How many subjects should i enroll?

I am planning a prospective study that examines whether adding an ultrasound machine to a general physician clinic reduces the amount of referrals and blood tests done by that same physician.
For several months I want to look at 4 different physicians, give 2 of them an ultrasound machine, and in the middle of the study period give the machines to the other 2 physicians. I would then compare each physician's amount of referrals with/without the machine.
1. Which statistical tests should I use? I was thinking about Mann-Whitney.
2. What is the amounts of subjects I need to enroll in order to get significant results?


Less is more. Stay pure. Stay poor.
Are you going to make any comparisons across physicians? If so, will you need to control for differences in patients per provider - issues with patient exchangeability?

Also, the biggest question is usually, what is the exact outcome and how big of an effect do you expect?