Demand Forecasting for Adaptive and Basket Trial Designs

A conventional trial gives your supply chain something precious: a plan that doesn't change. Fixed arms, fixed allocation, a known set of countries. Uncertain, yes — enrollment is always uncertain — but structurally stable. You're forecasting one trial.

Adaptive and basket designs remove that stability deliberately. The whole point is that the trial reshapes itself as data arrives: arms get dropped, allocation shifts toward what's working, cohorts expand, sample sizes get re-estimated. Scientifically, this is a significant advance. For supply planning, it means the thing you're forecasting is a moving target — and a static forecast built at study start is describing a trial that will stop existing at the first interim analysis.

Here's how to plan supply for designs that are meant to change.

What makes these designs different

Adaptive designs allow pre-specified modifications based on accumulating interim data — dropping ineffective arms, re-allocating patients toward promising treatments, adjusting sample size, or stopping early for success or futility.

Basket trials test one treatment across multiple diseases or tumour types sharing a common biomarker, with each cohort effectively a sub-study that may expand, close, or continue independently.

Umbrella trials invert this: multiple treatments tested within one disease, with arms opening and closing over time.

Platform trials go furthest, running as an ongoing infrastructure where new arms enter and old ones exit while the trial continues.

What they share is that the demand profile is a function of decisions not yet made. You can't forecast the trial that will happen, because it hasn't been determined yet.

Why static forecasts break

Applied to these designs, conventional forecasting fails in specific, predictable ways.

• Allocation shifts invalidate arm-level demand. If response-adaptive randomization moves patients toward a winning arm, that arm's demand can rise far above plan while others collapse — and you supplied to the original ratio.
• Dropped arms strand supply instantly. The moment an arm closes, everything packaged and positioned for it becomes waste. On a conventional forecast, that's a pure write-off.
• Cohort expansion creates step changes. A basket cohort that expands doesn't grow smoothly; demand jumps, often faster than manufacturing lead time can respond.
• Timing is unknown. You know an interim analysis is coming; you don't know exactly when, or what it will decide. Supply commitments must be made before the answer exists.
• Lead times outlast decision cycles. Manufacturing, packaging, and release lead times frequently exceed the interval between adaptations. You're committing to supply for a trial configuration that may change before the product arrives.

The approach: forecast scenarios, not a number

The mental shift is from producing a forecast to producing a set of forecasts, each tied to a decision the trial might make — and then planning supply that performs acceptably across all of them.

1. Map the decision tree. Adaptations are pre-specified in the protocol, which is a gift to supply planning. Enumerate the realistic branches: which arms could drop, which cohorts could expand, at what interim points, and what each does to demand.

2. Simulate across branches. Run demand simulations for each branch, incorporating the usual uncertainty in enrollment, drop-out, and titration on top of the structural uncertainty. You end up with a distribution of demand outcomes rather than a point estimate.

3. Weight by plausibility, but plan for range. Some branches are more likely than others, but the supply plan needs to survive the plausible ones — not just the expected one.

4. Identify the common core. A crucial and underused insight: some supply is needed in every branch. That's your low-risk commitment. Manufacture it confidently. The branch-dependent portion is what requires flexibility.

5. Stage commitments to decision points. Structure manufacturing and packaging campaigns so that irreversible commitments happen as late as possible relative to interim analyses — ideally after the decision, where lead times permit.

6. Build flexibility into the physical supply. Pooled labeling and bulk-then-package strategies let supply be directed after a decision rather than committed before it.

Practical levers that help

• Delay differentiation. Hold supply in a less-committed form (bulk, or packaged but not country-labeled) for as long as feasible, so it can be directed to whichever arm or cohort survives.
• Pool labeling aggressively. In designs where the country and cohort mix will shift, pooled booklet labeling is worth considerably more than in a fixed trial.
• Shorten lead times where you can pay for it. A more expensive supplier with a shorter lead time can be the cheaper choice in an adaptive design, because it lets you commit later and waste less.
• Re-forecast at every interim. Each analysis resolves uncertainty. The forecast should be rebuilt immediately after, not at the next scheduled planning cycle.
• Get supply into the design conversation. Protocol choices — the number of interim analyses, their timing, the adaptation rules — have direct and sometimes dramatic supply consequences. Supply teams who see the design early can flag choices that are scientifically neutral but operationally expensive.

The collaboration point

That last item deserves emphasis, because it's where the biggest wins live. In flexible designs, statisticians and clinical teams make decisions — visit intervals, interim timing, allocation rules — that determine supply cost far more than any downstream logistics optimization can. A conversation between the design team and the supply team before the protocol is locked routinely saves more than months of operational tuning afterward.