Back

Feasibility of linking markers of dementia-related health in primary care medical records to cognitive function assessed in a specialist dementia service

Marshall, M.; Campbell, P.; Bailey, J.; Chew-Graham, C. A.; Croft, P. R.; Frisher, M.; Hayward, R.; Negi, R.; Rathod-Mistry, T.; Singh, S.; Robinson, L.; Sumathipala, A.; Thein, N.; Walters, K.; Weich, S.; Jordan, K. P.

2022-10-13 psychiatry and clinical psychology
10.1101/2022.10.11.22279756 medRxiv
Show abstract

ObjectivesTo assess the feasibility of linking and comparing markers of dementia-related health recorded in primary care electronic health records (EHR) to assessments of cognitive function undertaken in a specialist dementia service. MethodsOne thousand patients in a UK secondary care specialist dementia service were invited to take part. Primary care EHR were requested from 72 general practices of consenting patients. Sixty-three previously established individual markers within 13 broader domains of dementia-related health were then extracted from primary care EHR and compared to cognitive assessments scores recorded in the dementia service EHR. Results258 (26%) patients consented to take part. At least one cognitive assessment score was recorded for 242 (94%) patients, but primary and secondary care EHR records could only be linked in 93 patients. 56 of these 93 patients had two cognitive assessments scores at least 12 months apart. In the patients with data available for analysis individuals with a higher number of markers and domains recorded in their primary care records had lower mean cognitive assessment scores (range 1.6-2.1 points), and after adjustment for earlier cognitive scores (range 2.0-2.5 points), indicating poorer cognitive function, although differences were not statistically significant. ConclusionThis feasibility study highlights the challenges in obtaining consent and linking primary and secondary care EHR in dementia, and in extracting cognitive function scores from dementia service EHR.

Matching journals

The top 4 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.