Taken statewide, the census data is just a constant and 1) can be done with FREQ. You can test the proportions against a hypothetical distribution (like uniform). If you have populaiton by year, you will get a more precise estimate.
Ditto 2). You will need to know when the new procedure started.
To look at 3) , I would use LOGISTIC regression; it does not require the Census data. You could use year as a continuous predictor. Then the type of procedure could be another binary predictor.
You may not have enough sample size to estimate hospital effect, let alone provider effect, in 3.
Doc Muhlbaier
Duke