Clinical data management turns what happens at a site into a dataset you can analyse and defend. This guide walks the steps in order, names the document produced at each one and shows where small teams can keep the process light.
Free sandbox · No credit card · 21 CFR Part 11 aligned
Month
DMP, CRF, specifications
Database build, testing
Entry, queries, reviews
Reconciliation, sign-offs
Exports, records
The process in brief
Overview
Clinical data management, often shortened to CDM, covers everything that happens to trial data between the protocol and the analysis dataset. It collects data, checks it, corrects it through documented queries, reconciles it with other sources and finally freezes it for analysis. The goal, in the language regulators use, is data that is attributable, accurate and complete, with a record of every change. The more your process depends on memory and email, the harder that goal is.
The process is the same shape for a 20-participant pilot and a pivotal Phase 3 study. What changes is the number of people involved and the amount of formality around each step. A small sponsor often has one person who is the data manager, the study coordinator and the person who answers queries. That is workable if the steps below are still done, even in compact form. For software that supports the process, see clinical data management software and software for clinical data managers.
The steps
Each step names the typical output so you can see what an auditor would expect to find.
Output: data management plan. Covers roles, systems, CRF completion rules, edit check strategy, query turnaround, external data, coding approach, reconciliation, lock criteria and archive. See the data management plan template.
Output: approved CRFs and completion guidelines. Map each protocol endpoint and safety need to a field, choose clear units and options, and write guidance for sites. The CRF builder guide and eCRF completion guidelines help.
Output: data validation specification. List every range, consistency and missing-data check, with its priority and message, plus calculated fields. See edit checks software.
Output: configured study. Forms, visit schedule, roles, sites and checks are set up from the specifications. AI can draft a first version from the protocol, and a human reviews it.
Output: test scripts and results. Run scenarios that include errors, missing data and rare visits. The UAT guide shows how, and the go-live checklist is the gate.
Output: release record and training log. Approved forms become live, users receive roles and a short training, and the plan for site support starts.
Output: query log, review reports, status metrics. Data are entered, edit checks fire, queries are raised and answered, and data managers review listings. See query management and clinical data cleaning.
Output: reconciliation records, deviation classification, sign-offs. Compare safety data with the safety database, bring in external data such as central lab results, close queries and obtain investigator sign-off.
Output: lock approval, final exports, archive package. Freeze the database, export the dataset with a data dictionary and file the records. See archiving and retention.
Build a study, enter test data, trigger an edit check, raise a query and export the result. Free, with every feature and no credit card.
Roles
In a larger organisation, a clinical data manager owns the plan, the specifications and the cleaning. A clinical programmer or the EDC builder configures the database. A biostatistician defines what the final dataset must contain and often reviews the plan. Clinical monitors look at source verification and site performance. Medical reviewers look at safety and coding. In a small study these roles collapse into two or three people, and the risk is that nobody is responsible for reviewing the data as a whole. Assign the review explicitly even if the same person does several jobs.
Role-based access should reflect these responsibilities. Capture offers researcher, site coordinator and admin roles, with participant identity visible to site coordinators and coded identifiers for researchers, so the people who clean the data do not need to see names.
Sites enter data and answer queries, so their experience is a cleaning metric in itself. Short, clear forms and good completion guidance reduce queries before they exist. If the same query comes back from every site, the form is the problem.
Documents
| Document | Created at step | Why an auditor wants it |
|---|---|---|
| Data management plan | 1 | Shows the process was defined before data arrived |
| CRF and completion guidelines | 2 | Shows what was collected and how sites were told to complete it |
| Data validation specification | 3 | Shows which checks existed and why |
| UAT scripts and results | 5 | Shows the system was tested for its intended use |
| Release and training records | 6 | Shows only approved forms went live and users were trained |
| Query log and review reports | 7 | Shows how discrepancies were found and resolved |
| Reconciliation and sign-off records | 8 | Shows the dataset was complete and confirmed |
| Lock approval and archive package | 9 | Shows the data were frozen deliberately and kept readably |
Requirements vary by regulator and sponsor contract. Treat this as a baseline, not a legal list.
In Capture
Draft the visits and forms from a protocol with the AI study builder, add skip logic, calculated fields and edit checks that raise an auto-query when a rule is broken, then test everything free. Forms move from draft to approved, and approved forms are locked for live use. Each field has an audit trail, SDV can be set per field, and exports include a data dictionary, with SDTM export as SAS XPT with Define-XML. Medical coding dictionaries are not covered on this page, so plan coding with your usual provider.
Pitfalls
Write a short data management plan before first enrolment and update it by version.
Trace each field to a protocol need and remove the rest.
Use scripted scenarios including wrong entries, then record results.
Review listings and open queries on a schedule during conduct.
Agree format and timing for lab and device data in the plan.
Use a checklist with owners and approvals.
Write the data management plan, design the CRF, specify checks, build the database, test with UAT, go live, collect and clean data during conduct, reconcile and prepare for lock, then lock, export and archive.
A document that defines how data will be collected, checked, corrected, reconciled and locked for a study, including roles, systems, query handling and lock criteria. It is written before enrolment and versioned.
It runs for the life of the study, with set-up before enrolment, cleaning during conduct and lock afterwards. Continuous cleaning shortens the time from last visit to lock.
Often one or two people cover several roles. The key is to assign data review explicitly so that someone looks at the dataset as a whole, not only at individual forms.
Remaining queries are closed, external data and safety data are reconciled, deviations are classified, investigators sign off and lock approvals are obtained. A checklist keeps it orderly.
Capture supports building and testing the study, edit checks with auto-queries, query handling, per-field SDV, a field-level audit trail, exports with a data dictionary and lock workflows. Try the steps free in the sandbox.
Keep exploring
Clinical data management software
Tools for the process.
Clinical data cleaning guide
Cleaning in depth.
Data management plan template
Start the plan.
Database lock checklist
The last step.
Software for clinical data managers
Daily workflow.
EDC UAT and sandbox
Testing before go-live.
CDISC SDTM export
Dataset delivery.
Free sandbox with every feature. No credit card, and you pay only when you go live with real participants.