Chapter 07 / Understanding Trial Designs Through Real Project Work / Paid beta preview

Chapter 7: Randomization, Blinding [ICH E9], Controls, and Comparators

How bias protection becomes operational work through randomization, blinding, controls, and supply. ## 7.1 Bias Protection Becomes Project Work Lauren Brooks once told me randomization and blinding [ICH E9] were "mostly statistics and pharmacy." She said it honestly, not carelessly. She knew randomization mattered. She knew blinding [ICH E9] mattered. She knew placebo-controlled trials were supposed to reduce bias. She could say the words correctly in a meeting. What she had not yet learned was that those design features become daily project work. The lesson came during CNS-119, a fictional Phase II/III trial in a symptomatic neurologic condition. The investigational therapy was intended to reduce episodes of disabling symptoms that fluctuated from week to week. Some outcomes came from patient reports. Some came from clinician-rated scales. Some came from wearable-device summaries. Rescue medication use mattered. Visit timing mattered. Expectation mattered. The placebo response was expected to be high. That is a dangerous setting for casual trial conduct. If patients guess their assignment, symptom reporting can change. If investigators guess assignment, assessment can change. If sites delay randomization because they want a particular participant to receive active treatment, the comparison can change. If the pharmacy can tell kits apart, behavior can change. If sponsor reviewers see grouped data before the right time, decisions can change. The trial may still enroll. The visits may still occur. The database may still lock. And the comparison may still be damaged. Claire Jiang, our executive director of biostatistics, explained it to Lauren this way: "Randomization and blinding [ICH E9] are promises. Operations either protect those promises or break them." That is the heart of this chapter. Randomization means assigning participants to study arms by a planned chance-based method, rather than by physician preference, participant preference, site convenience, sponsor pressure, or visible pattern. A study arm is a protocol-defined group, such as investigational treatment, placebo, active comparator, standard of care, dose level, or control strategy. Allocation means the assignment of a participant to one of those arms. Randomization does not guarantee perfectly balanced groups in every small trial. It improves comparability on average and helps reduce selection bias. Bias means a systematic distortion that can make a treatment effect look larger, smaller, safer, riskier, or more certain than it really is. Blinding [ICH E9], also called masking, means keeping certain people from knowing treatment assignment so that knowledge does not distort behavior, care, reporting, assessment, analysis, or decision-making. A blinded trial should specify who is blinded: participants, investigators, care providers, outcome assessors, adjudicators, data managers, statisticians, sponsor team members, monitors, pharmacists, vendors, or others. Labels such as single-blind and double-blind are useful only if the team explains who is actually blinded and how the blind is protected. A control or comparator answers the question, "Compared with what?" A trial may compare an investigational treatment with placebo, an active treatment, standard of care, another dose, an add-on regimen, no treatment, delayed treatment, or another protocol-defined strategy. These are not decorative choices. They shape the whole operating system. In CNS-119, Claire cared about whether the randomization schedule would protect comparability. Michael Tan cared about kit numbering, packaging, labeling, resupply, and emergency unblinding [ICH E9] in interactive response technology, or IRT. Victor Stein cared about role-based access in electronic data capture, or EDC, eCOA systems, safety systems, and blinded data outputs. Grace Kim cared about whether randomization, unblinding [ICH E9], deviations, and corrective actions could be reconstructed during inspection. Dr. Samuel Reeves cared about emergency unblinding [ICH E9] and participant care. Maggie Chen cared about whether sites could execute the workflow. Priya Raman cared about whether vendors understood the blind. Rafael Ortiz cared about whether the comparator would make recruitment honest and feasible. Lauren realized that randomization and blinding [ICH E9] were not sitting in one department. They were threaded through the trial. ## 7.2 Randomization: Fair Assignment Is an Operating System The first CNS-119 randomization readiness meeting sounded reassuring. The protocol had a 1:1 allocation ratio, meaning participants would be assigned equally between investigational therapy and placebo. Eligibility would be confirmed before randomization. The IRT vendor would hold the randomization schedule. Sites would randomize participants through the system. Supply would ship blinded kits according to IRT resupply rules. On the slide, the process looked clean: Screen. Confirm eligibility. Randomize. Dispense. Treat. Follow. Then Maggie asked, "Where can a site accidentally change the comparison?" The room became more useful. Sites could screen participants who were not appropriate because enrollment was behind. They could delay randomization for a participant whose symptoms looked severe because they hoped active treatment would become available. They could randomize before all eligibility data were confirmed. They could enter a stratification factor incorrectly. They could misunderstand replacement rules. They could call the help desk and reveal information they should not share. They could store or dispense kits in ways that made assignment guessable. Randomization is a statistical method, but its protection depends on ordinary conduct. Stratification is a randomization feature that helps balance important baseline factors across arms, such as disease severity, region, age group, or prior therapy. Block randomization is a method that helps maintain

...