Have surgeons predict their own chance of success. Preferably, gather multiple second++ opinions for each case.
The difficulty of the surgery is based on some combination of the different surgeon's predictions.
A surgeon's stats would then be based on his successful and unsuccessful surgeries, weighted against the difficulty.
Using this method, not only can you still rank the surgeons based on the successes of their surgeries, but also on the accuracy of their predictions, which will be in larger sample size and also perhaps a better indicator of their mastery of medical knowledge.