Outcomes of aortic aneurysm surgery in England: a nationwide cohort study using hospital admissions data from 2002 to 2015

Background The United Kingdom aortic aneurysms (AA) services have undergone reconfiguration to improve outcomes. The National Health Service collects data on all hospital admissions in England. The complex administrative datasets generated have the potential to be used to monitor activity and outcomes, however, there are challenges in using these data as they are primarily collected for administrative purposes. The aim of this study was to develop standardised algorithms with the support of a clinical consensus group to identify all AA activity, classify the AA management into clinically meaningful case mix groups and define outcome measures that could be used to compare outcomes among AA service providers. Methods In-patient data about aortic aneurysm (AA) admissions from the 2002/03 to 2014/15 were acquired. A stepwise approach, with input from a clinical consensus group, was used to identify relevant cases. The data is primarily coded into episodes, these were amalgamated to identify admissions; admissions were linked to understand patient pathways and index admissions. Cases were then divided into case-mix groups based upon examination of individually sampled and aggregate data. Consistent measures of outcome were developed, including length of stay, complications within the index admission, post-operative mortality and re-admission. Results Several issues were identified in the dataset including potential conflict in identifying emergency and elective cases and potential confusion if an inappropriate admission definition is used. Ninety six thousand seven hundred thirty-five patients were identified using the algorithms developed in this study to extract AA cases from Hospital episode statistics. From 2002 to 2015, 83,968 patients (87% of all cases identified) underwent repair for AA and 12,767 patients (13% of all cases identified) died in hospital without any AA repair. Six thousand three hundred twenty-nine patients (7.5%) had repair for complex AA and 77,639 (92.5%) had repair for infra-renal AA. Conclusion The proposed methods define homogeneous clinical groups and outcomes by combining administrative codes in the data. These methodologically robust methods can help examine outcomes associated with previous and current service provisions and aid future reconfiguration of aortic aneurysm surgery services.


Background
The United Kingdom had the highest mortality rate for the elective repair of aortic aneurysms (AA) compared to other western European countries in 2007 (7.9% UK vs 3.5% Europe) [1]. Improvement of outcomes such as postoperative mortality following AA repair was a major drive for vascular services reconfiguration in the National Health Service (NHS). Therefore, reliable and consistent methods to obtain comparable data on activity and outcomes could help the NHS measure the success and shortcomings of vascular services reconfigurations. A valuable resource to measure outcomes is the administrative data collected by NHS hospitals [2][3][4] and there is growing evidence that the quality of this administrative dataset has improved [5][6][7][8]. Furthermore, several studies reported that maximum use of the available information in the dataset can improve the validity of outcomes measured [7,[9][10][11].
NHS England generates Hospital Episode Statistics (HES) that include details of all inpatient admissions, outpatient appointments and accident & emergency attendances at all English NHS hospitals [12]. The HES database provides a detailed source of information regarding patient care across England [5][6][7][8]. HES data can be linked to the Office for National Statistics (ONS) mortality registry data and this can help to analyse survival post inpatient discharge. The basic unit of activity measured in HES is the finished consultant episode (FCE). This is a single period of care under one consultant and does not necessarily equate to a single hospital admission, which may comprise more than one episode if care is transferred between consultants or providers. Each FCE contains a primary diagnosis, up to 19 secondary diagnoses and 24 procedure fields. FCEs also include information such as patient demographics, type of admission, source of admission as well as length of stay in critical care and other important clinical and administrative information.
FCEs can be combined to generate a provider spell; this is the period of care when the patient remains in one hospital. This definition does not capture transfers between hospitals during the same stay. Therefore, to generate an accurate admission level dataset for a patient the FCEs need to be combined into a continuous inpatient stay (CIPS). A CIPS starts from the moment a patient is admitted under the care of a consultant in an NHS hospital and includes all the episodes during that admission including transfers to other hospitals [13].
The aim of this study was to develop standardised algorithms with the support of a clinical consensus group to identify all AA activity; classify the AA management into clinically meaningful case mix groups and define outcome measures that could be used to compare outcomes among AA service providers.

Methods
Vascular related inpatient HES data from the financial year 2002/2003 to 2014/2015 were acquired from NHS Digital data warehouse using broad filters including health resource groups (HRG) codes and office of population census and surveys (OPCS) codes; the OPCS codes are codes for interventions and procedures. The R© programme (Version 3.4.1) (R foundation, Vienna, Austria) was used to develop code to clean, validate, and explore this HES data extract.
From this broad extract of vascular episodes, specific filters were developed based on a combination of the OPCS codes, ICD10 diagnosis codes, and speciality codes to identify all AA-related episodes. The patient identification number (encrypted HESID) for the patients from this extract was used to identify all others episodes for each patient (Appendix 1). These patient identification numbers were also used to link patients to the mortality data from the Office of National Statistics.
The data had already been cleaned and validated to a certain degree at the data warehouse (NHS Digital) before it was passed to us [14]. Nevertheless, significant amount of data cleaning and validating was undertaken prior to analysis (Details of the cleaning and validating steps performed are reported in the Appendix 1 section 2). HES data are well-known for problems with missing data, duplicates, data formatting errors and invalid data [7,9], but accuracy can be improved by appropriate cleaning and validation.
Patients' pathways were described in terms of a series of admissions (CIPS) that include single or several episodes for each patient. The index admission was defined as the admission where patients' received their first AA repair or they died during the index admission secondary to AA without any repair in the current or prior admissions. Figure 1 illustrates a simplified single patient pathway in the data.
To capture all the AA cases and categorise them into clinically meaningful case mix groups an iterative process was employed with input from multidisciplinary group of vascular specialists and data analysts. The clinicians were presented with aggregate data as well as samples of fully anonymised individual records and, based on their recommendations, the algorithms used to define categories were modified. The clinical consensus group divided the AA groups into 'infra-renal repair', 'complex repair' and 'AA related death without repair'. The first two groups were subdivided into elective, emergency non-ruptured and ruptured repair subgroups and these were further subdivided into open surgical and endovascular repair (EVAR).
The initial algorithms for identifying the AA repair groups were based upon AA procedure codes (OPCS), whereas for the AA related death with no repair, the algorithm relied on diagnoses codes (ICD-10) (for details of the included codes refer to Tables 11-17 in Appendix 2. The subsequent alterations to the algorithms were based on input from the clinical consensus group and in these alterations, other information from HES and specific codes such as treatment speciality codes [15] (Table 18 in Appendix 2), admission method [16], discharge method [17] were used. These changes to the algorithms were required to overcome coding inconsistencies within AA HES dataset. Key issues identified by the consensus group were: -Categorisation of cases with multiple, potentially conflicting, OPCS codes. -Categorisation of admissions into elective and emergency in light of inconsistencies observed when cross tabulating OPCS and ICD-10 codes against admission method. -Identification of complex AA repaired by vascular specialists. -Identification of ruptured AA patients who died without AA procedure. -Distinguishing between of aortic bypass procedures for AA and peripheral arterial disease. -Identification of index admissions to describe the patient pathway accurately and detect related prior admissions, readmissions and complications.
Following the identification of all the cases within each of the AA case mix groups, outcomes such as length of stay, complications within the index admission, postoperative mortality and re-admission within 30 days of the index admission were calculated for each case mix group.

Results
The total number of inpatient episodes for vascular patients identified by the broad filters (Appendix 1)

Developing case mix groups
The development of the case-mix groups was based on the anatomy of the AA disease (infra-renal, complex), admission method (e.g. elective vs emergency), ruptured vs intact AA, type of the procedure (e.g. open repair vs EVAR), and a subgroup of patients dying in-hospital from AA with no previous AA operation. The anatomy of the AA disease was identified by specific OPCS codes that differentiate complex AA procedures from infra-renal AA procedures (See Tables 11-15 in Appendix 2). There were further issues that had to be resolved in the development of casemix groups.

Categorisation of elective and emergency
There were several potential methods to distinguish elective or emergency AA cases. Admission method (admimeth) in HES data defines elective and emergency admissions, OPCS codes differentiate between elective and emergency procedures (See Tables 11-15 in Appendix 2) and ICD-10 codes can describe whether AA is intact or ruptured (See Tables 16-17 in Appendix 2). Table 1 presents cross-tabulation of these three methods for the index AA operation episodes. The table demonstrate the degree of overlap between these methods in identifying elective and emergency episodes in the HES dataset.
Among cases with emergency admission indicated by the admission method, 36% of them did not have emergency operation or diagnosis codes and 58.5% had only emergency ICD10 codes. Among cases with elective admission indicated by the admission method, 97.2% of them did not have any emergency operation or diagnosis codes. Following discussion with the clinical consensus group based upon examination of sampled cases and aggregate data, admissions were divided into elective and emergency admissions based on admission method, irrespective of the categorisation of the procedure. Further analysis was carried out to consider the identification of ruptured AA based upon ICD10 codes.

Categorisation of ruptured aortic aneurysm
There were 27,359 ruptured AA cases identified using the ICD10 diagnosis codes for ruptured AA; of these  For ruptured cases with AA repair (16,809 cases) the in-hospital mortality based on their admission method and the delay (in days) from admission to procedure was investigated and compared to the in-hospital mortality of the non-ruptured AA repair cases. The results are shown in Table 2 and Table 3.
The mortality rates were significantly higher in those repair cases with the ruptured ICD-10 codes compared with those without these codes. Based on these results, the consensus group concluded that it is appropriate to include all cases that underwent repair and had a ruptured AA ICD-10 codes into a separate case-mix group (ruptured AA) regardless of admission method and delay between admission and operation, with the remaining emergency admissions being treated as a group of "emergency repair without mention of rupture".

Categorisation of open and endovascular AA
The specific procedure codes (OPCS) for endovascular aneurysm repair (EVAR) were introduced in 2005/ 2006 (See Table 12  for open AA repair the episode had codes for insertion of prosthesis into organ and/or aorta specific organ codes or arteriotomy codes (See Table 20 in Appendix 3) [18]). Upon investigating these methods, it was found that only the presence of the code for "insertion of prosthesis into organ" was useful in identifying EVAR cases prior to 2005/2006 (For more information see Table 21 in Appendix 3).
Aortic bypass OPCS procedures codes (Table 15 in Appendix 2) could be used to describe AA open repair, however they are also used to describe procedures for occlusive peripheral arterial disease. To separate cases undergoing bypass procedures for occlusive disease from cases of bypass for AA an additional filter was used. Cases with bypass procedures with AA ICD-10 diagnosis codes were added to the open infra-renal AA repair case-mix group. Cases with bypass procedures without such ICD-10 codes were categorised as aortic bypass for peripheral arterial disease.

Identifying complex AA procedures performed by vascular specialists
Complex AA open repair procedures could be performed by other specialists including cardiac surgeons, and the same procedure codes (See Tables 13 and 14 in Appendix 2) are used to document these procedures in HES. The cases performed by cardiac/cardio-thoracic surgeons were excluded by adding an extra filter for speciality fields in HES for cardio-thoracic surgery. This filter was applied when identifying AA-related episodes (see Fig. 1).
Identifying AA-related deaths without any repair AA-related deaths without any repair were defined as cases where patients were admitted with a diagnosis of AA and died within the admission, but there was no record of AA repair in that admission or in previous admissions. These cases were identified by including all admissions (CIPS) with a diagnosis of AA and discharge method indicating the patient died in the hospital with no record of AA repair within the admission. Based upon record linkage, those with AA repair within the same or previous admission were excluded.
In total 12,767 cases of AA-related death with no definitive repair were identified from the data between 2002/03 and 2014/15. Table 4 shows the length of time (in days) from the date of admission to the date of death for these patients.
A further investigation was carried to examine the presence of early interventions (not AA repair) in this group. Procedure fields of all episodes within the index admissions were investigated for evidence of early interventions (For more information about early intervention procedures see Table 19 in Appendix 2). The results revealed that only 6.9% of cases had evidence of early interventions (See Table 5).

Comparison of counts with NVR
The UK National Vascular Registry (NVR) is a dataset of vascular procedures performed by vascular surgeons; the information is uploaded voluntarily by vascular specialists and include details of the procedure, information specific to the patient as well as the disease [19]. The annual cases of infra-renal AA elective repairs from HES as identified by the methods described in this paper were compared to the numbers reported by NVR. This group was chosen since it was the only case-mix group consistently reported from 2009 to 2013 [19,20]. To allow comparability with the NVR cases, all AA operations (not just index admissions) for each patient within the calendar, rather than financial year, were identified. Table 6 shows the comparison between the numbers of elective infra-

Post-operative mortality
Post-operative mortality may be identified based upon in-hospital mortality, as defined by discharge method [17], or 30-day mortality based upon linked ONS data. The NHS definition of Continuous Inpatient Stay (CIPS) was used to define admission. This captures transfers between hospitals within the same stay. Thus, using this definition, a hospital death is defined as any death within the whole stay. This overcomes the issue where patients die after being transferred from the care of vascular specialist. The CIPS definition still categorises this as an in-hospital death whereas, the NHS definition of hospital spell or episode does not. Table 8 shows the results of mortality outcomes for patients with AA repairs between April 2002 and February 2015. In total 11,111 patients died in hospital following AA repair (13.2% of all AA repairs); however, 30-day mortality following AA was recorded among 10,096 patients (12% of all AA repair cases). The discrepancy between in-hospital mortality and 30-day mortality figures is due to additional mortality amongst those patients remaining in hospitals for longer than 30 days' post AA repair. Furthermore, using episode to define admission instead of CIPS can lead to significant under-estimation of in-hospital mortality following AA repair.

Re-admission within 30 days from discharge
Following the index AA repair admission 72,857 patients (86.8% of all AA repair cases) were discharged alive and 10,500 patients (14.41% of all patients discharged alive) had at least one re-admission within 30 days. There were 13,688 30-days readmissions, 9062 patients only had a single 30-day re-admission, and the remaining had more one re-admission within 30 days. Reasons for readmission were summarised by the ICD10 diagnostic blocks. Table 9 shows the main diagnoses for 30-day re-admissions for emergency and elective cases. Since most of the elective and emergency episodes were relevant to the index admission, the consensus group decided to include all of them in the 30-day readmission analysis.

Length of stay and other AA-procedure related complications
Length of hospital stay for an admission is defined as the period (in days) between the starting date of the admission and the discharge date. The short-term postoperative mortality (in-hospital mortality or 30-day mortality) and length of stay already captures certain aspects of the procedure-related complications. There are other methods to identify AA-procedure related complications in the index admissions including the presence of reoperation in the index admission (in addition to mortality and length of stay). Table 10 presents length of stay and reoperation in the index admission (together with the rates of readmissions within 30 days).

Discussion
This study developed standardised methods to identify AA activity and outcomes in England, based upon the HES administrative dataset. Using a stepwise approach with input from clinicians and data analysts, the solutions aimed to provide consistent and comparable methods to overcome some of the inherent pitfalls and ambiguities in identifying the activity and outcomes associated with AA case mix groups. Identified issues included invalid, missing, and duplicate information, the limitations of AA diagnostic and procedural codes in separating elective and emergency admissions consistently and the handling of potentially conflicting or overlapping codes [21][22][23][24][25][26][27][28][29][30][31][32][33][34]. Codes for admission methods were more consistent than clinical codes in differentiating between emergency and elective admissions. Mortality rates suggested that previously described methods of combining admission method and delay between admission and procedure [28,33] were less efficient at identifying ruptured AA repair cases than using the diagnosis codes alone.
Further challenges that were addressed were; changes in coding practices for EVAR prior to the introduction of codes specific to this procedure [28], identification of aortic bypass cases performed for AA as opposed to PAD, identification of complex AA repair treated by vascular specialists, and the identification of cases of AA-related death that had no definitive intervention. The study also demonstrated that potentially valuable co-morbidity data and outcomes measures can be obtained by record linkage, allowing re-operations within the same admission, transfers between specialities, prior and subsequent admissions and long-term mortality. A significant finding from this study is that the choice of episodes or hospital spell to describe index admission for AA repair [21][22][23][24][25][26][27][28][29][30][31][32][33][34] may lead to under-estimation of in-hospital mortality.
An important outcome of this study is the potential standardisation of definitions and algorithms for identifying AA activity and outcomes. Differing definitions used in the past may have resulted in ambiguity, doublecounting, misclassification or exclusion of some cases from appropriate case-mix groups, which may have affected the comparability of previous work that used the HES dataset [26,27,[30][31][32][33][34][35].
Currently the healthcare quality improvement partnership (HQIP) collects data about AA activity and outcomes through the National Vascular Registry (NVR). NVR dataset contains important information about case complexity as well as risk adjustment, which is not easily available from HES. However, NVR also has drawbacks as it is a voluntary, procedure-based registry, it does not provide information on cases of AA related death with no intervention and contains no data relating to readmissions, repeat procedures or post-discharge outcomes [36]. Although case ascertainment in NVR is high, even small amounts of selectively missing data may distort outcomes. Ideally, linkage between NVR and HES data could address many of the limitations of both datasets. Using the methods described in this paper will help identify AA activity (repair and no definitive intervention) and compare outcomes between different providers and across time.

Conclusions
HES is a rich source of data but has pitfalls and distortions as shown in this study. These can be overcome by developing consistent methods that rely on the data available within HES to identify and group all relevant AA activity. Many short and long-term outcomes can be analysed by linking admission level data to identify prior and subsequent admissions and by linkage to ONS data. Despite the potential of HES in examining AA activity if inconsistent methods are used, the results can be distorted. HES remains an underused resource for quality assessment of AA services and the use of proposed methods can help identify aortic aneurysm surgery activity and outcomes with greater precision.

Appendix 1
Section One Extract 1