Author: Rick Antezana, Chief Executive Officer
In pharmaceuticals and medical devices, the source document rarely holds still. A protocol gets amended. A notified body asks a question and the instructions for use change. Legal revises one sentence of a consent form three weeks after the file went out for translation into 14 languages.
The question worth asking before a project starts is not whether the source will change. It is what your process does when it does, and whether you can prove afterward which version of the source each language came from.
Why regulatory translation services strain under a mid-project change
By the time a change arrives, a multilingual project is not one job. It is a dozen jobs at different stages. Two languages are in translation, four are in review with in-country affiliates, three are in layout, and one has already been delivered. A single edit to section 4 of the English does not become an edit to section 4 of each target file. It becomes 14 separate decisions, each with its own cost and its own risk of being missed.
The failure that matters is quiet. Nothing looks wrong. One language ships against the superseded source, and nobody finds out until an audit, a recall, or a regulator comparing the local label against the approved text.
The three patterns that go wrong
Restart everything
The safest-looking option and often the most damaging. A one-sentence change triggers a full retranslation and a full review cycle in every language, the submission date moves, and the change itself gets blamed for a delay that the process created.
Patch the text, miss the downstream
The changed segment is updated and the translation memory is corrected, but files already exported into layout are never re-exported. The memory is right and the delivered PDF is wrong. This is the most common version of the problem and the hardest to see, because every system in the chain reports success.
Absorb the change informally
The change arrives in an email, someone agrees to it, the work continues. No change record, no version stamp, no re-approval. The translations are correct. What is missing is any evidence of which source they were made from, which is the part an auditor asks for.
What controlled change looks like mid-project
A process that can absorb a change without restarting has six things in place before the change arrives:
- A version-stamped source. Every file that goes out carries a version, and every delivery references the version it was produced from.
- An agreed freeze point. The project plan says when the source is final and what happens to changes that arrive after it, agreed by both sides rather than negotiated under pressure. After that point a change does not simply join the queue. It is raised as a change request, scoped and scheduled on its own, and any delivery date it affects is reset in writing.
- Change as a documented event. A change request with its own identifier, its own approval and its own record, not a thread.
- Scoped re-review. Review is redone on the changed content and the context around it, rather than on the whole document, with the scope decision written down and justified.
- Impact analysis across languages. Before anything is updated, a list of which languages are affected and at what stage, including any already delivered.
- Traceability at delivery. Each delivered language is linked to a specific source version, so the question "which English was this made from" has a one-line answer.
Dynamic Language runs a management system certified to the applicable standards, including ISO 13485 (medical devices quality management systems), ISO 17100 (translation services requirements) and ISO 18587 (post-editing of machine translation output). Those standards set the framework for revision and record-keeping. The six items above are how that discipline gets applied to a source change, and applying it is what makes them routine rather than heroic.
How a mid-project change is handled in practice
Take an informed consent form that is midway through 11 languages when a protocol amendment adds a newly identified risk to the risks section. Five languages are in translation, three are in back-translation, two are with the client ahead of ethics committee submission, and one has already been submitted. Here is how a change like that would move through the work.
The amendment would be logged as a new source version. Because the change is safety-relevant, the risks paragraph would be reopened in all 11 languages, with fresh back-translation wherever back-translation had already been completed.
The two languages sitting with the client would go back as red-line translations, showing the tracked change against the version the committee was due to see, so resubmission does not call for a second full review. The submitted language would be flagged for the sponsor to decide on, with a red-line prepared if the sponsor chooses to resubmit. Every delivery would name the amended version.
What to settle with your language partner before the project starts
Five questions, asked at kickoff rather than at the first change:
- How is a source change logged, and what identifier does it carry?
- Who decides how much review is redone, and on what basis?
- How are languages already in layout or already delivered identified and corrected?
- What does the delivery record show about the source version?
- What happens to a change that arrives after the agreed freeze point?
Frequently asked questions
Should we just wait until the source is final?
Where the timeline allows it, yes. Where it does not, the useful move is staging: send the sections that are stable, hold the sections still in review, and agree how the two will be assembled. That is a planning decision, and it is much easier to make at kickoff than after work has started.
Does every change require full re-review?
No, and treating it that way is expensive. What matters is that the scope of re-review is decided deliberately, recorded, and defensible. A corrected typo and a changed dosage instruction are not the same event.
Does machine translation change the risk?
It changes where the risk sits. Post-editing under ISO 18587 can move content faster, which means a change can propagate faster too, in either direction. The control that matters is still the version record rather than the production method.
Who should hold the version record, us or the vendor?
You should, in your own system, because it is your submission. A language partner should be able to hand you the record in a form you can file rather than one you have to reconstruct.
Before the next file goes out
Most of the pain around mid-project changes comes from treating them as exceptions. In pharmaceuticals and medical devices, source changes are normal. Documents that live under review change.
A process that assumes the source will change absorbs the change. A process that assumes a final source pays for the difference in delay, in rework, and in the occasional language that ships against superseded text.
If you want to look at how your current process would handle a change arriving next week, Talk to a Specialist.
Tags:
Life Sciences
Comments