Objective This study assessed the associations between hospital volume, resection rate and survival of oesophageal and gastric cancer patients in England.
Design 62 811 patients diagnosed with oesophageal or gastric cancer between 2004 and 2008 were identified from a national population-based cancer registration and Hospital Episode Statistics-linked dataset. Cox regression analyses were used to assess all-cause mortality according to hospital volume and resection rate, adjusting for case-mix variables (sex, age, socioeconomic deprivation, comorbidity and type of cancer). HRs and 95% CIs, according to hospital volume, were evaluated for three predefined periods following surgery: <30, 30–365, and >365 days. Analysis of mortality in relation to resection rate was performed among all patients and among the 13 189 (21%) resected patients.
Results Increasing hospital volume was associated with lower mortality (ptrend=0.0001; HR 0.87, 95% CI 0.79 to 0.95 for hospitals resecting 80+ and compared with <20 patients a year). In relative terms, the association between increasing hospital volume and lower mortality was particularly strong in the first 30 days following surgery (ptrend<0.0001; HR 0.52, (0.39 to 0.70)), but a clinically relevant association remained beyond 1 year (ptrend=0.0011; HR 0.82, (0.72 to 0.95)). Increasing resection rates were associated with lower mortality among all patients (ptrend<0.0001; HR 0.86, (0.84 to 0.89) for the highest, compared with the lowest resection quintile).
Conclusions With evidence of lower short-term and longer-term mortality for patients resected in high-volume hospitals, this study supports further centralisation of oesophageal and gastric cancer surgical services in England.
- Oesophageal Cancer
- Gastric Cancer
Statistics from Altmetric.com
Significance of this study
What is already known about this subject?
The recommendations outlined in the 2001 Improving Outcomes Guidance for upper gastrointestinal cancers have led to substantial changes in the services for oesophageal and gastric cancers over the last decade, with increasing centralisation of surgical referrals to nominated centres.
There is substantial evidence that increasing hospital and surgeon volumes are associated with improved 30-day or in-hospital mortality.
Fewer studies have considered the impact of hospital volume on long-term survival and these have shown conflicting results.
What are the new findings?
We found that increasing hospital volume was associated with lower mortality for resected patients.
The absolute risk of dying within 5 years ranged from 69% in the lowest-volume group to 61% in the highest-volume group. The risk of dying within 30 days ranged from 7% to 4%, respectively.
This association was, in relative terms, found to be particularly strong in the first 30 days following surgery, but a clinically relevant association remained beyond 1 year.
How might it impact on clinical practice in the foreseeable future?
With evidence of higher short-term and longer-term survival for patients resected in high-volume settings, our results support the recommendations set out in the Improving Outcomes Guidance for centralisation of oesophageal and gastric cancer surgical services in England.
In England, around 13 000 people are diagnosed with oesophageal and gastric cancer each year.1 The prognosis of these cancers is poor, with a 5-year relative survival rate of 12% and 16%, respectively.2 Surgical resection can offer long-term survival in appropriate patients.3 Only 20% of patients with oesophageal and gastric cancer in England underwent resection with curative intent in 2005, compared with 28% in 1998.4
In 2001, the Improving Outcomes Guidance proposed that specialist multidisciplinary teams located in cancer centres should aim to draw patients with oesophageal and gastric cancer from an area of at least one to two million people.3 The report stipulated that for a population of around one million, teams could expect to manage at least 100 patients with oesophageal cancer, and 150 with gastric cancer per year.3 These recommendations led to substantial changes in the delivery of upper gastrointestinal cancer services in England over the last decade, with increasing centralisation of surgical referrals to nominated centres.5
Centralisation of upper gastrointestinal cancer services is intended to improve outcomes for patients with oesophageal and gastric cancer.3 The results of studies considering hospital volume and survival for these cancers are heterogeneous. Differences in terms of study design, quality, sample sizes, cut-off values for annual volume, and ability to adjust for potential confounders may partly explain the variations in these findings.6 ,7 However, there is substantial evidence that increasing hospital and surgeon volume is associated with lower 30-day or inhospital mortality.7–20 Fewer studies have considered the impact of hospital volume on long-term survival, and these have shown conflicting results.6 ,7 ,17 ,19 ,21–23 Surgical resection remains the mainstay for cure and, therefore, resection rate may also need to be considered.
The purpose of this study was to assess the associations between oesophageal and gastric cancer hospital resection volumes (the annual number of resections), resection rates (the proportion of patients who are resected), and mortality rates for patients with these cancers in England. We hypothesised that increasing volume would be associated with lower mortality for resected patients, both in the short-term and long-term perspectives. We also hypothesised that as resection rates increase, mortality will be lower for all patients, and possibly higher among the resected patients, because higher-risk patients may be resected in areas with high resection rates.
Data on 64 711 oesophageal (ICD10 C15) or gastric (ICD10 C16) cancers diagnosed in England between 2004 and 2008 were extracted from the National Cancer Data Repository (NCDR). The NCDR contains information collected by the eight English population-based cancer registries on all cancer patients diagnosed in their catchment areas. Death information is supplied by the National Health Service (NHS) Central Register via the Office for National Statistics. These data are quality assured in each cancer registry before being combined. Registrations which only had information from death certificates (n=1611) and without an NHS number (n=174) were excluded. Only the first tumour for each patient was selected, which excluded a further 111 records on patients with multiple tumours. Finally, four patients were excluded, as their recorded dates of death were before their operation dates. The final dataset included 62 811 patients.
Patients were aggregated into 5-year age groups based on their age at diagnosis (<55 years, 55–59, …, 80–84 and 85+). Patients were grouped into quintiles of socioeconomic deprivation based on their postcode and lower super output area (each comprising a population of around 1500 people) of residence, using the income domain of the Indices of Deprivation 2007.24
Comorbidity information was obtained from a linked Hospital Episode Statistics admitted patient dataset supplied by the NHS Information Centre. For each patient, a comorbidity score was derived using diagnosis codes recorded within inpatient and day-case episodes between 2 years prior to and 3 months after the patient's date of diagnosis.25 Standard weights were assigned according to the severity of the condition,26 and the resulting scores were aggregated into four categories of increasing severity of comorbidity: 0 (no comorbid conditions), 1 (comorbidity score of 1), 2 (comorbidity score of 2), or 3 (comorbidity score 3 or higher). It was not possible to derive a comorbidity score for 2426 (3.9%) patients because they had no linked Hospital Episode Statistics record.
Surgery information was also obtained from the Hospital Episode Statistics dataset. Tumour resections from 1 month before to 12 months after the patient's diagnosis date were extracted. The resections for oesophageal or gastric cancers were defined as: oesophagogastrectomy, total or partial oesophagectomy, total or partial gastrectomy or other and unspecified total or partial excisions of the oesophagus or stomach. If a patient had more than one of these operations, the earliest operation was selected for the analysis.
The number of resections was available at the organisation level of NHS hospital trust. In England, an NHS hospital trust manages one or more local hospitals. In this paper, we refer to NHS hospital trusts simply as ‘hospitals’, and the annual number of resections in an NHS hospital trust is referred to as the ‘hospital volume’.
For each resected patient, hospital volume was computed as the number of oesophageal or gastric cancer resections carried out in the hospital in which they were treated, and in the same year as their diagnosis. Hospital volumes were aggregated into five groups: <20 resections per year, 20–39, 40–59, 60–79, and 80+ resections per year.
Hospitals contributed to more than one volume group when the number of resections in each hospital varied between diagnosis years. There were 144 individual hospitals included in the analysis, of which 108 individual hospitals contributed to the <20 volume group in at least one diagnosis year, and seven individual hospitals to the 80+ group.
The resection rate was defined as the resected proportion of patients resident in each geographical primary care trust area in each year of diagnosis. These proportions were then divided into quintiles representing areas with increasing proportions of resected patients.
The numbers and proportions of resected patients by hospital volume, resection quintile and case-mix variables, that is, sex, age, socioeconomic deprivation, comorbidity and type of cancer, were tabulated. p Values for trend or heterogeneity were calculated, as appropriate.
Univariate and multivariate Cox proportional hazards regression analyses were used to estimate the all-cause mortality HRs and 95% CIs according to hospital volume, resection quintile, and the case-mix variables. Analyses of resection quintile were performed separately for all patients and for resected patients.
For all patients, survival time was calculated from the date of diagnosis until death from any cause, and patients who were alive at the end of follow-up were censored on 31 December 2009. In the analysis restricted to patients who received surgery, survival time was calculated from the date of the operation. In the multivariate Cox proportional hazards analyses, hospital volume and resection quintile were adjusted for the available case-mix variables and mutually adjusted.
For each variable included in the models, p values for heterogeneity and trend were calculated as appropriate.
To assess the HRs according to hospital volume in different time periods after surgery, the follow-up period was divided into three predefined periods: short-term (<30 days postsurgery), medium-term (30–365 days postsurgery), and long-term (>365 days postsurgery). These analyses were adjusted for case-mix variables and resection quintile.
Of the 62 811 included patients diagnosed with oesophageal or gastric cancer in England between 2004 and 2008, 13 189 (21%) underwent surgical resection (table 1). The number of patients resected in each hospital in each diagnosis year ranged from 1 to 132. The proportion of patients who underwent resection in each of the 152 primary care trust areas of residence in each diagnosis year ranged from 4% to 46%. There was a positive association between hospital volume and resection quintile with patients operated in high-volume hospitals being more likely to live in areas where a higher proportion of patients were resected (Spearman's r=0.27, p<0.0001).
Surgical resection rates were lower in females than in males (17% vs 23%, p<0.0001), and for oesophageal cancer compared with gastric cancer (17% vs 25%, p<0.0001), (table 2). The proportion of patients who underwent resection decreased with increasing age, socioeconomic deprivation and severity of comorbidity (all with ptrend<0.0001).
Hospital volume and mortality of resected patients
Increasing hospital volume was associated with lower mortality (ptrend=0.0001), with a HR of 0.87 (95% CI 0.79 to 0.95) in hospitals carrying out 80+ resections compared with those carrying out less than 20 resections a year (table 3). Following adjustment for case-mix variables, hospital volume estimates did not change materially. Adjustment for resection quintile strengthened the association (HR 0.81, 95% CI 0.74 to 0.89), in hospitals carrying out 80+ resections compared with those carrying out less than 20 resections a year). Without any adjustment, the absolute risk of dying within 5 years in the five hospital volume groups ranged from 69% in the lowest group to 61% in the highest.
Mortality of resected patients, stratified by period of follow-up
The inverse association between hospital volume and mortality was, in relative terms, strongest during the first month after surgery in the fully adjusted model (ptrend<0.0001), with a HR of 0.52, (95% CI 0.39 to 0.70) in hospitals carrying out 80+ resections compared with less than 20 resections a year (table 4). A similar pattern was seen in the medium-term and long-term periods after surgery (ptrend=0.0370, HR 0.88, 95% CI 0.76 to 1.01, and ptrend=0.0011, HR 0.82, 95% CI 0.72 to 0.95, in hospitals carrying out 80+ resections compared with those carrying out less than 20 resections a year, respectively). Without any adjustment, the absolute risk of dying within 30 days in the five hospital volume groups ranged from 7.3% in the lowest volume group to 4.1% in the highest.
Resection rate and mortality of all patients
The overall mortality was higher in females than in males (HR 1.09, 95% CI 1.07 to 1.11, p<0.0001), (see web-only table 1). Mortality was higher with increasing age (ptrend<0.0001, HR 2.70, 95% CI 2.60 to 2.81, for patients 85+ compared with patients under 55 years old), in patients living in more socioeconomically deprived areas (ptrend<0.0001, HR 1.15, 95% CI 1.11 to 1.18, for those living in the most deprived areas, compared with the least deprived areas), and in patients with increasing severity of comorbidity (ptrend<0.0001, HR 1.49, 95% CI 1.43 to 1.54, for patients with the highest comorbidity score compared with no comorbidity). Mortality was similar in oesophageal compared with gastric cancer patients (HR 0.98, 95% CI 0.96 to 1.00).
Increasing resection rates were associated with lower mortality (ptrend<0.0001), with a HR of 0.86 (95% CI 0.84 to 0.89) in the highest resection quintile compared with the lowest resection quintile, (table 5). Adjustment for case-mix attenuated the association somewhat, but a relevant and statistically significant association remained (ptrend<0.0001, HR 0.90, 95% CI 0.87 to 0.92, in the highest resection quintile compared with the lowest resection quintile).
Resection rate and mortality of resected patients
In resected patients, the overall mortality was lower in females than in males (HR 0.88, 95% CI 0.83 to 0.93, p<0.0001) and higher with increasing age (ptrend<0.0001, HR 2.09, 95% CI 1.76 to 2.47, for patients 85+ compared with patients under 55 years old), and with increasing severity of comorbidity (ptrend<0.0001, HR 1.51, 95% CI 1.34 to 1.70, for patients with the highest comorbidity score compared with no comorbidity), (see web-only table 2). Mortality after surgery was similar between deprivation quintiles (p=0.1373) and between oesophageal and gastric cancer patients (p=0.6628).
Increasing resection rates were associated with higher overall mortality (ptrend<0.0001), with a univariate HR of 1.20, 95% CI 1.10 to 1.31 in the highest resection quintile compared with the lowest quintile (table 6). Following adjustment for case-mix variables, the HRs remained similar in each resection quintile (HR 1.20, 95% CI 1.10 to 1.30, in the highest compared with the lowest resection quintile). Further adjustment for hospital volume slightly strengthened this association (HR 1.26, 95% CI 1.15 to 1.37, in the highest resection quintile, compared with the lowest resection quintile).
We found that increasing hospital volume was associated with decreased mortality among the resected patients, which is consistent with previously published literature.6–17 19–22 In relative terms, the association between hospital volume and mortality was strongest in the first 30 days following surgery, but a clinically relevant and statistically significant association was also evident in the longer term. These findings were unaffected by adjustment for the available case-mix variables despite these variables, in general, being independently associated with the risk of dying. This is because these factors are not materially associated with hospital volume or resection rate. In absolute terms, the risk of dying within 30 days ranged from 7% in the lowest-volume group to 4% in the highest-volume group. The risk of dying within 5 years ranged from 69% to 61%, (these estimates are not adjusted for covariates and are, therefore, conservative).
Increasing resection rates for oesophageal or gastric cancer were associated with lower mortality among all patients, but higher mortality among those who underwent resection, which is consistent with our previous study of patients with lung cancer in England.27 These findings suggest that, in areas where a greater proportion of patients are resected, some higher-risk patients may be included in the resected population, leading to higher mortality in this group. Conversely, in areas where a lower proportion of patients are resected, some patients who are potentially resectable may not be included, therefore, leading to lower mortality. Future detailed investigation will be important in understanding the exact mechanisms driving this pattern. Therefore, at present, caution needs to be exercised as to whether it is possible that surgical resection could be offered to more patients. Recommendations should take into account tumour stage, use of combined chemotherapy modalities, and the quality of surgery as indicated by lymph node counts, resection margins, rate of completeness of the resection (R0), and local recurrence failure rates.
The population-based design and the large national sample of patients diagnosed with oesophageal or gastric cancer over a 5-year period across the whole of England are among the strengths of the study. This study also benefited from covering a period when centralisation of upper gastrointestinal cancer services in England was on-going, leaving a wide range of variation in hospital volumes.5 In our study, the percentage of patients resected in hospitals carrying out 80 or more operations increased from 7% in 2004 to 20% in 2008, with a corresponding decrease in the less than 20 volume group of 22% to 7%, respectively. The high volumes of resections conducted in many of the hospitals enabled us to analyse truly high-volume hospitals, including those resecting 80 or more patients per year. This study was also able to consider both short-term and long-term survival outcomes with good statistical power.
One of the limitations of this study is that our hospital volume measure is defined using the first relevant operation for each patient with cancer. A small proportion of patients (2%) had been resected more than once, and the hospital volume measure is therefore not a complete reflection of all resections carried out in each hospital. The surgery information in this study was taken from the Hospital Episode Statistics dataset. A systematic review found acceptable accuracy for procedure codes from NHS administrative data.28
There is inevitably some misclassification between oesophageal cancer and gastric cancer, particularly for the cancers occurring near the gastro-oesophageal junction. The analysis was repeated for patients with oesophageal cancer and patients with gastric cancer separately, and the results were not materially different from those reported overall (see web-only tables 3 and 4).
The comorbidity score and treatment information included in this study are obtained from an inpatient and day-case admissions dataset. Therefore, as the comorbidity score for each patient is only based on diagnosis codes recorded in this setting, it may under-ascertain the comorbidity in these patients. Not all patients receive chemotherapy as inpatients, and most radiotherapy is received in an outpatient setting which meant it was not possible to ascertain accurately how many patients received non-surgical treatment. Therefore, information on non-surgical treatments for these cancers was considered incomplete and was not included in this analysis. Also, there was no information available on the referral patterns of patients, and whether or not the proportion of patients referred was different between hospitals. Finally, tumour stage is strongly associated with long-term survival. Tumour stage is poorly recorded in the English cancer registries, and it was not possible to take account of stage in these analyses. Other known prognostic factors, including sex, age, socioeconomic deprivation and comorbidity were adjusted for, and did not materially change any of the main results.
Taken at face value, our data suggest that an increase of the annual hospital volume to about 60 or higher would be available to 70% of the resected patients, and represent a relative mortality reduction of up to 20%. If this was a medical treatment without major side effects, it would almost certainly become standard care. The study was not designed to identify an optimum number of resections in hospitals. The figure of 60 presented here is an arbitrary but predefined cut-off point for the analysis.
Hospital volume and outcome has been widely investigated since Luft et al first published their study in 1979 demonstrating lower mortality with increasing surgical volume for 12 surgical procedures of varying complexity in the USA.29 Many studies from Europe and the USA have found that, in general, centralisation of services into high-volume hospitals, 6–17 19–23 30–33 or to high-volume surgeons,7 ,9 ,12 ,14–16 ,18 is beneficial for complex surgery for various diseases including cancer.6–22 30–33 However, the methods of some of these studies have been criticised,34 and other studies have found no evidence of a benefit of high volume.13 ,23 ,34
We do not know the specific aspects of care in high-volume hospitals that act to improve short-term and long-term outcomes. Hospital volume is probably an indicator for institutional factors that may influence survival more directly. Two possible explanations of the association between high volume and lower mortality have been put forward.35 The first suggests that the ‘selective referral’ of patients to hospitals that already have superior outcomes would result in a higher volume of patients in these settings.9 ,35 The second explanation, ‘practice makes perfect’, proposes that more experience gained in hospitals that treat a greater number of patients could lead to improvements in the management of patients across the whole treatment pathway.35 For example, higher volumes could (1) improve the processes of care along defined protocols in a multidisciplinary approach; (2) further the experience in managing postoperative complications and (3) increase the technical experience of surgeons in performing complex oesophageal and gastric cancer surgery. Such factors may contribute to the lower postoperative mortality in high-volume hospitals. The improved long-term survival in high-volume settings may be due to more accurate cancer staging, appropriate use of combined oncological modalities, and superior surgical techniques.9 It is also possible that surgeons who already achieve better outcomes are attracted to work in higher-volume hospitals, which could result in the decreased mortality observed in these settings. If the latter point is true, merely increasing hospital volume will not necessarily lead to improved outcomes. Identifying the mechanisms that drive the improved outcomes in high-volume providers will be important in future and more detailed studies.
The recommendations of the 2001 Improving Outcomes Guidance for upper gastrointestinal cancers have led to increasing centralisation of surgical referrals to nominated centres.3 ,5 With evidence of lower short-term and longer-term mortality for patients resected in high-volume settings, this study lends support to further centralisation of oesophageal and gastric cancer surgical services in England.
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Files in this Data Supplement:
Contributors This study was conceived and designed by VHC and HM. VHC prepared and managed the dataset and VHC, RHJ and ML carried out the statistical analysis. VHC drafted the paper and all authors reviewed and revised the drafts and provided interpretation of results. All authors have approved the final version of the manuscript. VHC is the guarantor.
Funding This paper is a contribution from the National Cancer Intelligence Network and is based on the information collected and quality assured by the regional cancer registries in England (in http://www.ukacr.org; http://www.ncin.org.uk). This work was carried out by the Thames Cancer Registry, King's College London, which receives funding from the Department of Health. The research was supported by the National Institute for Health Research (NIHR) Biomedical Research Centre based at Guy's and St Thomas’ NHS Foundation Trust and King's College London. The views expressed are those of the author(s) and not necessarily those of the NHS, the NIHR or the Department of Health. Victoria H Coupland also received funding through a King's College London postgraduate studentship. The sponsors of this study had no involvement in the study design, data collection, data analysis, interpretation of data, writing the manuscript, or in the decision to submit this paper for publication.
Competing interests Mr William Allum received payment for a presentation at the 1st St Gallen Consensus Conference on GI Malignancy. All other authors declare no competing interests.
Ethics approval Cancer registries in England have approval from the National Information Governance Board to carry out surveillance using the data they collect on all cancer patients under Section 251 of the NHS Act 2006. Therefore, separate ethical approval was not required for this study.
Provenance and peer rerview Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.