Why Autoinjectors Fail After Phase III — and Why It’s Not a Device Problem

Key Takeaways

• Autoinjector and prefilled syringe failures often arise from system-level coupling, not standalone device malfunction.

• Viscosity, injection force, and delivery time form non-negotiable physical constraints that shape device reliability and patient experience.

• Late-stage incompatibilities surface when final formulations, fill volumes, and device requirements converge, revealing limits that were always present.

• Break-loose and glide force drift can quietly erode performance margins, triggering failures during validation or tech transfer.

• CDMOs serve as the proving ground where coupled drug–device assumptions are tested against real manufacturing variability.

Late-stage failures in autoinjectors and prefilled syringes (PFS) are frequently framed as device problems: a spring that underperforms, a syringe that sticks, an injector that fails to deliver the full dose. This framing is appealing because it localizes responsibility, but it obscures a more consistent pattern observed across combination products: these failures rarely originate in the device alone. They emerge when drug properties, container-closure performance, and injector mechanics are treated as separable components for too long, only to collide late in development or during tech transfer.

From a regulatory and quality perspective, combination products are already recognized as integrated systems rather than loose assemblies. The governing framework for combination product cGMPs was explicitly established to clarify how drug and device requirements apply together, with the goal of ensuring products consistently meet applicable requirements and specifications across their life cycle.¹ This systems orientation is not optional; it reflects the reality that performance at the point of use depends on the interaction of multiple constituent parts rather than on any single element in isolation.

Despite this, development programs often proceed as if modularity were achievable in practice. Drug formulation is optimized for stability, concentration, and exposure. The primary container is selected and qualified. The injector is chosen or developed in parallel. Each element may meet its own criteria, yet the combined system remains only partially understood. The assumption — sometimes explicit, more often implicit — is that integration can be finalized later through qualification and verification activities.

The physics of injection undermines this assumption. Injection performance is governed relationships among viscosity, force, time, and other variables. These relationships dictate how much energy an injector must supply, how long delivery will take, and whether dose completion is achievable within acceptable limits. They do not emerge from organizational structure or development sequencing; they are inherent to the system. As a result, late-stage failures are not simply the product of poor coordination or late testing; they are the inevitable consequence of treating a coupled physical system as if it were modular.

This reframing shifts the question away from why a device “failed” and toward why the system was ever expected to tolerate separation in the first place. Combination products break late not because teams ignored integration, but because integration was assumed to be deferrable. The reality is harsher and more instructive: the coupled nature of drug, container, and device asserts itself eventually, and when it does, it does so on the terms set by physics rather than by development plans.

Combination products are often managed as if they were assemblies: a drug product paired with a container and delivered through a device, each governed by its own specifications and qualification milestones. This mental model suggests that constituent parts can be optimized independently and integrated later through testing and documentation. In practice, this approach breaks down because it misrepresents how these products function at the point of use. Formulation, primary container, and device do not behave as discrete modules but rather as a coupled system in which changes in one domain propagate across the others.

Regulatory and quality frameworks already reflect this reality. The cGMP requirements for combination products were established to clarify how drug and device quality requirements apply together, with the explicit objective of ensuring that combination products consistently meet applicable requirements and specifications throughout their life cycle.¹ This framing recognizes that performance cannot be assured by evaluating constituent parts in isolation. What matters is whether the combined system performs as intended under real conditions of use.

Design control expectations reinforce this systems view. Design validation is defined as the process of providing objective evidence that device specifications conform to user needs and intended use, and it must be completed before commercial distribution.² For combination products, intended use is inseparable from the drug being delivered, the way it is presented in the primary container, and the manner in which the device administers it. Validation is thus not simply about whether a device meets its own requirements but whether the integrated system delivers the drug safely and effectively for its intended users.

Importantly, design control principles apply not only to initial device design but also to associated manufacturing processes and to modifications made over time.³ This means that changes in formulation, container components, or manufacturing conditions are not peripheral adjustments that remain confined to their originating domain. They alter the behavior of the coupled system and, by extension, the validity of prior design assumptions.

Seen through this lens, the notion that integration can be deferred until late-stage qualification becomes difficult to defend. Deferring device qualification presumes that the device can be meaningfully evaluated apart from the evolving drug and container system it must ultimately serve. The coupled-system model rejects that premise. It asserts that integration is not a final step but a continuous constraint, one that must be respected throughout development if late-stage failures are to be avoided.

The persistence of modular thinking in combination product development is not the result of inexperience or negligence. It is a structural outcome of how programs are organized and sequenced. Drug formulation, primary packaging, and device development are often not only advanced along parallel timelines, but each is given its own owners, deliverables, and success criteria. Within this structure, it becomes natural to treat the device as a component to be selected or optimized independently, with the expectation that integration can be finalized once the major uncertainties have been resolved.

Milestone-driven development reinforces this tendency. Early progress is often measured by discrete achievements (e.g., formulation stability established, container qualified, and device feasibility demonstrated). Each milestone reduces perceived risk within its own domain, creating confidence that the remaining integration work is largely procedural. Qualification and verification activities are then viewed as sufficient mechanisms to reconcile any remaining mismatches, even when the underlying system behavior has not been fully explored.

This confidence is further supported by the belief that rigorous testing can substitute for early integration learning. If each element meets its specifications (the logic goes), the assembled product should perform as intended once formally qualified. What this overlooks is that specifications themselves are often defined within a limited context. They capture acceptable behavior under assumed conditions, not the full range of interactions that emerge when formulation properties, container behavior, and device mechanics are combined and pushed toward commercial reality.

Deferral changes not only when integration risk is addressed, but how it is discovered. Early in development, mismatches between formulation and device are largely exploratory problems: parameters can be adjusted, alternatives evaluated, and tradeoffs explored with a reasonable degree of flexibility. When integration is deferred, those same mismatches surface later as constraints. By the time the full system is exercised, often during late-stage development, validation, or technology transfer, the formulation may be locked, the device architecture fixed, and the manufacturing pathway committed.

The result is not simply late discovery, but expensive discovery. Integration risk does not disappear when it is deferred; it accumulates. When it finally becomes visible, the cost of change is higher, the range of viable options narrower, and the tolerance for iteration significantly reduced. The modularity myth persists because it aligns with how teams are structured and rewarded. Its consequences emerge because the system itself does not respect those boundaries.

At the point of administration, drug delivery through an autoinjector or PFS is governed by mechanical constraints. Injection is fundamentally a force-and-time problem. A finite amount of energy must be transferred through the system to overcome fluid resistance, container friction, and needle flow limitations and deliver a defined volume within an acceptable timeframe. These relationships are dictated by rheology and flow resistance, not by the intent of the developer or patient-centric decisions.

Injection time is often treated as a downstream characteristic that can be observed, adjusted, or tolerated once a device is in hand. In truth, it is an engineered outcome emerging from the interaction of parameters including formulation viscosity, needle geometry, plunger friction, spring characteristics, and system losses. Because these parameters interact among each other, injection time cannot be inferred reliably from any single attribute in isolation. It must be predicted and controlled as part of system design.

Modeling work has demonstrated that injection time in spring-driven autoinjectors is sensitive to both formulation properties and component-level variability, and that these sensitivities can be quantified rather than treated as empirical surprises. Physics-based approaches show that parameters influencing injection time can be represented statistically and propagated through the system to predict not merely nominal performance but the expected range of outcomes due to component variability and manufacturing tolerances.⁴ This shifts injection time from a qualitative expectation to a measurable, designable property.

Rheology adds an additional layer of constraint. Many formulations exhibit shear-dependent viscosity, meaning their resistance to flow changes as shear conditions vary within the needle and syringe. Accurate prediction of injection behavior therefore depends on representing viscosity under relevant shear rates, not on static measurements taken under unrelated conditions. Modeling approaches that incorporate shear-dependent viscosity have shown that injection time prediction is tightly coupled to how formulation rheology is characterized and represented.⁵

Together, these considerations expose why injection performance so often becomes problematic late. When viscosity, force, and time are treated as loosely related attributes rather than as coupled design variables, systems may appear robust under early conditions but prove fragile when pushed toward final formulations, commercial fill volumes, or tighter performance expectations. The mechanics do not change late in development; they are simply revealed when the system is finally asked to operate within its true physical limits.

Viscosity is often framed as a formulation challenge that makes injection more difficult or less comfortable. This understates its impact. In an autoinjector system, viscosity is a governing variable that shapes the entire operating space within which delivery is possible. As formulation viscosity increases, the force required to drive flow through the needle and syringe rises accordingly, placing greater demands on the injector’s available energy and directly constraining how quickly delivery can occur.

Because autoinjectors rely on finite energy sources — typically springs with defined force–displacement profiles — the relationship between viscosity and force has immediate consequences. Higher viscosity consumes a larger fraction of the available energy budget simply to sustain flow. This leaves less margin to accommodate other system losses, such as plunger friction, component variability, or transient resistance at the start of injection. Injection time, in turn, is not freely adjustable; it emerges from how quickly the system can dissipate its stored energy against the combined resistances imposed by the formulation and hardware.

Shear-dependent viscosity complicates this picture further. Many injectable formulations do not behave as simple Newtonian fluids. Their apparent viscosity changes with shear rate, meaning that resistance to flow varies along the injection pathway and over the course of delivery. Within a spring-driven system, shear rates are not constant; they evolve as force decays and flow conditions change. Modeling efforts that account for shear-dependent viscosity show that injection behavior cannot be accurately predicted using a single viscosity value measured under arbitrary conditions. Instead, performance depends on how the formulation responds under the specific shear regimes encountered during injection.⁵

The practical implication is that viscosity does more than slow injection. It reshapes the feasible combinations of force and time that the system can support. As viscosity increases, acceptable operating windows narrow, and tradeoffs that were previously latent become unavoidable. A system that appeared robust at lower viscosities may cross a performance threshold with relatively modest formulation changes, not because the device has failed, but because the coupled system has been pushed beyond the limits imposed by its physics.

Patient comfort is often invoked as a design objective that can be optimized independently of device mechanics. Slower injections are presumed to feel gentler, lower forces less aggressive, and longer delivery times more tolerable. In contrast, device reliability is framed as an engineering problem: ensuring sufficient force margins, dose completion, and consistent performance across conditions. In practice, these objectives are not independent: they sit in structural tension because they draw on the same finite physical resources of the system.

Evidence from subcutaneous injection studies underscores that tolerability does not hinge on any single parameter in isolation. Injection comfort depends on the combined conditions of delivered volume, formulation viscosity, and injection time, rather than on viscosity or speed alone.⁶ This means that attempts to improve comfort by slowing delivery or increasing volume cannot be evaluated without considering how those changes interact with formulation properties and device capabilities. What feels acceptable under one set of conditions may become intolerable or unreliable when another parameter shifts.

From the device perspective, reliability depends on maintaining adequate force and energy margins to ensure complete delivery within defined time constraints. Slower injections require sustained force over longer durations, increasing exposure to variability in friction, component tolerances, and formulation behavior. As delivery time stretches, the system becomes more sensitive to small deviations that would have been inconsequential at higher flow rates. The same design choices that appear to favor comfort can therefore erode reliability margins, particularly as formulations evolve or manufacturing variability is introduced.

This tension is often treated as a late-stage optimization problem in which one can adjust the injection profile, tune the spring, or refine the device once comfort data are available. The reality is that comfort goals and device limits are coupled early through viscosity, force, and time. When these couplings are not explicitly acknowledged, systems may progress through development with latent conflicts embedded in their design. Those conflicts tend to surface late, when final formulations, fill volumes, and device constraints converge, and when the remaining room to maneuver has largely disappeared.

Formulation changes that appear incremental within a development context can have outsized effects once they propagate through a coupled drug–device system. Adjustments in concentration, excipient composition, or fill volume are often pursued for sound reasons: improved stability, reduced dosing frequency, or alignment with commercial presentation. Individually, these changes may fall within expected formulation variability. Collectively, they can reshape the mechanical demands placed on the injector in ways that are neither linear nor immediately apparent.

Viscosity drift is a common example. A modest increase in viscosity — whether due to higher concentration, temperature sensitivity, or formulation aging — does not simply slow injection. It increases the force required to sustain flow, drawing down the injector’s available energy margin. Because injection time is an emergent property of force balance rather than a directly controlled setting, this loss of margin can manifest abruptly. A system that previously delivered within its target window may suddenly approach or exceed its performance limits with little warning.

Changes in excipient composition can introduce similar effects through less obvious pathways. Interactions between the formulation and the primary container can alter lubrication behavior, increasing resistance at the plunger–barrel interface. These effects may not register as formulation failures in isolation, but they can directly affect the forces required to initiate and sustain injection. When combined with higher viscosity or longer delivery times, they can tip the system into a regime where reliable dose completion is no longer assured.

Fill-volume adjustments compound these sensitivities. Increasing the delivered volume extends the duration over which force must be applied, which exposes the system to cumulative losses and variability that were previously negligible. The injector does not experience this as a gradual degradation. Instead, performance margins are consumed until a threshold is crossed, at which point failures appear sudden and disproportionate to the apparent magnitude of the change.

This pattern explains why device-related issues so often emerge late. Systems appear stable not because they are robust, but because they have not yet been stressed into their true operating envelope. Small formulation changes consume margin rather than reveal weakness incrementally. By the time failure becomes visible, it reflects the cumulative impact of multiple “minor” decisions interacting within a tightly coupled system, rather than a single, identifiable misstep.

Late-stage incompatibilities in autoinjectors and PFS are often described as unexpected, yet their timing follows a consistent pattern. These failures tend to surface after phase III, not because something fundamentally changed in the underlying science but because the system is finally being exercised within the conditions it was always destined to face. Earlier stages of development typically operate in a different region of the design space that is more forgiving and less representative of commercial reality.

In early clinical programs, formulations are often less concentrated, viscosities lower, and delivered volumes smaller. Administration may occur via manual injection or through simplified delivery setups that mask the constraints of the intended final device. Under these conditions, force margins appear ample, injection times acceptable, and variability manageable. The system performs well, not because it is robust but because it has not yet been asked to perform at its limits.

The transition to late-stage development collapses this buffer. By the time phase III is complete, the formulation is usually locked to support regulatory submission and commercial manufacture. Fill volumes are finalized to match dosing strategy and patient convenience. Device requirements become fixed as human factors, labeling, and validation expectations crystallize. It is at this point —when formulation, container, and device converge into their final configuration — that the system enters its true operating envelope.

Within this window, the latent couplings described earlier assert themselves. Viscosity increases that were tolerable in isolation now interact with fixed force budgets. Longer delivery times amplify the impact of friction and variability. Assumptions made during parallel development are no longer abstract but embedded in hardware and specifications that cannot be easily altered. What appears as a late-stage failure is the first full expression of the system’s actual constraints.

This reframing removes the mystery from post–phase III incompatibilities. They are not aberrations introduced by scale-up or transfer alone. They are the predictable outcome of deferring integration until the moment when all degrees of freedom have been consumed. When discovery is postponed this far, it does not arrive as insight — it arrives as disaster.

Among the less visible contributors to late-stage failure in autoinjectors and PFS is the behavior of plunger forces over time. Break-loose and glide force are sometimes treated as narrow engineering metrics, relevant primarily to component selection or incoming inspection. In practice, they are system-level performance variables that directly constrain whether a dose can be delivered reliably under real conditions of use.

Break-loose force refers to the peak force required to initiate plunger movement from its resting position, while glide force describes the force needed to sustain plunger motion once static friction has been overcome. Both forces must remain within defined limits to ensure that the injector can initiate and complete delivery of the intended dose.⁷ If either exceeds the available force margin of the device, injection may be delayed, prolonged, or incomplete, even if the formulation and device architecture have not ostensibly changed.

What makes plunger-force behavior particularly insidious is its tendency to drift rather than fail abruptly. Time-dependent changes in container components, interactions between the formulation and the primary container, or cumulative variability across lots can incrementally increase resistance at the plunger–barrel interface. Individually, these shifts may remain within specification. Collectively, they can erode the force margin that the injector relies on to compensate for viscosity, flow resistance, and other losses.

This erosion often goes unnoticed until late-stage activities impose tighter constraints. During validation, stability studies, or technology transfer, systems are exercised more systematically and across a broader range of conditions. It is in these contexts that tolerance stacking becomes visible: a slightly higher viscosity, a marginally higher break-loose force, and a longer required injection time combine to push the system beyond its reliable operating range. The resulting failures can appear sudden, but they are the culmination of gradual, unobserved drift.

By framing break-loose and glide force as dynamic contributors to system performance rather than static acceptance criteria, this failure mode becomes easier to understand. Late-stage issues are not always the result of a single out-of-spec component. They often reflect the slow consumption of margin within a coupled system, revealed only when the system is finally asked to perform under its full set of constraints.

Contract development and manufacturing organizations (CDMOs) drive development stages where integration ceases to be theoretical and becomes operational. Fill-finish execution, component sourcing across multiple lots, process variability at scale, and disciplined test-method application converge in a way that rarely occurs in sponsor laboratories. This convergence makes CDMOs the first environment in which assumptions about a coupled drug–device system are exercised systematically rather than episodically.

In this setting, integration testing is not a discrete milestone but a lived reality. Formulation properties meet real syringes, real plungers, and real injectors under controlled yet variable manufacturing conditions. Small shifts that were invisible or tolerated earlier (e.g., lot-to-lot differences in components, subtle changes in friction, minor viscosity variation) become observable because they are no longer buffered by bespoke handling or informal adjustments. What survives this environment is what was genuinely designed to tolerate variation.

Test-method discipline amplifies this effect. Manufacturing contexts demand reproducible, validated methods that produce comparable data across time, operators, and sites. For needle-based injection systems, formal requirements and test methods exist precisely to evaluate whether devices can deliver discrete volumes reliably under defined conditions.⁸ When these methods are applied rigorously, they will expose whether prior development assumptions about force margins, injection time, or dose completion hold up outside of the original, controlled development settings.

This is why incompatibilities so often appear to “originate” at CDMOs, even though they typically originate long before the earliest tech transfer. CDMOs simply reveal these preexisting incompatibilities them because they operate where coupling is unavoidable. Integration that was previously implicit becomes explicit; margins that were assumed become measurable. The result is not new fragility, but the first clear view of whether the system was ever robust to begin with.

Seen in this light, CDMOs are not late-stage troubleshooters of device problems. Instead, they are the proving ground where combination products demonstrate whether they were designed as coupled systems capable of surviving real manufacturing conditions or merely assembled from parts that happened to work together under limited circumstances.

The idea that device qualification can be deferred until late in development is one of the most persistent and risky assumptions in combination product programs. It rests on the belief that device performance can be validated independently, once the drug product is finalized and the remaining uncertainties have been resolved. Regulatory design-control principles challenge this assumption directly by tying validation not to internal specifications alone, but to user needs and intended use, and by requiring that this validation be complete before commercial distribution.²

For combination products, intended use is inseparable from how the drug is delivered. Injection time, force requirements, dose completion, and usability are not abstract device attributes; they define whether the therapy can be administered as intended. When device qualification is postponed, these attributes are evaluated only after formulation, container, and manufacturing decisions have already constrained the system. Validation then becomes a test of whether late-stage reality conforms to earlier assumptions, rather than an opportunity to shape the system toward a viable outcome.

Design-control spillover amplifies this risk. Design controls apply not only to the device itself, but also to associated manufacturing processes and to changes made over time.³ This means that formulation updates, process adjustments, and component substitutions are not isolated events. Each change has the potential to alter the behavior of the coupled system and, by extension, the validity of prior design and validation conclusions. When such changes occur late (e.g., during scale-up, validation, or technology transfer), the cost of reconciling them with existing design controls can be substantial.

CDMOs operate squarely within the danger zone. They are responsible for executing manufacturing processes under design-control expectations while accommodating real-world variability and change. When integration has been deferred, CDMOs inherit systems that must simultaneously satisfy locked design requirements and evolving manufacturing realities. The result is a narrow path forward: either demonstrate that prior assumptions still hold under new conditions or confront the need for redesign at a point when flexibility is minimal.

The “we can qualify later” mindset treats design validation as a downstream gate. In practice, it functions as an upstream constraint that asserts itself late if it has been ignored early. Design controls do not merely document what was built; they codify what must continue to be true for the product to be acceptable. When integration is deferred, that truth is discovered only after the opportunity to change it easily has passed.

Reframing success in combination product development requires moving beyond milestone-driven validation toward a more demanding criterion: whether the system can sustain performance under realistic variation. Success is not demonstrated by the independent qualification of a formulation, a container, and a device, but by the ability of the integrated system to deliver reliably as those elements vary within expected manufacturing and life cycle bounds.

Under this lens, viscosity, force, and time are not late-stage tuning parameters. They define the design envelope within which the system must operate. Likewise, plunger-force behavior is not an isolated component attribute but a contributor to system margin that must be understood in the context of formulation rheology, injection profile, and device energy limits. When these factors are treated as upstream constraints rather than downstream checks, integration becomes a design activity rather than a validation exercise.

A credible program is one in which coupled-system behavior is characterized early enough that tradeoffs are explicit and margins are intentional. In such programs, late-stage surprises are rare not because testing was exhaustive, but because the system was never expected to behave modularly in the first place.

Combination products fail late when teams assume that coupling can be deferred. Autoinjectors and PFS do not break because devices malfunction in isolation but because the physical and functional links between formulation, container, and device eventually assert themselves. Physics enforces these links regardless of organizational boundaries, development timelines, or qualification strategies.

CDMOs sit at the point where these assumptions meet manufacturing reality. They do not introduce incompatibility; they expose whether a system was ever designed to withstand the variation inherent in real production. When integration has been treated as optional or postponed, that exposure arrives as constraint. When coupling has been acknowledged and designed for, it arrives as confirmation.

The distinction between these outcomes is not one of effort or intent, but of framing. Combination products succeed when they are developed as coupled systems from the outset, with performance defined by endurance rather than by elegance.

1. Current Good Manufacturing Practice Requirements for Combination Products: Guidance for Industry and FDA Staff. U.S. Food and Drug Administration. Jan. 2017.

2. “Design Controls” U.S. Food and Drug Administration. 28 Mar. 2023.

3. Design Control Guidance for Medical Device Manufacturers: Guidance for Industry. U.S. Food and Drug Administration. Mar. 1997.

4. Thueer, Thomas, Lena Birkhaeuer, and Declan Reilly. “Development of an advanced injection time model for an autoinjector.” Med Devices (Auckl). 11: 215–224 (2018).

5. Zhong, Xiaoxu, Ilias Bilionis, and Arezoo M Ardekani. “A framework to optimize spring-drive autoinjectors.” Int. J. Pharm. 617: 121588 (2022).

6. Berteau, Cecile, et al. “Evaluation of the impact of viscosity, injection volume, and injection flow rate on subcutaneous injection tolerance.” Med Devices (Auckl). 8: 473–484 (2015).

7. Jadhav, Santosh and Deepesh Bhatt. “Essential Performance-Requirement Assessment of Prefilled Syringes: Break-Loose and Gliding-Force Measurement and Analysis.” Bioprocess International. 15 Apr. 2025.

8. ISO 11608-1 2022: Needle-based injection systems for medical use — Requirements and test methods. International Organization for Standardization. 2022.