Screening guide

The guide states how records are screened and how decisions are exported for the analysis notebooks. The criteria come from the protocol and are a draft.

Stages

1. Title and abstract. Each deduplicated record gets one decision, include or exclude. When the title and abstract do not allow a decision, include the record.
2. Full text. Each included record’s full text is retrieved. A report that cannot be retrieved is counted as not retrieved. Each retrieved report gets one decision, and each exclusion gets one reason from the list below.

Each included report is assigned to a source of evidence. Reports of the same study share one source_id.

Criteria

A source is included when all of these hold.

• It concerns the sharing or reuse of research data.
• It reports factors that influence sharing or reuse, including barriers, facilitators, and governance mechanisms.
• It concerns health research (tier 1), or it is a review or synthesis of research in other fields or across fields (tier 2).

A source is excluded when it concerns only public health data, other health systems data, or only clinical care data. The rule for routine or clinical data reused for research is not yet set.

Exclusion reasons

Record one reason code per excluded report at full text.

Code	Reason
not-data-sharing	Not about the sharing or reuse of research data
not-factors	Reports no barriers, facilitators, or governance mechanisms
not-research-data	Concerns only public health, other health systems, or clinical care data
tier2-not-review	Research in other fields, and not a review or synthesis
language	Outside the language limit, if one is set
date	Outside the date limit, if one is set
other	Any other reason, explained in a comment

02-screening stops with an error when a full-text exclusion has a reason code that is not in this list.

Export formats

Export decisions from the screening tool into data/raw/screening_exports/ as CSV files with lowercase values. 02-screening reads them. The final files hold one row per record, with disagreements between reviewers already settled.

title_abstract.csv

Column	Content
record_id	The identifier from deduplicated_records.csv
decision	include or exclude

full_text.csv

Column	Content
record_id	The identifier from deduplicated_records.csv
source_id	The source of evidence the report belongs to. Optional. Defaults to record_id.
retrieved	yes or no
decision	include or exclude. Blank when retrieved is no.
reason	An exclusion reason code. Required when decision is exclude.

Reviewer files

When more than one person screens, export each reviewer’s own decisions as well. Name them title_abstract_<reviewer>.csv and full_text_<reviewer>.csv, for example title_abstract_r1.csv. They have the columns record_id and decision, and they may cover only a sample, such as a calibration set. With two or more reviewer files for a stage, the notebook reports percent agreement and Cohen’s kappa for each pair of reviewers and lists the records they disagree on. Reviewer files do not change the counts. The final files do.