Diabetes Clustering

M.van_Smeden@lumc.nl 2018-03-26
to diabetes-endoc., f.harrell, ddahly
Dear editor,

On behalf of my co-authors, we ask you to consider our commentary article (attached to this email). This comment is motivated by a recent publication in The Lancet Diabetes and Endocrinology (doi: 10.1016/S2213-8587(18)30051-2) in which 5 subtypes of diabetes were identified using cluster analysis. Because we anticipate an increase of such cluster analyses in search of more personalized diabetes diagnosis, we took the opportunity to reflect on the ability of such analyses to identify “true” diabetes subtypes. We argue that cluster analyses are likely to identify spurious (not “true”) clusters in any given data set where the variables included are correlated. We therefore advise to be cautious with interpreting clusters and provide suggestions for alternative approaches to identify important and “true” diabetes subgroups.
Sincerely,
Maarten van Smeden
(Email: M.van_Smeden@lumc.nl)

See annMed2018-05.pdf https://www.thelancet.com/journals/landia/article/PIIS2213-8587(18)30124-4/fulltext

Magical Clusters Blog

https://darrendahly.github.io/post/cluster/

Other

• https://psyarxiv.com/hqxzd
• Gustav Nilsonne gustab.nilsonne@ki.se Our letter is out now, see below for a “SharedIt link” from the publisher for social media etc. And here is the doi:
  https://doi.org/10.1038/s41587-020-00768-5
  And here is the authors’ response:
  https://doi.org/10.1038/s41587-020-0738-2
  I am happy to see that they have taken the criticism seriously concerning pre-post measures of the HAMD, and they have done a reanalysis following your recommendations, leading to an effect shrinkage from r = 0.60 to r = 0.49.
  For the external validation analysis, they are making the somewhat disingenious claim that they were never actually trying to assess predictive performance in the new sample (even though their fig 4 which shows these data is titled “prediction of treatment outcome”).
  Overall they are arguing with animation, and claim to refute our criticisms completely.