Table of Contents
Key Takeaways
- A statistician shapes the parts of a protocol that decide whether the trial can answer its research questions: the endpoint, the estimand, whose data gets analyzed, the sample size, and the decision rules. Those choices are best made while the draft is being written, not at review.
- Waiting until the protocol is finished or, worse, after the trial has begun, invites avoidable costs associated with amendments, underpowered studies, endpoints that do not support regulatory or business decisions, and questions from regulatory authorities during submission reviews.
- Regulatory authorities, like FDA and Health Canada, expect compliance with ICH principles like E8(R1) and E9(R1), which expect the estimand and the statistical analysis to be set out in the protocol itself.
The Statistician Should Be on the Protocol Team
Here is a common picture of how a statistician joins a study.
The protocol is nearly done. Someone sends it over with a note: “Can you check the stats section before we submit?” The statistician adds a sample size, tidies the analysis language, and signs off.
By then, many of the choices that decide whether the trial answers its question have already been made. That model treats the statistician as a proofreader. It is the wrong model, and it carries real costs.
The statistician should not be brought in only to approve a finished protocol. They should be one of the people helping to build it. The statistician should be on the protocol team.
Endpoints, the precise question the trial is meant to answer, who gets analyzed, how many patients are needed, and what the results trigger next are statistical decisions as much as clinical ones. They are easiest to get right while the protocol is still in draft, and hardest to fix once patients are enrolled.
Why Statisticians Need to Be Involved from Day One
Read a protocol closely and you will find a chain of choices resting on each other.
What are you measuring? Over what period? Using which measurement scale? In which patients? Against what comparison? What happens to the analysis when a patient stops treatment, switches treatment, uses rescue medication, or misses the main assessment?
Each answer constrains the next.
A primary endpoint can sound reasonable in a clinical discussion but still be difficult to measure or analyze cleanly. It can also produce a result that a regulator, partner, or investor cannot act on.
A statistician in the room while these choices are being formed can see the downstream effect of each one before it sets. Brought in afterward, the statistician can often only describe the consequences of decisions that are already locked.
That is the difference between shaping a trial design versus reacting to one.
Here is a simple example.
A trial is meant to show whether a treatment reduces pain after 12 weeks. But some patients take rescue medication before Week 12 because their pain is too high.
Now the team has a problem. If pain improves at Week 12, was it because of the study treatment, or because of the rescue medication?
How that question will be answered must be specified before the trial starts. The protocol needs a clear rule for how those patients will be handled in the analysis.
Without that rule, the study can produce a result that people interpret in different ways and will create credibility issues in the eyes of regulatory authorities. That is why the statistician needs to be involved while the endpoint and analysis strategy are being decided, not after the protocol is finished.
Involving a statistician only after the protocol is final is like asking an architect to review the plans after the foundation has already been poured.
A Statistician Shapes These Decisions Before the Draft Is Final
These are the choices that get worked out during clinical study design, while the protocol is still a draft.
Endpoints
The endpoint defines what counts as an effect.
A statistician helps choose one that is measurable, sensitive enough to detect a real difference, and accepted by the regulator for the claim the sponsor intends to make.
The wrong endpoint can pass internal review and still fail to support the next regulatory or business decision.
The Estimand
Under ICH E9(R1), the protocol is expected to state the precise treatment effect being estimated.
That means specifying the population, the treatment, the endpoint, how intercurrent events such as discontinuation or rescue medication are handled, and how the effect is summarized across patients.
This is not a formality.
Two trials with the same endpoint can be asking different questions depending on how these pieces are defined. The result that the trial produces must be the answer the next decision needs.
Analysis Populations
Who is included in the primary analysis, and on what basis, changes what the result means.
A per-protocol analysis and a full-analysis of the same data can point in different directions. These definitions belong in the protocol before anyone has seen the numbers.
Sample Size
Sample size follows from the endpoint, the expected effect, the variability in the measurement, the planned analysis method, and the question being asked.
Set it without that grounding and the study can run to completion and still finish without a clear answer. That means patients were enrolled, time was spent, and budget was used for a result that cannot carry the decision it was meant to support.
Randomization
Randomization determines how patients are assigned to treatment groups.
A statistician helps choose a method that fits the trial design, the planned sample size, the number of sites, and any factors that need to be balanced between groups.
This is not just an operational detail.
Too many stratification factors can weaken, or even render impossible, the balance that the design was meant to maintain. Enforcing balance within strata or subgroups can lengthen recruitment well beyond planned timeframes.
Those choices affect the sample size assumptions, the primary analysis, and the proper interpretation of the results.
Decision Rules
If the trial has interim looks, stopping boundaries, or rules for moving to the next phase, those need to be specified up front.
Added after the fact, they raise questions about whether the analysis was shaped by what the data showed.
None of these are sections to fill in at the end. They are part of the structure the protocol is built around.
Bringing a Statistician in Late Carries Real Costs
1. Protocol amendment.
When a design flaw surfaces after the protocol is approved, or after the first patient is enrolled, fixing it usually means an amendment. Amendments carry their own overhead: drafting, internal review, ethics and regulatory resubmission, site retraining, and delay while all of that happens.
Data collected under the earlier version may not combine cleanly with data collected under the new one. That can reduce the value of what the sponsor already paid to gather.
2. Underpowered study.
If the sample size rests on a weak assumption about the effect or its variability, the trial can complete its full enrollment and still produce a result too uncertain to act on. That is patient time and budget spent for an answer that is not useful when it arrives.
3. Statistical and Business Question Drift
The trial can meet its stated objective and still leave the sponsor short of what a regulator wants for a label, or short of what a partner or investor wants before committing the next round of funding.
The statistical question and the business question drifted apart during planning. No one connected them while there was still time to adjust.
4. Statistical considerations not specified clearly
When the statistical considerations were not specified clearly in advance, e.g. the estimand was vague about how intercurrent events would be handled, regulators ask questions. Those questions take time to answer.
They can at least mean additional analyses, a delay in the review cycle, or a more cautious reading of the data than the sponsor expected. At worst, this can render your study worthless in the eyes of the RA.
—–
Some of these are dramatic enough on their own and have the potential to limit the usefulness of a completed study to helping plan a study that corrects the deficiencies.
Regulators Expect This Thinking Inside the Protocol, Not After It
Regulatory authorities do not treat statistical planning as a separate exercise from protocol development. The guidance asks for it to be part of the same document.
ICH E8(R1) treats the statistical analyses of primary and secondary endpoints, the handling of interim analyses, and the justification of the sample size as items to be set out in the protocol itself.
ICH E9(R1) goes further. It asks that the estimand be defined and stated explicitly in the protocol, with planned sensitivity analyses to test the assumptions behind the main result.
For drug development programs working under ICH expectations, the estimand framework is now part of how regulators expect sponsors to connect the trial objective, design, conduct, analysis, and interpretation.
The FDA adopts these guidances. Health Canada, as a member of ICH, works from the same framework.
A protocol that defers its statistical thinking is therefore out of step with what agencies expect to see, and that gap tends to become visible at the least convenient moment.
Involving a Statistician Early Pays Off in Specific Ways
Some benefits are much easier to get when the statistician is part of the protocol build, not brought in after the draft is already finished.
A Statistical Strategy That Connects Each Phase
Early statistical input helps create a plan that links early and late studies, fits regulatory expectations, and respects business constraints like budget and timeline.
Without that input, each phase can start to feel like a separate statistical exercise.
Evidence Behind the Design You Chose
Simulation can help the team compare candidate designs before the protocol is locked.
It gives the team a numerical basis for choosing one design over another. That record can be useful in regulatory discussions and investor due diligence.
Smaller, Better-Aimed Late-Phase Studies
The right early design can shape the later design.
In some cases, what is learned in an early study can reduce the number of patients needed in a later trial while still answering the study question.
Shorter Gaps Between Phases
When the next study is being planned while the current one is still running, the program can avoid unnecessary idle time between steps.
Analysis-Ready Evidence
Early statistical planning helps make sure endpoints, estimands, analysis populations, missing data rules, and key outputs are defined before the study begins.
That makes the evidence easier to use for submissions, investor review, and later decision-making.
Early Statistician Involvement Means a Seat at the Planning Table, Not Just the Review
Early involvement does not mean sending the statistician a near-final protocol for review.
It means giving them a role in the planning work during the study concept and synopsis stages before the protocol is drafted.
That includes endpoint selection, estimand definition, sample size planning, analysis populations, interim analysis rules, missing data assumptions, and the statistical logic behind the next program decision.
The statistician should also be part of the conversations with clinical, regulatory, and operational leads.
That is how the team can see whether the design is scientifically sound, operationally realistic, and useful for the next regulatory or business decision.
For a small or mid-size sponsor without a statistician on staff, this does not always require a full-time internal hire. It does require access to statistical judgment early enough to shape the design, not just review the wording. That judgment can come from an outside biostatistics consultant brought in during planning, not at review.
A late review can tell the team where the protocol is weak. Early involvement can help prevent those weaknesses from being built into the protocol in the first place.
Ask the Following Before You Write the Protocol
Before protocol drafting begins, ask whether the following are pointing in the same direction:
- the clinical question
- statistical question
- regulatory question
- business question
If they are not, the protocol is already carrying risk.
A statistician is usually the person who catches that drift early, while it can still be fixed in the design.
Bring a statistician in once the design is settled, and you get a careful review of choices that are already made.
Bring them in while those choices are still open, and you have a hand in the design itself.
You may not notice the difference when you sign the first contract. You notice it a year later, in the amendment you never had to file and the regulator question you walked in already able to answer.
If you’re planning a study and want statistical input while the design is still open, that’s the work our team does with sponsors every week.
You can schedule a consultation to talk through the endpoint, estimand, and sample size before the protocol is written.
Frequently Asked Questions - FAQ
When should a statistician join a clinical trial?
A statistician should be involved before the protocol is locked, while the trial objective, endpoint, estimand, sample size, and main decision rules are still being decided.
That is the point where statistical input can shape the design rather than just check it. Bringing a statistician in after the protocol is written, or after the first patient is enrolled, means many of the decisions that matter are already set.
Do small or mid-size sponsors need a full-time biostatistician on staff?
Not always.
What matters is access to statistical judgment early enough to shape the design. That support can come from an internal hire, a consultant, or a CRO.
The cost to avoid is bringing that input in only at the review stage, once the design is already fixed.
What is an estimand, and why does it belong in the protocol?
An estimand is a precise statement of the treatment effect a trial is trying to estimate.
It combines the population, the treatment condition, the endpoint, how events such as discontinuation or rescue medication are handled, and how the effect is summarized across patients.
It belongs in the protocol because two trials with the same endpoint can answer different questions depending on how the estimand is defined.
How is the statistical analysis plan different from the statistical sections of the protocol?
The protocol sets out the main statistical features of the study. These include the endpoint, the estimand, the sample size, the main analysis approach, and key decision rules.
The statistical analysis plan, or SAP, is a more detailed document. It specifies the exact analysis methods, analysis populations, handling of missing data, sensitivity analyses, tables, figures, and listings.
For double-blind studies, the SAP should be finalized before treatment assignments are revealed. The point is to make sure the analysis is set before anyone sees information that could influence the result.
Can a statistician fix design problems after the trial has started?
Some issues can still be addressed through a protocol amendment or in the statistical analysis plan.
But core design choices are much harder to unwind once enrollment begins. A poorly chosen endpoint, an underpowered sample size, or a randomization scheme already in use can create problems that are expensive, slow, or difficult to interpret.
This is why early involvement does more for the program than a late review.
