Align multicentre data collection before pooling

Separate the common protocol from local implementation

A shared protocol does not guarantee that every site collects the same quantity in the same way. Local systems may differ in units, visit scheduling, available tests, coding conventions or responsibility for endpoint confirmation. Begin with a comparison of actual workflows rather than assuming that a common variable name establishes comparability.

The FDA E6(R3) Good Clinical Practice guidance provides a current framework for trial quality and reliable information. Apply the relevant study requirements and local approvals; this practical alignment exercise does not replace those obligations.

Build a site concordance table

For each critical variable, record the definition, source, timing, units, coding and responsible role at each site. Identify differences that are equivalent after a defensible conversion, differences requiring a protocol amendment and differences that cannot be harmonised without changing the question.

Include operational details such as specimen processing, device versions, recall periods and whether an endpoint is adjudicated centrally or locally. Version the agreed rules and distribute them through the study's approved process. Preserve local source values and the harmonisation mapping so transformations remain auditable.

Hypothetical example

Two fictional sites both submit a field called “baseline weight.” One measures it before treatment on a calibrated scale; the other imports the most recent value from a clinical record, sometimes months earlier. Converting kilograms to a common unit would not resolve the timing and source difference.

The coordinating team clarifies the intended baseline definition, checks whether compliant measurements can be recovered and records any remaining noncomparable observations. The example illustrates why semantic alignment must precede pooled analysis.

Test the shared rules before expansion

Use a small synthetic or appropriately governed test dataset from each site's workflow. Check valid ranges, missing codes, visit windows and endpoint derivations against the shared specification. Confirm that corrections can be traced and that sites receive feedback on recurring discrepancies.

Document deviations, their effect on analysis eligibility and any site-specific sensitivity analyses. Do not silently force all data into one schema when the underlying meaning differs. Use the data dictionary change-impact worksheet and collection device-change worksheet to preserve decisions. Report the participating sites, standardisation procedures and important remaining differences so the pooled result has an interpretable measurement basis.

Source verification: 6 November 2026. Examples are hypothetical.