Rectal cancer is a common disease in the UK with 13 000 cases and 5000 deaths per year.1 Surgery is very important in achieving cure so optimising its delivery is important. Evidence exists to suggest that there are significant variations in the quality of surgery performed and this, in turn, affects outcomes.2^–4

A number of operations including abdominoperineal excision (APE), anterior resection (AR) and Hartmann’s resection are available to excise rectal tumours. The traditional curative procedure has been the APE which requires the removal of the anal sphincter and, hence, necessitates a permanent stoma. Whilst this operation may be the only option in patients with very low rectal tumours, for the majority the more modern technique of AR can be performed which allows the retention of the sphincter and so may be preferable to the patient.

The introduction of modern techniques such as total mesorectal excision (TME) and the use of low AR has led to significantly improved local disease control and overall survival.5^–8 In contrast, evidence is accumulating to suggest that the APE is generally performed badly and is associated with higher rates of circumferential resection margin positivity, perforation, local recurrence and poorer survival than AR.9^–13 Rates of APE are slowly dropping14 but there is a concern that some surgeons may overuse this procedure. This will result in some patients not only receiving an avoidable colostomy but also undergoing a more deforming operation.

Since 1995 the UK Government has undertaken widescale reform of UK cancer services to ensure a “high quality of cancer care for all”15 and improve cancer survival in the UK. Quantifying and addressing variations in surgical practice is a key component in trying to achieve this. The Department of Health has published guidance16 stating that, wherever possible, surgeons should try to ensure anal sphincters are preserved. In addition, the Pelican Multidisciplinary Team Total Mesorectal Excision (MDT-TME) training programme17 18 was established to ensure all surgeons operating on rectal cancer patients in England offered the best surgical techniques. Significant investment has been made in these schemes but at present there is no data source that allows the variation in practice to be quantified or the impact of such interventions to be assessed. Recently, however, there has been a growing appreciation of the importance of undertaking national comparative audit to monitor performance and outcomes across the NHS and realisation of the potential for routinely available health datasets to achieve this.19^–21

In consequence, this retrospective population-based study sought, via linkage of cancer registry and Hospital Episodes Statistics (HES) data, to examine variation in rectal cancer surgery across England between 1998 and 2004. We sought to determine rates of use of APE and other rectal cancer surgery across the country at a population level and determine if any variation could be explained by differences in patient characteristics such as stage of disease, age, gender or socioeconomic deprivation. In addition we sought to quantify the extent of variation in practice between cancer networks, hospital trusts and surgeons.

METHODS

Basic information (age, gender, Dukes’ stage, dates of diagnosis and death, NHS number and postcode at diagnosis) for all rectal cancer patients (ICD C20) diagnosed between 1998 and 2004 in England were provided by the eight English cancer registries and pooled to produce a national dataset. Patients with multiple tumours were excluded.

These data were then linked (using all or combinations of the identifiers of NHS number, date of birth, postcode at diagnosis and gender) to an extract of HES covering the time period of April 1997 to March 2005 that included episodes of care for individuals who had presented with a diagnosis code of colorectal cancer in any episode over this time period.

Information about patient management was derived from HES. All the episodes for each patient identified in the registry data extract and linked to the HES dataset were searched to identify the initial surgical procedure for their rectal cancer. This was done by searching the episodes for the first OPCS4 coding of an APE (H331), AR (H332–4, H336), Hartmann’s resection (H335), excision of rectum unspecified (H338–9), panproctocolectomy (H041–9), total colectomy (H051–9), excision of sigmoid colon (H101–9) or left hemicolon (H091–9). If a patient received two of these operations within different episodes the first operation that occurred was used. Information regarding the method of admission to hospital, managing trust and the consultant in charge of each patient’s care was derived from the HES episode that detailed the patient’s initial surgical procedure.

Registry data enabled the identification of dates of diagnosis for each tumour. Within HES, however, we were able to see all treatments a patient received in the period April 1997 to March 2005. In some cases it was possible that the treatment information seen in HES related to tumours originating prior to the period covered by our registry extracts. For example, if a patient had a primary tumour in 1996 and a second primary in 2004 it was possible the linkage process would result in us taking the treatment for the cancer in 1996 as the treatment for tumour diagnosed in 2004. To avoid this, patients who were treated 31 days or more prior to their date of diagnosis were excluded. A short negative interval was allowed as in some cases dates of pathology were being used as dates of diagnosis in the registry which meant diagnosis appeared to occur after surgery. As histopathological reporting of colorectal specimens should be completed within a month, a 31 day threshold was used to clean the data.

Annual median rectal surgical workload was derived for each of the surgeons identified as operating on the patients included in our dataset. Annual median workloads greater than seven were used as a threshold to determine a specialist surgeon. This threshold was derived from the “Improving Outcomes Guidance” which stated the specialist MDT surgeons should be operating on a minimum of 20 colorectal cancer cases per year.16 22 As a third of colorectal cancers are rectal tumours a threshold of seven cases per year was chosen as the specialist workload threshold.

Patients who received a major abdominal procedure were categorised into four groups. These were APE, AR, Hartmann’s, and other operations. The percentage use of each of these operations was then calculated by patient age, gender, Dukes’ stage, quintile of the income domain of the Index of Multiple Deprivation 200423 (derived from each patient’s postcode of residence at diagnosis), method of admission to hospital, cancer network of treatment, managing hospital trust and operating surgeon. The statistical significance of any differences in rectal surgery rates across groups were assessed using the χ² test. Medians were compared using the Wilcoxon rank sum test.

A multilevel (random effects) binary logistic regression model was used to assess factors associated with the use of APE. In this analysis only patients for whom the managing consultant was known and who received an APE or AR were included (n = 26 097). The model was developed with a hierarchy of patients being clustered amongst surgical teams (level 2) and within trusts (level 3) so allowing for correlations among patient outcomes. The dependent variable, use of APE, was coded as a binary outcome; ie, patients who received an APE were coded as 1 and those who did not as 0. Covariates (explanatory variables) included age (per 10 year increase), gender, deprivation quintile,23 year of diagnosis, stage at diagnosis and annual median surgeon workload (fewer than seven, or seven or more cases per year).

RESULTS

Between 1998 and 2004 the English cancer registries recorded 58 290 patients with single rectal tumours. Of these, 52 297 patients (90.8%) were identifiable within the HES extract used. Eight hundred and eighty-seven (1.7%) of these patients were excluded due to differences between the date of diagnosis and the date of operation. Of these patients, 3456 (6.7%) who received a colostomy alone and 4653 (9.1%) who received a local excision were excluded, leaving 31 223 (60.7%) identified as receiving a major rectal cancer resection to form the basis of the study population. The characteristics of this population are detailed in table 1.

HES recorded this population as being managed by 1326 different consultant teams in 153 different hospital trusts. The consultant in charge of care was incorrectly or not recorded in 474 (1.5%) of cases. The overall annual median rectal surgical workload of the managing consultants was 2.5 (range, 1–30). Three hundred and forty-four of these consultants had an annual median workload of seven or more cases per year and their median workload was 10 (range, 7–30). Eighty consultant teams had a workload of 14 or more cases per year and their median workload was 16 (range, 14–30). In 1998, 32.9% of patients received their surgery from surgeons with an annual median workload fewer than seven cases but by 2004 this had decreased to 26.5%.

Over the study period there was a significant reduction (p<0.01) in the rate of use of APE amongst patients who received a major abdominal procedure. Overall, the rate fell from 30.5% in 1998 to 23.0% in 2004.

There was a significant difference between males and females in the use of APE: 28.5% of males received this operation compared to 25.3% of females (p<0.01). Conversely, AR was significantly more common in females than males (p<0.01).

No strong relationship was observed between type of operation and Dukes’ stage although patients diagnosed as stage D were significantly more likely to receive a Hartmann’s operation. Similarly, there was no relationship between the age of the patient and the use of APE although, again, Hartmann’s were used more frequently in patients over 80 years of age.

There was a statistically significant relationship between use of APE and the socioeconomic status of the patients. Those in the lowest IMD income quintile (ie, the most deprived) received an APE in 30.9% of cases compared to 24.1% in the most affluent quintile (p<0.01). Likewise, Hartmann’s operations were performed significantly more frequently in the most deprived compared to the most affluent quintile (p<0.01).

Overall, 5.8% of the population presented to hospital as an emergency. Stage differed between the two groups with a higher proportion (10.4%) of stage D tumours presenting as an emergency compared to 2.4% of Dukes’ A patients (p<0.01). Table 1 shows that AR and Hartmann’s procedures were used most frequently in patients presenting as emergencies.

Across the cancer networks rates of use of APE varied from 20.1 to 39.3% (fig 1). The variation was even more pronounced across hospital trusts (fig 2) with rates of use varying from 8.5 to 52.6% (p<0.01). Likewise there was extensive variation across surgical teams (fig 3) with rates varying (in those with an annual median workload of greater than seven cases) from 0 to 67.6% (p<0.01). The variation between surgeons who had an annual median workload of 14 or more cases per year was less extreme ranging from 5.4 to 44.3% (p<0.01).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1 Variation in the use of rectal cancer operation by English Cancer Network. APE, abdominoperineal excision; AR, anterior resection.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2 Variation in the use of rectal operations by English NHS hospital trust. APE, abdominoperineal excision; AR, anterior resection.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3 Variation in the rate of use of rectal operations by each English surgeon (with an annual median workload between 1998 and 2004 of at least seven rectal cancer cases per year). APE, abdominoperineal excision; AR, anterior resection.

The magnitude of the variation in APE use across networks, trusts and surgeons fell significantly over the study period (figs 4, 5 and 6). In the first time period of 1998 to 2001 the median rate of use of APE across all cancer networks was 30.2% (range, 21.4–43.1%) but this fell to 23.0% (range, 16.7–36.2%) in the second time period of 2002–2004 (p<0.01). Likewise, in the first time period, the median use across trusts was 28.3% (range, 4.6–60%) and this fell to 23.5% (range, 5.9–59.3%) in the second time period (p<0.01). Finally, the median use of the procedure amongst all surgeons fell from 27.3% (range, 0–100%) to 22.7% (range, 0–100%, p<0.01).

• Download figure
• Open in new tab
• Download powerpoint

Figure 4 Box and whisker plots showing the variation in abdominoperineal excision (APE) use by cancer network over the study period.

• Download figure
• Open in new tab
• Download powerpoint

Figure 5 Box and whisker plots showing the variation in abdominoperineal excision (APE) use by trust over the study period.

• Download figure
• Open in new tab
• Download powerpoint

Figure 6 Box and whisker plots showing variation in abdominoperineal excision (APE) use by surgeon over the study period.

The results of the model used to determine the odds of use of APE in relation to the characteristics of the population are presented in table 2. Each increasing year of diagnosis was associated with a statistically significant 4% reduction in the odds of use of APE [odds ratio (OR) 0.96, 95% confidence interval (CI) 0.94 to 0.97]. Increasing patient age had no association with the odds of use of the operation (OR 0.99; 95% CI 0.97 to 1.01 for each year). Females were significantly less likely to receive an APE compared to males (OR 0.85, 95% CI 0.80 to 0.89). There was no linear trend in the odds of use of APE in relation to Dukes’ stage. A very strong trend was observed in relation to the socioeconomic status of the patient. Those in the most deprived quintile were 37% more likely to receive an APE compared to those in the most affluent quintile (OR 1.37, 95% CI 1.24 to 1.50). Emergency admission into hospital was associated with an 18% reduction in the odds of receiving an APE (OR 0.82, 95% CI 0.71 to 0.94).

Finally, being operated upon by a surgical team whose lead consultant had an annual median workload of seven or more cases per year compared to one managing fewer than this was associated with a 23% reduction in the odds of use of APE (OR 0.77, 95% CI 0.71 to 0.83). Further multivariate analyses (results not shown) indicated the effects remained if the data were stratified across the two time periods of 1998–2001 and 2002–2004. Significant variation in practice still existed even in the most recent time period of the study.

DISCUSSION

This is the first study to demonstrate that it is possible to combine routine data sources to assess rectal cancer practice across the NHS. The linkage has revealed that substantial variation in rectal cancer surgical practice existed between 1998 and 2004. During this time the rate of use of APE significantly decreased but males, the economically deprived and those managed by low workload surgeons were significantly more likely to receive an APE. There were also statistically significant variations in the odds of receiving an APE between individual surgeons, hospital trusts and cancer networks even after adjustment for the case-mix factors of their respective populations.

Evidence is mounting to show that the standard APE procedure is inferior to AR with circumferential resection margin positivity, perforation and local recurrence rates being higher and survival lower with APE compared to AR.9^–13 Whilst some patients will always require an APE due to closeness of the tumour to the anal verge, in a larger proportion of patients either operation may be appropriate. It is concerning, therefore, that statistically significant variation, independent of patient case-mix, is being observed across the country in the use of this procedure. Ultimately, this may result in some patients needlessly receiving a colostomy (potentially leading to a poorer quality of life),24 have a higher risk of a local recurrence and ultimately poorer survival.9 It will also pose a significant financial burden to the NHS in terms of the ongoing maintenance costs of colostomies and treatment of local recurrence. These problems must be addressed if the Government is to attain its aim of “ensuring a high quality of cancer care for all”.15

One criticism of the study is that the results depend on the accuracy and quality of the coding within the linked HES and cancer registry data. The reliability of both data sources have been called into question and particular concern has been expressed about the consistency of HES.25 Work has, however, been undertaken to validate the use of HES for national audit by comparing it with information collected by clinical teams and submitted to the National Bowel Cancer Audit Project (NBOCAP).14 This voluntary colorectal cancer audit currently collects data, submitted by clinical teams, on around 30% of all English patients diagnosed with this disease26 and strong agreement has been demonstrated between these data and that in HES, particularly, for rectal procedures.14 This suggests that the linked dataset is representative of practice and these results are reliable. In addition, Aylin et al19 have investigated the ability of HES to predict mortality risk in colorectal cancer patients and, again, found a similar discrimination to the NBOCAP database. Finally, the process of linkage adopted enables the identification of duplicates in each dataset and hence their combination improves the overall quality of the data available. This, once more, supports the validity of this linked HES-registry dataset.

Other potential confounders of this study include the inability to adjust for the height of each rectal tumour. The choice of rectal cancer operation is dependent on the distance of the tumour from the anal verge but neither HES nor cancer registry data contain this information. Tumour height is collected as part of other audits14 27 and analysis of this information at a regional level suggests there are no fundamental geographical differences in the heights of tumours.28 Complete information is not available, however, and further work is required to determine if this could explain some of the variation in practice observed.

Another criticism of the study is that the adequacy of case-mix adjustment was limited due to the routine nature of the data. For example, staging information was only available in 74.5% of cases. This could be a flaw as it has been shown that population-based studies that fail to make adequate adjustment for case mix often produce inflated outcomes.3 Although the possibility of bias cannot be excluded entirely the high numbers included in the study ensure the results cannot be attributed to differences in case mix alone.

To date all the English population-based studies looking at the delivery and outcome of colorectal cancer care have been based on relatively small areas (eg, from single cancer registry regions13 29 30) or have relied on incomplete national data.14 31 Our findings are consistent with many of these studies. For example, rates of use of APE over time have been shown to be consistently dropping across all studies.13 14 30 31

Likewise, in this study, and others,31 male patients have been shown to receive APE more frequently than females. The reasons for this are unclear but it may be a result of the male pelvis being, generally, narrower than the female pelvis. This may make a successful AR for a low rectal tumour more technically challenging in males than in females and hence result in a higher rate of use of APE.

The trend towards the increased use of APE in the socioeconomically deprived has also been observed in other studies.31 The reasons behind these different surgical choices are not established, however. One theory is that patients from more deprived areas tend to present with later stage of disease but, given our observation that stage did not significantly influence the choice of operation, it seems staging differences may not account for this variation.

Over the study period the use of preoperative radiotherapy or chemoradiotherapy has increased in the management of rectal cancer as evidence has emerged of its ability to reduce local recurrence and improve survival.32 It is possible that such treatment could influence the type of operation used due to their ability to downstage tumours. National data are not available, however, to study patterns of preoperative radiotherapy or chemoradiotherapy use at a population level. Regional analysis of data from the Northern & Yorkshire Cancer Registry and Information Service (data not shown) demonstrated an increase in its use from 23.8% of cases in 1998 to 36.0% of cases in 2004. Marijnen et al33 have shown no change in the frequency of APE and no significant downstaging of rectal cancer after short course radiotherapy and immediate operation. The CR07 trial32 also shows no change in the frequency of APE. Downstaging does occur with chemoradiotherapy but no good evidence exists to support a change in operation type after such therapy. Further work on national datasets containing information on the use of oncological treatments is required to ultimately determine the influence of such treatment on operation use.

The operating surgeon has also been shown to have an important impact on operation choice and outcome in rectal cancer. Rates of APE have been shown to be reduced and outcomes improved when patients are operated upon by high-workload or specialist surgeons.13 29 34 35 This supports the data presented here showing that surgeon workload is associated with a significant reduction in the use of APE.

Rates of use of APE have been proposed as a quality indicator for rectal cancer services14 with a suggested maximum limit of 30%.36 This study shows this rate was exceeded in six cancer networks, 51 trusts and by 118 potential specialist colorectal surgical teams over this study period. The percentage excess across trusts exceeding this 30% threshold for appropriate use of APE equated to a total of 636 patients out of 8523 receiving this procedure who, potentially, received an unnecessary colostomy. Furthermore, many surgeons had APE rates much lower than 30% and if the threshold were equivalent to their rates the potential number of avoidable APEs would be much higher. Ensuring this variability is minimised is vital if NHS cancer care is to be improved and data such as these could help cancer networks identify and address apparent irregularities in practice.

In addition, there is now growing evidence to suggest inferior oncological outcomes after APE. Colorectal pathology is not available at a national level but across the Northern & Yorkshire regions Royal College of Pathologist minimum datasets are available for a large proportion of the population. Analyses of these data demonstrate that circumferential resection margin involvement rates are much higher in patients receiving an APE compared to an AR.28 National training programmes have optimised TME and AR5^–8 leading to improved results for patients where AR is possible but some low rectal tumours will always require an APE and with no similar focus on optimising this technique it is possible that the disparity in results between the two operations may increase. Work must be undertaken, therefore, to optimise the APE for patients with very low rectal tumours. There is now evidence to suggest this can be achieved through the use of a more radical operation, where such cancers are removed from below using a wide perineal approach37^–39 in an operation that resembles the procedure originally described.40 Training surgeons in this technique may improve outcomes for those patients whose tumour dictates APE as the only surgical option.

There is a growing realisation in the UK that population-based audit of cancer services is essential to ensuring high quality cancer care. For example, national comparative reviews of practice are demanded to monitor care and prevent postcode lotteries. Likewise, if the Department of Health’s Choice Agenda is to succeed, patients must know how their local hospital is performing. Similarly, audit is required to monitor the impact of costly changes in national cancer policy and good data are needed to enable the NHS to plan for cost-effective cancer services that will meet future need. This study demonstrates it is possible to combine routine colorectal cancer data sources to monitor cancer practice at a national level covering a population of 50 million people and fulfil all these aims. The methods are transferable to both other cancer sites and diseases. Whilst some improvement in routine data collection will increase the accuracy of audits we suggest that such data sources be used to routinely monitor practice within the NHS to improve patient care and outcomes.