Articles

What stands between you and a sterility failure

A working list of the controls that keep sterile product sterile, sorted by class, with how each one fails and what FDA has said about it in warning letters issued during 2026.

36 min read

A sterility failure is almost never one control failing. It is a source, a route, and a missing barrier, with a detection control that was not looking in the right place at the right time.

This is the list of what is standing in the way, sorted by class. For each control: what it does, how it fails, and where FDA has written about that exact failure during 2026. Every quotation is from a public warning letter and is linked at the foot.

Each entry also ends with Vioela’s note: the part that is not written in the regulation, that people usually learn the slow way. Where a note gives a number, the source is listed at the end.

Some notes are followed by a Picture this: a short scene, invented for illustration, showing how the problem tends to look in real life. If one of them sounds like your site, that is a coincidence, and also a hint.

Part one covers how to interrogate a control. Part two walks one investigation end to end. This is the reference you work through beside them.

Why the class matters

Class Works when Fails when
Engineering Always, whether or not anyone remembers The design was wrong, or it was qualified under conditions that do not match production
Administrative A person performs it correctly, every time, under pressure Any time attention lapses, which over enough repetitions is a certainty
Detection It is looking in the right place, often enough, with a method that can see the thing Sampling design misses the event, or the method cannot recover the organism
Systemic Authority and process let the other three be maintained honestly Schedule pressure, weak quality unit, investigations that close rather than conclude

Rank candidate causes by class before you rank them by plausibility. An administrative control that failed is usually a symptom of an engineering control that was never there.


1. Engineering controls

Facility design and physical separation

Keeps graded space graded, and keeps the critical zone protected from everything around it. Fails when: the critical area has no robust physical boundary, so protection depends on distance and behaviour. Documented: Eugia Pharma, August 2026: the lines “lacked robust physical separation between ISO 5 and surrounding ISO 7 areas,” and “appropriate ISO 5 conditions were not reliably maintained during these activities.”

Vioela’s note: Classification is a snapshot, not a guarantee. A room earns its grade from particle counts on the day it was tested, and that says nothing about how it behaves with six people in it and a jammed line. Walk two routes with the drawing in your hand: a component from the warehouse to the filling point, and a person from the street to the line. Every place where the clean route and the less clean route cross is a risk the drawing hides and a camera would show.

Unidirectional airflow over the critical zone

The single most important physical control in aseptic processing. Air moving down across open product sweeps contamination away from it. Fails when: air moves toward a return instead of across the zone, or entrains air from surrounding lower grade space. Documented: Hanlim Pharm, October 2026: airflow “failed to adequately protect the aseptic processing line,” with air “moving at a sharp angle upward toward return vents,” which “enables the entrainment of lower-quality air from the surrounding lower-classified” area.

Vioela’s note: The phrase to learn is first air: air that has come straight from the filter and touched nothing on the way. Sterile surfaces should only ever be bathed in first air. Anything placed between the filter and an open container, a hand, a tool, a machine guard, sheds into the air downstream of it, like a rock in a stream. Speed matters too, and it is judged where the work happens, not at the filter face. FDA’s guidance describes 0.45 meters per second, plus or minus 20 percent. Annex 1 gives 0.36 to 0.54 at the working position. Too slow fails to sweep. Too fast creates turbulence around equipment.

Picture this: A bowl of sterile stoppers, and directly above it, an operator’s forearm reaching across to nudge a sensor. The filter is doing its job beautifully. It is delivering perfectly clean air onto a sleeve, and the sleeve is delivering everything else onto the stoppers. That air was first air for about twelve inches.

Smoke studies as the qualification of that airflow

The evidence that airflow actually protects the zone during the work people really do. Fails when: performed at rest, or without the interventions that occur in every batch, or assessed from a report rather than the footage. Documented: Hanlim: “smoke studies were incomplete, as they failed to sufficiently evaluate airflow directionality.” Bausch and Lomb, September 2026: “your dynamic smoke studies failed to sufficiently demonstrate unidirectional airflow protection.” Pharmathen, May 2026: studies “failed to sufficiently evaluate removal of fallen vials.”

Vioela’s note: Smoke shows direction, not cleanliness, so the question while watching is always the same: after the air passes the operator’s hand, where does it go next? If the answer is toward open product, the intervention is unsafe however carefully it is performed. Watch the footage with the sound off, from more than one camera angle, and expect the study to be repeated whenever equipment is moved or added. A study filmed from an angle that cannot see the critical zone has demonstrated nothing, and that is more common than it should be.

Picture this: The report says “airflow acceptable, no turbulence observed.” You ask for the video. The video is fourteen minutes of the back of an operator’s head. Somewhere behind that head is the critical zone. The smoke may have behaved perfectly. We will never know, and neither will the person who signed the report.

Pressure cascade and filtered air under positive pressure

Keeps air flowing from cleaner to less clean, continuously. Fails when: qualified at rest and not maintained in operation, or monitored by spot readings rather than a trend that would show intermittent loss.

Vioela’s note: FDA’s guidance looks for 10 to 15 pascals between adjacent rooms of different cleanliness, and at least 12.5 where an aseptic processing room borders unclassified space. The expert question is not the setpoint. It is the logging interval and the alarm delay. Alarms are usually delayed to avoid nuisance triggers, which means a real loss of pressure shorter than the delay is silent, and a value averaged over fifteen minutes can hide a loss that lasted under a minute. Ask for both numbers, and ask who chose them and why.

Barrier systems, RABS and isolators

Separates people from product. Fails when: the barrier is open more than the design assumed, glove integrity is unproven, or airflow inside the barrier was never demonstrated to protect the zone. Documented: Pharmathen: “This problematic airflow pattern in your RABS creates significant potential for microbial contamination.”

Vioela’s note: Gloves are the weakest part of every barrier. A pinhole is invisible, so the real question after a failed glove test is how long that glove was in use and which batches were made in the interval. For isolators, Annex 1 expects glove integrity testing at least at the beginning and end of each batch or campaign. Two further points. Opening a RABS door turns the line into a conventional cleanroom until it is closed and disinfected again, so count door openings per batch. And vaporized hydrogen peroxide decontaminates the surfaces it reaches. It does not penetrate, so two items touching each other inside an isolator shield the spot where they touch.

Picture this: A glove fails its integrity test on Friday. Its last passing test was the Monday before. “So which batches were made this week?” Silence. “All of them?” More silence. The glove did not develop a hole on Friday afternoon out of respect for the testing schedule.

HEPA filtration and integrity

Delivers air of the required quality to the zone. Fails when: filters are loaded, seated badly, or leak at the frame; integrity testing is periodic and the failure is not.

Vioela’s note: Filters rarely fail through the middle. They fail at the edges, where the gasket meets the frame, and after anything that disturbs the housing or the ceiling. FDA’s guidance gives twice a year as an example frequency for leak testing in the aseptic processing room, and treats a single probe reading of 0.01 percent of the upstream challenge as a significant leak. Between tests, the pressure drop across the filter is your early warning. A slow rise means the filter is loading. A sudden fall means air has found an easier path.

Temperature and humidity

Comfort conditions are a contamination control, and are rarely treated as one. Fails when: the room runs warm or humid, particularly on long runs or at seasonal peaks.

Vioela’s note: A hot cleanroom is a dirtier cleanroom. Gowned operators who are too warm perspire, shed more, and hurry to finish. Moisture also decides whether mold can establish on a surface at all. When recoveries rise in summer, look at the room’s temperature and humidity trend before retraining anyone.

Picture this: It is August, the chiller is struggling, and the filling room is running warm. Everyone inside is wearing two layers, a hood, a mask and goggles. Now ask how well you would perform slow, deliberate aseptic movements while gently poaching. Fix the room before you fix the people.

Water systems

In an aseptic area, water is the one thing that supports a persistent population. Gram negative organisms live in water and nowhere else for long. Fails when: a loop, drain, condensate path, humidifier or wet storage point establishes a reservoir, and investigations treat each recovery as a new event. Documented: Bausch and Lomb: FDA cited failure to investigate “adverse trends in your water for injection system,” and said “Your investigations did not evaluate the persistent recovery of microorganisms over time.”

Vioela’s note: The word that explains most water problems is biofilm. Organisms attach to wet surfaces and build a protective layer that sanitization struggles to reach, then shed from it in bursts. That is why a sample point can pass ten times and fail on the eleventh with nothing having changed: the system was colonized the whole time. Three practical rules. Sample the way production actually draws water, because flushing before a sample when nobody flushes before use measures a system that does not exist. Treat hoses left connected and wet as part of the system. And remember that FDA’s guidance considers drains inappropriate, with rare exceptions, in aseptic areas cleaner than ISO 8, so standing water anywhere in the core deserves an explanation.

Picture this: The sampling procedure says flush the outlet for two minutes, then sample. Production connects a hose and starts filling immediately. So for years the laboratory has tested water from a careful two minute flush, and production has used water from the first three seconds. Both groups are very proud of the results.

Clean utilities: compressed gases, clean steam

Contact product or product-contact surfaces, and are routinely forgotten in contamination investigations. Fails when: point-of-use filtration is unqualified, or utilities are not monitored to the standard of the zone they enter.

Vioela’s note: Compressed air and nitrogen touch product and sterile surfaces as surely as water does, yet they are tested far less often. Gas that contacts product should pass a sterilizing filter at the point of use, and that filter needs integrity testing like any other. Vent filters on tanks deserve a specific look: a vent filter that gets wet blocks, and a blocked vent means the tank either pulls a vacuum or breathes around the filter. For steam, sterilization depends on the steam being dry and free of trapped air, which is a quality you have to test for and cannot see.

Equipment sterilization and depyrogenation

Autoclaves, dry heat tunnels, steam in place. Fails when: load patterns in use are not the load patterns validated, or cycles are not challenged at the hardest position.

Vioela’s note: An autoclave is validated for a specific load, not for the machine. A new item, a different wrap or a different orientation is a new load, and air trapped in tubing or a filter housing is the usual reason a cycle that passed in validation fails in routine use. A pack that comes out wet does not stay sterile, because moisture carries organisms through the wrap. Two points people miss. Steam kills bacteria but leaves their endotoxin behind, so a history of gram negative contamination matters even after the organisms are dead. And fitting sterilized parts onto the line is itself an open aseptic operation, frequently the riskiest manual task in the whole batch and frequently the least observed.

Sterilizing filtration

The sterility assurance step for anything not terminally sterilized. Fails when: the filter is not validated for the actual product, bioburden before the filter is not controlled or trended, or post-use integrity testing is treated as a formality.

Vioela’s note: “Sterilizing grade” has a precise meaning: the filter holds back a challenge of ten million organisms per square centimeter of a very small test bacterium. That has to be proven with your product under your conditions, because some formulations shrink bacteria or change how the membrane behaves. Time matters as well. Over a long filtration, organisms can work their way through a filter that would have held them for a short one, so the validated maximum filtration time is a real limit. So does the load arriving at the filter: European regulators generally expect no more than 10 organisms per 100 milliliters ahead of the final filter. Test integrity after use every time, since that is the only evidence the filter was intact while product passed through it. Annex 1 also expects a test before use, after the filter has been sterilized, because sterilization itself can damage a filter.

Container closure integrity

A sterile product in a non-integral container is not a sterile product. Fails when: the system is qualified once and leak rates are allowed to drift. Documented: Baxter, September 2026: “Loss of container-closure integrity is a direct cause of non-sterility,” and FDA found the firm “lacked effective CAPA to address the significantly increased rate of leak defects.”

Vioela’s note: A sterility test is not an integrity test, and neither is a visual check. Dye ingress, the traditional method, depends on the operator and misses small leaks. USP chapter 1207 favors methods that measure a physical quantity, such as vacuum decay, helium leak or laser headspace analysis. Know the vulnerable window on your own line: between stoppering and crimping, a vial is closed only by the friction of the stopper, and a stopper that rides up before the crimp is a leak that looks sealed afterward. For products stored frozen at ultra low temperature, including many cell and gene therapies, there is one more trap. Rubber stoppers stiffen as they approach their glass transition, which for common formulations sits somewhere around minus 55 to minus 70 degrees Celsius, and a stiff stopper can stop sealing until it warms. Integrity has to be demonstrated at the storage temperature, not just at room temperature.

Single-use systems and supplier-sterilized components

More and more sterile contact surfaces arrive already sterilized by someone else: bags, tubing sets, filters, ready-to-use vials and stoppers. Fails when: the supplier’s sterilization and packaging are trusted on a certificate, and nobody looks for damage on arrival.

Vioela’s note: With single-use systems, the sterility assurance of your process starts in your supplier’s plant, which makes supplier qualification a sterility control. The practical risks are mundane. Bags leak at welds and ports after rough transport, and a leak that lets product out lets organisms in. Confirm that the irradiation dose range was validated for that exact assembly, inspect packaging before it is opened, and for high value batches consider leak testing bag assemblies at the point of use.

Line design that forces interventions

Every reach into the critical zone is a contamination opportunity created by the equipment. Fails when: stoppages, falling vials and jams are frequent, and the response is to train operators to intervene well rather than to stop the stoppages.

Vioela’s note: Sort interventions into two kinds. Inherent ones the process requires: set-up, weight checks, monitoring. Corrective ones exist because something went wrong: jams, fallen vials, stuck stoppers. Then trend corrective interventions per batch. It is the best leading indicator in an aseptic plant, because it rises before the environmental data does. When it rises, look upstream at component quality, since a change in vial dimensions or stopper siliconization can double the jam rate without anyone on the line having changed anything.


2. Administrative controls

These depend on a person doing the right thing. Expect each to fail eventually and design accordingly.

Gowning, qualification and glove coverage

Fails when: qualification is a one-off, or technique leaves gaps. Documented: Pine Pharmaceuticals, August 2026: FDA cited “incomplete coverage of the backs of the hands and interdigital spaces,” and noted that the firm’s setup made “operator technique a primary sterility assurance control,” which is itself the finding.

Vioela’s note: People are the largest source of contamination in any cleanroom, and gowning reduces what they shed without stopping it. Shedding rises sharply with movement, which is why slow, deliberate motion is a control and not a courtesy. Two things experienced reviewers check. Whether gown qualification was performed on a calm first entry, when the real risk is the fourth entry of a long shift. And how many wash and sterilization cycles each garment has been through, because fabric loses its barrier with age and almost nobody tracks it.

Disinfection of materials and equipment entering the critical zone

Fails when: items pass in without sanitization, or the agent and contact time are not appropriate to what is on them. Documented: Pine: “operators failed to sanitize materials, equipment, and hands prior to introducing them into the ISO 5 critical area.” Bausch and Lomb: “Operators failed to disinfect” items before handling sterile supplies and between uses.

Vioela’s note: Alcohol does not kill spores. A wipe with 70 percent isopropyl alcohol deals with most ordinary bacteria, but a bacterial or fungal spore on that surface survives and rides in. That is why spore formers so often point to items entering the room, and why Annex 1 expects the disinfectants used in the highest grades to be sterile themselves. The other half is contact time. A disinfectant works only while the surface stays wet, so a wipe that flashes dry in twenty seconds has not delivered a one minute contact time, whatever the procedure says.

Picture this: An operator sprays a bag with alcohol, counts “one Mississippi,” wipes it, and passes it in. The procedure says one minute of wet contact. The bag was wet for four seconds. Somewhere on that bag, a Bacillus spore that can sit in alcohol all day long is having a perfectly nice afternoon.

Aseptic technique and behaviour in the zone

Fails when: first air is blocked, exposed units are not discarded, or people lean into the protected space. Documented: Pharmathen: an operator “blocked first air, and exposed vials were not discarded.” Eugia: “Operators reached their upper body (e.g., head and shoulders) over an open transfer hose.” Pine: “an operator rested their sleeved arms on ergonomic support bars” “without first sanitizing their sleeves.”

Vioela’s note: The rule underneath all aseptic technique is simple: never put anything between the filter and an open sterile surface, and never reach over one. Work from the side, use sterile tools in place of gloved fingers, and move slowly. When judging how people really work, remember that a media fill is performed by people who know they are being watched. Routine production at the end of a long shift is the truer picture, which is why reviewing video of ordinary batches teaches more than witnessing a qualification.

Picture this: Media fill day. Everyone moves as if defusing a bomb. Slow hands, perfect posture, a sterile tool for everything. Now pull the video from an ordinary Thursday at three in the morning, hour ten of the shift. Same people, same line. Different movie.

Interventions: defined, limited and qualified

Fails when: the list of permitted interventions does not match what happens in a real batch, and nobody has recorded the frequency.

Vioela’s note: A well controlled intervention has four things written down: how it is performed, with which tool and through which glove or door; the maximum number of times it may occur in a batch; which exposed units are discarded afterward; and who is qualified to do it. That last point is often missed. Only an operator who has performed that intervention in a successful media fill should perform it on product. And when something happens that is on no list, the right response is to stop and bring quality in before touching anything, because an improvised intervention has never been shown to be safe.

Aseptic connections

Fails when: the practice and the location of connections are not assessed as the hazard they are. Documented: Eugia: FDA required assessment of “All aseptic connections. Include an assessment of practices and locations at which connections are made.”

Vioela’s note: Count the open connections made per batch and treat the number as a design defect to be driven down. Each one can usually be replaced by something closed: a sterile connector, a tube welder, or a line steamed in place. Where a connection must stay open, the order of assembly matters, so that the sterile end is exposed for seconds and is never pointed at the operator. In cell and gene therapy, where many steps are still manual, the count of open manipulations is the most honest single measure of process risk.

Cleaning and disinfection programme

Fails when: agents are not effective against the organisms actually recovered, contact times are not met, rotation has no sporicidal step, or solutions are prepared with non-sterile water. Documented: Medline, May 2026: “Your manufacturing operators have routinely deviated from procedures for cleaning and disinfection,” in a facility where B. cereus was recovered “in at least five samples taken from your manufacturing environment since January 2025.”

Vioela’s note: Cleaning and disinfecting are two different jobs, and the first has to happen for the second to work, because disinfectants are neutralized by soil and by the dried residue of earlier disinfectants. Rotation is widely misunderstood. It is not about organisms developing resistance, as with antibiotics. It is about coverage: a routine agent for everyday organisms, plus a sporicide at a defined frequency for the spores the routine agent cannot kill. Three checks that pay off quickly. Was efficacy proven on your actual surface materials, using organisms recovered from your own facility? Does the surface stay wet for the full validated contact time? And was the sporicide used straight after the last shutdown, maintenance visit or construction work, which is when spores arrive?

Picture this: “We rotate three disinfectants so the bugs do not build resistance.” Lovely. Which of the three kills spores? Long pause. None of them. So the rotation is three different ways of not killing a spore. Variety is nice. Coverage is the point.

Personnel and material flow

Fails when: the designed flow is impractical, so people take the route that works rather than the route on the drawing.

Vioela’s note: Every room was qualified with a certain number of people in it, and that number is a limit, not a description. Each additional person adds particles and disturbs airflow, so an aseptic room during a shift change, or with engineers and observers present, may be operating outside anything that was ever tested. Watch the room at its busiest moment, and compare the head count with the maximum used in media fills.

Line clearance, campaign length and changeover

Fails when: campaigns run longer than the conditions validated for them.

Vioela’s note: Every limit in aseptic processing is also a clock. How long sterilized equipment may wait before use, how long bulk product may be held, how long a filling run may last, how long since the last sporicidal clean. Each of those should have been challenged at its maximum, and the media fill should run at least as long as the longest permitted batch. When an investigation finds nothing unusual, check the clocks. A hold time exceeded by a few hours is easy to miss and easy to prove.

Maintenance and return to service

Any shutdown, repair or construction work breaks the conditions the room was qualified under. Fails when: production resumes on a schedule rather than on evidence that the area has recovered.

Vioela’s note: Excursions cluster after shutdowns. Tools, parts and people who do not normally enter the area bring spores with them, ceiling work disturbs settled dust, and air handling units that were switched off shed what accumulated while they were idle. A disciplined restart has three steps: a full clean including sporicide, a recovery period with intensified monitoring, and a documented decision that results are back to baseline before the first batch. In any investigation, ask what maintenance took place in the weeks before the first recovery.

Picture this: The shutdown ends Friday. The contractors pack up Friday. Production restarts Monday, because the schedule says Monday. The spores that came in on the contractors’ toolboxes were not copied on the schedule.

Training and requalification

Worth stating plainly: training is the weakest control on this page. It is the right response to a knowledge gap and the wrong response to a design problem.

Vioela’s note: Training works when it teaches the reason and fails when it teaches the step. An operator who understands first air and how organisms travel can make a good decision in a situation the procedure never anticipated. One who has memorized steps cannot. The sign to watch for is drift, where a small shortcut that caused no harm becomes the normal way of working. Nobody decides to lower the standard. It lowers itself, one uneventful batch at a time.


3. Detection controls

These do not prevent anything. They tell you whether the controls above are holding, which means a weakness here makes every weakness above invisible.

Environmental monitoring: where you sample

Fails when: sample points reflect convenience rather than risk. Documented: SCA Pharmaceuticals, September 2026: the active air sampler was placed “at the far right of the hood, away from where aseptic operations occur.”

Vioela’s note: There are two kinds of sample map and they are easily confused. A classification grid proves the room meets its grade. A monitoring plan is supposed to look where product is at risk. Sites should be chosen by asking where sterile product or components are exposed and what could reach them, and the reasoning should be written down so it can be challenged. Never drop a site because it keeps giving results. A location that recovers organisms is the monitoring doing its job.

Picture this: One sample site keeps recovering organisms, so at the annual review it is “optimized” out of the monitoring plan. Results improve immediately. Everyone is pleased. This is the monitoring equivalent of curing a fever by throwing away the thermometer.

Environmental monitoring: when you sample

Fails when: sampling happens at rest or at shift start, describing a room that does not make product. Documented: SCA: the programme did not capture data “representative of aseptic production conditions.” NexCell Scientific, September 2026: “environmental and personnel monitoring was not performed in association with each production shift.”

Vioela’s note: The particle counter is the only monitoring that reports in real time, so use it as a clock. Lay particle spikes over the intervention log and you can see which interventions disturb the zone. Settle plates have a time limit of their own: left open too long, the agar dries and stops growing what lands on it, which is why Annex 1 caps exposure at four hours per plate and expects the chosen time to be backed by recovery studies. And check whether anything is sampled during set-up. It is often the highest risk activity of the batch and the least monitored.

Environmental monitoring: whether the method can see it

The most quietly damaging failure on this list, because it produces clean data. Fails when: media, incubation or recovery method cannot grow what is actually there. Documented: Bausch and Lomb: “Media used to support your EM program was observed to be inadequate to support growth of recovered contamination.”

Vioela’s note: Three quiet ways to get zero. Residual disinfectant on a surface is picked up by the contact plate and keeps killing on the agar, unless the media contains neutralizers. Plates read too early miss slow growers and molds, which need cooler and longer incubation than most bacteria. And in isolators, traces of peroxide left after decontamination can sterilize the plate itself. Every media lot should be shown to grow organisms recovered from your own facility, not just the standard laboratory strains. A site that never finds anything deserves the same scrutiny as a site that finds too much.

Personnel monitoring

Fails when: the technique makes a negative result meaningless. Documented: SCA: “Compounding technicians do not always carefully roll their individual fingers across the media touch plate.”

Vioela’s note: Timing decides whether a glove plate means anything. Taken straight after sanitizing, it will always be clean, which is a measurement of the alcohol and not of the operator. Sample on exit from the critical area and after critical interventions, before any sanitizing step, then change gloves, because the agar leaves nutrient on the fingertips. Read results per person over time. One recovery is noise. The same operator recovering repeatedly is information, and under Annex 1 the limit for gloves in the critical zone is no growth at all.

Picture this: “Our glove plates have been clean for two years.” Impressive. When are they taken? “Right after sanitizing, as the procedure says.” Congratulations. You now hold two years of excellent data on isopropyl alcohol. The operators remain a mystery.

Fails when: each excursion is investigated alone and the pattern across years is never assembled. Documented: Bausch and Lomb: investigations “did not evaluate the persistent recovery of microorganisms over time.”

Vioela’s note: Trend what was found, not just how many. A count of one that has changed from a skin organism to a waterborne one is a bigger signal than a count of five of the usual flora, and it will never breach a numeric limit. At the low counts of a clean area, averages are useless. Track the proportion of samples showing any growth at all, as USP chapter 1116 describes, and set alert levels from your own history, because a regulatory limit tells you when you are out of compliance, not when you have changed.

Picture this: Last quarter, five recoveries, all ordinary skin bacteria. This quarter, one recovery, a Pseudomonas. The dashboard turns greener, because five became one. Wonderful. Except the room just told you it has a water problem, and it said so very quietly.

Media fills and aseptic process simulation

The qualification of the whole aseptic process, and the clearest signal you will ever get. Fails when: the simulation does not represent real interventions, durations or conditions, or repeated failures are explained away individually. Documented: Bausch and Lomb: “your process simulations (media fill) were inadequate.”

Vioela’s note: The target is zero. FDA’s guidance scales its response to the size of the run: below 5,000 units, one contaminated unit is cause for revalidation after an investigation; above that, one unit means an investigation and two mean revalidation. Three places where a weak simulation hides. Reconciliation: every unit filled must be accounted for as incubated, or rejected for a reason that would also apply to product, because units set aside for cosmetic defects are where contaminated units disappear. Duration and staffing: the run should cover the longest permitted batch, the maximum number of people, and a shift change if production has one. And contact: units must be inverted or swirled so the medium touches every internal surface, including the stopper, or a contaminated closure will never show.

Picture this: 5,000 units filled, 4,870 incubated. Where are the other 130? “Rejected for cosmetic defects.” In a media fill. Nobody is buying this broth. It does not need to be pretty. It needs to be incubated.

Sterility testing, and its limits

Fails when: treated as proof of sterility rather than a sample-based test with poor sensitivity. Documented: Pine: “microbial contamination does not spread uniformly” and limited EM and sterility testing “cannot reliably detect low-probability contamination events.” Eugia: such data “cannot be solely relied upon as justification to release drug product batches.”

Vioela’s note: The arithmetic is worth doing once. A typical test examines 20 units. If one unit in every hundred in the batch is contaminated, the chance that all 20 tested units are clean is about 82 percent, so the batch passes four times in five. At one contaminated unit per thousand, it passes 98 times in 100. The test catches gross contamination and very little else. One more trap: some products inhibit microbial growth, so the method has to be shown to recover organisms in the presence of the product. Without that suitability work, a pass may only mean the product stopped the organism from growing in the test.

Laboratory investigation of a sterility failure

Fails when: a result is invalidated without a full investigation. Documented: Baxter: “You invalidated the sterility result without adequate scientific justification,” having concluded “there was no risk to the manufacturing process” “without conducting a Phase II investigation.”

Vioela’s note: The starting presumption is that the product is contaminated, and the burden of proof sits with anyone who wants to say otherwise. The pharmacopeia allows a sterility result to be invalidated only on narrow grounds: a demonstrated fault in the test environment, a demonstrated fault in how the test was run, growth in the negative controls, or an organism that can be tied unequivocally to the testing itself. Identify the isolate to species and compare it with what has been recovered in both the laboratory and the production area. Finding the same organism in production monitoring is strong evidence the failure is real. Finding a skin organism proves nothing either way, because production operators have skin too.

Picture this: The sterility test fails. Within the hour somebody says, “It must be laboratory error.” Based on what? “Well, the batch was fine.” And how do we know the batch was fine? “Because it would have passed, if not for the laboratory error.” That is not an investigation. That is a circle.


4. Systemic controls

These decide whether everything above is maintained honestly.

The contamination control strategy

The one document that is supposed to show how every control on this page fits together. Fails when: it is written once as an inventory of existing procedures and never used.

Vioela’s note: Annex 1 requires one for anyone supplying Europe, and the idea behind it, that controls must be designed as a connected system, applies everywhere. Its value is in the links, not the list. For each source of contamination it should show which control prevents it, which control would detect a failure of that prevention, and what the data from both currently says. A useful test: hand it to an operator and an engineer and ask each whether it describes their plant. If either hesitates, it is a compliance document and not a strategy.

Quality unit authority

Fails when: the quality unit cannot stop production or reject a conclusion. Documented: Bausch and Lomb: “your quality unit is not enabled to exercise proper authority.”

Vioela’s note: Authority is tested by history, not by the organization chart. Ask for the last time quality stopped a line or rejected a batch against production’s wishes, and what happened to the person who did it. If nobody can remember an example, the authority is theoretical. A second test is location. A quality unit that only reviews records at a desk learns about the batch from paper, while one that spends time on the floor during set-up and interventions knows what the paper leaves out.

Picture this: Ask a room of managers when quality last stopped a line. People study the ceiling. “2021, I think? Before the reorganization.” A fire alarm that has not sounded in five years means one of two things. Either there have been no fires, or somebody disconnected the alarm. Only one of those is good news.

Investigation quality

Fails when: investigations conclude at the behaviour and never reach the conditions. Documented: Pharmathen: investigations “did not include in-depth analysis of the conditions that create the circumstances for human error.”

Vioela’s note: Two numbers summarize the health of an investigation system faster than reading a hundred records. First, the share of investigations that conclude with human error or with no cause determined. Second, the share that close in the last few days before their due date. A high value for either one suggests records are being finished, not problems being solved. Neither number takes more than an afternoon to pull.

CAPA effectiveness

Fails when: the same action is repeated after it has already failed. Documented: Medline: “You have unsuccessfully attempted similar corrective actions in the past,” and the response “not explained why your previous CAPA attempts were not successful.”

Vioela’s note: An effectiveness check has to be able to fail. If an event occurred once in forty batches and the check is three clean batches, the check will pass 93 times in 100 even if the corrective action did nothing, because three clean batches were always the likely outcome. Size the window to how often the problem occurs, and pick a measure that moves before the failure does, such as corrective interventions, recoveries of the specific organism, or the process parameter that was drifting.

Picture this: The effectiveness check reads “no recurrence in three batches.” The event had happened once in the previous forty. That is declaring the roof repaired because it stayed dry for three sunny days. Wait for rain.

Records and data integrity

Fails when: the evidence an investigation would need is not contemporaneous, or no longer exists. Documented: Baxter cited “the lack of contemporaneous documentation.” SCA was cited under 211.22(d) over production video used in investigations not being retained.

Vioela’s note: Reading an environmental plate is one of the most exposed data points in a sterile plant: one person counts colonies on a plate that is then thrown away. Good systems add a second reader or photograph every plate, and reconcile plates so that each one issued is accounted for. Record interventions at the moment they happen, with the time, since that timestamp is what later allows product from before and after an event to be told apart. And if the line has cameras, decide in advance how long footage is kept, because the first thing an investigator will ask after a contamination event is to see it.

Product impact assessment

Fails when: the conclusion of no impact rests on the detection controls rather than on evidence about the event. Documented: Eugia: the firm based its no-impact conclusion “on EM, Personnel Monitoring (PM), and process simulations (media fill re-qualifications),” which FDA rejected.

Vioela’s note: Start from the event, not from the results. Ask what the worst credible exposure of product was, then ask what evidence could show the product was unaffected, and whether that evidence was capable of detecting the problem in the first place. A passing sterility test cannot carry that weight, for the arithmetic given above. Where a firm discards units around an intervention, the discard only protects the batch if the record shows exactly when the intervention happened and which units were on the line at that moment.


How to use this

Three ways, in rising order of usefulness.

As a gap assessment. Walk the list against one line. For each control, ask the five questions from part one: does it exist specifically, was it qualified for the condition that occurs, is it functioning, would it fire, does it depend on someone remembering.

As an investigation checklist. When contamination appears, the list is your candidate cause set. Work down it and record which ones the evidence excludes. An investigation naming one hypothesis is indistinguishable from a guess.

As a design review. Count how many of your sterility assurance controls are administrative. Every one of those is a control that works until somebody is tired. That count, more than any single finding, tells you how much of your sterility assurance rests on people not making mistakes.

One more use, for anyone leading a team. Vioela’s notes are short on purpose. Take one control a week into a team meeting, read the note aloud, and ask whether it is true of your line. Twenty minutes, no slides, and within a year the whole team has been through the list.

Sources

All quotations are from FDA warning letters issued in 2026:

Governing requirements referenced throughout: 21 CFR 211.42(c), 211.113(b), 211.192, 211.22, 211.67(b), 211.25(a), and FDA’s guidance on sterile drug products produced by aseptic processing.

Figures in Vioela’s notes come from:

  • FDA’s aseptic processing guidance (2004), linked above: air velocity, pressure differentials, filter leak testing, drains, media fill acceptance criteria, and the filter challenge
  • EU GMP Annex 1 (2022): air speed at the working position, settle plate exposure, the no growth limits for the critical zone, glove integrity testing, filter testing before use, sterile disinfectants, and the contamination control strategy
  • USP general chapters 71 (sterility tests), 1116 (microbiological control and monitoring of aseptic processing environments) and 1207 (package integrity evaluation for sterile products)
  • EMA, Guideline on the sterilisation of the medicinal product, active substance, excipient and primary container (2019): bioburden ahead of the sterilizing filter

The percentages in the sterility testing and CAPA notes are simple probability, worked from the assumptions stated in each note.

Warning letters state FDA’s position at the time of issue. They are not adjudicated findings, and firms respond to them.