Chapter 17 / Tools, Statistics, Vendors, Budgets, and Risk Control / Paid beta preview

Chapter 17: Statistics for Clinical Project Managers

How PMs can work fluently with statistical issues without pretending to be statisticians. ## You Do Not Need To Be The Statistician Lauren Brooks had survived database lock, first tables, listings, and figures (TLFs), and the first clinical study report (CSR) review cycle for DIAB-220. She had learned how fast a missing lab value or ambiguous participant status could become a timeline problem. Then Claire Jiang, Executive Director of Biostatistics, asked a question that made the room go quiet: "Before we discuss interpretation, are we aligned that the Week 12 HbA1c estimand [ICH E9(R1)] handles rescue medication the way the statistical analysis plan says it does?" Lauren understood about half the sentence and all of the risk. Statistics is the discipline of designing, analyzing, and interpreting data under uncertainty. A clinical project manager does not need to calculate the analysis, choose the model, or interpret the result independently. But the PM does need enough statistical literacy to manage dependencies, protect timelines, ask disciplined questions, and keep the team from turning outputs into claims the data cannot support. Chapter 16 taught that planning tools expose dependencies. Chapter 17 shows that statistical decisions are some of the most important dependencies in a trial. They affect protocol design, data collection, database lock, TLF review, interpretation, CSR language, top-line communication, and governance decisions. Lauren's first instinct was to ask, "Did the p-value pass?" By the end of the chapter, she asks a better question: > What was the study question, who was analyzed, how were missing data and intercurrent events [ICH E9(R1)] handled, what do the confidence intervals show, and what conclusions can we responsibly support? ## The STATS Frame Claire gave Lauren a frame for PM-level statistical thinking: | Letter | Meaning | PM Question | |---|---|---| | S | SAP And Study Question | Is the statistical analysis plan aligned with the protocol question? | | T | Targets: Endpoints And Estimands [ICH E9(R1)] | What exactly are we estimating, in whom, and under what treatment conditions? | | A | Analysis Populations | Which participants are included in each analysis and why? | | T | Trouble: Missing Data And Intercurrent Events | What happened after randomization that affects interpretation? | | S | Summaries, Significance, And Story | What do TLFs, p-values, confidence intervals, and limitations really support? | The statistical analysis plan, or SAP, is the controlled document that describes planned analyses, populations, endpoints, handling rules, and outputs. It should be finalized under the sponsor's process before unblinding [ICH E9] and before final analysis decisions can be influenced by results. The PM tracks SAP readiness, review ownership, version control, dependencies, and decision timing. Biostatistics owns the statistical methods. Medical, Clinical, Regulatory, Data Management, Safety, and other functions provide accountable input where their expertise is required. ## What PMs Need, And What They Do Not Need | PM Needs To Understand | PM Does Not Need To Do | |---|---| | Which statistical decisions affect protocol, data, lock, TLFs, CSR, and governance | Calculate statistical tests independently | | What the primary endpoint is and why it matters | Redesign the endpoint hierarchy alone | | What the SAP says about populations, missing data, and outputs | Choose analysis methods without Biostatistics | | Why estimands [ICH E9(R1)] and intercurrent events [ICH E9(R1)] matter | Invent ad hoc inclusion/exclusion rules | | What p-values and confidence intervals can and cannot support | Interpret efficacy or safety conclusions independently | | Why subgroup and exploratory findings need caution | Turn interesting patterns into claims | | How unblinding [ICH E9] should be controlled | Seek treatment-effect information before authorized access | The PM's statistical value is not mathematical bravado. It is process discipline. ## Endpoints: What The Trial Promised To Measure An endpoint is a defined outcome or measurement used to answer a trial question. The primary endpoint is the main endpoint the trial is designed to answer. Secondary endpoints support additional questions. Exploratory endpoints generate learning but usually have weaker claim strength unless planned and controlled appropriately. Safety endpoints describe safety, tolerability, adverse events, laboratory findings, and related risks. DIAB-220's endpoint map looked like this: The team used the data-flow map to turn the details into operating choices the PM could assign, monitor, and escalate: Main efficacy question: the accountable owner was Biostatistics/Clinical. This matters because drives data cleaning priority and primary TLFs.; endpoint was Change in HbA1c from baseline to Week 12; timing was Week 12; source was Central lab/EDC. Supportive glycemic control: the accountable owner was Biostatistics/Clinical. This matters because secondary hierarchy and lab reconciliation matter.; endpoint was Fasting glucose change; timing was Week 12; source was Central lab. Patient experience: the accountable owner was Clinical/eCOA vendor/Data. This matters because missing eCOA can affect interpretability.; endpoint was Treatment satisfaction diary score; timing was Month 3; source was eCOA. Safety: the accountable owner was Safety/Medical. This matters because coding and reconciliation must be complete.; endpoint was Hypoglycemia adverse events; timing was Throughout study; source was EDC/safety database/source. Operations: the accountable owner was Clinical Operations. This matters because May affect

...