The investigation they should have run
A reconstruction built on one public warning letter: recurring sterility failures blamed on operators, an airflow problem found by the regulator instead, and a step by step account of who you would actually need to talk to, in what order, to reach the real cause.
Part one set out the method. This is one event taken end to end.
A necessary caveat before anything else. FDA’s warning letter to Pharmathen International, dated 27 May 2026, records what the agency found. It does not publish the firm’s internal investigation records, so I cannot tell you what their investigation actually said. What follows reconstructs how an investigation reaching that conclusion typically runs, and then rebuilds it the way I would want it run. The facts I anchor to are the ones FDA put in writing. The weak investigation below is a composite of a pattern I have seen many times, not a claim about any particular firm’s documents.
What the letter establishes
Five things, in FDA’s own words or close paraphrase:
- Recurring sterility and media fill failures over five years, with operator error identified as a significant contributor.
- The firm’s investigations “did not include in-depth analysis of the conditions that create the circumstances for human error.”
- Gram negative organisms recovered in the aseptic processing room, which FDA called “highly atypical,” attributed by the firm to room disinfection practices.
- A RABS airflow problem: “This problematic airflow pattern in your RABS creates significant potential for microbial contamination.”
- Smoke studies that “failed to sufficiently evaluate removal of fallen vials,” while an observed operator “blocked first air, and exposed vials were not discarded.”
Read those together and the shape of the real event is already visible. Three separate things are being described: a source of organisms, a transport mechanism that lets them reach the critical zone, and an intervention that puts a person into that zone at the worst possible moment. The investigation reported one of the three.
The investigation as it probably ran
Reconstructed, and deliberately recognisable:
A sterility failure is confirmed. The batch is rejected. The investigation owner pulls the batch record, interviews the line, and finds that an operator retrieved fallen vials during the run. A 5 Whys is run in a meeting room with quality and production present. It terminates at aseptic technique. The cause is recorded as operator error, the CAPA is retraining plus a revised intervention procedure, effectiveness is a review of the next three media fills, and the record closes inside thirty days.
Separately, gram negative recoveries in the room are investigated as an environmental matter. The cause is recorded as a disinfection lapse. The CAPA is a revised cleaning schedule and retraining of the cleaning staff.
The two investigations never meet.
Everything in that paragraph is defensible on its own terms. Each conclusion is plausible, each CAPA is implementable, and both close on time. And five years later the same failures are still happening, which is the only evidence that actually matters about whether an investigation worked.
The structural error is at the end: two investigations, one room. A gram negative population in the aseptic area and a sterility failure in product made in that area are being treated as unrelated events by different owners on different timelines. Nobody ever asks whether the thing living in the room is the thing that reached the vial.
How it should have run
What follows is sequenced deliberately. Evidence before interviews, physical before behavioural, and the cheapest discriminating test first.
Step 0. Before you speak to anyone, pull the records
Two days of reading, and it determines every question you ask afterwards.
- Every EM result for that room for the full five years, not the quarter. Viable air, surface, personnel monitoring, with locations and dates.
- Every sterility failure and media fill failure in the same period, with the investigation conclusion recorded for each.
- The room’s qualification package: air balance, pressure cascade, recovery studies, the original smoke study report and video.
- Maintenance and change history for the air handling unit, the RABS, filters, dampers and the building management system.
- Water system records if the room has any water service, plus drain locations and cleaning equipment storage.
You are looking for one thing above all: does the pattern repeat, and does it repeat in the same places? Recurrence is the most informative fact in the file and it is only visible across years.
Vioela’s note: One thing has to happen even before the reading: protect the evidence. The natural reaction to a contamination event is to clean everything, and a thorough clean destroys exactly what you need. Before the room is reset, take extra environmental samples at and around the suspect locations, photograph the line as it stood, and tell the laboratory to keep every plate and isolate instead of discarding them on the normal schedule. An isolate thrown away in week one cannot be compared with anything in week three.
Step 1. The microbiologist
Why first. The organism is evidence about its own origin, and it is the cheapest test available for discriminating between hypotheses.
What to ask. Identify to species, not genus, for every recovery across the five years. Are isolates retained? Can the same species be compared between room recoveries, personnel monitoring and the failed sterility tests? Is there typing data, and if not, can it be obtained for retained isolates?
What it tells you. Gram negatives do not live on people and do not survive long on dry surfaces. They live in water. If the species recovered in the room is also the species in the sterility failure, you no longer have two investigations. You have one, with a route.
Vioela’s note: Ask how the identification was done. A species name from a biochemical test kit and one from genetic sequencing are not the same strength of evidence, and two isolates with the same name are not necessarily the same strain. When the question is whether the organism in the product is the organism in the room, ask for a strain-level comparison. It costs little, takes days, and can settle an argument that would otherwise run for months.
The question that reframes everything: if this organism needs water to persist, and the room is supposed to be dry, where is the water?
Step 2. The EM program owner
Why. They know where the sample points are, and more importantly where they are not.
What to ask. Plot every recovery on a room drawing. Are they clustered? Do they sit near a drain, a door, a transfer hatch, a sink, a piece of equipment with a condensate path? When were sample locations last reviewed against actual operations? Are samples taken during dynamic conditions or at convenient moments? What is the limit, and has it ever been changed?
What it tells you. A spatial cluster is a source. A diffuse pattern is a transport problem. If sampling only happens at the start of a shift, your data describes the room at rest, which is not the room that makes product.
Step 3. The operators, individually
Why here and not first. By now you can test what you are told. Interviewing before you have the data means accepting whatever account is offered.
How. Separately, never as a group, and never with their supervisor present. Ask what actually happens rather than what the procedure says. Make clear in the first sentence that you are mapping the process, not assigning blame, and then behave consistently with that for the rest of the investigation, because they will tell each other what the meeting was like.
What to ask. Walk me through the run. How often do vials fall? What do you do when one does? Who taught you that? Is it in the procedure? What is awkward about this line? What do you do when the pressure alarm sounds? What would you change if you could?
What it tells you. Frequency is the key number. If vials fall several times per run, the intervention is not an aberration, it is part of the process, and it should have been in the smoke study and in the media fill design. That single fact moves the finding from operator behaviour to process design.
Vioela’s note: Ask people to show you, not tell you. Stand at the line and have the operator walk through the task with their hands. People describe the procedure when they talk and reveal the practice when they move, and the gap between the two is usually where the answer is. Go on the shift where the event happened, too. Nights and weekends run with fewer people, less supervision and different habits, and a day shift interview about a night shift event is hearsay.
Step 4. The line supervisor and the intervention list
What to ask. What is the documented list of permitted interventions? Is removal of fallen vials on it? Was it included in aseptic process simulation? How many interventions are simulated in a media fill versus how many occur in a real run?
What it tells you. FDA’s letter says the smoke studies failed to evaluate removal of fallen vials. If the intervention also does not appear in the media fill design, then the qualification of the aseptic process never covered the thing operators do routinely. The process was validated for a run that does not happen.
Step 5. The facilities and HVAC engineer
Why. This is the conversation that quality investigations usually never have, and it is where the airflow finding lives.
What to ask. Show me the air balance, and tell me whether it was measured at rest or in operation. What is the room’s supply and extract volume? Does the RABS or any equipment in the room extract air, and was that extraction included in the balance? What are the differential pressures, how are they monitored, how often is the data reviewed, and what happens when they dip? Can I see the trend rather than the spot readings? Where does room air go when a door opens or a hatch is used?
What it tells you. Several candidates live in these records, and the engineer can usually rule most of them in or out in an afternoon: filter loading, damper drift, a return path blocked or modified, leakage through doors and hatches, equipment that draws air the balance never accounted for. What they have in common is that any of them can hold at rest and fail in operation, producing exactly the picture here. Intermittent, unexplained, worse during activity, invisible to anyone reading spot checks rather than trends.
The correlation to run: overlay the pressure trend on the EM recovery dates. Both are timestamped. If dips coincide with recoveries, you have an association worth pursuing. If they do not, exclude the hypothesis in writing and move on.
Step 6. Watch the smoke study, do not read it
Why. The report says pass. The video shows what actually happened.
What to ask for. The raw footage, not the summary. Then watch it specifically for: air sweeping over open containers, air moving toward return vents rather than down over the critical zone, entrainment from surrounding lower grade space, and what happens when an operator reaches in.
What to ask. Was the study done at rest or in operation? Were doors and hatches used during it? Which interventions were simulated? Was removal of a fallen vial among them? Who witnessed it and who approved it?
What it tells you. FDA’s letter answers this one already for this firm. The studies did not sufficiently evaluate the very intervention that was being blamed on the operator. If the study never demonstrated that the intervention could be performed without compromising first air, then telling the operator to perform it correctly is instructing someone to do something that has never been shown to be possible.
Step 7. The maintenance technician
Why. They service the equipment, they notice things nobody writes down, and they are almost never interviewed.
What to ask. When did you last work on this unit, and what did you find? Have you ever found standing water, condensate, corroded ductwork, a failed damper, a filter seated badly? Has anything about this room changed in the last five years? Does anything in here stay damp?
What it tells you. This is frequently where the water source turns up: a condensate tray that does not drain, a chilled surface, an unused drain with a dry trap, a cleaning trolley stored damp. A reservoir plus a transport mechanism is a complete contamination story, and you now have both halves.
Step 8. Cleaning and disinfection
Why. The firm’s own stated cause was disinfection practice, so it must be tested rather than dismissed because you prefer a different hypothesis.
What to ask. Which disinfectants, at what contact times, in what rotation? Is the efficacy data for the specific organism recovered? How is the solution prepared, and with what water? Where is the equipment stored between uses, and is it dry? Who performs it, and on what schedule relative to production?
What it tells you. Two outcomes, both useful. If the disinfectant was never validated against the recovered species, or the solution is prepared with non sterile water, the firm’s original cause was right but incompletely stated. If efficacy and practice are sound, you have excluded it on evidence rather than preference, and the record will show that you did.
Step 9. Validation, on media fill design
What to ask. How many units, what interventions, what worst case conditions? Are media fills run under the same dynamic conditions as production, including the equipment that draws air? Do they include the intervention frequency operators describe in step 3?
What it tells you. Media fill failures across five years are not bad luck. They are the qualification telling you the process is not in control, repeatedly, in writing, and being overruled each time by an investigation that found a person to attribute it to.
Step 10. The five year trend, with quality leadership
Why. The scope question is a management decision and should be taken at that level, explicitly.
What to put in front of them. Every failure, every recorded cause, every CAPA, and whether each recurred afterwards. If the same cause appears repeatedly with retraining as the action each time, that is a finding about the quality system rather than about any one event, and it needs stating in those terms.
The question to ask out loud: we have concluded operator error seven times in five years. Either we hire uniquely careless people, or the conditions make this error available. Which do we believe?
Step 11. Convene everyone, once, at the end
Not at the beginning. A cross functional meeting held before the evidence exists produces consensus around whoever speaks most confidently. Held after, with records on the table, it does the one thing it is good for: letting each discipline attack the others’ hypotheses.
Bring microbiology, engineering, production, validation and quality. Put up the candidate causes and the evidence for and against each. Leave with a written list of what has been excluded and why.
The cause statement that results
Something closer to this:
A water associated gram negative population established in the room, most probably at [location identified in step 7]. The airflow pattern within the barrier did not maintain protection of the critical zone during production, permitting entrainment of air from the surrounding lower classified area. Removal of fallen vials, an intervention occurring multiple times per run, was not evaluated in smoke studies and not represented in media fill design, so no qualified method existed for performing it without compromising first air. The combination of an environmental source, a transport route and an unqualified routine intervention produced recurring sterility and media fill failures. Operator technique was a contributing factor and not the root cause.
Longer than “operator error.” It is also testable, it explains all three documented observations rather than one, and it tells you what to fix.
The CAPA that follows
Note where each action sits in the hierarchy from part one, and that the engineering actions come first.
| Finding | Action | Class |
|---|---|---|
| Water source in the room | Eliminate the reservoir. Redesign or remove the wet feature, revise storage of cleaning equipment | Engineering |
| Airflow does not protect the critical zone in operation | Correct the airflow pattern, then requalify under production conditions | Engineering |
| Excursions invisible between periodic reviews | Continuous monitoring with alarms routed to someone who acts | Detection |
| Intervention never qualified | Smoke study in operation including fallen vial removal. If it cannot be done without compromising first air, eliminate the intervention by design | Engineering |
| Vials fall several times per run | Address the cause of falling vials: guide rails, line speed, format change parts | Engineering |
| Media fills do not represent the process | Redesign to include real intervention types and frequencies | Administrative |
| Technique | Retrain against the now qualified method | Administrative |
Retraining is still on the list. It is last, it is against a method that has now been shown to work, and it is not doing the load bearing work. That is the difference.
Effectiveness is not three clean media fills. It is EM recoveries of that organism in that room trending to zero and staying there, pressure differential holding under dynamic conditions across a defined period, and no recurrence of the intervention related failure mode. Define those before implementation, not when the check falls due.
What this costs, and why firms do not do it
Honestly: roughly two weeks of one competent person’s time, plus engineering hours, plus a requalification. Against a thirty day clock, with a batch rejected and pressure to close.
The weak version takes two days and closes on time.
That is the real decision, and it is not usually made consciously. It gets made by a due date, by whose calendar had room, and by the fact that nobody in the meeting had the authority to say the room needs rebalancing. Five years of repeat failures is what that decision costs when it is made the same way every time.
Related
What stands between you and a sterility failure is the companion reference: every control in this investigation, sorted by class, with how each one fails and what FDA has said about it.
Sources
- FDA warning letter, Pharmathen International S.A., 27 May 2026
- FDA, Sterile Drug Products Produced by Aseptic Processing: Current Good Manufacturing Practice, 2004
- 21 CFR 211.192 and 211.113(b)
The warning letter states FDA’s position at the time of writing and is not an adjudicated finding. The reconstructed investigation and the proposed sequence are my own, written as illustration, and are not a description of any firm’s actual records.