Introduction

Through the turn of the millennium, the USA has experienced a steady rise in the incidence of esophageal adenocarcinoma, with annual increases of more than 2% per year between 1998 and 2003.1 Age-adjusted incidence rates of esophageal cancer now approximate 4.5 cases per 100,000 population,2 placing it seventh among causes of cancer death.3

For over 30 years, surgeons have pondered the association between case volume and patient outcomes for high-risk surgical procedures.4–8 Esophagectomy, because of its high risk and relatively low volume, has been embraced as a procedure warranting regionalization of care within specialty centers.9,10

As a consequence, systems to drive cases to high-volume centers have emerged. For example, the Leapfrog group (Washington, DC), a collaboration of healthcare purchasing organizations that works to initiate improvements in the safety, quality, and affordability of healthcare,11 has established definitions for case volume requirements12 and tracks outcomes. Investigators have attempted to support or refute case volume thresholds for esophagectomy.13–15 The definition of what constitutes a high-volume center varies markedly in the literature and is usually arbitrarily defined.16,17

On the surface, recent data seem to support improved outcomes in the era of regionalization in esophageal surgery. High-volume centers show superior esophagectomy outcomes,18 with the best centers reporting mortality rates from 1% to 4%.16,19–25 However, as systems supporting regionalization gain traction, it remains vital to track national outcomes, since high-volume reporting centers may not represent the rate of actual mortality across the USA. To date, broad efforts to confirm case volume as a surrogate for quality have usually stratified hospital case volume as a categorical variable when comparing statewide or nationwide outcomes via administrative datasets.26–30

Paradigm shifts may bring unintended consequences. High-volume centers are also usually the seats of surgical training. Rising numbers of esophageal operations will require these institutions increase both clinical and educational missions. However, since esophageal surgery is often within the domains of specialist surgeons focused on minimally invasive, thoracic, and oncologic practice, the structure of advanced training is heterogeneous and difficult to evaluate. The impact of fellowship programs on patient outcomes after esophagectomy has not been evaluated outside of single-institution experience.31 The effect of general surgery training programs has rarely been assessed.

Finally, ongoing advances in both surgical and nonsurgical therapeutic modalities and protocols mandate periodic reassessment of our systems intended to regulate delivery of care. Therefore, we report the current state of esophageal surgery in this country with regards to national trends in provision and the impact of case volume and training programs on the safety of esophagectomy.

Methods

The most recently available Nationwide Inpatient Sample (NIS)32 databases covering the years 1998–2006 were queried. These are the largest all-payer inpatient care databases in the USA, containing data from approximately eight million hospital stays each year. The latest release, the 2006 database, contains all discharge data from 1,045 hospitals located in 38 states, approximating a 20% stratified sample of all nonfederal, short-term, general, and other specialty hospitals in the USA.32 A dataset was created by merging core and hospital files and filtered to identify esophagectomies using the ICD-9-CM procedure codes 42.4 (esophagectomy), 42.40 (esophagectomy, not otherwise specified), 42.41 (partial esophagectomy), 42.42 (total esophagectomy, excluding esophagogastrectomy), and 43.99 (esophagogastrectomy, also including complete gastroduodenectomy, esophagoduodenostomy with complete gastrectomy esophagojejunostomy with complete gastrectomy, radical gastrectomy, and other total gastrectomy). While these are standard codes for esophagectomy, they also include some gastrectomies without esophagectomy. To correct for this, gastric operations were assumed if associated with a diagnosis code for malignant neoplasm of stomach (151–151.9) or for gastric ulcer (531–531.9) and were excluded. Pediatric patients less than or equal to 17 years of age were excluded. To calculate nationwide case volume totals, the NIS-supplied discharge-level weight was applied. At all other times, the unweighted NIS cohort was utilized for calculating standard errors and performing regression analyses.

Information regarding the presence of a Fellowship Council (FC)-accredited fellowship program in each year of the study period was taken from the Fellowship Council’s webpage.33 The Fellowship Council is an association of minimally invasive, endoscopic, and combined gastrointestinal surgery fellowship directors formed to address the unique needs of fellowship applicants and programs. In 2006, there were 89 listed programs. Information regarding the presence of a thoracic surgery fellowship was taken from the National Resident Matching Program’s 2009 website34 and assumed the presence of such a fellowship throughout all the years of the study. There were 43 such fellowships identified. Information regarding the presence of a Society of Surgical Oncology (SSO) fellowship was taken from this society’s website35 and assumed the presence of such a fellowship throughout all the years of the study. There were 11 such fellowships identified.

A teaching hospital is defined within the NIS as a hospital with residents in any specialty and meeting any of the following criteria: Accreditation Council for Graduate Medical Education (ACGME) residency training approval, membership in the Council of Teaching Hospitals, or a ratio of full-time equivalent interns and residents to beds of 0.25 or higher. Hospitals having a surgical residency were defined as a subgroup. Details of such a surgical residency program were obtained by combining information from the American Medical Association’s Fellowship and Residency Electronic Interactive Database Access and the listings of accredited programs on the ACGME webpage.36,37 There were 192 identified accredited general surgery residencies. The NIS divides hospitals into size tertiles based on bed size, adjusted for region and teaching status.38

Comorbidity scores were applied to each inpatient stay record, using the Deyo adaptation of the Charlson comorbidity index.39 This validated index allocates a score between 0 and 35, with a higher score indicating more comorbidity. The comorbidities examined include myocardial infarction, congestive heart failure, peripheral vascular disease, cerebrovascular disease, dementia, pulmonary disease, connective tissue disease, peptic ulcers, chronic liver disease, hemiplegia, renal disease, diabetes, malignancy, leukemia, metastatic cancer, and acquired immune deficiency syndrome.

Perioperative complications were added based on ICD-9-CM codes, in a similar manner to that described by Santry et al.39 The diagnosis of “any complication” was made if the “died during hospitalization” field = 1 or if any of the NIS’s 15 diagnosis fields contained one of the following complication or procedure codes: abdominal drainage procedure (5491), acute cerebrovascular accident (43100–43191, 4330–4339, 4340–43491), acute dialysis (3895), acute deep venous thrombosis (4538, 4539), acute myocardial infarction (4100–4109), acute pulmonary embolism (4151, 41511, 41519), acute renal failure (5841–5849), acute respiratory failure (51881), adhesiolysis (5451, 5459), anastomotic leak (9986), bacterial pneumonia (481, 485, 486, 4820–4829), cardiac complications (9971), central nervous system complications (99701–99703), dialysis catheter insertion (3995), foreign body removal (5492), intraoperative hemorrhage (99811), laparotomy (5412), mechanical ventilation (967, 9671, 9672, 9673), postoperative shock (9980), reclosure of abdomen (5461), respiratory tract complications (99973), small bowel obstruction (5600–5609), splenectomy (4143, 415), splenic injury (8650–8651), tracheostomy (311, 3129), transfusion (9904, 9909), urinary complications (9975), wound dehiscence (9983, 99831, 99832), wound infection (9985, 99851, 99859), and wound seroma (99813).

Statistics

SAS 9.2 (SAS Institute, Cary, NC, USA) was used to analyze the data. Logistic regression modeling was performed using generalized estimating equations and assuming a binomial distribution of the data. This allowed control for certain covariables; thus, risk-adjusted outcome measures were calculated. Repeated measure analysis was performed with the experimental unit being hospital identification number clusters. The model was solved for empirical standard error estimates, and p values were based on these estimates. A p value < 0.05 was considered significant. Subsequently, the estimates were exponentiated to calculate an odds ratio (OR) and 95% confidence intervals. One of the authors (JAG) holds a Ph.D. in Biostatistics.

Results

Trends in Care

A total of 11,614 esophagectomies were recorded in the NIS database for the study period; this was the cohort utilized for subsequent analysis. NIS weightings indicate this cohort that represents 57,676 total esophagectomies performed in the USA during the 9-year study period of 1998–2006. With a nationwide weighted total of 6,425 esophagectomies being performed in 1998 and 6,032 in 2006, it is evident that the annual number of esophagectomies did not increase over this timeframe, despite the increasing number of new diagnoses of esophageal malignancy1 (Table 1). At the beginning of the study period, approximately 40% of these operations were performed in teaching hospitals, a proportion which remained constant throughout the study period. The majority of operations were performed in the largest third of hospitals (Table 2). The indications for surgery and the type of operations have remained similar over the same interval (Table 1).

Table 1 Indications for Operation
Full size table
Table 2 Characteristics of Hospitals Performing Esophagectomies from 1998–2006
Full size table

As illustrated in Table 3, high-volume centers for esophagectomy are variously described as performing at least 13 to 20 esophagectomies per year,12,17,25 and the number of surgical programs meeting these standards has remained stable over time. In 1998, 4.2% of hospitals performing esophagectomies completed 13 or more cases, and 1.2% of hospitals performing esophagectomies completed 20 of more cases. In 2002, these numbers were 7.5% and 2.5% and in 2006 were 12.4% and 5.8%.

Table 3 Hospitals in Each Annual Case Volume Group for Esophagectomies, n (%)
Full size table

Mortality Rates

Concurrent with the stable hospital case volumes, the in-hospital mortality rate for esophagectomies taken as a group has steadily decreased throughout the study period (Fig. 1). The mortality rate of all esophagectomies performed in the USA in 1998 was 12.1%. By 2002, it was 9.0%, and by 2006, it had reached 7.0%. As noted in Table 1, approximately 40% of the operations performed were esophagogastrectomies. Improvements of in-hospital mortality were quite impressive in this subgroup, decreasing from 12.3% at beginning of the study period to 8.9% in 2002 and to 7.8% in 2006. Just fewer than 40% of the operations were partial esophagectomies; mortality rates for this subgroup also fell, from 10.7% in 1998 to 8.6% in 2002 and to 5.9% in 2006. Approximately 16% of operations were total esophagectomies, and here too, mortality rates improved markedly over the study period—15.2% in 1998, 9.8% in 2002, and 6.3% in 2006. The only operation which increased in mortality was “Esophagectomy, not otherwise specified”. The numbers performed were small, with 35, 34, and 40 procedures coded in 1998, 2002, and 2006, respectively. Corresponding mortality rates were 11.4%, 8.8%, and 15.0%. These trends in mortality rate occurred synchronously with a steady decrease in every year of the mean Charlson comorbidity scores, from 4.5062 in 1998 to 4.2311 in 2002 and to 3.7997 in 2006.

Figure 1
figure 1

In-hospital mortality rate by year.

Full size image

With all esophagectomies considered together, there was noticeable variation in mortality rates according to expected primary payer status or by self-described racial group (Table 4). The largest three expected payer groups were private including HMO, Medicare, and Medicaid; unadjusted mortality was 5.2%, 12.2%, and 11.3%, respectively. Of the three largest racial groups in which a racial identity was specified, the mortality rates were White 8.9%, Black 12.5%, and Hispanic 7.1%.

Table 4 Unadjusted Mortality Rates by Primary Payer and by Racial Group
Full size table

Interestingly, as seen in Fig. 1, anastomotic leak rates were quite constant throughout the study period, with little variance about the mean of 1.53 (±0.29).

Effect of Hospital Case Volume

Table 5 examines the independent effect of annual hospital case volume on complication rates, after controlling for the improvements in outcomes seen over the study period and for Charlson comorbidity scores. That is, the risk-adjusted effect of increasing annual case volume is reported. In contrast to previously published studies, artificial case volume groups were not applied and the models were solved for case volume as a continuous variable. An odds ratio <1.0 signifies an inverse correlation between case volume and the complication under review. The odds ratios tend to be very close to 1.0 because the ratios represent the effect of increasing the annual volume by a single case. That is, the effect of each and every case on outcomes is reported. Nearly all analyzed complication categories trended toward an inverse correlation with case volume, with any complication, myocardial infarction, respiratory tract complications, bacterial pneumonia, acute respiratory failure, acute renal failure, postoperative shock, blood transfusion requirement, and splenectomy rates achieving statistically significant improvement. No complication was associated with increasing case volume.

Table 5 The Incremental Effect of Each Esophagectomy on Annual Outcomes, Controlling for Year, and for Charlson Comorbidity Scores
Full size table

Results which have been tabulated reflect modeling for the linear effect of the variables only in order to simplify presentation. For a more detailed examination of the effects of case volume specifically on mortality rates, modeling was also performed adjusting for year, case volume, and Charlson comorbidity score and additionally the quadratic of case volume. This examines the effect of very high case volume on mortality. When used as predictors in a logistic regression model, both the case volume (p < 0.0001) and the quadratic (p < 0.0001) achieved statistical significance, with the predicted trends plotted in Fig. 2. The improvement in mortality rate observed with increasing hospital case volume seems to level out at approximately 30–40 cases per year and then slowly increases after about 80–100 cases per year.

Figure 2
figure 2

Effect of case volume on mortality rates for esophagectomy, controlling for year, and Charlson comorbidity score.

Full size image

Effect of Training Programs

Table 6 examines the independent effect of fellowship programs on outcomes of esophagectomies. Forty-three hospitals submitting data to the NIS and offering National Residence Matching Program (NRMP)-affiliated thoracic surgery fellowship programs were identified. Examining the independent effect of the presence of a thoracic surgery fellowship, after controlling for yearly variations, annual case volume, and Charlson comorbidity score, it is seen that the rate of any complication was significantly better, as were rates of bacterial pneumonia and incidental splenectomy. Anastomotic leak rates were significantly worse in this group, being nearly double those in hospitals without a thoracic surgery fellowship program (OR 1.81808, 95% confidence interval [1.18347, 2.79297]). Eighty-nine NIS hospitals had Fellowship Council-accredited fellowship programs. Examination of the independent effect of a FC-accredited fellowship on esophagectomy outcomes, after controlling for yearly variations, annual case volume, and Charlson comorbidity score, revealed that anastomotic leak rate was significantly increased (OR 1.71926 [1.09136, 2.70843]). Eleven NIS hospitals offered a Society of Surgical Oncologists-administered fellowship program during the study period. Only a very small number of these institution performed esophagectomies ranging from one to three hospitals per year. In the years 1998, 1999, 2000, and 2004, one of these institutions also offered either a Fellowship Council-accredited fellowship or a thoracic surgery fellowship. There was no significant independent effect of an SSO-administered fellowship on any of the measured variables. There was no detrimental effect of any fellowship program on in-hospital mortality following esophagectomy.

Table 6 The Effect of Fellowship Programs on Outcomes, Controlling for Year, Charlson Comorbidity Scores, and Case Volume
Full size table

When all of the above fellowship programs were considered together, again controlling for yearly variations, annual case volume, and Charlson comorbidity score, it was noted that the presence of any fellowship program was associated with a decrease in the rate of any complication (OR 0.81655 [0.70613, 0.94425]) and an increase in rates of anastomotic leak (OR 1.64538 [1.12423, 2.40811], myocardial infarction (OR 1.47069 [1.02836, 2.10329]), and requirement for postoperative tracheostomy (OR 1.37774 [1.09114, 1.73961]).

The effects of the presence of an ACGME-accredited general surgical residency program in hospitals submitting data to the NIS are shown in Table 7. There were clear benefits in rates of any complication (OR 0.85656 [0.75270, 0.97482]), in-hospital mortality (OR 0.73408 [0.60460, 0.89128]), acute renal failure, acute respiratory failure (OR 0.77169 [0.63680, 0.93518]), and postoperative bacterial pneumonia (OR 0.70775 [0.60034, 0.83437]). The first column shows the independent effect of a surgical residency program, after controlling for yearly variations, annual case volume, and Charlson comorbidity score. The second column shows the effect of a surgical residency program after controlling for any fellowship in addition to the other controlled variables.

Table 7 The Effect of a Surgical Residency on Outcomes of Esophagectomy
Full size table

Discussion

With the incidence of esophageal adenocarcinoma in the USA increasing, demand for esophagectomies will persist for the foreseeable future. Several studies have reported an association between increasing hospital esophagectomy volumes and improved outcomes, and these data have been often cited by proponents of centralization of care. However, many of these studies have ignored case mix and comorbidity profiles. There has also been confusion in the studies between mortality rates attributed to institutions and those associated with individual surgeons, especially in hospitals where more than one division performs these operations.40 A further confounder of volume–outcome studies is the categorization of institutions into either low- or high-volume centers based on arbitrary case thresholds.41 Finally, interpreting such results is difficult when poorly described or suboptimal statistical methodology is utilized.42

Despite accruing evidence of the beneficial effects of case volume on cancer surgery outcomes since the end of the twentieth century,6,26,43 the percentage of esophagectomies being performed in higher-volume hospitals has not increased significantly over the study period. This single fact may explain the discrepancy between the best reported mortality rates and the latest US esophagectomy mortality rate of over 7%. Surgeons and patients discussing informed consent for esophagectomy outside high-volume centers should consider that one in every 14 patients undergoing esophagectomy in this country will die in-hospital.

Although it is encouraging that the mortality rate for esophagectomy has diminished by 60% over recent years, there is no clear association with the movement toward regionalization. Mortality improvement may be partly explained by the decreasing comorbidities of the patient population described above. There have been parallel improvements in perioperative care44,45 as well as staging and selection.46,47 Unfortunately, limitations of the NIS database prevent analysis of the effect of tumor stage on outcomes.

Case volume requirements have been determined by various organizations, such as the Leapfrog Group. To meet the standards of this group, at least 13 esophagectomies must be performed by an institution per year. According to the newest Leapfrog criteria, certain nonesophagectomy operations can also be counted toward esophagectomy, such as total gastrectomy and radical gastrectomy.11 As described above, by excluding operations performed for primary gastric diagnoses such as gastric malignancy or gastric ulcer disease, we have minimized the possibility of inclusion of any cases other than esophageal resection in our study group. Thus, the cohort we reviewed is equally sensitive and more specific for esophagectomy than that used by other groups. The most striking feature of these data is the beneficial effect evident for each and every increment in annual hospital case volume. Nearly every measured complication was seen to significantly improve with increasing annual volume, at least to volumes seen in nonoutlier hospitals. This has now been demonstrated in a very large administrative database, without recourse to artificial case volume groups. It appears that there may be a reversal of these positive volume–outcome associations when hospital volume exceeds 100 cases per year, with some evidence for rising mortality rates. However, the sample size of these very-high-volume hospitals is very small, which limits interpretation of this interesting and never previously reported finding. We plan further investigation to determine whether this effect is true or perhaps a consequence of case mix or other uncaptured variables.

It is becoming more evident that volume criteria are not the sole determinant of outcome.48,49 Even with equally experienced surgeons in a high-volume hospital, a variable that differs widely between institutions is the composition of the other members of the surgical team. No previous study has evaluated the effect of fellowship programs or general surgical residencies on outcomes after esophageal resection. If hospital case volume is used as a surrogate for the experience and capabilities of the perioperative team, particular scrutiny should be given to the effect of training programs, which involve multiple and variably rotating trainees in perioperative care of patients and which may sacrifice case volume for educational focus and academic inquiry.

We have identified an overall independent beneficial effect of a fellowship program in hospitals performing esophagectomies. If any fellowship program exists (thoracic, FC or SSO) at a particular hospital, the total numbers of complications decrease, though there is no way to verify from the NIS data whether the esophagectomies were performed by fellowship-affiliated surgeons. This limitation is probably more relevant with the Fellowship Council programs than the SSO or thoracic surgery programs, since the former places emphasis on minimally invasive gastrointestinal surgery and not necessarily surgery for malignancy in the chest. Of much more interest is the apparent increase in the serious adverse events of anastomotic leak, myocardial infarction, and tracheostomy associated with fellowship programs. While we intend to examine this further in future studies, our current hypothesis for the association between anastomotic leaks and fellowships is that, compared with private practice and resident training programs, the fellowship model puts trainees in the position of operating surgeon at crucial stages of an esophagectomy procedure. This complication is not associated with a greater death rate, perhaps as a consequence of better detection and management in these fellowship sponsoring hospitals, but clearly this is an area requiring further investigation.

Outcomes do not clearly stratify along surgical specialties. In this study, hospitals with fellowship programs administered by the Fellowship Council and the SSO had very similar outcomes, whereas a few outcomes, namely rates of incidental splenectomy and rates of bacterial pneumonia, were comparatively better in thoracic surgery fellowship program hospitals. Bias may have been introduced by the small sample size of SSO training hospitals, a result of nonreporting of many of such programs to the NIS. That said, surgeons identifying themselves as thoracic surgeons have been shown to have improved outcomes over those identifying as general surgeons,50 especially in low-volume centers.

In contradistinction to fellows, residents are usually supervised to a far greater degree during operations. This supervision has been thought to be the major means for ensuring safe outcomes in a teaching environment.31 In this study, we have shown that an ACGME-accredited general surgical residency program independently improves many of the measured complications, including rates of any complications, in-hospital mortality, acute renal failure, acute respiratory failure, and bacterial pneumonia. It has previously been reported that high-volume centers may minimize the effect of complications by earlier detection and more appropriate management.25 It is possible that the factor which enables earlier detection of problems is the presence of a strong residency program. The authors hypothesize that a larger house staff permits more frequent physician–patient contact and earlier management of adverse events. Higher deep venous thrombosis rate is reported, and this might be due to either longer operation time with resident training, or because of improved detection by residents in the postoperative period, a consequence of the aforementioned increases in contact with the patient.

Limitations exist in searches of administrative databases related to the accuracy of data entry by institutional coders. The accuracy of coding has previously been reported as suboptimal,51 though the detection of the presence of a particular diagnosis (as performed in this study) has been validated.52 It is conceivable that the programs with an active surgical residency might have better entry into the medical record of complications, with disproportionate capture of these measured outcomes in this group.53 Also, many hospitals are not represented in the NIS cohort, including some of the higher-volume esophagectomy centers in the USA. While attempts have been made to control for this statistically, a larger sample will always provide more accurate representation of the population as a whole. Finally, length of stay was considered by the authors as a variable dependent on the number and severity of complications and so was not used as a control variable in the mathematical modeling. It is, however, conceivable that length of stay is at least partly independent insomuch as the longer a patient remains in hospital the more time is available to capture complications for inclusion in the NIS.

Conclusion

The current 7% esophagectomy mortality rate of hospitals reporting to the Nationwide Inpatient Sample has improved but without evidence for measurable centralization of cases within high-volume centers. This rate remains higher than that reported in most contemporary series. In this model, the hypothesized positive volume–outcome relationship of esophageal surgery has been validated without the use of arbitrarily assigned case volume categories. This volume-related improvement in mortality is seen to taper with approximately 30–40 annual cases and may reverse in the highest-volume centers. The performance of esophageal resections in training hospitals is safe and with no increase in either mortality or total morbidity, though fellowship training may be associated with a higher anastomotic leak rate.