RCT Design Fundamentals in Oncology

Definition

A randomized controlled trial (RCT) is an interventional study in which eligible participants are allocated by chance to one of two or more treatment conditions, with at least one concurrent control group measured over the same period of time.

Per ICH E10 (Final, July 2000), the control group serves "one major purpose: to allow discrimination of patient outcomes (for example, changes in symptoms, signs, or other morbidity) caused by the test treatment from those caused by other factors, such as the natural progression of the disease, observer or patient expectations, or other treatment." The control group "tells us what would have happened to patients if they had not received the test treatment or if they had received a different treatment known to be effective."

Per ICH E8(R1) (Final, October 2021), rigorously designed studies require "careful attention to the design elements, such as the choice of study population, control group, response variables, randomization, blinding, and minimisation of bias."

Randomization and blinding are the two core structural tools that operationalize the RCT design: "Randomization and blinding are the two techniques usually used to minimize the chance of such bias and to ensure that the test treatment and control groups are similar at the start of the study and throughout the trial." (ICH E10)

Regulatory Position

RCTs are the evidentiary gold standard across all FDA approval pathways:

Approval Pathway	RCT Requirement	Notes
Regular Approval	Required (typically 2 adequate and well-controlled studies)	OS, PFS, or validated surrogate as primary endpoint
Accelerated Approval	Required; single trial acceptable	Reasonably likely surrogate acceptable
Breakthrough Therapy	Required; flexibility on single-arm for initial BT designation	Confirmatory RCT usually required post-BT
Fast Track	RCT preferred; SA acceptable for rare cancers with dramatic effect	Must confirm benefit

ICH E10 is a Final guidance (July 2000). ICH E8(R1) is a Final guidance (October 2021). Both are adopted by FDA, EMA, and other ICH member authorities.

Key regulatory requirement per ICH E10: "The choice of the control group should be considered in the context of available standard therapies, the adequacy of the evidence to support the chosen design, and ethical considerations."

When to Use

Parallel-group RCT is the standard design for virtually all Phase 3 oncology trials:

Metastatic solid tumors (NSCLC, RCC, breast, CRC): 1L, 2L, later lines
Adjuvant and neoadjuvant settings (breast, colon, melanoma): DFS/EFS/pCR as primary endpoints
Maintenance therapy (ovarian, myeloma, NSCLC): PFS in remission
Consolidation (AML, lymphoma): EFS following induction

Crossover design is essentially never used as the primary RCT design in oncology because:

Disease progression or death ends the crossover period prematurely in most patients
Carryover effects from the first treatment period confound the second period
Exception: crossover from control to experimental arm after progression (not a crossover design — this is a subsequent therapy intercurrent event handled analytically)

Non-inferiority RCT is used for de-escalation scenarios:

Shortening adjuvant chemotherapy duration (e.g., 3 vs 6 months CAPOX in stage III CRC)
Substituting less toxic regimens with hypothesized preserved efficacy
Biosimilar comparisons (equivalence design)
Radiotherapy dose reduction trials

Design Considerations

1. Trial Architecture: Parallel vs Crossover

Parallel-group: Patients randomized to arm A or arm B; both arms run concurrently for the duration of the trial. All Phase 3 oncology registration trials use this architecture.

Crossover: Each patient receives both treatments in sequence (A→B or B→A). Requires washout between periods. Infeasible in oncology for efficacy because:

Tumor progression after Period 1 changes the patient's disease state
The proportional hazards assumption is violated after crossover
Most patients will have progressed or died before completing Period 2

Practical implication: When the protocol allows control-arm patients to receive the experimental drug after confirmed progression, this is not a crossover design — it is a subsequent therapy intercurrent event. The estimand and censoring rules must pre-specify how this is handled (see Intercurrent Events section).

Randomized withdrawal design (section 2.1.5.2.4 of ICH E10): All patients receive open-label treatment initially; responders or stable patients are then randomized to continue or switch to placebo. Used in maintenance oncology settings (e.g., lenalidomide maintenance in myeloma) to enrich for a population likely to benefit.

2. Inferential Objective: Superiority vs Non-Inferiority vs Equivalence

The choice of objective determines the hypothesis, control type, sample size, and NI margin (if applicable).

Objective	Null Hypothesis	Test Treatment Goal	When Appropriate
Superiority	Δ ≤ 0 (or HR ≥ 1)	Demonstrate benefit over control	New agent vs placebo/SOC; add-on vs SOC alone
Non-inferiority	Δ < −M₂ (or HR > NI_HR)	Show loss not worse than M₂	De-escalation; when placebo unethical; novel route/formulation
Equivalence	\|Δ\| > M₂	Show difference within ±M₂	Biosimilar comparisons; generic approval

ICH E10 caution on NI: "There are circumstances in which a finding of non-inferiority cannot be interpreted as evidence of efficacy. Specifically, for a finding of non-inferiority to be interpreted as showing efficacy, the trial must have assay sensitivity" — meaning the active control must itself be capable of distinguishing effective from ineffective treatments in the current trial conditions.

Switching from NI to superiority: Pre-specified hierarchical test: first test NI (one-sided α = 0.025), then if NI is established, test superiority (one-sided α = 0.025). No alpha penalty if pre-specified. Cannot reverse (superiority first then NI) without inflation.

3. Active Control Selection: ICH E10 Framework

When placebo is unethical (effective standard therapy exists preventing death or irreversible morbidity), an active (positive) concurrent control is used.

Active control trial purpose (ICH E10 §2.4): Can support either (a) superiority claims if the test drug beats the active control, or (b) non-inferiority claims if assay sensitivity is established. Most Phase 3 oncology trials use active controls for superiority (experimental + SOC vs SOC alone, or experimental vs SOC).

Non-inferiority margin derivation — ICH E10 two-step framework:

Step 1 — Define M1 (entire effect of active control vs placebo): Estimate the full historical benefit of the active control over placebo from prior randomized trials. Use the lower confidence bound (conservative estimate) of the pooled treatment effect. M1 represents the maximum amount the test drug could be inferior to the active control while still being superior to placebo.

Step 2 — Define M2 (acceptable fraction of M1 to preserve): Specify the minimum fraction of M1 that must be preserved. Typically 50% in general therapeutics; 75–90% preservation is increasingly expected in oncology because patients should not sacrifice substantial survival benefit. M2 = M1 × (1 − fraction to preserve).

Example (per non_inferiority_oncology_summary.md): In adjuvant CRC, if M1 from historical data = 8% DFS benefit (lower CI bound), and 50% preservation required, then M2 = 4% — meaning the new regimen may not be more than 4% inferior in DFS.

Assay sensitivity requirement (ICH E10 §1.5): The trial must demonstrate "historical evidence of sensitivity to drug effects" — prior trials must have consistently shown the active control beats placebo. If this cannot be demonstrated (e.g., the active control has never been compared to placebo in the current setting, or its effect is unstable), non-inferiority cannot be reliably interpreted as evidence of efficacy.

Constancy assumption: The historical effect of the active control vs. placebo must remain constant in the new NI trial. This is threatened if:

Patient populations have changed substantially (e.g., more aggressive supportive care)
Concomitant therapies have improved
The active control's optimal dose/schedule has been revised

Control arm dose selection (ICH E10 §1.4.3.1): "When a comparative efficacy trial is conducted, both test and control drugs should be studied at doses and regimens that are optimal or represent the best available treatment." Deliberately underdosing the active control to make NI easier to achieve is a critical regulatory concern.

4. Randomization Methods

Purpose (ICH E10 §1.2.1): "Assurance that subject populations are similar in test and control groups is best attained by randomly dividing a single sample population into groups." "Randomization also provides a sound basis for statistical inference."

Method	Description	Use in Oncology
Simple randomization	Coin-flip equivalent at each allocation; no blocking	Rarely used; imbalance risk with small N
Block (permuted block)	Blocks of fixed or variable size ensure balance within blocks	Standard in Phase 3; variable block size prevents prediction
Stratified block	Separate randomization lists per stratum combination	Phase 3 standard; controls baseline prognostic imbalance
Minimization	Adaptive allocation minimizing imbalance across multiple factors simultaneously	Used when many stratification factors; must be pre-specified
Biased-coin	Increased probability toward underrepresented arm	Rarely used in oncology; more common in adaptive designs

Allocation ratio: Most Phase 3 trials use 1:1 (maximum power per patient). 2:1 allocation toward experimental arm is used when:

More safety data needed on new agent
Ethical preference to expose more patients to experimental treatment
Power loss is acceptable (inflation ≈ 11% for 2:1 vs 1:1 with same total N)

IWRS (Interactive Web/Voice Response System): Central randomization via IWRS is the regulatory standard. IWRS maintains allocation concealment, enforces stratification, produces audit trails, and can dynamically manage block sizes.

5. Stratification Factors

Purpose: Balance prognostic factors across arms at baseline; prevent confounding by known strong predictors; specified strata are typically included as covariates in the primary analysis (stratified log-rank test or Cox model).

Standard oncology stratification factors (by setting):

Setting	Typical Factors
NSCLC	ECOG PS (0–1 vs 2), histology (squamous vs non-squamous), line of therapy, biomarker status (PD-L1 %, EGFR/ALK)
Breast	HR status (ER/PR), HER2, menopausal status, geographic region
CRC	RAS/BRAF status, line, liver metastases only vs other
Myeloma	ISS stage, cytogenetics (high-risk vs standard), prior lines
Pan-tumor	Geographic region (regulatory requirement for multi-regional trials)

Selection criteria for stratification factors:

Must be a strong, documented prognostic factor for the primary endpoint
Must be assessable at randomization with high reliability
Creates cells with ≥10–15 expected events (avoid overstratification)
Cap at 3–4 factors maximum for a 2:1 ratio trial (too many creates sparse cells)
Must be recorded centrally and pre-specified in the SAP

Critical alignment rule: Stratification factors used at randomization must be used as covariates in the primary analysis (stratified log-rank test or Cox model with strata). Mismatch between randomization strata and analysis strata is a common FDA query.

ICH E9(R1) estimand implication: Stratification factors define subpopulations. If the estimand targets a specific subpopulation (e.g., PD-L1 high), the randomization must stratify on that factor and the primary analysis must be appropriately powered for that subgroup.

6. Blinding Strategies

Per ICH E10 (§1.2.2): "Clinical trials are often 'double-blind' (or 'double-masked'), meaning that both subjects and investigators, as well as sponsor or investigator staff involved in the treatment or clinical evaluation of patients, are unaware of the assigned treatment."

"Blinding is intended to minimize the potential biases resulting from differences in management, treatment, or assessment of patients, or interpretation of results that could arise as a result of subjects or investigators knowing the assigned treatment."

Specific blinding threats identified in ICH E10:

Observer ascertainment bias: "Observers might be less likely to identify and report treatment responses in a no-treatment group or might be more sensitive to a favorable outcome or adverse event in patients receiving active drug"
Concomitant therapy bias: "Knowledge of treatment assignment could affect decisions about whether a subject should remain on treatment or receive concomitant medications or other ancillary therapy"
Differential dropout: Unblinded patients may discontinue based on arm assignment rather than disease status

Blinding Level	Description	Oncology Use
Double-blind	Patient, investigator, sponsor assessors, IRB all masked	Ideal; required when endpoint is subjective (PRO, investigator-assessed response)
Double-dummy	Both arms receive one active + one placebo (e.g., oral TKI + IV placebo chemo)	Required when formulations differ (oral vs IV); used in ARCHER 1009 (NCT01360554): dacomitinib vs erlotinib, both with matching placebos
Single-blind	Patient masked; investigator aware	Used when investigator masking impractical but patient-reported endpoints are primary
Open-label	No masking	Standard in oncology when treatments are obviously different (IV vs oral), hard endpoints (OS, BICR-assessed PFS), or when blinding is infeasible

Open-label mitigation strategies:

Blinded Independent Review Committee (BIRC/IRC): For PFS and ORR endpoints, a central independent committee reviews imaging blinded to treatment assignment. Required when PFS is the primary endpoint and the trial is open-label (FDA guidance on NSCLC endpoints 2020). BIRC must have pre-specified charter, blinding maintenance procedures, and adjudication rules.
Central laboratory for biomarkers: Prevents site-level knowledge of randomization arm affecting biomarker collection
Objective endpoints: OS is unambiguous regardless of blinding; RECIST assessment by BIRC for PFS reduces bias

ICH E8(R1) on blinding: Among the design elements that minimise bias: "use of approaches to minimise bias, such as randomisation, blinding or concealment of allocated treatment."

7. Allocation Concealment

Allocation concealment prevents foreknowledge of the next treatment assignment before a patient is enrolled, which could allow investigators to selectively enroll (or exclude) patients based on expected allocation.

Methods:

Central randomization via IWRS: Gold standard. Allocation only revealed after eligibility is confirmed and consent obtained; prevents selective enrollment. All major Phase 3 oncology trials use IWRS.
Sequentially numbered sealed opaque envelopes (SNOSE): Acceptable for resource-limited settings; requires strict monitoring
Pharmacy-controlled: Dispensing pharmacy holds allocation; investigators unaware until drug dispensed

Distinction from blinding: Allocation concealment operates at the time of randomization (before treatment starts). Blinding operates after randomization (during treatment and follow-up). A trial can be open-label (unblinded) yet still have proper allocation concealment via IWRS.

Intercurrent Events

ICH E9(R1) defines intercurrent events (IEs) as "events occurring after treatment initiation that affect either the interpretation or the existence of the measurements associated with the clinical question of interest." The choice of RCT design directly shapes which IEs arise and which handling strategies are appropriate.

By Design Type

Superiority parallel-group (experimental vs placebo/SOC):

IE	ICH E9(R1) Strategy	Statistical Consequence	SAP Language Template
Treatment discontinuation (toxicity, non-compliance)	Treatment policy	Events included regardless of discontinuation; no censoring at discontinuation	"Patients who discontinue study treatment prior to the primary endpoint will continue to be followed for [OS/PFS] per protocol schedule. The primary analysis will include all events irrespective of treatment discontinuation."
Subsequent anticancer therapy (post-progression)	Treatment policy (OS primary); hypothetical or RPSFT (OS sensitivity)	ITT OS analysis includes all deaths regardless of subsequent therapy; RPSFT sensitivity estimates counterfactual OS without crossover	"Subsequent anticancer therapy will be captured on the CRF. The primary OS analysis (treatment policy) will include all deaths irrespective of subsequent therapies. A pre-specified sensitivity analysis will apply RPSFT to estimate OS in the absence of crossover."
Death before first post-baseline assessment (PFS)	Composite	Patient counted as having progressed (worst case)	"Patients who die before the first scheduled post-baseline tumor assessment will be counted as having experienced progression at Day 1."
Crossover (control → experimental at interim)	Treatment policy (primary); hypothetical (sensitivity)	OS ITT maintained as primary; crossover-adjusted analysis as sensitivity	Per above RPSFT template

Non-inferiority parallel-group (experimental vs active control):

IE	Strategy	Key Consideration
Protocol deviations / non-compliance	Per-protocol analysis as co-primary	NI interpretation requires both ITT and PP analyses to agree; non-compliance biases toward null (dilutes difference), which is conservative for superiority but anti-conservative for NI
Use of rescue/additional therapy	Treatment policy	Must be well-controlled; unbalanced rescue therapy use can mask NI
Switching to alternative active treatment	Hypothetical	If both arms allowed rescue with different agents, outcome differences may reflect rescue rather than study drug effect

Add-on placebo-controlled design:

IE	Strategy	Note
Discontinuation of background SOC	Composite (if required by protocol) or treatment policy	Must pre-specify whether SOC discontinuation triggers censoring or counts as event
Dose reduction of background SOC	Treatment policy	Typically ignored; sensitivity analysis with PP population

Regulatory Precedent

Confirmed NSCLC Phase 3 trials from ClinicalTrials.gov demonstrating key design features:

NCT#	Trial	Drug	Indication	Design Feature	Masking
NCT04129502	EXCLAIM-2	TAK-788 (mobocertinib) vs platinum-pemetrexed chemo	NSCLC EGFR exon20 insertion, 1L	Superiority parallel; active comparator (chemo)	Open-label (NONE); BIRC PFS primary
NCT01828112	ASCEND-5	Ceritinib vs chemotherapy	ALK+ NSCLC, 2L post-crizotinib	Superiority parallel; active comparator (pemetrexed or docetaxel)	Open-label; BIRC PFS primary
NCT02864251	CheckMate 722	Nivolumab+chemo or Nivo+Ipi vs chemo	EGFR-mut NSCLC post-TKI	3-arm superiority; active comparator	Open-label (NONE); BIRC PFS primary
NCT01360554	ARCHER 1009	Dacomitinib vs erlotinib	Advanced NSCLC	Superiority; active comparator; double-dummy blinding (placebo erlotinib + placebo dacomitinib)	Single (patient masked)
NCT00035152	Paclitaxel schedule	Weekly vs standard paclitaxel/carboplatin	Stage IIIB/IV NSCLC	Active comparator NI-like (dose schedule); parallel	Open-label

Note: Fewer than 3 examples with confirmed published NI designs found in this CTG NSCLC index. For NI trial precedent in oncology, see Article 35 (Non-Inferiority Designs) which includes de-escalation examples from CRC (IDEA trial) and breast (APT trial).

Limitations and Pitfalls

1. Over-stratification: More than 4–5 stratification factors creates sparse cells, makes IWRS management complex, and can result in an analysis model that is overparameterized relative to the sample size. Practical rule: limit to factors with HR contribution ≥ 1.5 for the primary endpoint.

2. Stratification–analysis mismatch: Randomizing by factor X but not adjusting for it in the primary model produces a type I error inflated analysis. FDA routinely queries this mismatch. If ECOG PS is a stratification factor, the stratified log-rank test must stratify on ECOG PS.

3. NI inflation by poor conduct: A poorly run trial (high dropout, non-compliance, concomitant therapy imbalance) biases toward the null, which inflates false NI findings. Unlike superiority trials where poor conduct is conservative (makes it harder to reject H₀), in NI trials poor conduct makes NI artificially easy to conclude. Both ITT and per-protocol analyses must agree for NI to be accepted.

4. Constancy assumption violations: If the active control's historical effect was established in a different patient population, era, or supportive care context, the M1 estimate may be inflated for the current trial. The NI margin then fails to protect adequately.

5. Open-label PFS assessment bias: Without IRC, investigator-assessed PFS in open-label trials is subject to ascertainment bias: investigators may scan more frequently, apply different RECIST thresholds, or interpret scans differently by arm. FDA expects IRC for open-label PFS primary endpoint trials (confirmed in NSCLC 2020 guidance).

6. Crossover contamination in OS: When control-arm patients cross over to the experimental drug at progression, ITT OS underestimates the true experimental drug effect. This is not a design flaw per se but must be anticipated in the estimand specification and SAP with pre-specified RPSFT or IPCW sensitivity analyses.

7. Allocation concealment erosion: In trials with unequal or distinctive physical appearance of treatments (different pill sizes, IV infusion durations), effective blinding is undermined. Double-dummy design (e.g., active tablet + IV placebo vs placebo tablet + active IV) should be pre-specified when treatments differ obviously.

Backlinks

Source: ICH E10 Choice of Control Group and Related Issues in Clinical Trials (Final, July 2000); ICH E8(R1) General Considerations for Clinical Studies (Final, October 2021) Status: Both Final guidances Compiled from retrieved FDA chunks + ClinicalTrials.gov NSCLC Phase 3 records + raw/literature/non_inferiority_oncology_summary.md