Chapter 08 / Understanding Trial Designs Through Real Project Work / Paid beta preview

Chapter 8: Complex and Modern Trial Designs

How adaptive, master-protocol, decentralized, digital, and pragmatic designs change project control. ## 8.1 Complexity Is Not the Same as Cleverness The first time Lauren Brooks heard the phrase "adaptive platform design with Bayesian borrowing," she wrote it down as if the words themselves might shorten the trial. They did not. The program was RARE-007, a fictional rare-disease development program. The eligible population was tiny. Disease expression varied widely. Some patients progressed quickly, others slowly. Families were organized and informed. Investigators were passionate. Leadership wanted faster answers. Regulatory wanted clarity. Statistics wanted simulations. Operations wanted to know how many versions of the protocol sites would have to live through. In the first design meeting, someone said, "A modern design could save us." Claire Jiang looked at Lauren and said, "Modern is not a method." That is the beginning of this chapter. Complex designs can be powerful. They may allow a trial to learn more efficiently, stop an unpromising treatment earlier, select a dose, enrich a population, share infrastructure, add or drop arms, study targeted subgroups, or use remote and routine-care features to reduce participant burden. They can also create delay, confusion, bias, amendment fatigue, data-access risk, site overload, vendor mismatches, and governance failure. The difference is not whether the design sounds advanced. The difference is whether the flexibility has rules. In Chapter 6, we learned that phases are different evidence machines. In Chapter 7, we learned that randomization, blinding [ICH E9], controls, and comparators are operating commitments that protect the comparison. Chapter 8 builds on both lessons. Complex and modern designs do not remove phase discipline or bias protection. They make both harder to manage. The PM question is not, "Is this design innovative?" The better question is, "Can this team operate the design without losing participant protection, data integrity, interpretability, and governance control?" For RARE-007, that question mattered because every participant counted. In a common disease, a weak operational assumption may be repaired by adding more sites or extending enrollment. In a rare disease, there may be no spare population. A confusing consent form, a delayed biomarker result, a broken data firewall, or a poorly explained adaptation can damage trust quickly. Lauren's first instinct was to treat complexity as a scheduling advantage. "If we can adapt," she said, "we can avoid waiting for the whole study to finish." Claire answered carefully. "Adaptive does not mean we change the trial whenever we feel smarter. It means we prospectively define what may change, when, based on what data, under whose authority, and with what protection against bias." That sentence belongs on the first page of every adaptive-design project plan. ## 8.2 Adaptive Designs: Flexibility With Guardrails An adaptive design is a trial design in which one or more specified features may change based on accumulating trial data, according to prospectively planned rules, while preserving trial validity and integrity. Plainly: the trial can learn while it is running, but only in ways the team planned before the learning began. Examples may include sample-size re-estimation, dropping an ineffective arm, selecting a dose, enriching enrollment in a subgroup, changing a randomization ratio, stopping for futility, stopping for success, or moving from an exploratory part into a confirmatory part under a seamless design. A stopping rule is a prespecified rule for pausing or stopping a trial, arm, dose, or enrollment path when defined conditions are met. Futility means the data suggest the trial or arm is unlikely to achieve its objective if it continues as planned. Sample-size re-estimation means reassessing the needed number of participants under prespecified rules. Response-adaptive randomization means future assignments may be shifted based on accumulating response data under a planned algorithm. Enrichment means focusing enrollment on a subgroup more likely to benefit or more relevant to the question. A biomarker is a measurable biological feature, such as a gene variant, lab value, protein, imaging signal, or physiological measure, that may help define risk, disease type, response, or eligibility. Those are statistical design features. They become project work quickly. For RARE-007, the team considered an interim review after a defined number of participants completed a key assessment. An interim analysis is a planned analysis before the trial is fully complete. The review could support dropping a poorly performing dose or increasing sample size within a prespecified range. Lauren built a tracker with the interim date highlighted in yellow. Victor Stein, our data and systems lead, asked, "What data have to be clean before that date?" Michael Tan asked, "If one dose drops, what happens to supply already packaged and shipped?" Priya Raman asked, "Do the IRT vendor, central lab, eCOA vendor, and statistical programming vendor all understand the same adaptation scenarios?" Grace Kim asked, "Can we reconstruct the decision without exposing unblinded information to the wrong people?" The tracker was not enough. Before the team could use the table below, Lauren had to make sure the basic systems language was clear. Interactive response technology, or IRT, helps manage randomization, treatment assignment, supply, and sometimes emergency unblinding [ICH E9]. Electronic data capture, or EDC, is the system where clinical trial data are entered and

...