SEARCH

SEARCH BY CITATION

Keywords:

  • centralization;
  • hospital volume;
  • oesophagectomy;
  • surgical volume;
  • systematic review

Abstract

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. References

Purpose:  This systematic review aims to assess whether overall survival, mortality, morbidity, length of stay and cost of performing oesophagectomy are related to surgical volume.

Methods:  A systematic search strategy from 1997 until December 2006 was used to retrieve relevant studies. Inclusion of articles was established through application of a predetermined protocol, independent assessment by two reviewers and a final consensus decision.

Results:  A total of 55 studies were identified of which 27 studies, representing 68 882 patients, met the inclusion criteria. Twenty-one of these solely examined hospital volume, 5 examined both hospital and surgeon volume, and 1 examined surgeon volume in isolation. All but one of the studies were retrospective in nature, and because of the heterogeneity of the literature, no meta-analysis could be performed. Of the studies exploring the relationship between hospital volume and mortality, 20 reported a statistically significant benefit to large volume centres. Five of six included studies showed significant evidence for a reduced mortality risk with greater surgeon volume.

Conclusions:  Based on the evidence from these retrospective studies, oesophagectomy performed in high volume centres would appear to be associated with better outcome compared with low volume centres.


Abbreviations
ASERNIP-S

The Australian Safety and Efficacy Register of New Interventional Procedures – Surgical

HVH

high volume hospitals

HVS

high volume surgeons

ICD

International Classification of Disease

LVH

low volume hospitals

LVS

low volume surgeons

MVH

medium volume hospitals

MVS

medium volume surgeons

NR

not reported

VHVH

very high volume hospitals

VHVS

very high volume surgeons

VLVH

very low volume hospitals

Introduction

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. References

Oesophagectomy is a procedure associated with significant morbidity and mortality. While there are continuing advances in surgical technique and perioperative management, much of the procedure has remained unchanged. It would seem logical that an increased level of experience of both the surgeon and the team involved in the perioperative period would decrease the risk associated with the procedure. There is a growing trend showing beneficial volume-based effects when applied to both major surgical procedures1 and selected medical conditions.2 Several variables have shown influence by volume, the most significant being operative mortality. Other factors demonstrating a volume-based effect include morbidity, length of stay and, ultimately, cost. While some studies have examined other markers of outcome, a paucity of data limits comparison between studies.

One of the main principles for examining volume-based outcomes is to establish whether centres performing a limited number of procedures should refer major cases to a centralized base to improve outcome. To verify this approach, we must first establish if outcome significantly improves in large volume centres. This does not assume that a small volume centre cannot produce results equivalent to a large volume centre. Instead, we aim to establish standards of care that can be applied to all centres. As, logistically, it would be difficult to justify randomization in this context, we can only compare outcome from centres as part of a systematic review of the literature on volume-based studies.

Centres performing a greater number of procedures may operate on a wider patient demographic with a greater number of risk factors, therefore generating selection bias. Figures must therefore be risk adjusted to ensure that any volume-based benefit is not obscured. We present the findings of the available evidence on volume-related outcome in oesophagectomy.

Methods

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. References

Literature search strategy

All studies that described different configurations of the following terms, ‘oesophagectomy’, ‘centralization’, ‘surgical volume’, ‘hospital volume’, ‘regionalization’, ‘treatment outcome’ and of ‘mortality’, were included.

Studies were selected, without language restriction, via CINAHL, Clinical Trials Databases, Current Contents Connect, Current Controlled Trials, EMBASE, MEDLINE, National Research Register, National Health Service Centre Research and Dissemination Databases, PubMed and The Cochrane Library between 1997 and December 2006.

Included articles must have provided information on at least one of the following outcomes: overall survival, mortality, morbidity, length of stay and costs. Data must have been presented in a format to allow comparison between articles. Studies where data were pooled with other procedures or were not specific to oesophagectomy were excluded.

Only full peer-reviewed articles were included. Abstracts failed to provide adequate detail on patient selection, treatment allocation, outcome, measurement methods and study design to allow an accurate, unbiased assessment and comparison of study results. Review articles that were not of a systematic nature were excluded from data extraction. Case studies were also excluded (see Table 6). Additional articles were identified through the reference sections of the studies retrieved (Fig. 1).

Table 6.  Excluded Studies
Exclusions
No applicable study data (n= 9)
Review articles (n= 4)
Data not relating to, or specific for, oesophagectomy (n= 3)
Case studies (n= 2)
Comment (n= 1)
No interhospital comparison (n= 1)
image

Figure 1. Process for selection of studies retrieved from electronic literature database search.

Download figure to PowerPoint

Only the most recent publication was included when selected articles used the same or overlapping data in multiple publications.

Study selection was conducted by three reviewers. Disagreements were solved by discussion.

Data extraction

Considerable variation in the documentation of patient selection was present in the publications. There was no reporting of International Classification of Disease (ICD) codes used to identify patients within 12 of the included studies, with another 4 reporting the ICD version used but not the specific codes. Of the studies to include this information, the majority used ICD-9-CM codes, while the remainder used the original ICD-9 coding.

Data from all included studies were extracted by one researcher and checked by a second using standardized data extraction tables that were developed a priori.

The majority of patient information reported in the included studies was from North America, using either nationwide or nationally representative databases. Studies from the Netherlands,3 the UK4 and Taiwan5 were also included. Each of these reports used patient data from nationwide databases.

Statistical analysis

The thresholds used to define high and low volume were heterogeneous among different studies. Because of this, a weighted meta-analysis of the data was not feasible and would have been unreliable. A positive correlation between the variable studied and outcome was considered significant if the P-value is <0.05 within each study.

Results

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. References

A total of 55 studies were identified of which 27 studies,1,3–28 representing 68 882 patients, met the inclusion criteria. Twenty-one1,3,5–7,9–13,15–20,22,24–27 of these solely examined hospital volume, 54,8,14,21,28 examined both hospital and surgeon volume, and 1 examined surgeon volume in isolation. All but one23 of the studies were retrospective, and because of the heterogeneity of the literature, no meta-analysis could be performed. Of the 25 studies examining the relationship between hospital volume and mortality, 20 reported a statistically significant benefit to large volume centres. There were five studies showing significant evidence for reduced operative mortality associated with increased surgeon volume.

Twenty-six of these studies were classified as National Health and Medicine Research Council (NHMRC) level III-3, and 1 was classified at level III-2.4 Accurate identification of patients across studies was particularly difficult as many studies did not report the use of ICD diagnostic codes. Additionally, studies that reported diagnostic codes used a range of different ICD versions, as well as diagnostic codes.

There was a large variation in the number of patients enrolled between studies, contributing to the range of statistical significance where calculated.

Patient covariates were often included in statistical calculations, however, hospital covariates were only utilized by one study.10

Statistical calculations were poorly reported in some of the included studies, with a variety of univariate or multivariate calculations being used. Identification of variables used in these calculations was difficult, making inter-study comparisons of outcomes additionally complicated. Calculations incorporating clustering for hospital outcomes were more commonly reported than those including clustering for surgeon outcomes.

A total of four included studies utilized data from patients enrolled before the 1997 publication limit.6,18,20,25 The study of Miller et al.23 was the only study reporting solely on the impact of surgeon volume and patient outcomes using data from patients enrolled before the publication limit. All studies reporting data prior to 1997 should be interpreted with caution. It is possible that data may be skewed by the inclusion of procedures or techniques that are not comparable with current practice.

All comparisons within this series are based on data collected concurrently, eliminating any chronological differences possible with split duration enrolment.

Discrepancy in the definition of mortality means that comparison between studies must be performed with caution. Patient mortality can be defined in a number of ways: in-hospital, within 30 days or within 60 days. Owing to this variation, detailed analysis and comparison of results between studies are much more difficult. This is of particular importance with in-hospital mortality, as this may confer a disadvantage to centres with a longer length of stay.

Morbidity rates are perhaps the most difficult to compare in surgical studies because of considerable variation in reporting. While there are recognized complications specifically associated with oesophagectomy, such as anastomotic leak, other more general complications associated with any surgical intervention are dependant on the authors' threshold for inclusion. Interestingly, ICD-9-CM coding had no specific complication code for anastomotic leak in oesophagectomy.

A point of potential bias in the study is that of authors reporting multiple studies. Birkmeyer, Dimick and Urbach have each published multiple studies, often producing several in 1 year. It appears that calculations used in each of these studies are based on the same group, or subgroup, of patients, respectively. Although each of these studies has been included on its own merits, each publication should be interpreted with caution because of the similarities with other published studies.

Hospital volume outcomes

Mortality

Twenty-five studies performed 29 statistical comparisons of interest comparing hospital volume to four different types of mortality: operative, operative plus 30 days, in-hospital and overall (Table 1). Twenty reported a statistically significant association between increased hospital volume and decreased patient mortality. Although four studies performed statistical comparisons, no significant association was reported. Five comparisons did not test for statistical significance.

Table 1.  Summary of included studies
StudyProcedureVolume typePatients (n)Time periodLength of follow-upICD version/codes
  • Overlap of study cohorts Dimick (2003b, 2003c, 2005).

  • ‡Overlap of study cohorts Dimick (2001, 2003d);

  • §

    §Overlap of study cohorts Urbach (2003, 2004, 2005). ICD, International Classification of Disease; NR, not reported.

Bachmann et al. 20024OesophagectomyHospital volume and surgeon volume78107/1996–06/19971NR
Begg et al. 19986Partial oesophagectomy and total oesophagectomyHospital volume67821984–19939ICD-9-CM
42.40
42.41
42.42
Birkmeyer et al. 20021Partial oesophagectomy and total oesophagectomyHospital volume63371994–19995ICD-9
42.40
42.41
42.42
43.99
Birkmeyer et al. 20038Partial oesophagectomy and total oesophagectomyHospital volume and surgeon volumeNR1998–19991ICD-9
42.40
42.41
42.42
43.99
Birkmeyer and Dimick 20047OesophagectomyHospital volume43502000–20033NR
Birkmeyer et al. 200610OesophagectomyHospital volume43491994–20017NR
Birkmeyer et al. 20069OesophagectomyHospital volumePopulation data were pooled with excluded procedures2000–20022NR
Dimick et al. 200112Partial oesophagectomy and total oesophagectomyHospital volume113601/1984–09/199915.6ICD-9-CM
42.41
42.42
Dimick et al. 200315Partial oesophagectomy and total oesophagectomyHospital volume12261996–19971ICD-9-CM
42.40
42.41
42.42
Dimick et al. 200313OesophagectomyHospital volume30231995–19994NR
Dimick et al. 200316Partial oesophagectomy and total oesophagectomyHospital volume3661994–19995ICD-9-CM
42.40
42.41
42.42
Dimick et al. 200511Partial oesophagectomy and total oesophagectomyHospital volume86571988–200012ICD-9-CM
42.40
42.41
42.42
Dimick et al. 200514Partial oesophagectomy and total oesophagectomyHospital volume and surgeon volume19461998–19991ICD-9-CM
42.40
42.41
42.42
Elixhauser et al. 200317OesophagectomyHospital volumeNR20001NR
Finlayson et al. 200318OesophagectomyHospital volume52821995–19972NR
Goodney et al. 200319OesophagectomyHospital volume11251994–19995NR
Gordon et al. 199920OesophagectomyHospital volume5181990–19977ICD-9-CM
Ho et al. 200621Partial oesophagectomy and total oesophagectomyHospital volume and surgeon volume10 0231988–200012ICD-9-CM
42.40
42.41
42.42
Kuo et al. 200122OesophagectomyHospital volume119301/1992–12/19997ICD-9-CM
42.4x
42.52
42.62
43.5
Lin et al. 20065OesophagectomyHospital volume667401/2000–12/20033ICD-9-CM
Miller et al. 199723OesophagectomySurgeon volume741989–19934ICD-9-CM
Patti et al. 199824Partial oesophagectomy and total oesophagectomyHospital volume15611990–19944ICD-9
42.40
42.41
42.42
42.43
42.44
42.45
42.46
Swisher et al. 200025OesophagectomyHospital volume3401994–19962ICD-9
Urbach et al. 200328OesophagectomyHospital volume61301/04/1994–31/03/19995NR
Urbach and Baxter 200427OesophagectomyHospital volume61301/04/1994–31/03/19995NR
Urbach and Austin 200526OesophagectomyHospital volume and surgeon volume61301/04/1994–31/03/19995NR
van Lanschot et al. 20013OesophagectomyHospital volume13001993–19985NR
Morbidity

Four studies performed one statistical comparison of interest each. Two of these studies reported a significant association (Table 2).

Table 2.  Hospital volume morbidity outcomes
StudySignificant volume–outcome relationship reportedVolume threshold (cases per annum)
  1. HVH, high volume hospitals; LVH, low volume hospitals; MVH, medium volume hospitals; VHVH, very high volume hospitals; VLVH, very low volume hospitals.

Dimick et al. 200514<6 (LVH) versus >6 (HVH)
Dimick et al. 200316<34 (LVH) versus >34 (HVH)
Patti et al. 1998241–5 (VLVH) versus 6–10 (LVH)
11–20 (MVH) versus 21–30 (HVH) versus >30 (VHVH)
Swisher et al. 200025<5 (LVH) versus ≥5 (HVH)

Dimick et al.15 reported a statistically significant relationship between hospital volume and patient morbidity. In total, nine different patient complications were recorded. Within these complications, five (aspiration, pulmonary failure, renal failure and septicaemia) were shown to decrease significantly with increased hospital volume. However, the four remaining morbidity outcomes (cardiac complications, pneumonia, post-operative infection and surgical complications) showed no statistically significant decrease.

Dimick et al.16 also compared the incidence of specific complications at high and low volume hospitals. High volume centres showed a statistically significant reduction in the incidence of complications. Specifically, pulmonary failure, renal failure, aspiration, reintubation, surgical complications and septicaemia were all less commonly observed at high volume hospitals (Table 3).

Table 3.  Hospital volume mortality outcomes
StudySignificant volume–outcome relationship reportedVolume threshold (cases per annum)
  • Significant only in unadjusted relative risk calculations.

  • ‡Volume thresholds based on patient numbers over a 4-year period. HVH, high volume hospitals; LVH, low volume hospitals; MVH, medium volume hospitals; NR, not reported; VHVH, very high volume hospitals; VLVH, very low volume hospitals.

Bachmann et al. 200247–32 (LVH), 35–52 (MVH), 60–83 (HVH)
Begg et al. 199861–5 (LVH) versus 6–10 (MVH) versus ≥11 (HVH)
Birkmeyer et al. 20021<2 (LVHV) versus 2–4 (LVH) versus 5–7 (MVH) versus 7–19 (HVH) versus >19 (VHVH)
Birkmeyer et al. 20038<5.0 (LVH) versus 5.0–13.0 (MVH) versus >13.0 (HVH)
Birkmeyer and Dimick 20047<13 (LVH) versus >13 (HVH)
Birkmeyer et al. 200610NR<1.3 (VLVH), 1.3–2.0 (LVH), 2.1–3.0 (MVH), 3.1–7.3 (HVH), >7.3 (VHVH)
Birkmeyer et al. 20069NRNR
Dimick et al. 200112†‡≤3 (LVH) and 4–15 (MVH) versus >15 (HVH)
≤3 (LVH) versus >15 (HVH)
Dimick et al. 200315NR<6 (LVH), >6 (HVH)
Dimick et al. 200318<3 (LVH) versus >16 (VHVH)
3–5 (MVH) versus >16 (VHVH)
Dimick et al. 200316†‡<34 (LVH) versus >34 (HVH)
Dimick et al. 200511NR<6 (LVH), >6 (HVH)
Dimick et al. 200514<5 (LVH) versus >12 (HVH)
Elixhauser et al. 200317<7 (LVH) versus >7 (HVH)
Finlayson et al. 200318<4 (LVH) versus >9 (HVH)
Gordon et al. 19992010 (VLVH) versus >201 (HVH)
11–20 (LVH) versus >201 (HVH)
21–50 (MVH) versus >201 (HVH)
Ho et al. 2006213.2 (LVH), 3.5 (MVH), 3.8 (HVH)
Kuo et al. 200122<6 (LVH) versus >6 (HVH)
Lin et al. 20065NR<78 (VLVH), 78–135 (LVH), 136–235 (MVH), 236–346 (HVH), >346 (VHVH)
Patti et al. 1998241–5 (LVH) versus >30 (HVH)
Swisher et al. 200025<5 (LVH) versus ≥5 (HVH)
Urbach et al. 2003282.8 (LVH) versus 8.8 (MVH) versus 16.6 (HVH) versus 19.0 (VHVH)
Urbach and Baxter 200427<8.8 (LVH) versus >8.8 (HVH)
Urbach and Austin 2005260.2–2.1 (LVH), 2.2–7.0 (MVH), 7.1–12.0 (HVH), 12.1–14.4 (VHVH)
van Lanschot et al. 200131–10 (LVH) versus 11–20 (MVH) versus >20 (HVH)

Patti et al.24 reported no significant difference in morbidity rates across each of the five volume categories.

Two surgeon volume categories, as well as hospital volumes, were analysed by Swisher et al.25 This study reported a positive trend towards increased hospital volume and reduced patient morbidity (chi-squared analysis, P= 0.06).

Length of stay

Nine studies performed one statistical comparisons of interest each (Table 4). Six of the nine comparisons reported a statistically significant association between increased hospital volume and decreased patient length of stay. One study performed a statistical comparison where no significant association was reported. Three comparisons did not test for statistical significance.

Table 4.  Hospital volume length of stay outcomes
StudySignificant volume–outcome relationship reportedVolume threshold (cases per annum)
  1. HVH, high volume hospitals; LVH, low volume hospitals; MVH, medium volume hospitals; NR, not reported; VHVH, very high volume hospitals; VLVH, very low volume hospitals; †Significant only in unadjusted relative risk calculations.

Dimick et al. 200112≤3 (LVH) versus 15 (HVH)
Dimick et al. 200313<3 (LVH) versus >16 (VHVH)
Dimick et al. 200316NR<34 (LVH) versus >34 (HVH)
Dimick et al. 200511NR<6 (LVH) versus >6 (HVH)
Goodney et al. 200319<2 (VLVH) versus 2–4 (LVH) versus 5–7 (MVH) versus 8–19 (HVH) versus >19 (VHVH)
Kuo et al. 200122<6 (LVH) versus >6 (HVH)
Patti et al. 1998241–5 (VLVH) versus 6–10 (LVH) versus 11–20 (MVH) versus 21–30 (HVH) versus >30 (VHVH)
Swisher et al. 200025<5 (LVH) versus ≥5 (HVH)
Urbach and Austin 2005260.2–2.1 (LVH), 2.2–7.0 (MVH), 7.1–12.0 (HVH), 12.1–14.4 (VHVH)

Surgeon volume outcomes

Mortality

Six studies each performed one statistical comparison of interest, encompassing operative, operative plus 30-day mortality, in-hospital and 30-day mortality (Table 5). Five of six comparisons confirmed a statistically significant association between increased surgeon volume and decreased patient mortality. One comparison did not test for statistical significance.

Table 5.  Surgeon volume mortality outcomes
StudySignificant volume–outcome relationship reportedVolume threshold (cases per annum)
  1. HVS, high volume surgeons; LVS, low volume surgeons; MVS, medium volume surgeons; NR, not reported; VHVS, very high volume surgeons.

Bachmann et al. 200241–6 (LVS) versus 7–14 (MVS) versus 17–48 (HVS)
Birkmeyer et al. 20038<2.0 (LVS) versus 2.0–6.0 (MVS) versus >6.0 (HVS)
Dimick et al. 200514<2 (LVS) versus >5 (HVS)
Ho et al. 2006211.8 (LVS), 1.9 (MVS), 2.0 (HVS)
Miller et al. 199723≤5 (LVS), ≥5 (HVS)
Urbach and Austin 2005260.6–2.3 (LVS) versus 2.4–4.5 (MVS) versus 4.6–6.8 (HVS) versus 6.9–15.8 (VHVS)
Morbidity

Miller et al.23 were the only authors to examine the relationship between surgeon volume and morbidity (specifically, anastomotic leakage). This study reported no statistically significant association between surgeon volume and rates of anastomotic leakage.

Length of stay

The study of Urbach and Austin26 was the only included study to measure the relationship between patient length of stay and surgeon volume. This study reported a statistically significant association between surgeon volume and patient length of stay; however, because of the paucity of the reported data, the precise nature of this association could not be clearly reported.

Discussion

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. References

The review of the evidence from these retrospective studies suggests that oesophagectomy performed in high volume centres is associated with better outcome compared with low volume centres. While the evidence for each of the parameters of outcome varies, most of the results show a positive trend.

The majority of included studies reported a statistically significant association with either hospital or surgeon volume and patient outcomes, however, there was little methodological consistency between these studies, undermining any common statistical association. Despite the discrepancies in the recording of mortality figures, it would appear that there is a significant hospital volume-based advantage for oesophagectomy.

As surgeon volume was not commonly reported, no conclusive statement regarding the association between surgeon volume and patient outcomes can be confidently reported. While evidence of positive outcomes by experienced surgeons in a low volume setting has been published,29 the small number of patients involved means that the incidence of complications cannot be accurately assessed.

With regard to hospital stay, there would appear to be a positive volume-based effect, however, more studies are required to further assess this variable as only nine of the included studies included data to support this. In four of these five studies, only a modest clinical difference was found.

Other outcome factors such as quality of life, or longer term measures of outcome such as recurrence, are more difficult to assess because of the paucity of data.

Overall, no evident difference in costs between high and low volume hospitals was found.

Another problem encountered when evaluating volume-based outcome studies is the parameters used to define volume categories. Depending on the study size, the ranking of volume can vary considerably among centres. While there is overlap between some studies' definitions, the most significant comparisons are made between equivalent volume measurements. The question also posed is, at what threshold should a centre be classified as large volume? It would be very difficult to answer this question in absolute terms as setting an arbitrary figure would not take into account the population that the health care system serves. A potential alternative could be expressed as a percentage of the total number of procedures performed nationally or within the region in question. This is with the aim of achieving a threshold figure at which outcome has been shown to improve. Conversely, the same technique could be applied to define a low volume limit at which outcome has been shown to be significantly worse. The authors concede that the lack of absolute figures to define volume categories may leave readers frustrated at the lack of simple guidelines by which treatment can be planned. However, the decision to refer a patient to a higher volume unit can only really be performed with a knowledge of the facilities available locally. Without a realistic alternative for the patients' treatment, it is pointless to set referral criteria that cannot be achieved. The institutions involved should have audit processes in place so that patient outcomes can be easily compared between sites and guidelines that are relevant to their practice established. When looking on a national basis at the available resources and population size, using a ranked volume analysis can provide useful data for the planning of health care provision as international regionalization is not feasible.

One of the commonly encountered problems when reviewing patient outcomes is the use of administrative data to identify both patients and complications. As mentioned previously, coding techniques were different among the studies included in this review, which will undoubtedly have an affect on reporting of patient covariances, and especially patient morbidities. This is further confounded by the potential for coding errors, which may not be recognized by clinical staff.

An aspect not fully explored by this review is the underlying cause for any volume-based effect. Undoubtedly, causation is multifactorial, and any volume-based effect is the reflection of a synergy in care provision on a number of levels. While analysis of multiple hospital and health care professional characteristics may highlight some statistically significant aspects of care, these measures are much more complex and difficult to measure than volume alone. A significant measure of outcome for a large volume centre may not necessarily be transferable to a low volume setting. For example, while a prolonged Intensive Therapy Unit (ITU) stay in a large volume centre may be associated with a poorer outcome,30 in a low volume centre, the only reason for a prolonged ITU stay could be a lack of step-down care before return to the ward.

It has been shown that results in large volume centres can be replicated by importing the necessary expertise and care pathways.31 It is difficult to say that this would always be achievable, particularly in a rural setting, and as such, the predominant evidence would suggest that oesophagectomy should be performed in high volume centres.

References

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. References