Stability Chambers: The Records Nobody Reads Until an Audit

Stability data is read as a result. It is produced as an experiment, and the experiment rests on equipment that has to be qualified, mapped, monitored and calibrated before its output means anything. That equipment is where the least-read pages of a stability file live.

The failure mode is not usually a broken chamber. It is a chamber that ran for three years with a sensor in the wrong place, an alarm nobody answered, and a chart with two gaps that were never explained. Nothing in the result table shows any of it.

Four things a chamber must demonstrate

Control. The set point is maintained within the study's tolerance, not merely programmed. A chamber that reads 25.0 °C on its display may hold 27 °C at the top shelf, because the display sensor sits where it was fitted at build.

Distribution. Temperature and humidity vary inside a loaded chamber. Mapping establishes where, and which positions are worst case. Loading pattern changes the answer, which is why a chamber mapped empty and then filled to a different density is only partly characterized.

Detection. Excursions are noticed, not discovered later in the data. That requires alarms that reach a person, and a log that preserves the event rather than overwriting it with the next reading.

Traceability. The sensor that recorded the number is calibrated against a standard that is itself traceable, with a certificate covering the period of use. A calibrated sensor whose certificate expired mid-study produces data with a hole in its pedigree.

Mapping and monitoring are different activities

	Mapping	Monitoring
Purpose	Characterize the chamber	Confirm it stays characterized
Duration	A defined study, days	Continuous, for the life of the study
Sensor count	Many, in a grid	Few, at worst-case positions
Output	A report with a position map	A chart and an alarm log
Repeats	On change, on repair, periodically	Never stops

The common error is to treat mapping as a one-time installation formality and monitoring as the real evidence. They are sequential: monitoring is only meaningful at positions that a mapping study identified as worst case. Move the sensor and the monitoring measures a different chamber.

The four records a buyer should ask for

A mapping report that names the positions, the load condition and the duration. A calibration certificate for each sensor in use, covering the study dates. A continuous record with no unexplained gap. An alarm log with the response written next to each event.

Most suppliers can produce the first and the second. The third and fourth are where answers become vague, and the vagueness is informative. A site that has never had an excursion has either very good equipment or a log that does not record them.

Excursions, and why the average is the wrong instrument

An excursion is assessed on two dimensions: peak and duration. The tolerance in a stability protocol is usually expressed as a cumulative allowance, which is a way of saying that a short deviation at a modest peak is tolerable and a long deviation at the same peak is not.

A monthly average destroys that information. A month that spent four hours at 32 °C and the rest at 25 °C averages close to 25, and the summary looks clean while the event sits unassessed in the raw data. This is the same arithmetic problem as a transit excursion, applied to a fixed installation: the mean is not the exposure.

The correct treatment is boring and specific. Identify the event, quantify peak and duration, identify which studies were in the chamber, assess whether the affected samples remain usable, and record the decision. Where the decision is to keep the data, say why. Where it is to repeat a time point, say that instead.

Where the condition itself is the mistake

A perfectly run chamber can still produce unusable data if it was set to the wrong condition. A study conducted at 25 °C and 60% relative humidity answers a question about a temperate market. If the destination sits in a hotter, more humid zone, the long-term condition is different and the study is answering something the customer did not ask.

The same applies to light. A chamber is dark by design, so a study that needs a photostability arm will not get one from a chamber, however well it is mapped.

Both are planning failures, and both are invisible in the result table. The table reports numbers that are internally consistent and externally irrelevant.

What a defensible programme looks like

Chambers qualified on installation with a documented mapping, including loaded conditions. Sensors at mapped worst-case positions, calibrated on a schedule that never lapses mid-study. Continuous recording with alarms that route to a named person and a response log. An annual review of excursions against the studies they touched. And a stability protocol that names the condition, the duration, the test points and the acceptance criteria before the first sample goes in.

That is the whole of it, and none of it appears in the certificate. It appears in the ongoing programme when a customer asks whether the three-year data describes material stored the way their warehouse stores it.

We describe chamber evidence as an input to the stability conclusion rather than as a formality, and we distinguish a study we can defend from a table we can quote. The expectations for the conditions themselves are set out in the ICH Q1 series, with the zone definitions carried in WHO guidance and the facility expectations in the European GMP text.

The question to ask is not whether the chamber is calibrated. It is: show me the excursion log for the period this study covers, and tell me what was decided about the samples that were in there.

What an auditor reads, and in what order

A chamber file is read in a sequence, and the sequence matters because each document depends on the one before it. The mapping report comes first, because it defines the positions. The sensor list comes second, because it should match the mapped worst cases. The calibration certificates come third, because they establish that the sensors were reading something traceable during the period. The chart comes fourth. The alarm log comes last, and it is the one that produces findings.

Reading in that order makes a specific failure visible: a chart that looks perfect, recorded by a sensor that is not where the mapping said the worst case was. The chamber was controlled. The evidence is about a different position in the chamber than the one the study needed.

The two findings that recur

Sensor placement. The probe sits near the door because that is where the cable reaches, or near the top because that is where the bracket was installed. Both are convenient and neither is a mapped worst case. A chamber monitored at a convenient position is monitored at a position chosen for convenience.

Excursion without assessment. The log shows an event, the event has a duration, and there is no record of what was decided about the samples that were in the chamber. The alarm worked. The system did not. An unassessed excursion is a data point whose effect on the study is unknown, which makes the study's status unknown.

A third finding is less common and more serious: the study protocol names a condition, the chamber is set to that condition, and the condition does not correspond to the market the customer is in. That is not an equipment problem and no amount of mapping will catch it, because the chamber is doing exactly what it was told.

The loaded-chamber question

Mapping is often performed on an empty chamber, and an empty chamber is the easiest case to control. Loading changes the airflow, creates thermal mass, and in humidity-controlled chambers changes the local moisture profile around the load. A chamber mapped empty and then filled to a different density than the mapping assumed has been characterized for a condition it no longer runs.

The practical compromise most sites adopt is to map empty, then verify at one or two load conditions, and to define a maximum load in the procedure. That is defensible if the maximum is respected. It stops being defensible when the chamber is filled beyond it during a busy quarter, which is precisely when nobody is looking at the procedure.

What to ask for, and what the answer means

Ask for the mapping report and the sensor positions, in that order, and check that the positions on the drawing match the positions on the sensor list. Ask for the calibration certificates covering the study dates, and check the dates rather than the existence. Ask for the excursion log for the same period.

A supplier who produces all three in one response has a programme. A supplier who produces the mapping report and then looks for the rest has a file that was assembled for the last audit. The difference is not honesty; it is whether the records are kept in a system or in a folder.

We treat chamber evidence as part of the stability conclusion rather than as supporting material, and we say which condition each study was run at, because a well-run study at the wrong condition is a well-run answer to a question the customer did not ask. The condition definitions themselves come from the ICH Q1 series and the zone framework carried in WHO guidance, with facility expectations in the European GMP text.

Frequently asked questions

What is the difference between mapping and monitoring?

Mapping is a study: a grid of sensors characterizes how temperature and humidity distribute inside an empty and then a loaded chamber, and identifies the worst-case positions. Monitoring is routine: fixed sensors at those positions record continuously and raise alarms. A chamber with monitoring but no current mapping is recording a condition nobody has shown to be representative.

How often should a chamber be mapped?

On installation, after any repair or relocation, after a significant change in load pattern, and periodically thereafter. The interval is a risk decision written into the site's programme, not a fixed number, but a chamber whose last mapping predates a compressor replacement is unqualified until it is repeated.

Does an alarm need a response record?

Yes. An alarm without a documented response only proves the alarm worked. The record should say when it fired, what the condition was, how long it lasted, what was in the chamber, and what was decided about the affected study. That last item is what an auditor is looking for.

Can a short excursion be averaged away?

No. Mean kinetic temperature and monthly averages are summary statistics, and a summary can hide four hours at 34 °C inside a month that averages 25 °C. The excursion has to be assessed on its own duration and peak, then judged against the study's tolerance.

What if the chart has a gap?

The gap is an event. It should be documented with its cause, its duration and the maximum plausible drift during it, and the affected study should be assessed. A chart with a silent gap is worse than a chart with a labelled one, because the reader cannot tell whether the condition was maintained.

Do I need to see the chamber records?

If the stability data is going into your regulatory file, yes, or at least a summary of the mapping, the calibration traceability and the excursion log for the period the study covers. The result table alone does not establish that the condition existed.