Why can't you have both? Maybe a different design would become apparent while developing the product, as opposed to being discovered in the analysis phase. (I'm assuming you are talking about a process with a large and separate analysis phase).
The worst case is millions of dollars spent to build a product only to discover that there are issues late in the testing phase. Wouldn't is be less risky to experiment and validate assumptions earlier in the cycle?
Agile as opposed to methodology X is a bit fuzzy here, as I'm sure testing is extensive and expensive and the feedback cycle is inherently longer, so change is more expensive.
However, it seems that early feedback and validation of design is even more important in these circumstances, since changes late in the product development process and orders of magnitude more expensive than changes early in the process.
Yeah, I'm not talking about process so much as languages anyway though. For all I know (I've never actually developed software for a medical device) its worthwhile to build an actual prototype in a more flexible language first. But if my code could kill someone, I want my tools to optimize correctness over flexibility in the final implementation.
