How to Define Raw Data in a GLP Study
Decide what the original record is, before an instrument decides it for you.
What a how-to is not
A how-to is SPEQ’s practitioner method, not a procedure. It does not replace your own SOP, it is not a validated approach, and the judgement calls in it belong to your quality unit.
Raw data are the original records and documentation from which the study report is derived. In a paper facility this was obvious. With instruments that acquire, process and re-process, it stopped being obvious, and the question "which of these is the raw data" now has to be answered deliberately for each system. A facility that has not answered it will find its archive holds processed output and no original.
- 1
Identify, per system, what the original record is
For each instrument and application, state whether the raw data are the acquired file, the processed result, the printout, or a combination. Write it down per system rather than as a facility-wide sentence, because the answer genuinely differs between a balance and a chromatography data system.
- 2
Treat the acquired file as raw data where processing is applied
Where a system acquires a signal and then applies integration, calibration or transformation, the acquired data and the processing parameters are both needed to reconstruct the result. Retaining only the processed output makes the result unreproducible, and it is the most common version of this failure.
- 3
Ensure the record is attributable and contemporaneous
Each entry must identify who made it and when, at the time of the observation. Shared logins defeat attributability entirely, and a system where several analysts appear as one account cannot support any of the conclusions drawn from its data.
- 4
Enable and review the audit trail
The audit trail records changes, who made them and why. It has to be switched on, it has to be protected from the users it records, and it has to be reviewed — an audit trail nobody reads is a control that exists on paper only.
- 5
Define how a correction is made
Corrections preserve the original entry, identify the person, the date and the reason. Overwriting is the defect; a visible correction trail is the expected state, and this applies identically to paper and electronic records.
- 6
Archive in a form that can still be read
Electronic raw data must remain retrievable and readable for the retention period, which outlives most software. Decide whether that means maintaining the system, migrating with verification, or retaining a validated rendering — and record the decision, because discovering it at retrieval time is too late.
- !Raw data defined once for the whole facility, when the answer differs between instrument types.
- !Only processed output retained, so a result cannot be regenerated from the acquisition.
- !Shared instrument logins, which make every record from that system unattributable.
- !Electronic records archived in a format tied to software that will not outlive the retention period.
How to Define Raw Data in a GLP Study: frequently asked questions
Common questions on define raw data in a glp study.
Is a printout the raw data?
Only if it is the original record and nothing needed to reconstruct the result lives elsewhere. Where an instrument acquires a signal and processes it, the printout is a derived view — the acquired data and the processing parameters are what allow the result to be regenerated, and those are what have to be kept.
Why does a shared login matter so much?
Because attributability is one of the properties that makes a record trustworthy, and a shared account destroys it for every record the system produced. It is not a control weakness to be compensated elsewhere; it removes the basis for saying who made an observation.
How do you archive electronic data for decades?
By deciding the approach in advance and validating it: maintain the reading system, migrate with verified fidelity, or retain a rendering that preserves the content and its audit trail. What does not work is archiving in a live proprietary format and assuming the software will still be there.