The question almost always arrives too late: the layout is frozen, the budget approved, and someone asks whether the still can be dropped in favour of a membrane train. Or the reverse: the project starts out membrane-based and, during due diligence, an auditor asks which control strategy demonstrates equivalence. Either way the discussion drifts in the same wrong direction — which of the two technologies is "better". That is not the question. The pharmacopoeia does not favour one route over the other: it defines an outcome and leaves the site to demonstrate that the chosen process achieves it reproducibly.
What follows compares the two routes against the same yardstick, separating three planes that are routinely conflated in practice: what the pharmacopoeia permits, what the authority expects as a control strategy, and what is simply a site engineering choice. Confusing them is why so many technology evaluations end up as ideological arguments rather than as a defensible dossier.
Whoever decides is not choosing a machine. They are choosing, in one move, a microbiological risk profile, a sanitisation model, a set of utilities, a redundancy level, a multi-year cost structure and a monitoring workload that will follow the system for its whole service life. The decision is also asymmetric in time — a sizing error can be corrected, a technology-route error touches layout, utilities, qualification and, in the EU, an interaction with the supervisory authority — which is why it belongs before the URS is frozen, as a chapter of source water and pretreatment design.
What the pharmacopoeia permits: the two texts that matter
Ph. Eur. monograph 0169 Water for injections was revised by adoption at the 154th session of the European Pharmacopoeia Commission (March 2016) and is effective from 1 April 2017: it allows WFI to be produced by a purification process equivalent to distillation. This was a structural change, not a clarification — before that date, in Europe, the membrane route was not an admissible route for WFI. The same revision made monograph 1927 Water, highly purified redundant; it was suppressed (160th session, March 2018) with effect from 1 April 2019.
In the United States, the USP Water for Injection monograph reads, verbatim: "Water for Injection is water purified by distillation or a purification process that is equivalent or superior to distillation in the removal of chemicals and microorganisms." Two elements matter. Equivalence is referred explicitly to the removal of chemicals and microorganisms, not to water quality in general. And the monograph controls conductivity (<645>) and TOC (<643>), while USP states verbatim that "Because of the various uses of these waters, microbial requirements are not included in these monographs": the microbiological burden does not sit in the monograph, it sits entirely in the site's control strategy.
On the GMP side, Annex 1 of EudraLex Volume 4 (in operation since 25 August 2023) states at clause 6.10 that WFI be produced by distillation or by a purification process equivalent to distillation, from water meeting specifications defined during qualification, and be stored and distributed so as to minimise microbial growth (the text cites constant circulation above 70 °C as an example). Clause 6.15 adds that WFI systems must include continuous monitoring such as TOC and conductivity. Neither clause distinguishes between generation routes: both apply identically.
Regional differences: where the symmetry breaks
On the compendial plane both routes are admissible in Europe and in the United States, and WHO has devoted a dedicated document to the subject — WHO TRS 1025 Annex 3 (2020) Production of water for injection by means other than distillation — which requires validated thermal and/or chemical sanitisation at specified intervals, the intervals being set by the manufacturer. The regional difference therefore does not lie in admissibility but in the administrative path and in supervisory expectations. Guideline EMA/CHMP/CVMP/QWP/496873/2018 (in force since 1 February 2021) recognises the Ph. Eur. 0169 revision at section 4.2 and requires prior notification to the supervisory GMP authority before introducing RO. It should be described for what the text says: a prior notification, not a formal authorisation, not a licence, not an approval that is "obtained". It remains, however, a calendar event that must enter the project plan, because it brings the system to the authority's attention before it is in operation.
A site supplying markets outside the EU, the US and the WHO framework must verify the applicable pharmacopoeia before freezing the choice: the symmetry described above does not extend automatically, and that verification belongs in the design rationale rather than being assumed.
What the authority expects: the Q&A on non-distillation routes
Document EMA/INS/GMP/443117/2017 Production of WFI by non-distillation methods – reverse osmosis, biofilms and control strategies (in force since 1 August 2017) translates compendial admissibility into operational expectations. Its key points should be read as a description of risk, not as a sanction against the technology.
- RO typically operates at ambient temperature, which is an ideal environment for biofilm formation. That is the central observation: the risk is not the membrane as such, it is the thermal regime in which it works.
- It expects extensive testing, with daily testing of all critical points in the initial phase and data collected over roughly one year to capture seasonal variation.
- It mentions destructive testing of the membrane to confirm the absence of biofilm.
- It states verbatim that "Increasing of such limits is not good practice and may mask a failing system": raising limits once a system begins to drift is not an acceptable response.
- It mentions thermal treatment above 75 °C, leaving contact times to the manufacturer's validation.
- As a minimum, it calls for an annual assessment of monitoring effectiveness.
Seasonality is not an isolated theme: Annex 1 clause 6.8 requires water systems to be qualified and validated taking seasonal variation into account. The difference is that on a membrane route the requirement is more demanding to satisfy, because source water variability propagates more directly to the finished water.
Three architectures, described against the same yardstick
Membrane-based generation
The typical architecture couples single- or double-pass reverse osmosis with a polishing stage — EDI, ultrafiltration or nanofiltration — along the combinations the EMA guideline recalls at section 4.2. Ion removal rests on the RO/EDI combination; retention of endotoxins and microbial residues typically rests on the final membrane stage. The process runs at ambient temperature unless membranes and modules are selected for hot water sanitisation: that is a specification choice, not a given property.
The design consequences are internally consistent: microbiological control rests on a sanitisation programme on a predetermined schedule (Annex 1, 6.12) rather than on operating temperature; integrity of the final stage becomes a critical attribute to be verified, not assumed; source water quality and pretreatment stability enter the product control chain directly. ISO 22519:2023 (2nd ed.) Membrane-based generation of water for injection (WFI) is the published technical reference dedicated to this route — issued, however, under ISO/TC 282 (Water reuse) rather than a pharmaceutical committee, with the 2019 first edition withdrawn. The ISPE Good Practice Guide Membrane-Based WFI Systems (May 2022) covers the same architecture but remains industry guidance.
Multiple-effect distillation
Multiple-effect distillation arranges a cascade of evaporation and condensation stages fed by plant steam: latent heat released in one effect feeds the next, and the separation between liquid and vapour phase provides a physical barrier against endotoxins and non-volatile contaminants. The distillate is produced hot, a condition that in itself disfavours downstream microbial proliferation, and many configurations allow pure steam to be co-produced.
The design consequences are equally consistent: the system depends on the availability, pressure and — an attribute often overlooked in the URS — quality of plant steam, including boiler water treatment and additive management; it requires cooling capacity; and it shifts part of the control burden from the sanitisation programme to the thermal regime. It does not remove it: Annex 1 clause 6.12 still requires sterilisation, disinfection or regeneration on a predetermined schedule and as a remedial action, whatever the generation route.
Vapour compression distillation
Vapour compression distillation uses a mechanical compressor to raise the pressure and temperature of the vapour generated, reusing its latent heat within the same exchange body. The most significant architectural consequence is the shift of demand from plant steam to electrical energy: a site without an adequate steam network finds here an alternative that does not require the same infrastructure. The machine also tolerates reduced-flow operation better than a multiple-effect cascade, which matters in a plant that never runs at nominal load.
The consequences to put on the table are equally concrete: the compressor is a highly critical rotating item, with its own maintenance plan, spares chain and failure profile; noise and vibration are an installation topic; and unit availability depends on a single component in a way a multiple-effect cascade does not. These points do not make the technology inferior — they make it different, and they must be compared against the corresponding points of the other routes, not against their absence.
The real discriminator: source water
No route is indifferent to feed water, but the sensitivity differs in nature. A membrane route propagates source water variability — organic and particulate load, seasonal temperature, residual disinfectant from the municipal network, fouling index — directly onto the performance and service life of the membrane stages, and hence onto the stability of the product. A distillation route tolerates microbiological and organic variability in the feed better, but is sensitive to scaling, carry-over and feed water chemistry, and still sits downstream of a pretreatment that must be designed with the same rigour.
The operational point is that source water characterisation cannot be a nameplate figure. It requires a dataset covering at least one full seasonal cycle, including abnormal events on the supply network and not only average conditions, related to the specifications defined during qualification that Annex 1 clause 6.10 refers to. Without that dataset the technology evaluation is not yet possible: any comparison is an exercise on assumptions. This holds for both routes, but an error here is paid for more dearly on the membrane route.
Microbiological and endotoxin control
On the membrane route, microbiological control is active: it depends on faithful execution of the sanitisation programme, on the integrity of the final stage, on the absence of stagnant zones in the generation train and on early detection of drift. The Q&A observation on biofilm at ambient temperature is not theoretical: a mature biofilm does not show up in online parameters before it has consolidated, which is why the document points to destructive testing of the membrane as a means of confirmation. Endotoxin retention depends on a functioning, verified unit, not on a physical principle that operates regardless.
On the distillation route, microbiological control is largely intrinsic to the process and its thermal regime, but it is neither free nor absolute. Droplet carry-over — the mechanism through which endotoxins can cross the barrier — must be controlled, as must the condenser, the sampling arrangements, the points where hot distillate meets colder surfaces, and cold start-up, which is typically the widest risk window. The thermal advantage does not remove the need for a control programme: it relocates it.
In both cases the continuous TOC and conductivity monitoring required by Annex 1 clause 6.15 must be designed as part of the system rather than added afterwards; alert levels must be built on initial qualification data and periodically reviewed (6.13), with excursions documented and investigated, distinguishing an isolated event from an adverse trend (6.14). How these controls are set up in practice is covered in the article on microbiological and endotoxin control of PW and WFI.
Sanitisation, redundancy, energy, water
Sanitisation. Annex 1 clause 6.12 requires sanitisation on a predetermined schedule for any water system; WHO TRS 1025 Annex 3 requires, for non-distillation routes, validated thermal and/or chemical sanitisation at intervals specified by the manufacturer. On the membrane route, frequency, method and material compatibility become URS parameters: a membrane that cannot be hot-water sanitised commits the site to a chemical programme, with everything that follows in rinsing, residue verification and downtime. On the distillation route, sanitisation concerns mainly the cold sections, the pretreatment and the downstream distribution. Different situations, but neither is exempt. Building and qualifying the cycles is covered in the article on qualification of sanitisation cycles.
Redundancy. The right question is not "how many units" but "which event do we want to absorb without stopping production". A membrane route is modular by nature and lends itself to partial parallel configurations; a distillation route concentrates capacity in one machine, but recovery time after a microbiological event is typically easier to govern when the process runs hot. Different properties, not one superior.
Energy and water. The consumption profiles are structurally different — plant steam for multiple effect, electrical energy for vapour compression, electrical energy and feed water for the membrane route, plus the cooling required by the distillation routes and any heat recovery. No numerical comparison is meaningful outside the specific case: it depends on capacity, draw-off profile, temperature required at the point of use, source water quality and local utility costs. The only defensible way to handle the point is to ask suppliers for mass and energy balances calculated on the site's real data, with assumptions stated and verifiable, and to compare them under the same boundary conditions. A catalogue figure without its assumptions is not a comparable figure.
Lifecycle and TCO
Purchase cost is the least informative line in the comparison. Over a multi-year horizon consistent with the expected service life, what must be reconstructed is: utility consumption at the real load profile, not the nominal one; consumables and planned spares, with attention to periodic replacement of membrane stages and to maintenance of rotating equipment; planned downtime for sanitisation and maintenance, valued as lost availability; the analytical monitoring workload, which in the initial phase of a membrane route is heavier because of the Q&A expectations; qualification and requalification; obsolescence of components and of the automation layer; and the technical time of regulatory interaction, including preparation of the prior notification.
Two recurring errors: calculating TCO at nominal load instead of at the real draw-off profile, and ignoring that the initial monitoring workload is not permanent but is not negligible either. The model must be built on site data, with assumptions declared and challenged, and reviewed over time as part of periodic review and retrofit assessment.
A vendor-neutral evaluation matrix
The matrix is delivered with the weighting column empty: weights depend on the site context and must be assigned and justified by the team before scores are collected, not after the results are visible. Weight the criteria, then evaluate the options; never the other way round.
| Criterion | Evidence required to assign a score | Weight | Membrane route | Distillation (MED / VC) |
|---|---|---|---|---|
| Compendial admissibility in the markets served | List of markets and applicable monograph for each, verified and documented | |||
| Interaction with the authority | Plan and timing of prior notification where applicable, with responsibilities assigned | |||
| Source water suitability | Dataset over at least one seasonal cycle, abnormal events included | |||
| Robustness of microbiological control | Documented control strategy; early detection mechanisms; response to drift | |||
| Endotoxin control | Barrier mechanism and how its effectiveness is verified over time | |||
| Sanitisation programme | Method, proposed frequency, material compatibility, impact on availability | |||
| Continuous monitoring and data | Online TOC/conductivity architecture, alarms, audit trail, trending | |||
| Utilities required and available | Mass and energy balances on real site data, assumptions stated | |||
| Redundancy and recovery time | Defined failure scenarios and the system's response to each | |||
| Maintainability, spares, competence | Maintenance plan, component criticality, supply chain, on-site competence gap | |||
| Qualification and analytical burden | Extent of IQ/OQ/PQ, duration of seasonal data collection, initial vs routine sampling plan | |||
| TCO over a defined horizon | Cost model at the real load profile, assumptions declared |
It must be completed under the same boundary conditions for every option, with the source of each score traceable. A cell filled in without evidence is an empty cell in disguise.
Worked example: Site Delta
Site Delta is a fictitious site, used here only as an example. It manufactures sterile forms for the EU and US markets and must replace a WFI generator at end of life. The team opens with an upstream question: which points of use genuinely require WFI and which are covered by PW. The check reduces the demand and changes the starting sizing before technology is even discussed.
Source water characterisation shows marked seasonal variation in organic load and high summer feed temperatures. The team does not conclude that the membrane route is excluded: it concludes that, if selected, it requires a more robust pretreatment, hot-water-sanitisable membranes and an initial monitoring phase consistent with the Q&A expectations, on a data collection plan covering the full seasonal cycle. In parallel it verifies that the plant steam network, plant steam quality and cooling capacity are sufficient for a multiple-effect unit, and that the installed electrical power is adequate for vapour compression.
All three options enter the matrix at the same level of detail, and the weights are assigned and minuted before scores are collected. The final rationale, whatever it is, is documented with the evidence that supported it: it is that document, not the technology, that will be asked for during an inspection.
Common mistakes and red flags
- Treating "equivalent to distillation" as a label acquired with the technology rather than an outcome to be demonstrated with site data.
- Presenting the prior notification as an authorisation obtained, or conversely omitting it from the project plan because "it is not an approval".
- Comparing options using catalogue figures stripped of the assumptions that generated them.
- Evaluating on a source water characterisation limited to average conditions and a single season.
- Treating distillation as exempt from a sanitisation programme, or the membrane route as inherently non-compliant.
- Raising alert limits once the data begin to drift — a practice the Q&A addresses explicitly.
- Sizing redundancy by analogy with other sites instead of on your own failure scenarios, or deferring the monitoring architecture until the layout is frozen.
Red flags: nobody can say what sanitisation frequency is proposed and on what basis; the TCO exists in a single version with no sensitivity analysis; the choice has already been made and the matrix serves to justify it; the team includes neither microbiology nor the function that will operate the system.
The sequence that holds up in inspection
Define the real WFI demand by point of use and time profile; characterise the source water over a full seasonal cycle; list the markets served and the applicable monographs; build the matrix and minute the weights; ask suppliers for balances and performance calculated on site data, with explicit assumptions; assess the control strategy before cost, because a weak control strategy makes any economic advantage irrelevant; document the rationale with the evidence; put any prior notification into the plan with an owner and a date. The wider picture of generation technologies is developed in the article on pharmaceutical water generation technologies; the full set of topics sits in the Pharmaceutical Water & WFI Systems hub.
If you work on decisions of this kind, The Pragmatic GMP collects technical and regulatory analysis on GMP systems in the same register.
Key takeaways
- Ph. Eur. 0169, in the revision effective from 1 April 2017, permits WFI production by a purification process equivalent to distillation; the USP WFI monograph permits distillation or a process equivalent or superior in the removal of chemicals and microorganisms.
- Annex 1 clause 6.10 recognises both routes without treating them differently; clause 6.15 requires continuous monitoring such as TOC and conductivity for WFI systems, whichever route is used.
- EMA guideline 496873/2018, section 4.2, requires prior notification to the supervisory GMP authority before introducing RO: it is a notification, not an authorisation, and it must be managed as a project activity.
- EMA Q&A 443117/2017 sets the operational expectations for non-distillation routes: extensive testing, daily testing of critical points in the initial phase, data over roughly one year, destructive testing of the membrane, and an explicit position against raising limits.
- Source water characterisation over a full seasonal cycle is a prerequisite of the evaluation, not an output of it; no energy or TCO comparison is meaningful outside the site's real data.
- No route is universally superior: the control mechanism, utility profile, redundancy structure and cost distribution change. The decision is a site decision and must be documented as such, with matrix weights assigned before scores.
Regulatory and technical references
- Ph. Eur. 0169 Water for injections — revision adopted at the 154th EPC session (March 2016), effective 1 April 2017; monograph 1927 Water, highly purified suppressed (160th session, March 2018; effective 1 April 2019).
- USP Water for Injection monograph; chapters <643> and <645>; informational chapter <1231>.
- EudraLex Volume 4, Annex 1 (C(2022) 5938 final), in operation since 25 August 2023 — clauses 6.7-6.15. health.ec.europa.eu
- EMA/CHMP/CVMP/QWP/496873/2018, in force since 1 February 2021 — section 4.2 (ema.europa.eu); EMA/INS/GMP/443117/2017 Q&A Production of WFI by non-distillation methods, in force since 1 August 2017.
- WHO TRS 1025, Annex 3 (2020) Production of WFI by means other than distillation; WHO TRS 1033, Annex 3 (2021) GMP: water for pharmaceutical use.
- ISO 22519:2023 (2nd ed.) Membrane-based generation of water for injection (WFI) — ISO/TC 282. iso.org
- ISPE GPG Membrane-Based WFI Systems (May 2022) and Baseline Guide Vol. 4 Water and Steam Systems, 3rd ed. (2019) — industry guidance, not regulatory texts.
- PIC/S PI 009-4 Inspection of Utilities, rev. 4, in force since 1 January 2021 (picscheme.org); ICH Q9(R1) (Step 4, 18 January 2023).
