Clinical Supply Forecasting and IRT: Reducing Overage Without Risking Stockouts

Clinical Supply Is an Asymmetric Risk Problem

Ask a clinical supply manager what keeps them awake and you will not hear about waste. You will hear about a site in Poland with two patients due for their week 12 visit and one kit on the shelf. That is the correct thing to worry about. It is also the reason the numbers on the other side of the ledger get so large.

The two failure modes are not comparable, and pretending otherwise produces bad decisions in both directions. A stockout at a site is a patient-level event. Depending on the protocol and the therapeutic area, a missed dose may be recoverable with a rescheduled visit, or it may break the dosing window, generate a protocol deviation, compromise the participant’s evaluability, and in some designs remove them from the primary analysis population. In an oncology or rare disease study where each participant took months to identify, that is a serious matter. It is also visible: the site reports it, the monitor documents it, the medical monitor is consulted, and the sponsor’s clinical operations leadership hears about it within days.

Overage is invisible until the very end. The material is manufactured, packaged, labeled, released, distributed, and then, at the close of the study, reconciled and destroyed. There is no single moment where someone has to explain it. It shows up as a line in a reconciliation report eighteen months after anyone could have done anything about it.

The trade-off structure, stated plainly

Most supply plans carry an unstated exchange rate. It sounds something like this: we are willing to manufacture a great deal of extra material in order to make the probability of a site stockout very small. That is a defensible position. What is not defensible is refusing to say what “a great deal” and “very small” mean, because without those two numbers there is no way to tell whether the plan is well calibrated or wildly conservative.

Consider the arithmetic in a published example. A global study with 100 sites and 1,000 patients receiving monthly kits for a year needs 12,000 kits. At a hypothetical $100 per kit, moving from a 200% overage assumption to a 50% overage assumption is a difference of roughly $1.8 million in material.¹ That is the entire question in one line. Is the incremental protection bought by the extra 150% worth $1.8 million? Sometimes yes. For a first-in-class oncology asset with a six-month manufacturing lead time and no ability to make more, quite possibly yes. For a well-characterized oral solid in a Phase 3 program with existing commercial-scale capacity, almost certainly not.

Nobody can answer that question if the two sides are never placed next to each other. And they usually are not, because the two sides are owned by different people. The stockout risk belongs to clinical operations and the medical monitor. The material spend belongs to clinical supply and finance. The IRT configuration belongs to a vendor or a small internal team. Each group is optimizing its own visible failure mode.

Why the asymmetry justifies conservatism, but not this much of it

It is worth being clear that the asymmetry is real. We are not arguing that supply teams should accept more stockout risk. The argument is narrower and more useful: a large share of the overage currently carried does not buy any reduction in stockout risk at all. It sits in the wrong country. It sits at a site that will never enroll. It expires before the patient who would have used it arrives. It is the wrong dose strength for the participants who actually enrolled.

Material that cannot reach the patient who needs it is not protection. It is inventory. The distinction between the two is the whole subject of this article, and it is a distinction that a single blended overage percentage is structurally incapable of expressing.

Writing the Asymmetry Down: The Supply Plan as a Risk Decision

Every organization we work with has a validation plan, a data management plan, a monitoring plan, and a risk management plan. Very few have a document that states, in one place, what stockout probability the study is willing to accept and what it is prepared to spend to get there. The supply plan exists, but it usually reads as a set of logistics arrangements rather than a risk decision.

This matters more now than it did five years ago. ICH E6(R3), the current Good Clinical Practice guideline, is built around a risk-proportionate approach: sponsors are expected to identify the factors critical to trial quality, assess the risks to those factors, and apply controls in proportion to the risk.¹¹ Investigational product availability is plainly critical to quality. A study that cannot dose participants on schedule cannot produce interpretable data. So the guideline’s own logic points toward documenting the supply risk decision, not just the supply logistics.

What a documented supply risk decision contains

None of this is exotic. It is the same discipline applied to any other risk-based decision in a regulated program: state the risk, state the tolerance, state the controls, state who owns it, and revisit it. The reason supply escapes that discipline is largely historical. Supply planning grew up as an operations function and inherited operations habits, while risk management grew up in quality and inherited quality habits. They have not fully met.

What the IRT System Actually Controls

This is the most actionable section of this article, and it is the one most often skipped in strategy discussions about clinical supply, because it looks like a configuration detail rather than a leadership issue.

The manufacturing plan determines how much material exists. The IRT configuration determines where it goes, how much of it sits at each site, how often it moves, and how much of it will still be usable when a patient needs it. In practice, the second set of decisions drives more of the end-of-study destruction number than the first. And unlike the manufacturing plan, which is reviewed by several functions and approved formally, the IRT resupply settings are frequently set once during study build and left alone.

The four parameters that do most of the work

Commercial IRT platforms differ in naming, but the underlying resupply model is broadly consistent. Oracle’s published IRT documentation describes it directly.¹² Four parameters carry most of the weight:

Two resupply strategies sit on top of these. In a buffer strategy, only the minimum and maximum apply and patient visits do not influence orders at all. In a projection strategy, the full model applies and the system orders against the visit schedule.¹² A buffer strategy is simple and safe, and it is also the configuration most likely to strand material at sites whose enrollment never materialized, because it holds stock at every site regardless of whether anyone is coming.

The settings that quietly decide where material is stranded

Beyond the core four, most systems expose country and depot level controls that determine allocation. Two in particular deserve attention because they are adjustable during the study and rarely adjusted: the values that prevent shipping to a given country or site, and the values that exclude certain inventory from availability calculations. A 2024 analysis of supply chain optimization describes dynamic adjustment of exactly these controls at country and visit level as one of the mechanisms behind a case where a single study saved approximately $6 million, about a quarter of its total supply budget.³

The system also decides which lot to allocate. Standard practice is to release the earliest-expiring available lot first.¹² That sounds obviously correct, and it is, but it interacts with everything else: if short-dated material is pushed out to sites under a generous buffer setting, it will sit there and expire rather than being consumed centrally where demand is pooled.

Why the settings drift out of alignment

The drift is not carelessness. It has a specific mechanism. At study build, nobody knows which sites will enroll well, so the configuration is set uniformly and conservatively. Then the study starts, real enrollment data accumulates, and the picture changes completely: a handful of sites carry most of the participants, a third of the sites carry none, and the country mix looks nothing like the feasibility assessment. But the configuration was never framed as a parameter to be managed. It was framed as a build artifact. Changing it requires a change control, a vendor request, possibly a validation touchpoint, and a person willing to own the decision. In the absence of a defined review cadence, the path of least resistance is to leave it alone.

The same industry commentary that reported the 62% unused figure is blunt about the consequence. Resupply settings can trigger shipments that go unused even by the time the next shipment arrives, and the objective should be to send more material per shipment without raising site buffers, as late as possible while still guaranteeing that no participant is ever without drug.¹ That is a description of a tuning problem, not a manufacturing problem.

Where the Overage Actually Comes From

If you want to reduce overage without touching stockout risk, you have to know which part of the overage is doing protective work and which part is not. Four drivers account for most of the non-protective share.

Driver 1: Enrollment uncertainty, especially sites that never enroll

This is the largest single source and the least discussed in supply terms. Medidata’s analysis, drawn from more than 2,000 studies, found that nearly one-third of sites that actively participated in a trial never recruited a single patient. Regional variation was substantial: Asia Pacific performed best at just over 17%, while some geographies exceeded 40%.²

Now combine that with a uniform minimum buffer. Every one of those non-enrolling sites received an initial shipment and was maintained at its buffer level for as long as it stayed active. That material was never protection for anybody. It was allocated against a patient who never existed, and in most cases it will be returned and destroyed. If a third of your sites are in that category, a third of your site-level buffer inventory is pure waste, and no amount of improvement in the demand forecast will touch it, because the forecast was not the problem. The allocation policy was.

This is why site performance analytics and supply configuration should be connected functions rather than separate ones. The data that tells you a site is not going to enroll is generally available months before the supply configuration reacts to it, if it reacts at all.

Driver 2: Expiry and retest date management

Investigational material has a shelf life, and in early studies that shelf life is often short because the stability program has not matured. The functional shelf life is shorter still, because sourcing, packaging, release, and distribution lead times all consume it before the material reaches a site. One published comparator case described material with only ten months of usable life remaining after quality person release.¹⁴

Short-dated material interacts badly with generous buffers. Material pushed to sites early, against a demand that arrives late, expires in place. Then it has to be replaced, which means a second manufacturing and packaging campaign, which means the total quantity produced for the study rises well above the number of doses any patient will ever take. Expiry-driven replacement is one of the clearest cases where overage is not protection at all: it is the same protection, purchased twice.

Driver 3: Country and depot allocation that strands supply

Global studies distribute material through regional depots, and material released to a country is generally committed to that country. Import licenses, country-specific labeling, and local regulatory requirements make it difficult or impossible to move stock from a country with slow enrollment to one with fast enrollment. The result is a familiar pattern: a study is simultaneously over-supplied in aggregate and at risk of stockout in the one country that is enrolling ahead of plan.

This is the single most persuasive argument for treating supply as a network problem rather than a per-site problem. A blended overage percentage applied evenly across a network guarantees that some nodes are over-stocked and others are under-stocked, because the percentage carries no information about where demand will actually appear.

Driver 4: Titration and weight-based dosing

Fixed-dose studies have a narrow distribution of what each participant will consume. Titration studies and weight-based or body-surface-area dosing do not. Every participant may need a different strength, and the strength may change over the course of treatment. If the supply plan covers the worst case for every participant at every strength, the total quantity balloons.

The clearest published illustration comes from a pediatric study enrolling participants from newborn to eighteen years with weight-based dosing across multiple locations. The initial distribution was sized to cover worst-case requirements across all treatment doses. In the end, roughly 6,000 kits were needed out of 55,000 distributed, leaving about 49,000 kits destroyed.³ That is not a forecasting failure in the ordinary sense. It is what happens when the planning method is worst-case arithmetic applied independently to every dose strength at every site.

What Actually Reduces Overage

Having separated protective overage from stranded inventory, the levers that matter become clearer. Each of the following reduces the quantity of material required without increasing the probability that a participant misses a dose.

Lever 1: Pooling supply across studies and countries

Pooling means holding material centrally in a form that can serve more than one destination, rather than committing it to a specific study, country, or site at packaging. Interactive response technology supports this by managing shared drug types, depots, lots, and country-specific label groups across multiple studies. One published scenario describes a sponsor that would have packaged 10,000 kits for each of three studies, 30,000 in total, and instead packaged 25,000 for the program as a whole.¹⁵

The mechanism is straightforward statistics. Demand variability across a pool is smaller in relative terms than the sum of variability across its parts, so the buffer needed to cover the pool at a given service level is smaller than the sum of the individual buffers. Nothing about the participant’s experience changes. The protection is identical. The inventory required to deliver it is lower.

The constraints are real and worth naming. Pooling requires a vendor whose platform supports it, it requires supply planning at the program level rather than the study level, and it requires decisions made for the first study in a pool that will constrain later protocols.¹⁵ That last point is the one that stops most pooling programs: the organizational planning horizon is shorter than the pooling horizon.

Lever 2: Just-in-time and late-stage labeling

Labeling is the step that commits material to a destination. Under EU rules, labeling requirements are extensive and country-specific, which is why booklet labels covering many languages became standard practice. If labeling happens at packaging, the material is locked to a country long before anyone knows where the participants will be.

Moving labeling later, to the depot and close to the point of dispatch, keeps material fungible for longer. It converts country-committed inventory back into pooled inventory, which is the same statistical benefit described above applied to geography instead of protocol. It also reduces the waste inherent in producing large multilingual booklets for material that will only ever go to one country.

Digital display labeling is an emerging variation on the same idea. An electronic label with an e-paper display replaces the printed label, and content is written to it over near-field communication. Published work through the ISPE Clinical Supply Leadership Forum describes this as reducing relabeling cycle times from weeks to days and supporting both supply pooling and just-in-time labeling, while noting that universal regulatory guidance has not been established globally.¹⁸ That last caveat is important. This is a technology to evaluate deliberately, not to assume.

Lever 3: Expiry extension through an ongoing stability program

Extending the expiry date on material already in the field is the most direct way to avoid a second manufacturing campaign. It also has the most regulatory structure around it, and the structure differs by region.

The MHRA has stated the position clearly for UK trials. Extension of shelf life is a substantial amendment unless there is a prior agreement in the approved investigational medicinal product dossier to extend when further stability data becomes available. The regulator must be informed through a variation to the clinical trial authorization, and the Product Specification File must go through a controlled change so manufacturing sites and qualified persons can act.⁸ The words that matter most in that sentence are “unless you have previous agreement”. Building the extension mechanism into the dossier before the study starts converts a substantial amendment into a planned update. Very few teams do this, and it is close to free at the point where it has to be decided.

The relabeling operation itself is more flexible than many assume. MHRA guidance indicates relabeling may be performed at the clinical site, and that qualified person certification is not required for stock already at a trial site if the relabeling is carried out by or under the supervision of a pharmacist, while stock not yet shipped to an investigator site does require new certification after relabeling.⁸ Continuing to use material past its labeled expiry date is a different matter and is usually not acceptable except where the supply is critical to participant care and has been agreed with the regulator in advance.⁸

The underlying labeling requirement sits in Annex 13 of the EU GMP guidelines, which calls for the period of use to appear on the label, expressed as a use-by date, expiry date, or retest date as applicable.⁹ Under the Clinical Trials Regulation, the labeling particulars are set out in Annex VI, which distinguishes between outer packaging and immediate packaging and defines the limited circumstances in which particulars may be reduced.¹⁰ Teams planning any expiry strategy should read the applicable annex rather than relying on institutional memory, because the memory in most organizations predates the current rules.

Lever 4: Simulation instead of deterministic worst-case arithmetic

This is the difference between a forecast and an optimization, and the distinction is more than semantic. A forecasting tool takes overage as an input, usually derived from a rule of thumb or a comparable past study, and calculates a plan from it. A risk-based optimization tool runs simulations across the uncertain parameters and produces overage as an output, together with a stated risk level.¹³

That inversion is the whole point. When overage is an input, the number is an assumption nobody can challenge, because there is no evidence attached to it. When overage is an output, it comes with a risk figure, which means a leadership team can look at two plans and ask a real question: this plan carries 20% less material and raises the modeled probability of a dispensing failure from 0.4% to 0.6%. Do we accept that? Now the conversation is about the trade-off rather than about whose instinct is better.

One published comparator sourcing case illustrates the progression across successive optimization passes: an initial requirement of 65% overage, reduced to 52%, and finally to 35%.¹⁴ The material need fell by nearly half without a change in the protocol or the enrollment plan. What changed was the method used to compute it.

Lever 5: Adaptive settings driven by real enrollment

The final lever is the one that ties the others together, and it gets its own section because it is where most of the accessible value sits.

Revisiting IRT Settings on a Defined Cadence

A supply plan built at study start encodes the best available guess about enrollment. Within three months of first patient in, that guess has been replaced by data. Within six months it has usually been contradicted. The question is whether the configuration is allowed to move in response.

Vendors have moved in this direction. Current commercial platforms describe continuously evaluating supply conditions across a study and dynamically adjusting resupply strategies, with the explicit aim of reducing the need for excessive buffer stock while minimizing stockout risk and aligning supply more closely with true patient demand.¹⁶ The technology is available. What is often missing is the organizational routine that uses it.

A workable review cadence

Making the review possible, not just scheduled

A cadence on paper accomplishes nothing if the review has no authority. Three things make it real.

A named decision owner with the authority to change settings. Not a committee. One accountable person who can approve a configuration change and who is measured on both stockout events and end-of-study destruction, so that neither failure mode can be optimized in isolation.

A change path that is proportionate. Resupply parameter changes within a pre-approved range should not require the same effort as a system change. Define the range in the validation documentation up front, so routine tuning sits inside the qualified envelope rather than triggering a fresh assessment each time. This is exactly the kind of proportionality ICH E6(R3) contemplates.¹¹

Reporting that shows both sides. Most supply dashboards show inventory position and shipment activity. Very few show projected end-of-study destruction alongside projected stockout risk. Until those two numbers appear next to each other in the same review, the asymmetry stays invisible and the decision stays unmade. Organizations moving toward unified trial technology platforms have an advantage here, because the enrollment data and the supply data are already in the same place.

Constraints That Can Rule Out an Otherwise Sensible Strategy

Everything above assumes room to maneuver. Often there is not. Four constraints can eliminate an otherwise sound strategy, and a supply plan that ignores them is not conservative, it is simply wrong.

Comparator and co-therapy sourcing

Comparator material is bought, not made, and that changes the entire calculation. Tufts Center for the Study of Drug Development research covering 11 major pharmaceutical companies and 370 studies found participating companies spending an average of $50 million per year on clinical supplies, with half of the entire clinical supply budget going to comparator drugs and co-therapies, and individual company spending ranging from $10 million to $120 million.⁴ Nearly 90% of studies used higher-priced branded comparators.⁴

The operational constraints are more limiting than the spend. Sponsors frequently cannot obtain comparators directly from the competitor and must go through wholesalers, third-party buyers, or local pharmacies. Availability is unpredictable, particularly for products that represent a large share of the originator’s revenue. Obtaining the required documentation, including certificates of analysis, certificates of conformity, and temperature excursion data, is repeatedly cited as a top cause of delay.⁴ Sourcing managers report particular difficulty when supply is limited or when the available material carries short expiry dating.⁴

The practical consequence: for comparator material, a just-in-time strategy may be unavailable because you cannot reliably buy on demand. Buying in larger, earlier tranches may be the only way to guarantee availability, which pushes overage up for reasons that have nothing to do with poor planning. The same Tufts work notes that participating companies could not reliably report how much comparator material went unused, though senior sourcing managers estimated anecdotally that somewhere between 30% and 55% is left over when studies complete or terminate.⁴ That range should be treated as what it is, an informed estimate rather than a measurement, and it is a fair indication of how little visibility exists.

Temperature-sensitive material and cold chain

Biologics, cell and gene therapies, and many vaccines carry storage requirements that constrain every part of the strategy. Cold chain distribution is more expensive per shipment, which pushes toward fewer and larger shipments, which pushes site buffers up. Excursions during transit can quarantine material pending quality disposition, which means the nominal inventory at a site is not always the usable inventory. Depot capacity for controlled storage is finite and cannot be expanded quickly.

Pooling is harder for the same reason: material held at a central location still needs qualified storage, and the number of qualified locations is limited. Teams working through cold chain traceability approaches generally find that visibility improves decision quality but does not remove the physical constraint.

Regulatory, import, and labeling requirements

Country-specific import licenses, local language labeling, and national requirements for release all commit material to a jurisdiction. Under the Clinical Trials Regulation, labeling particulars are prescribed in Annex VI, and the circumstances in which particulars may be reduced are defined rather than discretionary.¹⁰ Some countries require in-country qualified person release or local testing. Others restrict the movement of investigational material once imported.

On the United States side, the investigator is required to maintain adequate records of the disposition of the drug, including dates, quantity, and use by participants, and unused supply must be returned to the sponsor or otherwise disposed of under defined conditions.¹⁷ These accountability obligations are not obstacles to reducing overage, but they do mean that any strategy involving reallocation or relabeling has a documentation trail that has to be designed, not improvised.

Blinding

Blinding constrains supply in ways that are easy to underestimate. If active and placebo are visually distinguishable in bulk, pooling across studies with different randomization ratios can create inference risk. If a titration schedule means the number of kits dispensed reveals dose level, then adjusting site inventory by dose strength can reveal assignment. Kit allocation logic itself has to be designed so that inventory position at a site does not leak treatment information to the pharmacist or the monitor.

This is a genuine reason that some otherwise attractive optimizations are not available. It is also a reason to involve the statistician in the supply configuration discussion, which almost never happens. In adaptive designs, where allocation ratios may change at an interim analysis, the interaction between blinding, adaptation, and supply is more complex still and should be worked through at protocol design rather than discovered during conduct.

A Practical Sequence for the Next Study

None of this requires a transformation program. It requires a different sequence of decisions on the next study that starts, and a willingness to look at one study already in flight.

For a study in design

For a study already running

Start with the single highest-value question: which active sites have received material and have never screened a participant? Reduce or suspend their buffers. This requires no new tooling, no simulation, and no regulatory interaction. It is a configuration change against data the organization already has, and given that roughly a third of sites in a typical study never enroll anyone, it is usually the largest available reduction.²

Then look at expiry. Identify material that will expire before the projected end of enrollment in its assigned country. Decide, deliberately, whether to pursue an extension, reposition the material where that is permitted, or accept the replacement campaign. The point is to make the decision consciously rather than discovering it as a shortage six months later.

Then set the review cadence for whatever remains of the study. Even a study three-quarters enrolled has material decisions ahead of it, and the expiry and destruction consequences of those decisions land after the last patient visit, when nobody is looking.

What this connects to

Clinical supply optimization is often treated as a niche operational concern, separate from the broader question of how a sponsor uses its clinical data. It is not separate. The inputs to a good supply decision are enrollment data, site performance data, visit compliance data, and dose distribution data, all of which live in other systems. The organizations that make progress here are generally the ones that have already done the work of connecting those systems, which is why supply optimization tends to arrive as a consequence of better demand sensing and inventory practice rather than as a standalone initiative.

Conclusion

The asymmetry between a stockout and an overage is real, and it should shape the supply plan. A missed dose can end a participant’s involvement in a study that took years to enroll. Destroyed material is money. Those are not equivalent, and any framework that treats them as equivalent deserves to be rejected. But the correct response to an asymmetric risk is a calibrated margin, not an unexamined one. The evidence that the current margin is uncalibrated is not subtle: 62% of packaged material unused across 200 studies, a third of sites that never enroll a patient still carrying buffer inventory, and single studies where tens of thousands of kits were distributed against a need in the low thousands.

The most useful thing a life sciences leader can do about this is not to buy a forecasting tool. It is to insist that two numbers appear in the same review: what this plan is likely to destroy, and what risk it is actually carrying. Then to make sure the settings that drive both are treated as parameters someone owns and revisits, rather than build artifacts nobody touches. The IRT configuration is where most of the accessible value sits, it is adjustable during the study, and in most programs it has not been looked at since first patient in.

Sakara Digital works with pharma and biotech organizations on the data, systems, and governance decisions behind clinical operations, including how supply, enrollment, and site performance data connect well enough to support decisions like these. If you are looking at a study where the end-of-study destruction number is uncomfortable, or you want an independent read on whether your supply plan is calibrated or just conservative, we are happy to have that conversation.

For Further Reading