The Cochrane Dementia and Cognitive Improvement Group maintains a specialised register of dementia trials.
How is the register maintained?
The register is maintained by the information specialists of the group: Anna Noel-Storr and Candida Fenton.
Step 1: the search for studies
Every month the following sources are searched:
- Medline (via Ovid SP)
- Embase (via Ovid SP)
- PsycINFO (via Ovid SP)
- Web of Science Core Collection (via Clarivate Analytics)
- WHO’s ICTRP meta-registry
- CENTRAL (via the Cochrane Library)
- LILACs (via Bireme)
View the Cochrane Dementia and Cognitive Improvement Group search here.
Step 2: importing and de-duplicating the results
The search results are then imported into Cochrane’s Register of Studies (CRS web) and de-duplicated.
Step 3: screening the results
The searches that are run in the sources listed above are ‘sensitive’ searches. This means that we’ve gone to some effort to try to ensure that no relevant studies are missed. This sensitivity comes at a cost because it means that we also retrieve a fairly high proportion of ‘noise’ (irrelevant records).
The screening process is where we try to weed out the records that are not within domain scope (topic area) of the group, or records that are not randomised or quasi-randomised trials.
We use a combination of machine learning and manual effort in the screening process. Thanks to Cochrane’s Project Transform, we are able to save a good proportion of human effort by using the RCT classifier. This machine learning model has been built to identify RCTs from non-RCTs, and we have been using it since April 2018. Once we’ve imported and de-duplicated the search results, the first part of the screening process is to get the remaining records classified by the RCT Classifier. Records that get a very low likelihood score of being an RCT, get rejected (i.e. no further action is taken on these records as they are assumed to be not RCTs). The remaining records are then manually screened using the triage function in the CRS web. This manual screening is based on assessing the titles and abstracts of records.
Step 4: adding relevant records to the register and to CENTRAL
We then add relevant records to the register and submit them for publication in CENTRAL.