The Regulatory Landscape: ICH Q13, FDA, and PMDA Alignment

The single most consequential development for continuous manufacturing validation was the publication of ICH Q13 at Step 4 in November 2022. The guideline formally recognizes CM as a legitimate manufacturing paradigm and describes the scientific and regulatory considerations for its development, implementation, operation, and lifecycle management. It applies to CM of drug substances and drug products for both chemical entities and therapeutic proteins, whether used for new products or for conversion of existing batch processes.2

Q13 does something subtle and important. Rather than mandate a single validation strategy, it describes three CM implementation approaches and integrates the concepts of Q8 (Pharmaceutical Development), Q9 (Quality Risk Management), and Q10 (Pharmaceutical Quality System) into a coherent framework. The guideline explicitly facilitates continuous process verification as an alternative to traditional batch-based validation, which the ICH parties acknowledge can reduce validation complexity and clerical burden when properly executed.3

The FDA operationalizes these principles through its final guidance on Quality Considerations for Continuous Manufacturing, issued in March 2023. That document addresses CM in NDAs, ANDAs, DMFs, BLAs, and non-application OTC products, and it clarifies that a single continuous run can, under the right conditions, provide both process and product validation evidence.4 The 2011 FDA Process Validation Guidance still supplies the three-stage lifecycle framework (Process Design, Process Qualification, Continued Process Verification) that anchors every CM validation strategy in the United States.5

USD 12.09B Projected global CM market by 2035, up from USD 3.28B in 20251
70% Equipment footprint reduction reported for well-designed CM lines1
3-5x Volumetric productivity increase versus batch, driving cost-of-goods reductions1

The EMA’s position tracks closely with FDA and ICH but adds two useful nuances. First, its Biologics Working Party is developing supplementary guidance on process validation filings that will touch impurity clearance, hold times, reprocessing, and batch selection for continuous biologics.6 Second, the revised EU GMP Annex 15 on Qualification and Validation now expects lifecycle validation thinking to apply to active substance manufacturers, not just finished products, closing a gap that was awkward for continuous API operations.7

PMDA has been an active participant in ICH Q13 and has issued its own points-to-consider documents through the Innovative Manufacturing Technology Working Group. Practitioners preparing Japanese submissions should note that PMDA inspectors tend to focus on lifecycle thinking, integration of the Pharmaceutical Quality System with CM control strategies, and demonstrable data integrity in the historian layer.8 Superficial root cause investigations and weak Product Quality Review processes are common inspection findings and are more visible in continuous operations because the data volume is so much higher.

SD Perspective. Global regulatory convergence on CM is real, but it is not the same as regulatory uniformity. The safest strategy is to design a validation package that satisfies the strictest expectation in your submission portfolio, then subset it for other agencies. Trying to design three parallel validation packages for FDA, EMA, and PMDA is expensive, redundant, and often introduces internal inconsistencies that inspectors will find.

Continuous Process Verification vs Traditional PPQ

The most consequential validation choice for a CM line is how to handle Stage 2 of the FDA process validation lifecycle. In the traditional model, Stage 2 is a discrete campaign of three Process Performance Qualification (PPQ) batches at commercial scale, executed against a pre-approved protocol with tightly bounded acceptance criteria. This model was designed around batch processes where the concept of a “batch” is unambiguous and where a small number of runs can plausibly demonstrate reproducibility.

For continuous processes, the batch concept requires redefinition. ICH Q13 permits batch to be defined by time (a defined run duration), quantity (mass or units produced), or an equipment-based rationale. Once the definition is fixed, PPQ can still be executed but often takes the form of a smaller number of longer runs during which the process is intentionally challenged at edge-of-design-space conditions.2

Continuous Process Verification (CPV, sometimes called Continued Process Verification) is the alternative Q13 explicitly enables. Rather than segregating validation into a discrete PPQ campaign followed by ongoing monitoring, CPV treats validation as continuous. Every commercial run contributes evidence to the state-of-control demonstration. This is possible because CM lines instrument the process at a density traditional batch equipment does not, and because RTRT-grade PAT models make in-process measurements informative about final product quality.9

TRADITIONAL PPQ

Discrete Validation Campaign

Three qualification batches at commercial scale, tight acceptance criteria, offline release testing. Regulatory expectation for batch products; still permitted for CM but often inefficient given data density.

HYBRID PPQ + CPV

Anchored Runs with Continuous Verification

A small PPQ campaign (often 2-3 runs) establishes the state-of-control baseline; continuous verification takes over for lifecycle. Common for first-generation CM lines.

PURE CPV

Continuous Verification from Day One

No discrete PPQ campaign; validation is demonstrated through instrumented continuous operation, RTRT, and rigorous SPC. Enabled by ICH Q13 but requires mature PAT and control-strategy justification.

SINGLE-RUN QUALIFICATION

One Extended Run as Validation Evidence

Emerging approach where a single, long, well-characterized run provides both process and product validation. FDA’s 2023 CM guidance acknowledges this is possible; practical adoption is limited to greenfield sites with exceptional PAT maturity.

The tradeoffs are practical. A traditional PPQ campaign on a CM line is expensive because the runs are long, the material demand is high, and the process is running slowly relative to its capability during qualification. But it produces a document set that maps cleanly onto agency reviewer expectations. A pure CPV approach requires more up-front investment in PAT models, control system engineering, and statistical infrastructure, but the ongoing validation cost is dramatically lower and the response to process drift is faster.10

Statistical Process Control Models in a Continuous World

Statistical process control is the connective tissue between the validation lifecycle stages. In batch processes, SPC typically operates on inter-batch data: batch means, batch capability indices, batch-level trend charts. In continuous processes, the same principles apply but the data cadence is intra-run, often at frequencies of seconds or minutes. The volume difference is roughly three orders of magnitude, and the analytical approach has to change accordingly.

Four SPC methods dominate continuous manufacturing validation packages:

1

Shewhart Individuals and Moving Range Charts

The workhorse for CQA time series. Effective for detecting large step changes but insensitive to slow drift. Practitioners typically pair them with EWMA or CUSUM to catch small persistent shifts before they cross specification.

2

Multivariate SPC (Hotelling T-squared, PCA-based)

Essential when the process has strongly correlated CQAs, which continuous lines almost always do. Multivariate methods detect out-of-control signatures that univariate charts miss, especially when correlations shift without any single variable exceeding its individual limit.

3

Capability Indices Adjusted for Autocorrelation

Traditional Cpk and Ppk assume independent observations, which is almost never true for high-frequency in-line data. Autocorrelation-corrected capability metrics, or explicit time-series models with residual analysis, are the technically correct approach.

4

Common-Cause vs Long-Term Common-Cause Discrimination

A well-documented failure mode in biopharmaceutical CPV is misclassifying long-term common-cause variability as special cause. Robust models separate short-timescale noise from campaign-level drift and flag only the latter for investigation.

The BioPharm International and BioProcess International CPV literature is unusually candid about this. Continued Process Verification for biopharma consists of collecting parameter data, trending against statistical limits, and calculating capability at defined intervals or after every few batches. The FDA and EU GMP Annex 15 both require manufacturers to keep processes in a continual state of control, but they do not prescribe specific statistical methods. That freedom is helpful but places the burden of scientific justification firmly on the manufacturer.11

Common pitfall. Applying batch-era Cpk and Ppk to continuous data without correcting for autocorrelation produces capability indices that look great but do not reflect actual process performance. If your CPV report shows a Cpk of 4.0 on a continuous line, the number is almost certainly wrong. Fix the statistics before an inspector notices.

Real-Time Release Testing as a Validation Backbone

Real-Time Release Testing changes the validation conversation because it changes the evidence base. In classical release testing, product quality is verified by end-of-process laboratory testing on a representative sample. In RTRT, product quality is verified continuously during manufacturing through validated PAT models that predict CQAs from in-line measurements. The validation package must demonstrate that the PAT model is at least as reliable as the reference laboratory method it replaces.12

The FDA has significant regulatory experience with RTRT in NDA submissions, particularly for dissolution testing on solid oral dosage forms. A recent review of that experience noted that PAT-enabled RTRT for dissolution has been successfully implemented across multiple NDA approvals, but the validation packages must address analytical correlation, robustness across the design space, and ongoing model lifecycle monitoring.13

Three integration points matter for validation strategy:

NIR SPECTROSCOPY

Blend Uniformity and Content

Near-infrared spectroscopy is the workhorse for solid-dose CM lines. Well-validated NIR models replace offline assay and content uniformity testing at feed frames and post-compression. Model transfer between instruments remains a lifecycle challenge.

RAMAN SPECTROSCOPY

Polymorph and Chemical Composition

Raman is preferred where API polymorphism, cocrystals, or hydrate/anhydrate distinctions must be tracked. Increasingly deployed in flow chemistry for API synthesis and in bioreactor headspaces for perfusion monitoring.

FBRM AND PARTICLE VISION

Crystallization Endpoint Detection

Focused beam reflectance measurement and particle vision measurement are used for continuous crystallization endpoint validation. Critical for continuous API where polymorph and particle size distribution drive downstream processability.

PROCESS MODELS

Soft Sensors and Predictive Release

First-principles or hybrid models can act as soft sensors that predict CQAs from process states without a dedicated spectroscopic probe. Requires rigorous model qualification and versioning discipline.

PAT model lifecycle management deserves particular attention. A PAT model is not qualified once; it is qualified initially and then continuously re-verified against reference method data. Model drift, reference method drift, and equipment changes all require model updates, and each update needs its own change control and re-validation evidence. The best-run CM sites treat PAT model management as a formal engineering discipline with the same rigor they apply to analytical method transfer.14

Model-Informed Process Design and Digital Twin Coupling

The most sophisticated CM validation packages now include model-informed process design (MIPD) and digital twin coupling as first-class validation evidence. A digital twin, in this context, is a computational representation of the continuous line that receives real-time data from the physical process, updates its internal state to mirror actual operations, and generates predictions that guide manufacturing decisions.15

For validation, a digital twin can serve three distinct roles:

  • Design space characterization. Hybrid flowsheet models combining first-principles equations with data-driven components can explore process behavior across regions that would be uneconomic to characterize with physical experiments alone. This is particularly valuable for continuous direct compression, where residence time distribution behavior drives batch definition and traceability.
  • Real-time state estimation. A digital twin coupled to the historian can estimate CQAs that are not directly measured, effectively acting as a soft sensor with a physical basis. Some regulatory submissions now cite digital twin outputs as supporting evidence for RTRT model qualification.
  • Deviation investigation. When a CM line experiences an excursion, a digital twin can be used to reproduce the event, test hypotheses about root cause, and simulate corrective actions before applying them to the physical line. This dramatically shortens investigation cycle time and improves the quality of change controls.

What “model-informed” really means. The phrase is used loosely across industry. In a validation context, model-informed means the model is qualified, versioned, tied to a formal control strategy, and integrated into the Pharmaceutical Quality System. A spreadsheet that a scientist runs on their laptop is not a validated model; it is a working file. The distinction matters when an inspector asks.

Residence time distribution (RTD) modeling deserves separate mention because it is uniquely important in CM validation. RTD models describe how long material resides inside a unit operation or the entire line, and they enable both batch definition and raw material lot traceability. Convolutional RTD models can trace a specific raw material lot from receipt through to individual tablets, satisfying the regulatory requirement to demonstrate material traceability that is trivial in batch processes but non-trivial in continuous ones.16

Continuous Biologics: Perfusion, Integrated Downstream, and Validation Nuances

Small-molecule continuous manufacturing has clear precedent and a maturing toolkit. Continuous biologics is a different story. The regulatory framework is the same at the highest level (ICH Q13 applies to both drug substances and therapeutic proteins), but the operational realities of perfusion bioreactors, integrated downstream unit operations, and continuous capture chromatography introduce validation challenges that solid-dose CM never had to solve.

Sanofi’s integrated continuous biomanufacturing platform, using perfusion with a chemically defined medium and a purpose-built cell line, reported roughly a 100-fold improvement in productivity over the legacy fed-batch process and up to 80 percent reduction in process cycle time. Amgen’s Singapore facility demonstrated a 75 percent footprint reduction and 70 percent lower water and energy usage compared with traditional plants. Both cases show what is possible; both also illustrate that the validation package is not a copy-paste from small-molecule CM.22

Three validation challenges are distinctive to continuous biologics:

IMPURITY CLEARANCE

Steady-State vs Transient Behavior

Batch chromatography validates impurity clearance per cycle. Continuous capture validates clearance in steady state. The transient at start-up and shut-down is where residual impurities and host cell protein excursions are most likely, and the validation package must characterize both regimes.

RESIN LIFETIME

Cycles per Batch vs Cycles per Campaign

Continuous capture resins may see hundreds of cycles per campaign rather than a handful per batch. Resin lifetime studies must be re-scoped to reflect the total exposure and interrogate carryover risk in ways traditional resin validation did not.

HOLD TIMES

Redefined for Flowing Material

Traditional hold-time studies assume discrete volumes held under defined conditions. Continuous processes hold material in transit at varying residence times, requiring RTD-based hold time characterization and biological stability data across the full residence time distribution.

BATCH DEFINITION

Bioburden and Sterility Boundaries

The batch definition must satisfy not only regulatory identification but also bioburden risk management. Long continuous campaigns raise the possibility that a single bioburden excursion could span multiple defined batches, and the strategy must address that risk explicitly.

The EMA Biologics Working Party has been unusually explicit that its evolving process validation filing guidance will address impurity clearance, column and membrane lifetime, hold times, reprocessing, pooling of intermediates, and batch selection specifically in the biologics context. Manufacturers preparing continuous biologics submissions should expect these topics to be scrutinized more closely than they would be for a batch biologics filing.6

The perfusion validation shortcut that isn’t. A frequent misstep is to treat a perfusion bioreactor as though it were a very long fed-batch. The steady-state productivity numbers look great, but the validation package has to explain what happens during start-up, medium switches, cell bleed adjustments, and controlled shut-down. Each transient is its own validation question, and the CPV data model needs to distinguish steady-state operation from transient regimes so that SPC is not confused by expected variability.

For biologics leaders evaluating continuous options, the practical guidance is to start with the unit operation where the productivity gain is largest and the validation risk is best-understood (typically perfusion upstream with fed-batch downstream) and add downstream continuity in later phases. Fully integrated end-to-end continuous biologics is achievable but requires a level of investment and regulatory sophistication that few organizations can support on a first program.

Comparative Case Studies: Vertex, Janssen, and Eli Lilly

Three published implementations illustrate the range of validation approaches available. All three are small-molecule solid oral dosage products, but each represents a different starting point and a different balance between traditional and modern validation.

Vertex Pharmaceuticals: Greenfield CM for Orkambi

Vertex is the pathfinder for continuous drug product manufacturing. Orkambi (lumacaftor/ivacaftor) was, in 2015, the first product approved with a fully continuous drug product manufacturing line, and Vertex has since added Symdeko/Symkevi and Trikafta to its CM portfolio. All three are cystic fibrosis therapies with limited patient populations, which made continuous manufacturing economically compelling despite the up-front investment.17

Vertex’s approach combined a partnership with GEA on the ConsiGma continuous direct compression platform with a robust internal PAT and control strategy program. From a validation standpoint, Vertex leaned heavily on model-informed development: a formulation could be finalized early using minimal API, and as more API became available a data-rich design space was constructed for the process. The validation package included RTRT and a formal approach to post-approval PAT model maintenance.18

Janssen: First Batch-to-Continuous Conversion at Commercial Scale

Janssen’s April 2016 FDA approval to switch PREZISTA (darunavir) 600 mg tablets from batch to continuous manufacturing at its Gurabo, Puerto Rico facility remains the industry reference case for a conversion approval. The line was developed through a five-year partnership with Rutgers University and the University of Puerto Rico that integrated weighing, milling, blending, compression, and coating into a single continuous line.19

The validation strategy for Prezista is instructive because it had to demonstrate equivalence to the incumbent batch process. That equivalence framework shaped both the analytical strategy (paired testing during transition) and the regulatory approach (a supplemental filing rather than a new application). The EMA subsequently approved continuous manufacturing with real-time release testing for the same product, making it a rare example of trans-Atlantic harmonization on a CM conversion.20

Eli Lilly: CM in the Commercial Portfolio

Eli Lilly’s Verzenio (abemaciclib) is the third widely cited example. Verzenio was Lilly’s first solid oral dosage form produced by continuous manufacturing and is now used in early breast cancer, where volumes have grown substantially since the initial 2017 approval. Lilly’s validation strategy, discussed in industry conferences and DCAT commentary, emphasized continuous process verification and integration of the CM line into the broader commercial network rather than treating it as a research showcase.21

The comparison across the three programs is summarized below.

Attribute Vertex (Orkambi) Janssen (Prezista) Eli Lilly (Verzenio)
Starting point Greenfield CM at NDA Batch-to-continuous conversion Greenfield CM at NDA
Approval year 2015 (FDA) 2016 (FDA), 2017 (EMA) 2017 (FDA)
Primary validation approach Model-informed design + RTRT + PPQ Equivalence to batch + RTRT + PPQ CPV-forward + RTRT
PAT emphasis NIR-heavy, formal model lifecycle NIR + traceability via RTD Integrated PAT + digital enablement
Batch definition Time-based, with defined run duration Quantity-based to match batch analog Time-based, with campaign structure
Post-approval strategy Established PAT model maintenance program Cross-Atlantic CM approvals for related products Integration into commercial supply network

What is striking about the comparison is not the differences but the convergence. All three programs invested heavily in PAT, all three formally addressed batch definition and material traceability, and all three built RTRT into the release strategy. The differences are matters of emphasis and starting condition, not fundamental disagreement about what a good CM validation package looks like.

A Validation Strategy Decision Tree

The choice among validation approaches is not arbitrary but it is genuinely product- and site-specific. The decision tree below is a working synthesis we use with clients to structure the first strategy conversation.

1

Is this a greenfield CM line or a conversion?

Greenfield lines have more design freedom and can adopt CPV-forward strategies more easily. Conversions must demonstrate equivalence to the incumbent process, which drives paired testing during transition and often a hybrid PPQ + CPV approach.

2

Small molecule or biologic?

Small molecule CM is well-supported by ICH Q13, FDA guidance, and established PAT toolkits. Continuous biologics (perfusion-based upstream, integrated downstream) require heavier reliance on process models and are still on the regulatory learning curve. Plan for more regulator interaction.

3

What is the PAT maturity of the site?

RTRT and CPV both require validated in-line measurements. If the site does not have an established PAT team and model lifecycle program, plan for a hybrid approach and use the first commercial years to mature the PAT foundation before removing offline release testing.

4

Which agencies are in the filing portfolio?

FDA and EMA are well-aligned via ICH Q13; PMDA is aligned in principle but places specific weight on PQS integration and data integrity. Design the validation package for the strictest agency’s expectations and subset it for others, rather than building parallel packages.

5

How mature is the digital twin and model layer?

If a validated digital twin exists, it can support state estimation, deviation investigation, and design space characterization. If it does not, do not manufacture one for the filing; overclaimed models are worse than no models.

6

Is the batch definition defensible?

Every CM validation package rests on a defined batch. Confirm the definition is documented, tied to the control strategy, aligned with RTD-based traceability, and consistent with the regulatory submission wording before finalizing validation protocols.

What good looks like. The best CM validation packages we see are legible: a reviewer can trace from control strategy to CQAs to PAT models to SPC methods to batch definition without having to interpret ambiguous cross-references. Legibility is not a nice-to-have; it is what allows the package to survive personnel turnover, agency reviewer changes, and lifecycle updates.

Where Programs Get Into Trouble

Three failure modes recur in our client work:

  • PAT models that outrun their qualification. A model built during development is quietly updated for commercial use without the qualification package following behind. When an inspector asks for the current qualification report, the paper trail does not exist. The fix is a versioned PAT model registry with change control tied to the Pharmaceutical Quality System.
  • Batch definitions that drift from control strategy. The filing states one batch definition; the actual line uses a slightly different one because of a change to the control system. RTD-based traceability calculations become inconsistent. The fix is a periodic reconciliation between the regulatory submission, the batch record, the control system configuration, and the CPV package.
  • Statistical methods that were correct in Stage 2 but not in Stage 3. Autocorrelation-corrected capability was applied for PPQ, then quietly replaced with standard Cpk in the routine CPV package because the automated tools calculated it that way. The fix is a documented statistical methods standard for the site that is applied consistently across the validation lifecycle.

Lifecycle Governance: Post-Approval Change and Continuous Improvement

An underappreciated dimension of CM validation is how the operating model changes the post-approval change management story. A traditional batch process, once validated, tends to stay fixed because every meaningful change triggers a supplemental filing and a comparability exercise. A continuous process, if it has been established with a robust control strategy and mature PAT, has genuine room to improve over time within its established design space without a new filing being required for every adjustment.

This is one of the more meaningful, and least discussed, business cases for CM. The ICH Q12 lifecycle management framework and the FDA’s Established Conditions guidance both create explicit pathways for managing post-approval change through the Pharmaceutical Quality System rather than through supplemental filings, provided the manufacturer has done the up-front work of characterizing which conditions are established and which are flexible. CM operations are natural candidates for this treatment because the data density they generate makes the state-of-control demonstration straightforward.

SD Perspective. The most durable CM validation packages we see treat the initial approval as the start of a lifecycle, not the end of a project. That framing shapes governance choices: how the PAT model registry is maintained, how the digital twin is versioned, how CPV findings feed continuous improvement pipelines, and how the site handles the tension between “keep the process fixed for regulatory simplicity” and “improve the process because we can.” Sites that resolve that tension explicitly, with a documented change governance rubric, tend to age much better than sites that resolve it implicitly and rediscover the choices in the middle of an inspection.

Three governance disciplines matter more than any single technical decision:

  • Established Conditions clarity. The regulatory submission should be explicit about which operating parameters, models, and control settings are established conditions and which are flexible under the PQS. Ambiguity here is the source of most post-approval friction with agencies.
  • Change categorization discipline. Every proposed change should be routed through a documented rubric that maps change type to regulatory pathway (annual report, changes-being-effected, prior approval supplement, or PQS-only). The rubric should be reviewed annually against actual change history.
  • CPV-to-improvement pipeline. CPV data should be used not only to detect out-of-control conditions but also to inform continuous improvement. This requires an intentional handoff between the monitoring team and the process engineering team and a governance forum where improvement candidates are evaluated against the risk of triggering unnecessary regulatory work.

None of this is required by ICH Q13 or FDA guidance. All of it is what separates CM programs that keep delivering value over a decade from programs that ossify shortly after approval and become expensive to change.

Conclusion

Continuous manufacturing validation is now a solved problem in the sense that the regulatory framework exists, the toolkit is mature, and there are published implementations to learn from. It is not a solved problem in the sense that any single approach fits all products, sites, and agency portfolios. The comparative review above suggests three things: the ICH Q13 framework has genuinely opened space for approaches beyond traditional PPQ; the winning validation packages treat RTRT, CPV, and digital twin coupling as integrated rather than layered; and the differences between Vertex, Janssen, and Lilly are less about disagreement on principles than about starting conditions and product economics.

For senior leaders evaluating a first CM program or a batch conversion, the highest-leverage early decisions are about PAT model lifecycle, batch definition, and the statistical foundation of the CPV program. These decisions look technical but they determine whether the package is legible to inspectors five years later, when the original engineering team has moved on and the regulatory reviewer has never seen the site. Legibility beats sophistication when a lifecycle question arrives on a Friday afternoon.

Sakara Digital works with pharma and biotech organizations building continuous manufacturing validation strategies across small molecule and biologics portfolios. If you are evaluating a first CM program, a batch conversion, or a lifecycle upgrade to an existing continuous line and want an independent perspective on where to start, we are happy to have that conversation.