The ideas presented here seem pretty sensible. However, there is a risk that if such data is collected, it may later be used to compare surgeons to each other.

The ideas presented here seem pretty sensible. However, there is a risk that if such data is collected, it may later be used to compare surgeons to each other. One of the reasons that is risky is that good surgeons tend to deliberately take on more complex and riskier operations than those less capable, and therefore have statistically worse outcomes. There can also be geographical biases if statistics are summarised, such as certain hospitals having poor outcomes because they happen to get a lot of gunshot wounds, or stabbings, or the local population is older than average, or childbirth rates are higher, etc. It would difficult to control for all possible such variables, and ever more difficult to explain the methodology to the public. It's probably a universally good idea as long as there are safeguards to ensure that the data is not shared generally, which might lead to misinformed reactions and subsequently, target-chasing (see: Goodhart's law). Using the data purely for personal improvement seems like it should be effective. I think it's reasonable to assume that most surgeons would want to self-improve if they can. Also surgeons tend to be especially competitive, so creating a competition against themselves could be a good motivator in itself. The data could perhaps be used to weed out poorly performing surgeons too. I believe the right way to do that would be to share the data anonymously with peers who are able to interpret it properly. These peers could then flag any worrying anomalies in the data that can't be explained away, which should trigger a follow-up investigation of the individual concerned. This could perhaps stop the next Harold Shipman, as well as weeding out incompetence. Of course, it may not be possible to usefully anonymise case data since some of the more complex cases may be sufficiently unique that the surgeon could be personally identified from the description. This would only work if case can be classified broadly enough to avoid personal identification and narrowly enough to provide enough information for peer review.