Run-time

43

admin — Fri, 20 Mar 2020 15:42:58 +0000

43 admin Fri, 03/20/2020 - 22:42

Known/emergent issues with the data dictionary are documented and reported in an accessible manner.

Run-time

Representation

42

admin — Fri, 20 Mar 2020 15:42:06 +0000

42 admin Fri, 03/20/2020 - 22:42

Data irregularities are reported back to data collectors in a systematic and timely process. There is a standard operating procedure for data irregularities to be reported back to the data collectors and for documentation of the resolution of the issue.

Examples

1: Queries in place to identify the following types of scenarios: For 9 of 10 study sites a mandatory and critical variable is only missing in 5% of the subjects. For site 10 data is found to be missing in 30% of the subjects. For a single site study, several personnel have been collecting longitudinal study data with low numbers of missing data. A pattern of increasing numbers of missing data has been identified with a newly hired data collector.

2: Variables known to be problematic are regularly monitored by curation team and delays in site resolution of issues are reported to governing committee.

3. Aggregate data are examined for inconsistencies such as lower average mortality in patients with GCS 3 than GCS 4-5.

Run-time

Correctness

41

admin — Fri, 20 Mar 2020 15:41:36 +0000

41 admin Fri, 03/20/2020 - 22:41

Scoring of tests is checked. Scoring is performed automatically where possible.

Run-time

Correctness

40

admin — Fri, 20 Mar 2020 15:40:59 +0000

40 admin Fri, 03/20/2020 - 22:40

Reliability checks have been performed on variables that are critical to research hypotheses, to ensure that information from multiple sources is consistent.

Examples

1: Database rule checks in place to identify conflicts in data entries for related or dependent data collected in different CRFs or sources.

2: Inter-rater scoring of MRIs is done independently by 3 experts on 10 cases per month.

Run-time

Plausibility

39

admin — Fri, 20 Mar 2020 15:40:21 +0000

39 admin Fri, 03/20/2020 - 22:40

Source data validation procedures are in place to check for agreement between the original data and the information recorded in the database.

Examples

1: Site visits for audit of subject hospital records against CRF and eCRF entries.

2: Submission of redacted hospital records for audit against eCRF entries.

3: Independent double entry of any data transferred to the database from paper form, either for full dataset or for representative subset.

Run-time

Correctness

38

admin — Fri, 20 Mar 2020 15:39:36 +0000

38 admin Fri, 03/20/2020 - 22:39

Systematic and timely measures are in place to assure ongoing data accuracy.

Examples

1: Schedule of standard reports for data quality checks.

2: Ad hoc reports for identification of new issues.

Run-time

Plausibility

37

admin — Fri, 20 Mar 2020 15:39:00 +0000

37 admin Fri, 03/20/2020 - 22:39

Individual data elements should be checked for missingness. This should be done against pre-specified skip-logic / missingness masks. This should be performed throughout the study data acquisition period to give accurate ‘real time’ feedback on completion status.

Examples

1: Rules in place to check whether the apparently missing element should be completed or not for a particular study stratum or time point.

Run-time

Plausibility

36

admin — Fri, 20 Mar 2020 15:36:44 +0000

36 admin Fri, 03/20/2020 - 22:36

There is centralized monitoring of the completeness and consistency of information during data collection.

Run-time

Completeness

35

admin — Fri, 20 Mar 2020 15:35:59 +0000

35 admin Fri, 03/20/2020 - 22:35

Automated variable transformations are documented and tested before implementation and if modified.

Examples

1: For example, testing is carried out if fields such as ‘years of education’ and ‘level of education’ are automatically combined into a new variable summarising educational status.

2: Coding is tested on a subset of data, ideally by multiple members of the study team.

3: There is a centralised log / version management of all transformations used.

Run-time

Representation

34

admin — Fri, 20 Mar 2020 15:35:28 +0000

34 admin Fri, 03/20/2020 - 22:35

Proxy responses for factual questions (such as employment status) are allowed in order to maximize completeness.

Examples

1: Protocol is designed to allow for surrogate to respond to selected questions when patient is too impaired to respond. CRF indicates when surrogate is responding.

Run-time

Completeness