Provided the data includes good representative sampling of all degrees of severity (from 0% to 100%), then the above makes very good sense. Cluster analysis is a data mining technique that starts off with no preconceptions, teasing out data clusters that may not be at all obvious from conventional analysis; highlighting similarities within each cluster, whilst highlighting differences between clusters. But of course the "cr*p in, cr*p out" principle applies, so the information out can only be as good as the data in. Would have to be very careful that those at the high end of the severity spectrum are adequately represented, and not just a token representation. And because there must be no preconceptions about where the spectrum terminates at either end, the low-to-no severity end of it must also be sensibly represented. The point here is that the cluster analysis itself should be teasing out what the spectrum comprises, and what its limits are, and not be presumed by researchers constraining the input data according to their own preconceptions.
It is pretty much agreed by all that ME/CFS encompasses a number of sub-types ... trouble is no-one has a clue what they are or how many. Cluster analysis could well unlock the secret of what those subsets are, along with what distinguishes them from each other, which would of course be a huge leap forward.
The minutes make good sense and are actually really encouraging, so it is just such a pity that MEGA is dragged right down by the presence of EC, and Prof Holgate's continuing support for her and her ways.