Discussion

Dr Gregorio Sicard (St. Louis, Mo). I really congratulate Dr Bush for this excellent paper. The NSQIP database is serving as a model for outcomes assessment. It has the uniqueness of risk adjustment, which is one of the issues that has always haunted outcomes assessment. I would like to state that I am really glad to see that the results of Dr Bush’s evaluation were not that dissimilar to the results we found when we looked at the high-risk cohort of patients in the IDE trials, (ie, the 29% mortality in the EVAR arm of the high-risk surgical group.).

I do have a question for Dr Bush. When you looked at the EVAR versus open at one year, were you able to determine if this was all-cause mortality or aneurysm-related mortality?

Dr Ruth L. Bush. The death data in the BIRLS database, the veterans’ beneficiary database, only answers the question, “Is the veteran alive of deceased?” The database does not contain any information on what caused the death. We have proposed looking at the National Death Index, which is available via linkage with Social Security numbers to determine cause of death. However, the problem even with this data set is that, for determining aneurysm-related death, autopsy rates are so low that a lot of assumptions are made and speculations on what is the cause of death, so we decided to only analyze all-cause mortality.

Dr Sicard. And since this is a data set that is very robust in risk adjustments, did you do any univariate or multivariate analysis to identify what were those risk factors that could predict or that have an association with mortality?

Dr Bush. In this particular study, we did not. In our first study, now published in the Journal of the American College of Surgeons, we looked at the endovascular versus open repair in all elective cases listed in the NSQIP database. In this study we looked at factors that may be independent predictors of outcome. We found that advanced age, especially over 80, was associated with mortality as was a history of stroke, smoking, and liver disease. And we created a separate variable for hospital volume. In this study, low-volume hospitals, in univariate analysis, were predictive of mortality. However, in our multivariate analyses, this variable was not significant.

Dr Sicard. Again, congratulations, and I hope all the members that are here will attend this Saturday’s session because we will be looking at different levels of evidence exactly on this issue of high-risk patients.

Dr Robert Cambria (Bangor, Me). I had a simple question about the definition of high risk. Looking at your slides, I don’t think I see any low-risk patients based on those criteria. How many procedures were excluded, both in the EVAR and open category, for not being high risk, if you have that information?

Dr Bush. I do. During the time frame we looked at, from May 2001 through December 2004, there were about 3400 elective aneurysm repairs performed at that time in VA hospitals. Of these, 2400 patients fit our high-risk criteria. We found that in our first study, again, when we looked at the whole cohort together, probably 75% of our patients in that original cohort could be identified as a high risk, so we made the definition fairly strict for this study. VA patients have been suggested to have more comorbidities and this may account for our relatively high-risk population overall.

Dr George Geroulakos (London, United Kingdom). I wonder whether you considered that perhaps the population of high risk, as you defined it, is not entirely comparable to the EVAR 2; therefore, you cannot draw the conclusion that is written in the abstract that high-risk patients could be considered safely, or relatively safely, for intervention. To give you an example, you said that a history of coronary artery bypass graft was considered as one of the risk factors that makes patients to belong to the high-risk group; while, in fact, what we’re trying to do is to convert these high-risk factors to low-risk factors. So if a patient had coronary artery bypass grafts and remained asymptomatic, he would not be considered anymore, as far as risk factor is concerned, to a high-risk group.

Dr Bush. You are absolutely right. One of the problems with looking at administrative databases is, indeed, what is the definition of the variables being entered into the database as well as the consistency of data entry? Furthermore, the data entry relies on the people that are putting in the data. For example, a patient who is undergoing coronary revascularization, or has had this procedure, may or may not be completely now symptom-free or free from coronary disease after their bypass.

The NSQIP database, it is a robust validated database. The data are being entered by trained nurse abstractors who are continually audited by the Department of Veteran Affairs, and they have strict definitions of actually what goes into the database. So it’s more than just ICD-9 codes within that database.

But you are correct, I do not know if the patients are symptom-free or still symptomatic following their coronary revascularization.

Dr Michel Makaroun (Pittsburgh, Pa). To my knowledge, this is probably the first time I see a long-term survival curve from all-cause mortality with a significant advantage of the endovascular group over the open group long term. And I did not even see the early separation in the first year that is very typical of these curves. Do you have an explanation why your results are so different than the usual analysis of one versus the other?

Dr Bush. That’s a very good point, and we did see our patients demonstrating a benefit of endovascular repair and we also saw that in all the patients as well, not just those that fit our high-risk criteria. With the BIRLS database, we had death information on almost every patient. We found that very few patients had unavailable information. So our number of patients at risk at the beginning is the same as the number of patients at the end of the 2 years who are at risk. So it may be part of the lack of data censorship in our survival curves that makes this difference.