Repository logo
 

Evaluation of routinely collected records for dementia outcomes in UK: a prospective cohort study.

Published version
Peer-reviewed

Change log

Authors

Luben, Robert 
Khaw, Kay-Tee 
Brayne, Carol 

Abstract

OBJECTIVES: To evaluate the characteristics of individuals recorded as having a dementia diagnosis in different routinely collected records and to examine the extent of overlap of dementia coding across data sources. Also, to present comparisons of secondary and primary care records providing value for researchers using routinely collected records for dementia outcome capture. STUDY DESIGN: A prospective cohort study. SETTING AND PARTICIPANTS: A cohort of 25 639 men and women in Norfolk, aged 40-79 years at recruitment (1993-1997) followed until 2018 linked to routinely collected to identify dementia cases. Data sources include mortality from death certification and National Health Service (NHS) hospital or secondary care records. Primary care records for a subset of the cohort were also reviewed. PRIMARY OUTCOME MEASURE: Diagnosis of dementia (any-cause). RESULTS: Over 2000 participants (n=2635 individuals) were found to have a dementia diagnosis recorded in one or more of the data sources examined. Limited concordance was observed across the secondary care data sources. We also observed discrepancies with primary care records for the subset and report on potential linkage-related selection bias. CONCLUSIONS: Use of different types of record linkage from varying parts of the UK's health system reveals differences in recorded dementia diagnosis, indicating that dementia can be identified to varying extents in different parts of the NHS system. However, there is considerable variation, and limited overlap in those identified. We present potential selection biases that might occur depending on whether cause of death, or primary and secondary care data sources are used. With the expansion of using routinely collected health data, researchers must be aware of these potential biases and inaccuracies, reporting carefully on the likely extent of limitations and challenges of the data sources they use.

Description

Keywords

Journal Title

BMJ Open

Conference Name

Journal ISSN

2044-6055
2044-6055

Volume Title

Publisher

BMJ
Sponsorship
Medical Research Council (G1000143)
Cancer Research Uk (None)
Medical Research Council (MR/N003284/1)
MRC (MC_UU_00006/1)
Medical Research Council (MC_UU_12015/1)