4.1 Bias, Randomisation and Comparator Controls

Bias, Randomisation and Comparator Controls

In cluster randomization, entire social or organizational units (e.g., medical clinics, hospital wards, or geographical communities) are randomized rather than individual patients. This design is deployed when interventions cannot be easily restricted to individuals (e.g., hospital hygiene protocols, educational interventions) to avoid contamination between trial arms.

Bias Control Framework: Operationalizing ICH E9

Bias is any systematic tendency in design, conduct, analysis, or evaluation that causes trial estimates of a treatment effect to deviate systematically from the true value. ICH E9 outlines four primary operational biases and their counter-measures:

Intention-to-Treat (ITT) vs. Per-Protocol (PP) Analysis

• Intention-to-Treat (ITT): Analyzes all randomized subjects according to their assigned treatment group, regardless of whether they received the medication, adhered to the protocol, or dropped out early. The ITT principle helps preserve the prognostic balance created by randomization and estimates an effect consistent with the defined estimand.
• Per-Protocol (PP): Restricts analysis only to subjects who completed the full prescribed course of therapy without major protocol deviations. While PP addresses an effect within a prespecified adherent subset, selection of a per-protocol subset can introduce bias because patients who drop out due to lack of efficacy or adverse events are excluded.

Randomization Methodologies & Allocation Concealment

Properly implemented concealed randomization reduces selection bias and establishes the theoretical foundation for statistical hypothesis testing.

• Allocation Concealment: The critical procedural safeguard ensuring that investigators, research coordinators, and participants cannot foresee or decipher the upcoming treatment assignment prior to irreversible enrollment.
• Centralized Interactive Response Technology (IRT): Interactive Voice/Web Response Systems (IVRS/IWRS) manage centralized randomization. Site staff enter validated eligibility criteria, and the IRT dynamically assigns a masked medication kit number, supporting allocation concealment when access, programming and site processes are controlled.

Blinding Levels & The Double-Dummy Technique

Blinding (masking) prevents performance bias and ascertainment bias.

Levels of Blinding

1. Open-Label: Both participant and site investigator are aware of treatment assignment. Used when blinding is physically impossible (e.g., surgical vs. medical management) or unnecessary.
2. Single-Blind: The term may refer to one blinded group; specify the actual blinded parties.
3. Double-Blind: Neither the participant nor the investigator/site clinical team knows the treatment assignment. The protocol should identify exactly which roles remain blinded.
4. Triple-Blind: The term can refer to several blinded roles; the protocol must define actual access and timing.

The Double-Dummy Technique

When comparing two active treatments that have visibly different pharmaceutical forms, administration routes, or dosing schedules (e.g., an oral tablet vs. a subcutaneous injection), masking cannot be achieved simply by packaging. A double-dummy design is deployed:

Every subject in both groups receives both an oral formulation and an injection, supporting masking; adverse effects or other clues can still create unblinding risks.

Control Groups and Comparative Hypotheses

Common control choices relevant to study design include the following; ACRP's reference list, rather than every related ICH guideline, defines the examination references:

• Placebo Concurrent Control: Controls for the placebo effect, spontaneous disease remission, and observer bias. Highly sensitive for demonstrating assay superiority, but requiring ethical justification under the Declaration of Helsinki, with no exposure to additional serious or irreversible harm from withholding proven treatment.
• Active Concurrent Control: Compares the investigational product to an established standard-of-care comparator.
  • Superiority Trial: Hypothesizes that the new agent is statistically and clinically superior to the active control.
  • Non-Inferiority Trial: Aims to show that the new agent is not clinically worse than the active control by more than a predefined, justified non-inferiority margin ($\Delta$). Often utilized when the new agent offers non-efficacy advantages, such as superior safety, reduced dosing frequency, or oral instead of intravenous administration.
• Historical (External) Controls: Compares trial participants to patients treated outside the study in the past. Highly vulnerable to severe selection bias, temporal shifts in medical care, and differing diagnostic criteria; requiring careful justification of comparability and bias controls.

Official sources

Checked October 18, 2026: ICH E6(R3), Principles and Annex 1, ICH E9, ICH E9(R1), ICH E8(R1).