On the ICH E20 draft guideline (2025) for adaptive clinical trials

The ICH E20 guideline, released as a Step-2 draft in June 2025, establishes globally harmonised principles for the planning, conduct, analysis, and interpretation of confirmatory clinical trials employing adaptive designs, including Type I error control, appropriate estimation of treatment effects, and trial integrity. The draft consolidates concepts previously addressed in regional guidance from the FDA and EMA and provides a unified regulatory lexicon, heightened clarity on documentation standards, and expanded recognition of Bayesian and enrichment-based designs.

This article reviews the scope, key principles, methodological expectations, and regulatory implications of ICH E20, and offers considerations for sponsors planning adaptive confirmatory trials under this emerging international framework.

Previous vs new guidance on adaptive designs

Adaptive clinical trial designs have become integral to modern drug development, allowing for prospectively planned modifications based on accruing interim data while maintaining validity, integrity, and statistical rigour. Historically, guidance in this area has been region-specific, for example, the FDA’s 2019 guidance on adaptive designs and the EMA’s 2007 reflection paper. The publication of ICH E20 (Step-2 draft, June 2025) represents a major shift towards a globally harmonised framework for confirmatory adaptive trials, offering a common set of expectations across regulatory agencies.

The guideline specifically targets confirmatory (pivotal) investigations and provides detailed expectations for planning, statistical methodology, preservation of Type I error, trial integrity safeguards, and documentation packages required for regulatory review. The draft also integrates the estimand principles introduced in ICH E9(R1), reinforcing the need for alignment between adaptation decisions and the clinical questions of interest.

Regulatory context and scope

The ICH E20 draft was endorsed by the ICH Assembly for public consultation in June 2025. The scope encompasses Phase III or other confirmatory trials, including those embedded within master protocols, multi-arm/multi-stage (MAMS) studies, or seamless phase II/III strategies.

The following are explicitly excluded from the guideline’s scope: a) early-phase exploratory or dose-finding studies, unless they form part of a seamless confirmatory program, b) trials involving unplanned modifications (e.g., IDMC-initiated protocol amendments not prospectively defined), c) routine operational monitoring (e.g., accrual tracking, data quality, missing data reviews) not linked to pre-specified adaptive rules.

Novel contributions of ICH E20

ICH E20 advances the regulatory science of adaptive design in several ways: global harmonisation of terminology; clear documentation expectations; expanded recognition of modern design classes; and adaptive design definitions and examples.

Global harmonisation of terminology

For the first time, regulators adopt a unified definition of adaptive design as a prospectively planned approach permitting modifications to key design features based on interim data, provided Type I error is controlled. The harmonised terminology and documentation expectations in E20 can shorten sponsor–regulator iteration cycles. Using a single, traceable lexicon for terms such as adaptation, interim analysis, information time, and unblinded statistician across the protocol, SAP, and Simulation Report reduces definitional debate and keeps review focused on operating characteristics. Aligning internal SOPs to this vocabulary improves cross‑functional consistency and auditability.

Clear documentation expectations

The draft guideline outlines a structured documentation package that includes: a detailed Simulation Report describing operating characteristics across a relevant scenario grid, a Statistical Analysis Plan (SAP) with fully pre-specified decision rules and an Integrity and Blinding Plan delineating the separation of roles between the sponsor team and the IDMC.

Expanded recognition of modern design classes

ICH E20 formally acknowledges: a) Master protocols (basket, umbrella, platform trials), b) Bayesian designs; provided prior specification, calibration, and operating characteristics are transparent and justified, and c) adaptive enrichment strategies for population refinement. E20’s recognition of Bayesian designs and adaptive enrichment is fundamentally about transparency and calibration. Sponsors should define decision thresholds in clinically interpretable terms and show how posterior or predictive criteria via simulation preserve strong Type I error under a realistic scenario. For enrichment, pre‑specification is essential to protect interpretability if enrichment is not triggered.

Strengthened requirements for prospective planning

The guideline emphasises complete pre-specification of adaptive algorithms, simulation-supported evaluation of operating characteristics and scrutiny of estimand alignment. Simulation Reports should show distributions, not just summary parameters, for key metrics (sample size, number of interims, false‑positive selection, estimation bias) and include stress tests: data gaps or lags, protocol deviations, event‑time mis‑calibration, and boundaries under non‑proportional hazards. Tying each adaptation to the primary estimand and summarising estimator properties after adaptations (bias, variance, coverage) anticipates common reviewer questions.

Expert insight – Additional considerations

Expect earlier time and effort before first‑patient‑in: validated simulation engines, version‑controlled decision tables, and mock interims should be ready at startup. The budget shifts earlier, but execution should be smoother with cleaner review packages.

Adaptive design definitions and examples

ICH E20 describes adaptive designs as those in which prospectively planned statistical methods allow modifications during an ongoing trial while ensuring control of the Type I error rate. Examples include: group-sequential designs with stopping rules for efficacy or futility; sample size re-estimation (SSR), blinded or unblinded; adaptive dose-finding methods linked to confirmatory stages; multi-arm selection/dropping, including MAMS trials; adaptive population enrichment (biomarker-guided); Bayesian adaptive designs, including predictive-probability stopping; seamless phase II/III designs combining learning and confirmatory objectives.

These designs must be fully pre-specified in the protocol, SAP and adaptation algorithm tables.

Advantages and challenges in confirmatory settings

ICH E20 addresses both potential benefits and significant challenges in several ways.

Potential advantages

The literature consistently reports several benefits of adaptive confirmatory designs, such as: improved statistical efficacy by increasing power; ethical advantages by reducing exposure to inferior or unsafe regimens; potential reductions in sample size and development timelines; ability to focus on patient subgroups more likely to benefit (adaptive enrichment); and enhanced learning about treatment effects through interim decision-making.

Adaptive rules can convert accruing information into conditional power or predictive‑probability gains, concentrating resources on promising regimens or enriched populations. Even when total enrolment resembles a fixed design, information is better allocated, raising the chance of detecting clinically meaningful effects while protecting participants through pre‑specified futility rules.

Key challenges

ICH E20 highlights significant challenges, including: complex statistical modelling and simulation requirements; risks to trial integrity due to insufficient controls on unblinded interim data access; potential estimation bias, particularly following arm selection or sample size adaptation; operational complexities (drug supply, enrolment management); and need for robust justification of adaptation timing and decision thresholds.

Bias can arise from arm selection, sample‑size re‑estimation based on unblinded effects, or over‑tuning interim rules to a narrow scenario set. Mitigations include combination‑test frameworks that maintain strong Type I error despite sample‑size changes, bias‑reduced or shrinkage estimators after selection, and reporting unconditional coverage (not just mean bias) across the scenario grid in the Simulation Report.

Operational risks and statistical risks include Interim timeline slippage (e.g., data lag, SDTM mapping delays), which can force decisions on incomplete or inconsistent datasets and inadvertently change the estimand. Build realistic pre‑specified data‑cut rules, and make sure adaptation‑critical variables have passed predefined quality checks before any unblinding.

Ethically, futility thresholds should be clinically interpretable, linked to minimal important differences (MID) and disease indication, rather than defined only by statistical p-values. This framing resonates with investigators and patients and remains faithful to the estimand.

Core principles emphasised in the guideline

The ECH20 draft outlines key principles for acceptable adaptive confirmatory trials:

Adequacy of the development programme

Adaptive confirmatory trials must be grounded in a development programme that adequately characterises dose–response, identifies the appropriate target population, selects clinically meaningful endpoints, reliably confirm efficacy, support safety, and provide adequate benefit-risk assessment.

Prospective planning

Adaptations must be fully pre-specified, including the type, number, and complexity of adaptations with justifications especially the timing of interim analyses. Choose an error‑control framework that fits the adaptation set. Pre‑specify triggers in operational terms (e.g., when a target number of events is locked and critical queries are closed), rather than calendar dates, to reduce ambiguity when accrual or maturation deviates from plan.

Type I error control

A central requirement is the maintenance of strong control of Type I error of the primary estimand at a pre-specified threshhold, typically through alpha-spending functions, combination tests, or Bayesian calibration to maintain trial integrity and validity of results. Group‑sequential alpha spending is intuitive and operationally mature, but can be cumbersome with multiple adaptation types (e.g., arm selection plus sample‑size re‑estimation). Combination tests offer modularity when information fractions change or arms are added/dropped, while Bayesian calibration can be persuasive when posterior or predictive thresholds are translated into frequentist error control via simulation. Document why the chosen framework is fit‑for‑purpose and show robustness to deviations from assumptions (e.g., mis‑specified event‑time distributions).

Estimand alignment

Adaptation rules must be defined consistently to be aligned with the estimand of interest under ICH E9(R1), particularly when population enrichment or treatment selection is possible. Estimand alignment is the glue. Cross‑check every adaptation against the primary estimand (population, variable, intercurrent event handling, summary measure). For enrichment, define both the prospective enriched estimand and a fallback all‑comers estimand, with decision tables that preserve interpretability regardless of path. State explicitly how intercurrent events (switching, discontinuation, rescue therapy) are handled before and after adaptation to avoid estimand drift.

Preservation of trial integrity

The ICH E20 guideline requires minimising operational bias with strict role segregation, especially regarding IDMC interactions, unblinded statisticians, and sponsor personnel. Integrity is a process; integrity should be designed, not promised. Firewalled statistical roles, independently validated interim code, and decision summaries for the sponsor team (e.g., continue/stop, expand/stay) reduce operational bias and ease concerns about information leakage.

In addition to role segregation, implement eligibility freeze windows around interims to prevent selection bias from knowledge of pending adaptations; maintain audit‑ready access logs to interim outputs; and reference version‑locked decision tables in both the SAP and Simulation Report so reviewers can verify alignment.

Practical implementation insights & considerations

While ICH E20 provides clear expectations for statistical validity and preservation of trial integrity, real‑world execution remains one of the strongest determinants of success in adaptive confirmatory trials. Even when adaptation rules are fully pre‑specified and supported by simulations, operational constraints often drive the feasibility, timeliness, and interpretability of the design.

In practice, sponsors must ensure that data management, statistical programming, statistical analysis, and clinical operations are prepared for the demands associated with interim decision-making. This includes establishing adequate lead time for interim data cuts, maintaining strict version control for adaptation algorithms, and potentially preparing sites for rapid implementation of post‑interim changes. These activities frequently require more planning effort than traditional trials and should be built into early development timelines.

Cross‑functional readiness, especially clear separation of blinded and unblinded roles, reduces the risk of errors that could compromise integrity or introduce avoidable delays. Attention to these practical considerations early in the design process helps ensure that adaptive trials are meeting expectations internally and for regulators and stakeholders.

Oncology-specific and time-to-event issues

ICH E20 devotes specific attention to adaptive oncology trials due to their reliance on time-to-event endpoints. Key considerations include: a) Information-time alignment: interim analyses must be triggered by accrued information, rather than calendar time; b) Non-proportional hazards: weighted log-rank tests or alternative estimands may be needed; c) Enrichment and early stopping: thresholds must be simulation-supported and integrated into alpha-spending plans.

In event‑driven oncology trials, interim timing must follow information time, not calendar time. Build event forecasts with variance, not just a point estimate, and pre‑define tolerance bands to avoid ad hoc decisions when accrual diverges from plan.

Under non‑proportional hazards, classical group‑sequential boundaries tied to log‑rank statistics can misbehave. Consider weighted log‑rank and pair it with an estimand (e.g., RMST difference, milestone survival, etc.) that remains interpretable when hazards are not constant. Simulations should vary the shape of treatment effects (delayed onset, waning benefit, etc.) and confirm that Type I error and power hold.

Rapid accrual can cause over‑run between data cut and IDMC review. Pre‑specify whether late‑arriving events are ignored, include a new information recalculation and clarify how this choice affects alpha spending or combination‑test weights, so integrity and error control are preserved.

For biomarker‑driven enrichment, take into account model marker measurement error, screen‑fail rates, and potential assay drift over time. Provide operational plans for re‑consent and site communication, so that post‑interim eligibility changes can be implemented without compromising blinding or fairness to ongoing participants.

Common pitfalls identified by regulators

ICH E20 highlights pitfalls that may compromise validity or regulatory acceptability: unplanned sample size adjustments; dropping or adding treatment arms not prospectively defined; post-hoc population modifications; informal, non-pre-specified stopping rules; inadequate control of trial integrity or unblinding risks; insufficient simulation scenarios or sensitivity analyses; and estimand misalignment due to inconsistent or incorrect definitions.

Regulatory interaction strategy

Sponsors are strongly encouraged to: engage early with FDA/EMA on adaptation types, estimands, and simulation frameworks; submit SAP, Simulation Report, Integrity, and Blinding Plan as part of regulatory packages; provide IDMC charters and decision-making documentation to regulators; and discuss implications of adaptations for interpretability, patient heterogeneity, and benefit–risk assessment.

Make the agency’s job easy with a pre‑read that mirrors ICH E20’s structure: (1) concise descriptions of each adaptation and its estimand, (2) version‑identified decision tables, (3) Simulation Report highlights with scenario heatmaps and stress tests, (4) the Integrity and Blinding Plan, and (5) mock shells for adaptive outputs (e.g., decision memos, CONSORT flow under selection). This mapping speeds reviewer navigation.

Ask crisp, decision‑oriented questions such as: Is the proposed unblinded‑statistician firewall adequate given our IDMC cadence? Are the scenario grids sufficient for the selection plus sample‑size re‑estimation design? Does the proposed fallback estimand preserve interpretability if enrichment is not triggered? Clear questions invite clear feedback.

Document traceability: map each adaptation rule to the simulation object that evaluated it, the SAP section that implements it, and the programming module that will execute it. Traceability reduces concerns that the executed trial may diverge from the reviewed design.

For global programmes, anticipate regional nuance. Even with harmonisation, review cultures differ. Optional appendices that emphasise frequentist calibration for Bayesian designs or expanded IDMC charter excerpts can shorten follow‑up cycles and de‑risk late‑stage surprises.

ICH E20 represents a significant advance in the international harmonisation of adaptive clinical trial methodology. By articulating clear expectations for design justification, operating characteristics, estimand alignment, and preservation of trial integrity, the guideline helps ensure that adaptive confirmatory trials remain both methodologically robust and clinically interpretable. As adaptive designs become more widely adopted, especially in oncology, rare diseases, and biomarker-guided development, consistent global standards will facilitate regulatory review and accelerate access to effective therapies.

Sponsors are encouraged to evaluate their development programmes in the context of ICH E20 and to consult regulators early when planning adaptive confirmatory trials. I believe this important release of the ICH E20 guideline will lead to further global increase in the use of adaptive trial designs by sponsors.

References

About the author

Bob Chastain is biostatistics SME at Navitas Life Sciences. He brings 20+ years of extensive experience in clinical research, adaptive and traditional trial design, regulatory submissions, DMC oversight, and statistical programming. He has supported diverse therapeutic areas and contributed to numerous NDA/BLA packages, ISS/ISE analyses, and pivotal Phase I–IV clinical studies.