Building eCRFs and data dictionaries to a standard

A 231 variable data dictionary and the case report forms built from it, aligned to CDISC conventions and designed to be reused.

Two audiences, one artefact

A case report form has to work for the coordinator filling it in at a clinic and for the statistician analysing it eighteen months later. Those two want different things, and most badly designed forms are badly designed because somebody served only one of them.

The coordinator wants the fewest fields, in the order the visit actually happens, with nothing that requires leaving the room to answer. The statistician wants structured values, consistent naming, no free text where a code would do, and variables that map cleanly to an analysis dataset without a translation layer.

Standards exist to settle that argument before it starts.

What alignment to CDISC actually buys

Working to CDASH conventions is often treated as a compliance tax. It is not. It is the difference between a dataset that maps to SDTM mechanically and one that needs a bespoke transformation written, validated and maintained for every study.

Naming and structure decided at dictionary stage cost nothing extra. The same decisions deferred to database lock cost a mapping specification and the risk of getting it wrong under time pressure.

The dictionary before the form

Across studies, the sequence I hold to is dictionary first, forms second.

Every variable defined with its name, label, type, format, permitted values, units, required status and derivation. Controlled terminology selected rather than invented, with any study specific extension documented as an extension. Only then are forms built, as a rendering of a dictionary that already exists.

Building forms first and documenting afterwards produces a dictionary that describes what was built rather than what should have been built, and the difference shows up as inconsistency across studies that nobody has time to reconcile.

Designing for reuse

A 231 variable dictionary for one study is a deliverable. The same dictionary structured so that its common domains carry into the next study is an asset.

Demographics, medical history, concomitant medications, adverse events and vital signs are broadly the same everywhere. Treating them as study specific work every time is the most common waste in clinical data management. Standardising them once and varying only the study specific domains turns study build from authorship into assembly, and means the reused parts have already been tested against real data.

The judgement call that recurs

Sponsors ask for study specific variations on standard domains, often for reasons that are real but minor. Every accepted variation weakens the standard for every study after it.

My default is to push back once, with the cost made explicit, and then accept the variation gracefully if the sponsor still wants it. Winning that argument every time is not the goal. Making sure it is a decision rather than a drift is.