Do the "standard clinical tests" include subjective questionnaire-derived data? If so, given the layers of people involved in collecting data from such a huge number of people, let alone training some to collect it, it feels like the data itself would almost certainly be subject to considerable interpretative variation at source. Presumably this would then skew any reported clustering; I do not know how catastrophically. Also, I do not know how well cluster analysis copes with subjectively derived data anyway.
I would think it essential to have a data scientist on the team with a proven track record in data mining technologies, at least in a consultancy role, throughout the whole study life cycle. Too much at stake, and too many chances to mess up, to get that aspect wrong.