Mortality and morbidity in gastro-oesophageal cancer surgery: initial results of ASCOT multicentre prospective cohort study

Data source

The ASCOT (Assessment of Stomach and Oesophageal Cancer Outcomes from Treatment) database was developed in 1998 by the British Oesophagogastric Cancer Group to provide comprehensive and accurate data on stage, comorbidity, and outcome for cases of oesophageal or gastric cancer.4 The dataset comprises 54 data fields divided into sections dealing with demographic data, preoperative assessment, operative details, postoperative course, and pathology. From 1 January 1999 to 31 December 2002, 32 hospitals in England and Wales (about 10% of all acute trusts doing major cancer surgery) registered with ASCOT on a voluntary basis. Members agreed to enter all details of all patients with gastric or oesophageal cancer referred to them, whether patients underwent resection or not, but relied on their own resources to do this.

Inclusion and exclusion criteria

In this study of surgical outcomes, we included data on patients undergoing resectional surgery for oesophageal and gastric cancer. We excluded patients who underwent non-surgical treatment (palliative chemotherapy, radiotherapy, endoscopic stenting), patients who underwent surgery without resection (palliative gastrojejunostomy, laparotomy only), and patients whose outcome was not recorded.

End points and risk factors

The primary outcome was mortality in hospital after resection of gastric or oesophageal cancer, defined as any death during the admission when the operation was performed. This was more reliably quantified than 30 day mortality. The secondary end point was morbidity in hospital. We performed separate analyses of complications apparently related to surgical technique (for instance, anastomotic leakage, fistula, abscess, haemorrhage) and other “medical” complications. The risk factors studied included age, sex, comorbid status, tumour site and stage, surgical approach, and annual case volume (defined as the number of oesophageal and gastric resections undertaken in each unit per year and categorised into three groups; see table 5). We classified tumour site as oesophageal, gastric, or junctional. Data on preoperative health status were collected by using the physiological part of the POSSUM (physiological and operative severity score for the enumeration of mortality and morbidity) score,5 the ECOG (Eastern Cooperative Oncology Group) performance status,6 ASA (American Society of Anesthesiology) grade,7 and a surgeon's simple assessment score (see table 3).

Statistical analysis

Continuous risk factors were grouped into subcategories of increasing operative risk. We used a multilevel logistic regression analysis with backwards stepwise variable selection to identify independent risk factors for operative mortality and morbidity. The standard errors of the model estimates were adjusted for the clustering of patients within hospitals. To maximise the information extracted from the predictor variables, we used a median imputation technique to substitute for incomplete data.8

Model validation —The adjusted odds ratios were based on the multivariate analysis of the cases submitted during the first three years of the study (n = 773). The model was then tested on the data from the final year of the study (n = 222). Model performance was evaluated by the area under the receiver operating characteristic (ROC) curve and the Hosmer Lemeshow test.8

Data validation —The ASCOT data were internally validated by requesting duplicate information from the information technology, pathology, and theatre departments of participating hospitals. To test for bias from incomplete reporting, we compared the case volume and operative mortality for oesophageal and gastric resections reported to the ASCOT database with data from the hospital episode statistics (HES), obtained independently for 12 participating units for 1999-2000.

Statistical software —We used STATA version 6.0 software (STATA Corporation, College Station, TX), SPSS version 11 for Windows (SPSS, Chicago, IL), and MLwiN version 1.2 (University of London) in the analysis of results.

Surgical treatment

In 23 hospitals 1251/2087 patients (60%) underwent surgical treatment and 955 (46%) underwent tumour resection. Three hospitals submitted data but had no patients undergoing resection. Demographic details for patients undergoing resection are shown in table 1. Operative volume ranged between 2 and 39 cases a year. Nine hospitals performed 10 or fewer resections a year and three hospitals performed more than 30 oesophagogastric resections a year. Of the patients, 590 underwent gastric resection (resection rate 54%), 365 underwent oesophageal resection (resection rate for oesophageal or junctional cancers 37%), and 255 underwent non-resectional palliative or exploratory procedures. Only 62 (6%) resections were considered palliative in intent (table 2). There were 254 total gastrectomies, 262 distal gastrectomies, and 365 oesophagectomies. The chest was opened in 337 procedures. Overall, 232 (64%) of oesophagectomies incorporated a 2-field lymphadenectomy, and 40% of gastrectomies had “D2” lymphadenectomy. Additional organ resection was performed in 224 patients (23%).

Contributing factors

Factors that may contribute to high operative mortality include a high frequency of unrelated systemic disease, liberal selection criteria for operation, low case volumes in individual units, poor quality or inappropriately radical surgery, and inadequate standards of postoperative care. Occult cardiac disease is higher in the United Kingdom than in many other European countries,14 and, by international standards, UK surgeons work with limited postoperative nursing and intensive care unit support.15 16 Patients of surgeons who operate on a high proportion of referred cases have higher mortality,17 but the 46% curative resection rate in this series was not unusually high. Radical lymphadenectomy was not associated with increased mortality in our series, although it was associated with medical and surgical complications.

We found that factors related to patients—the stage and site of the tumour and general health and fitness—were the strongest predictors of postoperative death and complications. The type of resection is determined largely by the site of the tumour, but the exposure used, the extent of node dissection, and, to a lesser extent, excision of additional organs are aspects of the operation over which the surgeon has some control and that independently predicted death or complications. In our study, total gastrectomy was as strongly associated with mortality as oesophagectomy. Only 15 of the participating units performed any oesophagectomies, so this procedure may be more widely recognised as requiring specialist skill and resources. “Medical” complications were predicted by two measures of patient fitness, by sex, and by more extensive surgery, but surgical complications were predicted only by tumour stage and additional organ resection, although radical node dissection came close to independent significance. Thus the robustness of the patient is not a factor in predicting these complications.

High case volumes in specialist cancer surgery units are associated with superior outcomes in other healthcare systems,4 18 19 but evidence about this association in the UK context is limited. The Scottish audit did not find a clear effect of case volume on mortality,13 while an audit in the South West region found an effect that just achieved significance.20 We found a trend in favour of a weak volume effect compatible with the latter study. The presence of an association does not prove causality, and the proposition that increasing average unit case throughput would improve mortality remains unproved. The test set results showed the reliability of the model developed from the first three years of data by their confirmation of the predictive associations noted.

Problems with data collection

Submission of data to ASCOT was voluntary and essentially unfunded, and, not surprisingly, the completeness of data recording varied considerably. Rates of omission of clinical data were reduced by internal verification work to between 5% and 10%.21 Overall, ASCOT recorded two thirds of the cases recorded by hospital episode statistics, although figures for individual trusts varied considerably. Comparison of mortality data with figures from hospital episode statistics showed no evidence of overall bias, even among trusts who contributed few cases to ASCOT. The uneven quality of the hospital episode statistics data reflected the experience of others.22 Large clinical databases are robust and generally produce effect estimates close to the true effect in the face of substantial omissions.23 24 Multiple databases gave similar results for cardiac surgery outcomes.25

Our study was limited in scope to the elucidation of factors associated with death and complications. Systems like ASCOT will permit the development of accurate predictive scoring models and of risk adjusted comparisons of performance between hospitals and surgeons, but for these purposes near complete data will be required. We have shown both the potential value and the problems of large prospective databases as tools for surgical audit and research. Resources should be provided to improve the ability of multidisciplinary teams to record complete, comprehensive, and accurate data on all patients undergoing major surgery for cancer.

The ASCOT project was developed by the membership of the British Oesophago-gastric Cancer Group. We thank all the consultants who contributed patients to the study and the data collection officers for their assistance. Abdullah Jibawi, David Corless, and David Cade of the Surgical Research Group at Leighton Hospital, Crewe, provided and prepared the HES data for external validation of the ASCOT dataset.Consultants at the following trusts registered with the ASCOT study: Addenbrooke's NHS Trust, Cambridge; Aintree University Hospital, Liverpool; Airedale NHS Trust, Keighley, North Yorkshire; Barnet General Hospital, Hertfordshire; Bishop Auckland General Hospital, County Durham; Broomfield Hospital, Chelmsford; Chorley General Hospital, Lancashire; East Cheshire NHS Trust; Macclesfield District General Hospital, Macclesfield; Essex Rivers Healthcare, Colchester General Hospital, Essex; Furness General Hospital, Cumbria; Glenfield Hospital, Leicester; Harrogate District Hospital, Harrogate; Huddersfield Royal Infirmary, Huddersfield; Leicester Royal Infirmary, Leicester; Leighton Hospital, Crewe, Cheshire; Maidstone General Hospital, Maidstone; Morecambe Bay NHS Trust; Royal Lancaster Infirmary, Lancaster; Newham General Hospital, London; Norfolk and Norwich NHS Trust, Norwich; North Staffordshire City General Hospital, Stoke on Trent; Queen Elizabeth Hospital, Birmingham; Royal Bolton Hospital, Bolton; Royal Free Hospital, London; Royal Hull Hospitals NHS Trust, Hull; Ipswich Hospital, Ipswich; Royal Bournemouth Hospital, Bournemouth; Watford General Hospital, Hertfordshire; West Wales General Hospital, Carmarthen.