Why Science? | ICH E10 Choice of Control Group and Related Issues in Clinical Trials

E10 says the major purpose of a control group is to distinguish outcomes caused by the test treatment from outcomes caused by natural history, expectations, other care or other influences. We cannot observe the same participant both receiving and not receiving the treatment at the same time, so the design creates a comparison group intended to represent that unobserved alternative.

A control group makes the control group represents a missing counterfactual a causal question rather than a before-and-after story. E10 says the major purpose of a control group is to distinguish outcomes caused by the test treatment from outcomes caused by natural history, expectations, other care or other influences. We cannot observe the same participant both receiving and not receiving the treatment at the same time, so the design creates a comparison group intended to represent that unobserved alternative. Keep the comparison honest with this boundary: A comparator is credible only when it approximates the relevant alternative closely enough for the intended inference.

A strong science note separates observation from interpretation. State the treatment decision, then write the counterfactual outcome the control group is meant to estimate. Record the source, version or date, population or device context, method, result and one uncertainty. If the official record does not show a field, write “not shown” instead of filling the gap from memory.

Contents · Next chapter

FDA identifies E10 as final guidance dated May 2001 under docket FDA-1999-D-1874, issued by CDER and CBER. It assists applicants in choosing a control group for efficacy trials, discusses related design and conduct issues and describes what each design can demonstrate. It does not establish the regulatory requirements of every region.

For the may 2001 e10 guidance is final, ask what experience the comparator is meant to represent. FDA identifies E10 as final guidance dated May 2001 under docket FDA-1999-D-1874, issued by CDER and CBER. It assists applicants in choosing a control group for efficacy trials, discusses related design and conduct issues and describes what each design can demonstrate. It does not establish the regulatory requirements of every region. The inference remains inside this guardrail: A final general guidance still needs product-specific, disease-specific and regional interpretation.

A strong science note separates observation from interpretation. Create a source receipt with title, date, final status, docket, issuing centres, purpose and stated scope boundary. Record the source, version or date, population or device context, method, result and one uncertainty. If the official record does not show a field, write “not shown” instead of filling the gap from memory.

Contents · Previous chapter · Next chapter

Random assignment divides one sample population into treatment groups without using investigator preference or participant prognosis. It reduces systematic baseline differences in known and unknown factors and provides a sound basis for statistical inference. Allocation concealment protects the random sequence before assignment so enrolment decisions cannot anticipate the next treatment.

The design choice in randomisation protects baseline comparability changes both participant experience and the claim available at the end. Random assignment divides one sample population into treatment groups without using investigator preference or participant prognosis. It reduces systematic baseline differences in known and unknown factors and provides a sound basis for statistical inference. Allocation concealment protects the random sequence before assignment so enrolment decisions cannot anticipate the next treatment. Preserve this limit: Randomisation does not repair poor follow-up, biased measurement or post-randomisation treatment differences.

A strong science note separates observation from interpretation. Draw the enrolment and allocation flow, then mark every point where the next assignment could be predicted or altered. Record the source, version or date, population or device context, method, result and one uncertainty. If the official record does not show a field, write “not shown” instead of filling the gap from memory.

Contents · Previous chapter · Next chapter

Blinding can keep participants, investigators, care teams, outcome assessors or analysts unaware of treatment assignment. E10 explains that knowledge may influence symptoms reported, ancillary care, enthusiasm for follow-up, decisions to discontinue and interpretation of subjective outcomes. When full blinding is impossible, blinded endpoint assessment and other controls can still reduce specific biases.

A control group makes blinding protects care, measurement and analysis a causal question rather than a before-and-after story. Blinding can keep participants, investigators, care teams, outcome assessors or analysts unaware of treatment assignment. E10 explains that knowledge may influence symptoms reported, ancillary care, enthusiasm for follow-up, decisions to discontinue and interpretation of subjective outcomes. When full blinding is impossible, blinded endpoint assessment and other controls can still reduce specific biases. Keep the comparison honest with this boundary: A blinded assessor cannot remove every behavioural or treatment difference created by an open-label design.

A strong science note separates observation from interpretation. List who could know assignment, which decisions they make and one safeguard for each decision that could bias the result. Record the source, version or date, population or device context, method, result and one uncertainty. If the official record does not show a field, write “not shown” instead of filling the gap from memory.

Contents · Previous chapter · Next chapter

In a placebo-controlled trial, participants are usually randomised to the test treatment or an identical-appearing inactive treatment. Both groups may still receive common standard care in an add-on design. The contrast can estimate the absolute effect of adding the test treatment under trial conditions while helping blind expectations and assessment.

For placebo concurrent controls isolate pharmacologic effects, ask what experience the comparator is meant to represent. In a placebo-controlled trial, participants are usually randomised to the test treatment or an identical-appearing inactive treatment. Both groups may still receive common standard care in an add-on design. The contrast can estimate the absolute effect of adding the test treatment under trial conditions while helping blind expectations and assessment. The inference remains inside this guardrail: Placebo use must remain ethically acceptable and may not provide direct comparison with the best active therapy.

A strong science note separates observation from interpretation. Explain the care received by both arms, the scientific need for placebo and the protections against avoidable harm. Record the source, version or date, population or device context, method, result and one uncertainty. If the official record does not show a field, write “not shown” instead of filling the gap from memory.

Contents · Previous chapter · Next chapter

A no-treatment concurrent control compares the test intervention with the absence of study treatment. Because participants and investigators usually know assignment, differences in expectations, care and reporting can enter. E10 therefore points to settings where blinding is impractical and outcomes are sufficiently objective, while encouraging blinded evaluation when feasible.

The design choice in no-treatment controls need objective safeguards changes both participant experience and the claim available at the end. A no-treatment concurrent control compares the test intervention with the absence of study treatment. Because participants and investigators usually know assignment, differences in expectations, care and reporting can enter. E10 therefore points to settings where blinding is impractical and outcomes are sufficiently objective, while encouraging blinded evaluation when feasible. Preserve this limit: Objective endpoints reduce some assessment bias but do not eliminate differences in co-intervention, retention or follow-up.

A strong science note separates observation from interpretation. Make a bias map for an open comparison and assign a prevention or detection control to each pathway. Record the source, version or date, population or device context, method, result and one uncertainty. If the official record does not show a field, write “not shown” instead of filling the gap from memory.

Contents · Previous chapter · Next chapter

Participants may be randomised to several fixed doses, concentration targets or regimens of the study treatment. A placebo or active comparator may also be included. The design can show whether larger exposure produces larger benefit or harm and can identify a useful dose range without treating dose selection as a separate afterthought.

A control group makes dose-response controls test how effect changes with exposure a causal question rather than a before-and-after story. Participants may be randomised to several fixed doses, concentration targets or regimens of the study treatment. A placebo or active comparator may also be included. The design can show whether larger exposure produces larger benefit or harm and can identify a useful dose range without treating dose selection as a separate afterthought. Keep the comparison honest with this boundary: A flat dose response may reflect saturation, poor adherence, narrow exposure separation or an insensitive outcome rather than no pharmacologic relationship.

A strong science note separates observation from interpretation. Sketch expected benefit and harm across doses and list the data needed to distinguish a real plateau from design failure. Record the source, version or date, population or device context, method, result and one uncertainty. If the official record does not show a field, write “not shown” instead of filling the gap from memory.

Contents · Previous chapter · Next chapter

An active-control trial compares the test treatment with a treatment already expected to work. It may seek superiority, estimate comparative efficacy or safety, or support efficacy through a non-inferiority conclusion. The purpose determines the margin, analysis and interpretation. Comparable dosing, populations, endpoints and follow-up are essential for a fair contest.

For active controls can support superiority or non-inferiority, ask what experience the comparator is meant to represent. An active-control trial compares the test treatment with a treatment already expected to work. It may seek superiority, estimate comparative efficacy or safety, or support efficacy through a non-inferiority conclusion. The purpose determines the margin, analysis and interpretation. Comparable dosing, populations, endpoints and follow-up are essential for a fair contest. The inference remains inside this guardrail: Similarity to an active treatment is evidence of efficacy only when the trial could have detected a meaningful difference.

A strong science note separates observation from interpretation. Write separate success statements for superiority and non-inferiority and identify the evidence each needs. Record the source, version or date, population or device context, method, result and one uncertainty. If the official record does not show a field, write “not shown” instead of filling the gap from memory.

Contents · Previous chapter · Next chapter

External or historical controls come from people outside the current trial, either earlier or in another setting. E10 warns that differences in diagnosis, prognosis, supportive care, measurement and time can create serious bias. Such designs may be persuasive in unusual circumstances with predictable outcomes and large treatment effects, but general medical impressions are unreliable comparators.

The design choice in external controls are vulnerable to changing context changes both participant experience and the claim available at the end. External or historical controls come from people outside the current trial, either earlier or in another setting. E10 warns that differences in diagnosis, prognosis, supportive care, measurement and time can create serious bias. Such designs may be persuasive in unusual circumstances with predictable outcomes and large treatment effects, but general medical impressions are unreliable comparators. Preserve this limit: Statistical adjustment cannot guarantee comparability when important factors were absent, measured differently or changed over time.

A strong science note separates observation from interpretation. Build a comparability table for eligibility, prognosis, care, outcome definition, calendar time and data quality. Record the source, version or date, population or device context, method, result and one uncertainty. If the official record does not show a field, write “not shown” instead of filling the gap from memory.

Contents · Previous chapter · Next chapter

A trial may include placebo, active treatment and several doses. A three-arm design can estimate the test drug’s absolute effect, compare it with active therapy and demonstrate assay sensitivity through the active-versus-placebo contrast. The richer design can clarify an ambiguous result, although it adds recruitment, analysis and interpretation demands.

A control group makes multiple control groups can answer more than one question a causal question rather than a before-and-after story. A trial may include placebo, active treatment and several doses. A three-arm design can estimate the test drug’s absolute effect, compare it with active therapy and demonstrate assay sensitivity through the active-versus-placebo contrast. The richer design can clarify an ambiguous result, although it adds recruitment, analysis and interpretation demands. Keep the comparison honest with this boundary: More arms create more formal comparisons and can introduce multiplicity that must be planned.

A strong science note separates observation from interpretation. Draw every arm and label the exact question, error control and interpretation supported by each planned contrast. Record the source, version or date, population or device context, method, result and one uncertainty. If the official record does not show a field, write “not shown” instead of filling the gap from memory.

Contents · Previous chapter · Next chapter

Control choice depends on available effective therapy, risk from delay or omission, disease seriousness, duration and rescue provisions. Placebo can sometimes be acceptable when added to standard care or used briefly with escape criteria. Active control may be ethically attractive, yet a poorly sensitive active-control design can expose participants without producing interpretable evidence.

For ethics and inference must be designed together, ask what experience the comparator is meant to represent. Control choice depends on available effective therapy, risk from delay or omission, disease seriousness, duration and rescue provisions. Placebo can sometimes be acceptable when added to standard care or used briefly with escape criteria. Active control may be ethically attractive, yet a poorly sensitive active-control design can expose participants without producing interpretable evidence. The inference remains inside this guardrail: Ethical acceptability does not automatically make a design scientifically informative, and scientific efficiency does not override participant welfare.

A strong science note separates observation from interpretation. Create an ethics-and-inference table showing burden, safeguards, rescue, information gained and uncertainty for each control option. Record the source, version or date, population or device context, method, result and one uncertainty. If the official record does not show a field, write “not shown” instead of filling the gap from memory.

Contents · Previous chapter · Next chapter

Assay sensitivity is the ability of a trial to distinguish an effective treatment from a less effective or ineffective one. A superiority result directly demonstrates separation in that trial. A successful non-inferiority result does not, because poor adherence, weak measurement or high placebo response can make treatments look similar even when both performed poorly.

The design choice in assay sensitivity means the trial can reveal an effect changes both participant experience and the claim available at the end. Assay sensitivity is the ability of a trial to distinguish an effective treatment from a less effective or ineffective one. A superiority result directly demonstrates separation in that trial. A successful non-inferiority result does not, because poor adherence, weak measurement or high placebo response can make treatments look similar even when both performed poorly. Preserve this limit: A result showing no important difference is persuasive only if the design and conduct were capable of revealing one.

A strong science note separates observation from interpretation. Identify historical and within-trial evidence that the chosen endpoints, population and conduct preserve assay sensitivity. Record the source, version or date, population or device context, method, result and one uncertainty. If the official record does not show a field, write “not shown” instead of filling the gap from memory.

Contents · Previous chapter · Next chapter

A non-inferiority margin specifies how much loss versus the active control can be tolerated while preserving an effect that matters. Its justification relies on reliable historical evidence of the control’s effect and on constancy: the current setting should resemble the trials that established that effect. The chosen margin also requires clinical judgement about acceptable loss.

A control group makes non-inferiority margins need historical and clinical justification a causal question rather than a before-and-after story. A non-inferiority margin specifies how much loss versus the active control can be tolerated while preserving an effect that matters. Its justification relies on reliable historical evidence of the control’s effect and on constancy: the current setting should resemble the trials that established that effect. The chosen margin also requires clinical judgement about acceptable loss. Keep the comparison honest with this boundary: A convenient margin chosen to reduce sample size can approve a treatment that preserves too little benefit.

A strong science note separates observation from interpretation. Trace the margin from historical effect through uncertainty and clinical preservation to the final numerical rule. Record the source, version or date, population or device context, method, result and one uncertainty. If the official record does not show a field, write “not shown” instead of filling the gap from memory.

Contents · Previous chapter · Next chapter

Nonadherence, crossovers, concomitant treatment, outcome misclassification, missing follow-up and broad eligibility can dilute contrasts. In a superiority trial dilution usually makes success harder; in a non-inferiority trial it can make treatments falsely appear acceptably similar. Monitoring should therefore focus on processes that preserve the planned comparison and record deviations transparently.

For trial conduct can erase differences, ask what experience the comparator is meant to represent. Nonadherence, crossovers, concomitant treatment, outcome misclassification, missing follow-up and broad eligibility can dilute contrasts. In a superiority trial dilution usually makes success harder; in a non-inferiority trial it can make treatments falsely appear acceptably similar. Monitoring should therefore focus on processes that preserve the planned comparison and record deviations transparently. The inference remains inside this guardrail: The same operational error can bias different trial objectives in different directions.

A strong science note separates observation from interpretation. For each major deviation, predict its likely effect on superiority and non-inferiority conclusions separately. Record the source, version or date, population or device context, method, result and one uncertainty. If the official record does not show a field, write “not shown” instead of filling the gap from memory.

Contents · Previous chapter · Next chapter

Comparative claims depend on using appropriate doses, regimens, patient populations, endpoint definitions and assessment times for both treatments. An active comparator used below its effective dose can favour the test drug; an unnecessarily toxic dose can distort safety. Differential titration, follow-up or supportive care can make the label “active controlled” look fairer than the actual experiment.

The design choice in fair comparisons require comparable dose and timing changes both participant experience and the claim available at the end. Comparative claims depend on using appropriate doses, regimens, patient populations, endpoint definitions and assessment times for both treatments. An active comparator used below its effective dose can favour the test drug; an unnecessarily toxic dose can distort safety. Differential titration, follow-up or supportive care can make the label “active controlled” look fairer than the actual experiment. Preserve this limit: Random assignment cannot make an inappropriate comparator regimen clinically fair.

A strong science note separates observation from interpretation. Audit both arms for dose, route, duration, titration, rescue, adherence support and outcome timing. Record the source, version or date, population or device context, method, result and one uncertainty. If the official record does not show a field, write “not shown” instead of filling the gap from memory.

Contents · Previous chapter · Next chapter

Placebo control can estimate an absolute treatment effect under trial conditions. Active control can compare treatments and, with adequate assay sensitivity, support superiority or non-inferiority. Dose-response control can reveal exposure relationships. External control can provide context when comparability is unusually strong. Each design leaves questions unanswered, so reporting should keep the inference boundary beside the result.

A control group makes the chosen control determines what the result can say a causal question rather than a before-and-after story. Placebo control can estimate an absolute treatment effect under trial conditions. Active control can compare treatments and, with adequate assay sensitivity, support superiority or non-inferiority. Dose-response control can reveal exposure relationships. External control can provide context when comparability is unusually strong. Each design leaves questions unanswered, so reporting should keep the inference boundary beside the result. Keep the comparison honest with this boundary: A statistically valid result should not be translated into a comparison the study did not make.

A strong science note separates observation from interpretation. Write one supported conclusion and one tempting but unsupported conclusion for each control type. Record the source, version or date, population or device context, method, result and one uncertainty. If the official record does not show a field, write “not shown” instead of filling the gap from memory.

Contents · Previous chapter · Next chapter

The receipt records the clinical question, counterfactual, control type, ethical rationale, standard care, randomisation, concealment, blinding, comparator dose, endpoints, assay-sensitivity evidence, non-inferiority margin if used, deviations, missingness, effect estimates and interpretation limits. It lets another reader see why the trial’s comparison deserves trust.

For a complete control-group receipt makes the comparison auditable, ask what experience the comparator is meant to represent. The receipt records the clinical question, counterfactual, control type, ethical rationale, standard care, randomisation, concealment, blinding, comparator dose, endpoints, assay-sensitivity evidence, non-inferiority margin if used, deviations, missingness, effect estimates and interpretation limits. It lets another reader see why the trial’s comparison deserves trust. The inference remains inside this guardrail: A complete receipt cannot turn an unsuitable comparator into a valid scientific reference.

A strong science note separates observation from interpretation. Give the receipt to an independent reviewer and ask whether the design can support every sentence in the planned conclusion. Record the source, version or date, population or device context, method, result and one uncertainty. If the official record does not show a field, write “not shown” instead of filling the gap from memory.

Contents · Previous chapter