Influence of Surgical Subspecialty Training on In-Hospital Mortality for Gastrectomy and Colectomy Patients

  Table 1. Mortality and Provider Characteristics for Colectomy and Gastrectomy Procedures in New York State-1998 to 2001

  Table 1. Mortality and Provider Characteristics for Colectomy and Gastrectomy Procedures in New York State-1998 to 2001

  Table 2. Patient Characteristics: Prevalences and In-hospital Mortality Rates-1998 to 2001

  Table 3a. Colectomy: Patient Characteristics, Hospital Volume, and Surgeon Volume Stratified by Subspecialty Versus Nonsubspecialty surgeons-1998 to 2001

  Table 3b. Gastrectomy: Patient Characteristics, Hospital Volume, and Surgeon Volume Stratified by Subspecialty Versus Nonsubspecialty Surgeons-1998 to 2001

  Table 3a. Colectomy: Patient Characteristics, Hospital Volume, and Surgeon Volume Stratified by Subspecialty Versus Nonsubspecialty surgeons-1998 to 2001

  Table 3b. Gastrectomy: Patient Characteristics, Hospital Volume, and Surgeon Volume Stratified by Subspecialty Versus Nonsubspecialty Surgeons-1998 to 2001

  Table 4a. Colectomy: Number of Hospitals, Number of Patients, and Mortality Rates for Categories of Hospital and Surgeon Volume in New York State-1998 to 2001

  Table 4b. Gastrectomy: Number of Hospitals, Number of Patients, and Mortality Rates for Categories of Hospital and Surgeon Volume in New York State-1998 to 2001

  Table 2. Patient Characteristics: Prevalences and In-hospital Mortality Rates-1998 to 2001

  Table 4a. Colectomy: Number of Hospitals, Number of Patients, and Mortality Rates for Categories of Hospital and Surgeon Volume in New York State-1998 to 2001

  Table 4b. Gastrectomy: Number of Hospitals, Number of Patients, and Mortality Rates for Categories of Hospital and Surgeon Volume in New York State-1998 to 2001

  Table 5. Colectomy and Gastrectomy: Number of Patients and Mortality Rates for Subspecialty and Nonsubspecialty Surgeons in New York State-1998 to 2001

  Table 2. Patient Characteristics: Prevalences and In-hospital Mortality Rates-1998 to 2001

Authors and Disclosures

Mark A. Callahan, MD*, Paul J. Christos, MS, MPH*, Heather T. Gold, PhD*, Alvin I. Mushlin, MD, ScM*, John M. Daly, MD ^†

From the *Department of Public Health, Weill Medical College of Cornell University and New York Presbyterian Hospital, New York, NY; and ^†School of Medicine, Temple University, Philadelphia, PA.

Sidebar: Discussion

Dr. Timothy J. Eberlein (St. Louis, Missouri): This is an excellent presentation. This Association has had several papers demonstrating improved outcomes for either high volume surgeons or high volume hospitals actually presented by several of the past presidents of this Association. However, the manuscript that opened this meeting demonstrated the difficulty in interpreting administrative databases and the fact that thoughtful analysis is required to use these databases in establishing substantive norms and setting standards.

In the present study the authors use an administrative database to assess the relationship of volume (both surgeon and hospital) and specialty training as well as with an end-point of in-hospital mortality. Therefore, patient selection and detailed comorbidities are not well assessed.

My first question deals with the issue of the criteria for defining subspecialty training.

The authors found 61 of 2,651 surgeons were subspecialty trained for colectomy as defined as membership in the Society of Colorectal Surgeons. How did they define members of the Society of Surgical Oncology who also perform colectomies?

Similarly for gastrectomy. 68 of 1,387 surgeons were members of the Society of Surgical Oncology. But what about other subspecialty organizations such as SAGES?

Similarly, what is the accuracy of the comorbidity data used in this administrative database? For example, in my own institution, we have had extensive difficulty having the hospital document comorbidities accurately for our billing purposes.

While intuitively I agree with the authors that subspecialty training or specialization as well as volume clearly makes a difference in outcome, why wouldn't gastrectomy have shown a greater significant difference?

Finally, how much impact did the 1 highest volume hospital have in the overall results? We saw from Dr. Brennan's Presidential Address the outstanding results his institution demonstrated having high volume surgeons, a high volume institution, as well as subspecialty training.

I want to congratulate the authors for presenting a very well-written manuscript and thank the Association for the privilege of discussing this paper.

Dr. Mark A. Callahan (New York, New York): Let me first address the issue of defining subspecialty training. We agree there are limitations with our definition. We used the database of membership in these societies to assign subspecialty versus nonsubspecialty, realizing there is some overlapping in both the skill set and the training for these 2 groups. However, that served as a proxy for our analysis.

That relates to your third question about why we didn't see a more significant difference in the gastrectomy data. Gastrectomy data was limited by the fact that, number one, we did have a relatively limited number of surgeons performing that we called subspecialty surgeons. And secondly, there are many fewer of those procedures done annually in New York state and therefore the number of mortality events was lower, and that affects our statistical power. Now, the odds ratio we found was .75, which says that you have about a one-third less chance of dying if a subspecialty trained surgeon does your operation by our definition, and had we had more observations I think we would have achieved significance statistically.

As far as the accuracy of comorbidity data, I am shocked to hear that a hospital doesn't accurately code that, actually. We have never seen that in our institution. The reality is that all these data sets suffer from the same problem, all these administrative sets. There are biases to overcoding and undercoding for comorbidities obviously driven by staffing issues, accuracy issues, and payment pressures.

All administrative data sets do suffer from those issues. The fact that we had a fairly large data set with 55,000 collective patients across 200-some-odd hospitals, you hope that to some degree you can wash out the over- and the undercoding within the data set.

That being said, there were a couple of interesting things that came out. Subspecialty trained surgeons tend to operate on more patients with metastases in our data set. And that came through very clearly and I think that is probably a real observation.

Now, that may be driven by the issue of high volume hospitals and how do they impact on these outcomes? I am at Cornell, and right next to us is Sloan-Kettering. Obviously that is a very high quality institution which is doing a number of these operations.

It turns out that the high volume hospitals, which there is a limited number of them across New York state, which tend to be university hospitals, had the best outcomes, had the most subspecialty trained surgeons. So there is a clearly effect there of being at those institutions and having those surgeons operate on you.

Dr. Carlos A. Pellegrini (Seattle, Washington): I am still troubled by the same segue that Dr. Eberlein brought up. I am not convinced that you can isolate in your study the training issue from the volume issue. Instead I see subspecialty training as 1 component of outcome where volume, dedication to a disease process and other factors also play a role. The only way that you can demonstrate your point would be if you had somebody with subspecialty training who was working in a low-volume hospital, was doing a low volume of operations and still had excellent outcomes. Did you find any such situation in your study?

I believe that when you look at a high volume surgeon in a high volume hospital, if you add subspecialty training you may observe better results but I think that is a function of the volume. Were you ever able to see an independent effect of subspecialty training when an individual had not frequently practiced a given operation?

Dr. Mark A. Callahan (New York, New York): That is a very important question that Dr. Pellegrini is raising. It goes to the heart of the type of analysis that we were doing statistically. And in a multivariate analysis, you have got subspecialty trained physicians - not a lot of them, but we have some in our database - who are doing low volumes of these procedures. And you can compare that individual statistically to a nonsubspecialty trained physician also doing low volume and see a difference in their mortality.

You have also got the additional from high volume people and they are getting even better outcomes because of the subspecialty training. That is the advantage of the type of analytic approach that we took, because you have got surgeons who cross all our volume domains and who also cross all our training domains, if you will, and you are allowed to sort of tease that out statistically within the data set.

We are limited by the number of surgeons who can go into each of these categories statistically. And there was a very good paper presented earlier this morning where the discussion was we can only put 8 or 9 variables into our model because of the limited number of patients. We had a larger number of patients and we still were limited by the number of variables we can analyze statistically.

Dr. Henry A. Pitt (Milwaukee, Wisconsin): When we looked at the volume effects for liver resection in the state of Maryland a number of years ago, we subdivided our patients into major versus minor operations, which was important in showing differences. Therefore, my question is whether you have data on the subset of total gastrectomy or the subset of proctocolectomy patients where you would anticipate worse outcomes? What was the effect of subspecialty training in those patients undergoing even higher risk operations?

Dr. Mark A. Callahan (New York, New York): An important question. We did do a subgroup analysis based on some of the codes but not all of the codes. We did see the effect in some of the more complicated operations. The problem, of course, is that you lose numbers in those, you lose power statistically, and therefore our confidence intervals got wider.

Dr. Paris Tekkis (Cleveland, Ohio): I would like to congratulate the authors for this excellent presentation. The Association of Coloproctology of Great Britain and Ireland have recently performed a similar study and I would like to share the similarities of their results in relation to this presentation.

A total of 8,077 major colorectal procedures were evaluated with similar outcomes, however in addition to the risk factors used in your presentation, 2 further variables were found to be important in the risk stratification process. The first risk factor was cancer staging. Stage IV disease was associated with a 2-fold increase in operative mortality in comparison with Stage I, II or III. The second risk factor was mode of presentation. Emergency surgery was associated with a doubling of the operative mortality over elective procedures.

With regards to specialist and nonspecialist surgeons there was no difference in operative mortality for the emergency group of operated patients, but a significant difference in operative mortality was evident in a selective group of elective colorectal resections. I wonder whether similar results were obtained in your study?

Dr. Mark A. Callahan (New York, New York): Thank you. An important question.

We don't have cancer staging because of the database limitations. You cannot get cancer stage out of administrative data that we have. So I can't answer the question. Although I suspect that your observation would hold true in our data as well if we had that.

As far as the urgent versus nonurgent, we did separate out the urgent versus the nonurgent at 1 point in time. We didn't see a difference. We saw an effect in both. Now, on the collections, we have 55,000. And if you had 8,000, then whatever your subgroup of urgents was with some fraction, you may not have had enough power to be able to do the observations because of the number of patients.

Dr. Martin S. Litwin (New Orleans, Louisiana): I felt obligated to rise and ask a question which seems obvious to me from both the abstract and the manuscript. The overall mortality for gastrectomy patients was 8.4%, with the adjusted for subspecialty-trained surgeons of 5.4%. For the colectomy patients it was 4.6%.

These seem like exceedingly high mortality rates to me. I wonder if you have had the opportunity to go into the data a little further and determine why those mortalities are so high. Those are high mortalities for these patients. Those are mortalities that are even high for congenital heart surgery. I wonder if you could comment on that?

Dr. Mark A. Callahan (New York, New York): I really can't tell you the 'why.' I can tell you that unfortunately this is the state of the art of surgery in our state at this point in time, or actually the observation of what is actually going on in our state at this point in time.

Most of us here are at very sophisticated academic medical centers where our experiences are quite different. And indeed when you stratify your data out and look at the high volume academic medical centers, we are running 2 to 3% mortality rates for these types of procedures, considerably lower than the statewide average.

The reality is that the folks in this room do not do the majority of procedures in their states. There are a lot of procedures being done at other institutions where the experience is quite different.

Dr. Marshall Z. Schwartz (Philadelphia, Pennsylvania): I want to expand on Dr. Pellegrini's question further and address the issue of not just mortality but the cause of death. Were you able to evaluate the cause of death and whether the mortality was directly related to a surgical problem such as an anastomotic leak or perhaps unrelated such as a myocardial infarction.

The reason for addressing the cause of death is that in a higher volume environment you are likely to have better ICU care, better cardiology, pulmonology, et cetera. Thus, it is possible that the decreased mortality in higher volume environments may be related to better ancillary support and have less to do with surgical skills.

Dr. Mark A. Callahan (New York, New York): Thank you. That is a very important question.

We cannot differentiate the cause of death from the administrative records. However, again, by doing the analytic approach that we did, we can look at a low volume place where you have got a high volume surgeon - those types of places exist - or a low volume place where you have got a subspecialty surgeon and statistically tease out the independent effect of that specialty training or of that volume for that individual surgeon on the outcomes.

Now, the presumption is that that low volume place doesn't have the rich ancillary support, ICU care, et cetera, that the higher volume place does. And that is probably true. We can at least mathematically pull it out, but I can't tell you for sure - we haven't gone to each hospital and said: What services do you not have? And also, what is the actual cause of death of these patients?

Dr. Michael G. Sarr (Rochester, Minnesota): Let's come back to specialty. You talked about specialty training. But yet your definition is membership in a society. That doesn't always equate. Why don't you go back and look at specialty training and then redo your calculations and not base your argument just on 'membership' in a society?

Dr. Mark A. Callahan (New York, New York): A very important point. In the manuscript we refer to specialty training or interest. I think that is reflected by being a member of 1 of these societies. But it is not a 100% definition that you have had specialty training. We agree with that. That is 1 of the limitations we acknowledge in our paper.

Dr. Ronald V. Maier (Seattle, Washington): One last quick question. As you know, at the American Board of Surgery there is currently ongoing significant debate about reengineering the training of the general surgeon and where subspecialty training fits.

As an extension of the previous question, is the improvement in outcome due to subspecialty training or is it merely additional training? In other words, if you take your proxy group with advanced training in surgical oncology and looked their outcomes following colectomy instead of gastrectomy, do you find the same improvement in outcome? That may not be the best crossover procedure to compare, but can you use your database to analyze the impact of the additional year of training on improving survival across the board and that it doesn't matter necessarily what you did the additional training in, just that you did it? Are the current traditionally trained surgeons inadequately trained?

Dr. Mark A. Callahan (New York, New York): We haven't looked at our data that way. I think that is an excellent suggestion for the next step in this work is to sort of see.

There are 2 points there. One is that the extra training makes you better. The second thing is there is sort of an issue of lifetime volume of procedures. Is it just another year of training means you just have that many more procedures? We can't really tease that out at this point in time. My own gut feeling is that there is a role for that extra training in improving technique specifically. And that is, I think, part of the results that we are observing.