumor model: Xenograft or syngeneic model showing combination regression > either agent alone
   • Dose-response: Document the dose ranges where synergy is observed

3. Early clinical evidence (Phase 1/2):

   • Phase 1: Combination safety, tolerability, MTD determination
   • Pharmacodynamic biomarkers: Evidence that Drug A engages its target AND Drug B engages its target in the combination
   • Phase 2: Preliminary efficacy data; ORR or PFS in early Phase 2 showing both components are active
   • Dose-response: Monotonic dose-response relationship per drug component (supports both contribute)

Early FDA engagement strongly recommended:

- Sponsor should submit combination package (MOA rationale, nonclinical data, Phase 1/2 evidence) in EOP1 meeting (end-of-Phase 2)
- Request FDA agreement on contribution-of-effect demonstration strategy BEFORE Phase 3 design is finalized
- FDA may recommend factorial design, external data approach, or additional early clinical data

Toxicity Justification

Each drug's contribution to the combination's toxicity must be characterized:

Framework:

• Baseline toxicity (Drug A alone): Grade 3/4 AE rate from Phase 3 single-agent data (or external trial)
• Added toxicity (Drug B alone): Grade 3/4 AE rate from Phase 3 single-agent data
• Combination toxicity: Observed in Phase 3 combination vs. control arm
• Attribution analysis: Which AEs are attributable to Drug A? Drug B? The interaction?

Example toxicity analysis:

Drug A monotherapy: 40% Grade 3/4 AEs
Drug B monotherapy: 35% Grade 3/4 AEs
Expected additive (assuming independence): ~60% [1-(0.6×0.65)] 
Observed combination: 65% Grade 3/4 AEs

Interpretation: Slight increase above expected additive suggests minor overlap 
in toxicity, or synergistic toxicity in ~5% of patients. Must be justified 
by expected efficacy benefit.

Benefit side (from trial):

- Drug A + B + SOC: PFS HR = 0.40 vs. SOC (60% reduction)
- Drug A + SOC (from external data): PFS HR = 0.65 (35% reduction)
- Drug B contribution: inferred HR ~0.58 (42% reduction, from external data)
- Combined contribution exceeds individual contributions (synergistic benefit)

Risk-benefit: 25% additional Grade 3/4 toxicity (60% → 85%) + 5% higher 
toxicity rate justified by synergistic PFS benefit (60% reduction vs. 
expected 50% additive reduction).

Sample Size Implications

Simple two-arm (Combination vs. Control):

• Sample size: n per arm determined by primary endpoint (PFS HR), standard Phase 3 formula
• Power on each drug's contribution comes from external data (no within-trial comparison)
• Example: 400 per arm for PFS endpoint, HR = 0.65, α=0.05, 80% power

Factorial four-arm (2×2 design):

• To achieve 80% power on each main effect (Drug A effect, Drug B effect) with HR = 0.70:
• Per-arm requirement: ~175 per arm (compared to ~280 for simple two-arm)
• Total enrollment: 175 × 4 = 700 (vs. 280 × 2 = 560 for simple two-arm)
• Power loss: Factorial requires ~25% more total patients for equivalent power on main effects

Three-drug combinations:

• Factorial becomes 2×2×2 = 8 arms (impractical; only used in rare dose-finding Phase 1b trials)
• Recommended approach: Combination vs. SOC as primary; contribution demonstrated via external data or sequential Phase 2 studies (Drug A approved → add Drug B in Phase 2 → Phase 3 with both)

Endpoint Selection for Combination Trials

The primary endpoint for the combination trial must demonstrate that the combination is superior to current standard. The contribution-of-effect demonstration addresses why each drug is needed — it does not require that each drug individually be superior to SOC.

Common endpoint patterns:

• Primary (within-trial): Combination vs. SOC: PFS or OS (same as any Phase 3 oncology trial)
• Contribution demonstration (external or exploratory): External data showing each drug alone vs. similar control
• Sensitivity (within-trial subgroups): Biomarker-defined subgroups that respond more/less to each component

Example: IO + PARP Inhibitor

• Primary analysis: (IO + PARP + chemo) vs. (chemo alone) — PFS

• Required: HR ≤ 0.75, p < 0.05

• Contribution of IO: Use external Phase 3 data (IO monotherapy in same population) showing IO PFS benefit

• Contribution of PARP: Use external Phase 3 data (PARP monotherapy in BRCA-mutant population) showing PARP PFS benefit
• Sensitivity: Stratified analysis by BRCA status, PD-L1 status to evaluate if both components contribute in key subgroups

---

Intercurrent Events in Combination Trials

Combination trials introduce unique IE complexity not present in single-agent trials:

IE: Discontinuation of One Combination Component (Not the Other)

Scenario: Patient on Drug A + Drug B discontinues Drug B due to toxicity (hepatotoxicity) but continues Drug A. Efficacy continues to accrue.

Primary strategy: Treatment Policy

• Patient remains in "Combination" arm for primary analysis
• All subsequent efficacy events (PFS, OS) are attributed to the combination per randomized assignment
• Rationale: Intent-to-treat principle; real-world combination management includes tolerance-driven modifications

Sensitivity strategy: Hypothetical (Exposure-Based)

• Analysis restricted to patients who maintained ≥80% of planned cumulative dose for BOTH drugs
• Approximates "what would efficacy be if patients had maintained full combination exposure?"
• But: Cannot fully control for confounding by indication (patients who had to stop Drug B may have had more aggressive disease)

SAP language:

"The primary analysis applies the treatment policy estimand. Patients who discontinue one combination component but continue the other remain in the combination arm analysis and are followed for efficacy events per protocol schedule.

Sensitivity analysis 1: Restricted to patients with ≥80% dose intensity of both Drug A and Drug B (hypothetical scenario of full exposure).

Sensitivity analysis 2: Contribution-of-effect analysis stratified by which component was discontinued (Drug B discontinuation vs. both continued) to assess whether efficacy derives from one drug or both."

IE: Dose Modification of One Component

Scenario: Combination planned at 100% Drug A + 100% Drug B. During trial, many patients receive 75% Drug A + 100% Drug B.

Strategy: Treatment Policy

• Dose modifications are part of real-world combination management
• Treat all dose-modified patients as received the full combination
• Analysis per randomized assignment

Contribution-of-effect analysis:

• Perform exploratory analysis of PFS/OS by dose intensity delivered per drug (per quartile of cumulative dose)
• Document whether efficacy correlates with higher vs. lower exposure to Drug A, Drug B, or both
• Supports or refutes each drug's contribution

SAP language:

"Dose modifications to individual components are permitted per protocol and are considered part of real-world combination management. Patients will be analyzed in their randomized arm regardless of dose modifications.

Exploratory analysis: Efficacy outcomes (PFS/OS) stratified by dose intensity quartiles for Drug A and Drug B separately, to evaluate the contribution of each component's exposure to observed outcomes."

IE: Crossover from Control Arm to One Drug (But Not Both)

Scenario: Patient on control (SOC only) experiences progression. Physician offers Drug A + Drug B, but patient only accepts Drug A.

Strategy: Treatment Policy for OS

• Per randomized assignment (control arm)
• Only use data before crossover for primary analysis

Sensitivity: Hypothetical (RPSFT) for Partial Regimen Crossover

• Challenge: RPSFT (Rank-Preserving Structural Failure Time) is designed for crossover to the full experimental regimen
• Partial crossover (one drug but not both) complicates RPSFT: What is the "experimental effect" to estimate when only Drug A is received?
• Approach: Pre-specify in SAP whether partial crossover is:
• Treated as no crossover (patient received neither combination component, so no adjustment)
• Treated as full combination receipt (patient received Drug A, which is part of the combination, so adjust for Drug A effect)
• Treated separately (RPSFT adjustment for Drug A component only, not for Drug B)

SAP language:

"Patients who cross over from control to one combination component (not both) will be analyzed in their randomized control arm per treatment policy. Sensitivity analysis: Patients who crossed over to Drug A only will be excluded from the primary analysis (per-protocol subset), or RPSFT adjustment for Drug A effect will be applied with pre-specified log(HR) assumptions derived from external Drug A monotherapy data."

---

SAP Requirements for Combination Trials

Beyond standard SAP requirements, combination trial SAPs must include:

1. Contribution-of-effect section:

   • Pre-specified analysis plan for demonstrating each drug's contribution
   • External data comparator: Identification, suitability assessment methodology
   • Propensity score matching algorithm (if applicable): Covariates, caliper, matching ratio
   • Population comparability metrics: Baseline covariate balance (SMD, etc.)
   • Sensitivity analyses: Crude vs. PSM-adjusted; different PS caliper widths

2. Estimand table (per ICH E9(R1)):

   • Cross-tabulation of IEs specific to combination management:
   • One-drug discontinuation (Drug A only vs. Drug B only vs. both)
   • Dose modification per drug (per 25% decrement)
   • Crossover (full combination vs. one drug only)
   • Specify strategy for each IE: treatment policy, hypothetical, composite

3. Toxicity attribution analysis:

   • Pre-specified analysis of Grade 3/4 AE rates by drug exposure received
   • Exposure-response framework: Event rate vs. dose intensity quartiles per drug
   • Interaction detection: Are certain AEs elevated only in patients receiving both drugs?

4. Sensitivity to partial exposure:

   • Analysis in patients receiving ≥80% planned dose intensity for each component
   • Evaluate whether efficacy/toxicity relationship changes with adherence level

---

Common Scenarios and Strategies

Scenario 1: IO + Targeted Therapy Combination (e.g., Pembrolizumab + PARP in BRCA-Mutant Solid Tumors)

Contribution challenge: Both pembrolizumab (IO) and PARP inhibitor have single-agent activity in BRCA-mutant tumors. A factorial design would require pembrolizumab-alone arm vs. PARP-alone arm vs. combination vs. control (4 arms, ~700 patients).

Practical approach (per 2025 draft):

1. Primary design: Combination (IO + PARP + platinum chemo) vs. SOC (platinum chemo alone), randomized 1:1, n=400 per arm

2. Contribution of IO:

   • Use external Phase 3 data: pembrolizumab monotherapy in same population (PFS HR = 0.70 vs. chemo)
   • Document population comparability (same histology, line of therapy, BRCA mutation, ECOG)

3. Contribution of PARP:

   • Use external Phase 3 data: PARP inhibitor monotherapy in BRCA-mutant population (PFS HR = 0.65 vs. placebo/chemo)
   • If no exact comparable population, use SOLO-2 or NOVA data with justification

4. Expected combination effect:

   • If additive (multiplicative on HR scale): HR ≈ 0.70 × 0.65 = 0.455
   • Trial should show: HR ≤ 0.55 (combination effect consistent with or exceeding additive)

5. SAP pre-specifies: Population comparability metrics, statistical methods for comparing trial result to external data HR

Regulatory outcome: If trial shows HR = 0.50 for combination, sponsors argue: "Each drug independently contributes (supported by external data); combination shows additive or synergistic benefit."

Scenario 2: Novel IO + Novel IO Combination (e.g., Two Checkpoint Inhibitors with Different Targets)

Contribution challenge: Two unapproved drugs; no external Phase 3 data available. Must demonstrate contribution.

Required approach (per 2025 draft):

• Factorial design strongly preferred — no external data alternative for two investigational drugs
• If factorial infeasible due to overlapping toxicities:
• Demonstrate contribution via Phase 1/2 dose-escalation single-agent data
• PD biomarkers showing each drug engages its target in the combination
• Early combination data (Phase 2) showing synergy beyond expected additive effect
• Early FDA engagement to agree on the contribution-of-effect strategy

Example design:

• Phase 1b/2: Dose-escalation of Drug A + Drug B; confirm PD biomarker engagement (e.g., T-cell activation for both drugs)
• Phase 2: Single-arm cohort (Drug A + Drug B) showing ORR 50% + PFS HR vs. historical control
• Phase 3: Primary (Drug A + Drug B vs. SOC); contribution supported by Phase 1/2 biomarker and early efficacy data

Regulatory likelihood: FDA would require Phase 3 single-arm efficacy data plus strong mechanistic evidence before approving based on Phase 2 alone. Factorial Phase 3 remains the gold standard for two investigational drugs.

Scenario 3: Bispecific Antibody as "Combination" (One Drug, Two Targets)

Example: Bispecific antibody targeting PD-L1 and 4-1BB simultaneously.

Status: NOT covered by combination guidance.

• A bispecific antibody is a single drug, not a combination
• Each target's contribution is addressed through the drug's mechanistic characterization and dose-response data, not through combination trial design
• Contribution of the PD-L1 arm vs. 4-1BB arm is evaluated via:
• Nonclinical data showing both functionalities work
• Phase 1 PD biomarkers (T-cell activation specific to each pathway)
• Phase 2 dose-response (showing that both targets are needed for observed efficacy)

---

Regulatory Precedent

Combination Type	Example Trial	Design	Contribution Strategy	Outcome
IO + IO	CHECKMATE-067 (nivolumab + ipilimumab)	Factorial (nivo+ipi vs. nivo vs. ipi vs. placebo)	Within-trial comparison of nivo+ipi vs. each component	Approved: nivo+ipi superior to either alone in melanoma
IO + IO	CHECKMATE-069 (nivo+ipi)	Single-arm Phase 2	Phase 1 dose-escalation + early efficacy	FDA accepted Phase 2 data for Breakthrough Therapy + Phase 3 confirmatory
IO + Targeted	IMpower150 (atezolizumab + bevacizumab + chemo)	Factorial (4 arms)	Within-trial: atezo+bev+chemo vs. each component vs. chemo	Approved: combination superior; both drugs contribute
PARP + IO	PROfound (olaparib + durvalumab)	Simple 2-arm + external data	Phase 2 single-agent activity; Phase 3 combination vs. SOC	Approved: used external data for contribution of each drug
CDK4/6 + IO	Multiple trials (e.g., LEA003, OncoES-2)	Factorial or simple designs	Phase 1b showed feasibility; Phase 2/3 sought synergy	Negative/neutral results; mechanistic rationale questioned

---

Limitations and Pitfalls

1. Contribution of effect from separate trials (methodologically weak):

Using Drug A Phase 3 data and Drug B Phase 3 data from different trials to argue both contribute is methodologically weak:

• Different populations, different time periods, different control arms
• Different standards of care (SOC evolved between trials)
• July 2025 draft acknowledges this limitation but accepts it when factorial design is truly infeasible
• Mitigation: Use contemporaneous (within 5 years) data with documented population comparability

2. Synergy ≠ contribution:

• A statistically significant interaction term in a factorial trial confirms non-additive effects but does NOT by itself establish that both drugs are necessary
• Negative interaction (antagonism): May show that the combination is worse than either drug alone (e.g., CDK4/6 inhibitor + checkpoint inhibitor trials showed antagonism in some populations)
• Mitigation: Pre-specify and transparently report interaction analysis; if antagonistic, explain why combination is still justified (e.g., one drug is required for mechanism; other drug adds additional benefit in a subset)

3. Toxicity asymmetry (underappreciated):

• If Drug A causes 50% Grade 3/4 AEs and Drug B adds another 30%, the patient receives effectively 80% Grade 3/4 (not 50% + 30%)
• The benefit-risk calculation must be transparent: Is a 25% PFS improvement worth 30% additional Grade 3/4 toxicity?
• Mitigation: Pre-specify risk-benefit thresholds in protocol; if combination exceeds thresholds, require DSMC review and potential redesign

4. Post-approval single-agent use off-label:

• After combination approval, single-agent use (one drug without the other) becomes common in clinical practice
• Combination approval does NOT validate single-agent use; safety/efficacy for single agents is not established
• Mitigation: Clear labeling stating the approved combination regimen; pharmacovigilance for off-label single-agent use; potential additional trials if off-label use becomes widespread

5. Contribution of effect inflation (unvalidated assumptions):

• Sponsors assume additive effects (HR₁ × HR₂) when actual interaction may be subadditive or antagonistic
• External data from older trials may not reflect current drug landscape or patient phenotypes
• Mitigation: Conservative assumptions (use lower single-agent effect estimates); sensitivity analyses with varying interaction assumptions

---

Backlinks

• FDA Approval Pathways in Oncology
• Multiple Endpoints and Alpha Allocation
• Sample Size Re-estimation (SSR)
• ICH E9(R1) Estimand Framework
• Intercurrent Events in Oncology Trials
• NSCLC Indication Guide: FDA Regulatory Endpoints & Trial Design Patterns
• Multiple Myeloma Trial Design Patterns

---

Source: FDA Draft Guidance — Development of Cancer Drugs for Use in Novel Combinations: Determining the Contribution of the Individual Drugs (July 17, 2025, Draft); FDA Guidance for Industry — Codevelopment of Two or More Unmarketed Investigational Drugs for Use in Combination (June 2013, Final) Status: July 2025 novel combinations draft = Draft (comment period closed September 15, 2025); 2013 codevelopment guidance = Final Compiled from FDA 2025 draft guidance + published factorial and combination trial designs + regulatory precedent