As the rate of morbid obesity in the US increases, the number of patients who would benefit from bariatric surgery continues to grow. The increasing number of training programs dedicated to bariatric surgery and the increasing number of institutions performing these procedures reflect the importance given to this problem by the medical community.
The management of the bariatric patient is often quite complex so realizing the best possible outcomes for these patients may require changes to standard institutional facilities, and to the practices of staff and physicians. Institutions must have furniture, hospital equipment and imaging suites suitable for high body mass index patients. Staff members need to be competent and compassionate when dealing with the medical and psychological complications of obesity. Surgeons must have an understanding of the perioperative and long-term risks of bariatric surgery, and be experienced in the operative care of the morbidly obese 5, 11
A strong association between case volume and outcomes has been demonstrated for many surgical procedures. Bariatric surgery was expected to demonstrate a similar relationship, and early studies suggested that this might be true 12–14
. Most early reports divided centers into arbitrarily determined case volume categories, frequently using an annual case volume of 100 as the distinction between low-volume and high-volume centers, and examined very few complications. Nonetheless, these findings soon led to widespread belief in the necessity of ensuring bariatric cases were directed to high-volume centers. Soon thereafter, center-of-excellence accreditation programs were devised, based on these arbitrary thresholds, and this development was quickly linked to remuneration by third-party payers. However, there has been very little published validation of this paradigm, and post-COE-introduction outcomes have not been reported.
The vast number of cases captured in the NIS dataset allows very small differences between groups to be detected. With over 100,000 bariatric cases available for review, low frequency outcomes such as death, pulmonary embolism and others can be examined. The numbers also allow modeling with continuous variables rather than with the arbitrary assignment of categorical groups. The use of the NIS dataset does entail some limitations, one of the most important of which is the absence of a longitudinal identification number for patients. Thus, once a patient is discharged from hospital, that patient is lost to follow-up through the database; any re-admission will not be captured.
The most striking finding of our analysis is the association of case volume with outcomes. The magnitude of the positive association of volume with outcomes was small and required the large numbers of the NIS dataset for detection, but was evident for total complications, death, requirement for postoperative percutaneous abdominal drainage, acute pulmonary embolus, cardiac complications and postoperative in-hospital myocardial infarction, incidental splenectomy, post-operative physiological shock states, renal failure, postoperative respiratory failure and bacterial pneumonia. These improvements in outcomes increased with increasing case volume. However, and very importantly, it appears that the positive association of many outcomes with volume, including the outcome of mortality rates, improves continuously with ever-increasing case volume. While it may be reasonable to establish case volume thresholds tied to “acceptable” complication rates for credentialing purposes, the authors emphasize that complications continue to improve at still higher case volumes. These improvements, though statistically significant, are nonetheless small in magnitude.
It might be argued that the study design, encompassing many years, might have some limitations, with the recent increases in annual case volumes being reflected in increasing experience, and the early years not being representative of the current environment. This argument is countered by the inclusion of year as a covariable in the analysis in a speculative attempt to negate this time-related effect. Additionally, it might be argued that any risk-stratification not including BMI as a predictor might limit accurate prediction of complications. However, since BMI is uncommonly reported by coders to the NIS, we determined it could not usefully be added to the risk stratification model. Irrespective, the use of the Charlson Comorbidity Score is a multiply-validated technique of risk stratification of the surgical patient, and in spite of not including BMI as a predictor, is thought to be a valid and robust prognostic index to use.
It is interesting to note that the independent association with outcomes of both ACS and ASMBS designated Centers of Excellence was quite small, a finding supported by a previous study by Livingston 15
. Even though lower rates for total complications and deaths were measured in Centers of Excellence, no statistical differences were identified. Center-of-excellence comparisons are limited by the small size of these cohorts relative to the total study population, with the margin for error therefore being increased. Though these COE credentialing programs require peripheral aspects (such as psychological counseling, nutritional support, etc) for bariatric program accreditation, it appears that the only real independent variable affecting outcome in a center-of-excellence program is the case volume. Evaluation of a single year, 2006, showed an improvement in certain outcomes in Medicare-recognized bariatric facilities, namely in postoperative shock and acute renal failure. Though this single year analysis eliminates the need to consider program designations across years, it does limit the period of study and magnifies potential errors. We contend our decision to extrapolate ACS and ASMBS COE designations across the entire study period likely affects the results little, as similar support and surgical experience both before and after the credentialing period are probably present. When the non-gastric band cohort was analyzed independently so as to definitively negate the aforementioned effect of missing outpatient gastric band data, all reported outcomes were unchanged except that the improvement in “any complication” rate of ACS-designated COE achieved statistical significance. The odds ratio was nearly unchanged.
It has previously been reported that fellowship training programs remove the “learning curve” period of skills acquisition for bariatric surgery 16
, while not increasing “any perioperative complications” 17
, but studies in this area are small and limited to single institutional cohorts. In our study of the actual situation present in the wider US surgical community, these training programs have now been confirmed to be safe for patients, with no difference in total complications or death rates, and a very large and strongly significant improvement in rates of intraoperative splenectomy during the bariatric procedures and of postoperative bacterial pneumonia rates. With the majority of bariatric operations now being performed through laparoscopic access methods, it is possible that advanced laparoscopic skills among fellows and their mentors improves these specific technical and postoperative outcomes. Decreased splenectomy rates were also noted to be independently associated with the presence of an accredited surgical residency program. It should be clearly noted that the performance of a bariatric operation in a hospital offering Fellowship Council-approved fellowship training does not guarantee that the fellow performed or was even a participant in the procedure; this limitation of the dataset will necessitate caution in the interpretation of associations between Fellowship training per se
and outcomes, but will not nullify the association between the parent institution and the outcomes under consideration.
Examination of the influence on outcomes of surgical residency programs was concerning. Some major complications were significantly increased in this group; acute pulmonary embolism, acute myocardial infarction and other cardiac complications. Respiratory complications were lower in this subgroup, as were total complications. The cause for these observations is unknown. Sites of surgical training are usually sites of training for other specialties; perhaps the concurrent involvement of more junior trainees from other specialties is a confounder which results in higher perioperative medical complications such as these. Regardless, the findings warrant both concern and a need for urgent further investigation, preferably in a prospective and controlled manner. Small published series have previously reported no difference in complications in operations with resident participation as compared to those performed by an attending alone 18 19
; the present NIS-derived series, much larger and inclusive of more complications, shows that safety may indeed be compromised in teaching hospitals with surgical residency programs, though the cause of this is not able to be determined from the available NIS data.
Limitations exist in searches of these administrative databases. The major problem is identification of the procedures of interest, in this situation, bariatric operations. These datasets use codes from the International Classification of Disease, Ninth-Revision, Clinical Modification (ICD-9-CM). However, codes do not always exist for all procedures of interest. This is especially true early in the acceptance of a procedure. For example, though the first laparoscopic adjustable gastric band was placed in the U.S. in 2002, an ICD-9 code was not established until October 2004. Thus, a search of any ICD-9-based administrative dataset prior to this date would not have detected the band placement. The alteration and addition of ICD-9-CM codes during a given calendar year further complicates analysis of newer procedures. A previous publication by the authors highlights the difficulty in accurately determining case volume of bariatric operations 9
. While the search strategy utilized herein has been previously widely used by many investigators, the potential problem of specificity of the codes still arises.
Length of stay was considered by the authors as a variable likely to be dependent on the number and severity of complications - the more complications, the longer the stay - and so was not used as a variable in the modeling. Nonetheless, modeling was performed with the assumption that length of stay was independent; this showed minimal differences to the results above, and has not been included in the interests of brevity. Other studies have used total costs as a surrogate for complications 15
, but this is heavily dependent on individual hospital variation, and would penalize institutions providing more expensive, but not inferior care.
A further limitation of the NIS is related to the accuracy of data entry by hospital medical coders. The accuracy of coding has previously been reported as suboptimal 20
, though the detection of the presence of a particular diagnosis (as performed in this study) has been validated 21
. It is conceivable the programs with an active surgical residency might have better entry into the medical record of complications, adversely affecting the outcomes of this group. Furthermore, even if coding is correct, once a patient is discharged from hospital, that patient is essentially lost to follow-up. Readmissions are not captured.
Due to the statistical sampling methods used by the NIS, not all states and not all institutions are represented in the dataset, a problem encountered with most large administrative databases. However, a major advantage of the NIS is the completeness of information from individual reporting hospitals, which facilitates volume-outcome analysis 15