Cluster analysis with longitudinal data

[K MacQuarrie]

Are there commands in Stata that are similar to the kml3d package in R (see e.g. Genolini et al 2015) for cluster analysis of longitudinal data? I would like to conduct unsupervised (to begin with) machine learning to identify an unknown number of clusters based on common characteristics of women's trajectories, analyzing approximately 60 months of data for each case. My understanding is that using the cluster kmeans commands (Makles 2012) is inappropriate with longitudinal data and won't help me achieve my research objectives. Have there been any developments since cluster kmeans were introduced? I couldn't find what I am looking for elsewhere on this forum. I'm hoping to use Stata for this study rather than R since that is what the rest of my team uses and I prefer it as well. Any leads? Thanks!

Kerry MacQuarrie,
Avenir Health

Genolini, C., X. Alacoque, M. Sentenac, and C. Arnaud. 2015. "Kml and Kml3d: R Packages to Cluster Longitudinal Data." Journal of Statistical Software 65 (4): 1-34
Makles, A. 2012. "Stata Tip 110: How to Get the Optimal K-Means Cluster Solution." The Stata Journal 12 (2): 347-351.

[Jesper Eriksson]

Also very interested if there exists a similar function in Stata. I have the same problem, 700 patients observed once daily during 28 days, where each day have data on 7 different measurements. My approach has been to (in Stata) first perform PCA and then kmeans. I prefer Stata as well so I'm hoping there is a way to do longitudinal clustering. Anyone knows?

All the best,

Jesper Eriksson
Stockholm, Sweden