Start with a source map, not a folder tree
Most study folders grow by convenience: a folder for raw files, one for datasets, one for documents, and a few extra folders created when a new transfer arrives. That structure is easy to make and hard to defend later. A better starting point is a source map. Before the first upload, agree on the five groups your study team will use: EDC, SDTM, ADaM, lab and documents. Write down what belongs in each group and, more importantly, what does not. EDC holds the clinical data collected in the study system; SDTM holds submitted datasets; ADaM holds analysis datasets; lab holds laboratory transfers; documents hold protocols, SAPs and annotated CRFs. A good test for clinical trial data management software is whether the source group and data cut are visible before anyone opens a file. If source and cut are buried in a filename or left to memory, the team will eventually compare the wrong things.
Group each source without breaking the cut
Source grouping and cut versioning are two different controls, and they should not be collapsed. Source grouping answers “what kind of file is this?” Cut versioning answers “which snapshot of the study does this file belong to?” A study can have several EDC extracts, several SDTM deliveries and several lab transfers, and they may not arrive on the same rhythm. Do not merge an EDC extract from one cut with a lab transfer from another just because the filenames look similar. Keep each source group tidy, then keep every cut inside it as its own version.
Use one list to check every incoming package:
- EDC: CRF data and related extracts, labeled with the cut they represent.
- SDTM: submitted datasets, kept with the cut they belong to.
- ADaM: analysis datasets, versioned alongside the SDTM cut they derive from.
- Lab: laboratory transfers, grouped separately from EDC and analysis data.
- Documents: protocols, SAPs and related document files, stored with the cut they describe.
The list is deliberately simple. Its job is to stop two common failures: mixing source types in a single “latest” folder, and losing the cut label when a file is renamed. If a file is a derived dataset, it does not belong in EDC. If a file is a lab transfer, it does not belong in documents. If a document describes a specific cut, it should not float above the version structure as an unlabelled attachment.
Register each data cut with its checksum
A data cut is not just a date on a folder. It is a version of the study that people will refer to, compare and defend. For each cut, record a clear label, the source groups included, the upload event and the checksum that belongs to each file revision. The checksum is the fingerprint. It lets the team see whether a file with the same name really contains the same content. If a new upload arrives, it becomes a new revision with its own checksum. Nothing should be overwritten. Older cuts should remain available so that a reviewer can return to the analysis that was run at the time.
This is where many manual filing systems fail. They overwrite the “final” folder, rename the old file to “final_v2” and hope the team remembers which version was used. A stronger pattern is to keep every cut in order, let each person choose which cut they are looking at, and use compare mode between cuts when something needs to be explained. If a checksum changes without a corresponding new cut, that is a signal to stop and investigate before anyone runs an analysis. The same discipline applies to documents: a protocol or SAP page can change, so the document version and the data cut need to be connected in the workspace, not just in an email chain.
Version SDTM, ADaM, lab and documents together
SDTM and ADaM are often discussed as separate deliverables, but in daily review they are used together. A reviewer may start with an ADaM dataset, trace back to SDTM, then check the source in EDC or a lab transfer. That chain only works if every link carries the same cut label. When a new SDTM delivery arrives, do not let it silently become the default while ADaM and lab remain on an older cut. Version the related source groups together, even if their files arrive at different times, and make the cut selection explicit for each person who opens the workspace.
Documents need the same treatment, with one extra requirement: page-level context. Protocols, SAPs and annotated CRFs are not just attachments; they are the place where definitions, visit schedules and analysis rules live. Store them with the cut they describe, and use page-by-page search so a reviewer can find the relevant section without downloading the whole file. Notes and issues should attach to the exact form, field, row or document page they concern. That keeps a question about a lab value next to the lab value, not in a separate spreadsheet. Blinding also has to survive the filing structure. Unblinded content needs its own grant. Without that grant, those files, records and search results should stay hidden, whatever a person’s role.
Verify the source-grouped cut before handoff
Before a cut goes to a broader review, run a short verification pass. Open each source group and confirm the cut label is the one you expect. Check that every file revision has a checksum and that no earlier revision has disappeared. Use compare mode to see what changed between the previous cut and the current one, then confirm that the change is explained by the new transfer or document version. Search a protocol or SAP page by page and confirm that the result appears beside the data it describes. Open a note or issue and check that it is attached to the exact form, field, row or document page. Finally, confirm access by study. People should see only the studies they belong to, with roles such as viewer, collaborator or owner, and unblinded content should require its own grant.
A source-grouped, checksum-versioned workspace is not about making filing look complicated. It is about making the study’s history readable. When EDC, SDTM, ADaM, lab and document files keep their source identity and every cut keeps its own checksum, the team can move from a question to the right file, the right version and the right page without guessing. That is the practical standard to aim for, whether the study is in startup, interim review or final analysis.



