Thread by @melodem_group

announcing this week's topic - #validity of dementia diagnosis, and how does it relate to use of administrative databases for #dementia research!

1/n bit of basic #epidemiology first: Gold standard measures are used to assess the validity of alternative measurement #epitwitter

2/n BUT in #dementia research there is no clear gold standard --> so the question remains - what do we typically ise as the "gold standard" and what does it say about the alternative =administrative dabatases measure? Stay tuned for more!

3/n in practice the #dementia gold standard is neurocognitive&physical assessment, complemented with proxy interviews, &diagnosis assigned by consultation of multiple specialists. so as @alexjweigand wrote! #biomarkers are not typically not part of it

4/n the above described gold standard diagnosis is - time and resources expensive! thus applied in #longitudinal #research cohort samples

5/n arriving at first issue - #selectivesurvival! those who die between follow-ups are missing the gold standard assessment & death is random! So how does this missing not at random affect affect using the gold standard comparison for validity?

6/n study enrollment, follow-up attrition, selective survival change the composition of the sample, in which we do the gold standard assessesment - participants are more likely healthier, wealthier and more educated

7/n some findings of validity studies?ncbi.nlm.nih.gov/pmc/articles/P… which finds that those with lower education were less likely to have dementia diagnosis in Medicare claims and on death certificates. So low educated ppl - more likely false negatives

8/n ncbi.nlm.nih.gov/pubmed/19542620 newer study comparing Medicare & Aging Demographics and Memory Study: education was not predictor of agreement between data, but again majority of false -ves in medical claims were those with low education

9/n --> those with lower ed more likely to be missed in administrative data --> this could result in underestimation of #health #inequalities in dementia and other issues. We know even less ab validity of data sources with regards to #ethic #racialminorities

10/n however several studies report that non-overlap in dementia cases between data sources eg. ncbi.nlm.nih.gov/pubmed/14505776 or ete-online.biomedcentral.com/articles/10.11…

11/n these studies recommend exploiting the diversity of the sources "each source contributes unique and essential information to the identification of dementia" which may be true but how does it help is to know what dementia is?

12/n and that gets us full circle - hard to know since we do not have a "perfect" gold standard. so maybe triangulation of findings is the way? Thoughts?

13/13 here is a review of 27 studies looking at #dementia cases identification in administrative #healthcaredata - PPVs range 33%–100%, majority >75%; sensitivity range 21% to 86%.
sciencedirect.com/science/articl…