Figure. A funnel plot demonstrating
institutional caseload and percentage of patients who died after reintervention.
Almoudaris AM, Mamidanna R, Bottle A, et al. Failure to rescue
patients after reintervention in gastroesophageal cancer surgery in
England. JAMA Surgery. doi:10.1001/jamasurg.2013.791.
eTable 1. Modified Clavien-Dindo grading system of surgical
complications and how they relate to the study outcome measures
eTable 2. Final multiple regression model of independent predictors
and the relative odds risk of death in hospital at 30 days for the
whole study cohort
eTable 3. Demographics between the mortality groups
Customize your JAMA Network experience by selecting one or more topics from the list below.
Almoudaris AM, Mamidanna R, Bottle A, et al. Failure to Rescue Patients After Reintervention in Gastroesophageal
Cancer Surgery in England. JAMA Surg. 2013;148(3):272–276. doi:10.1001/jamasurg.2013.791
Author Affiliations: Department of
Surgery and Cancer, Faculty of Medicine, St Mary's Hospital (Messrs
Almoudaris, Mamidanna, and Faiz and Prof Hanna), Dr Foster Unit, Department
of Primary Care and Public Health (Drs Bottle and Aylin), and Centre
for Patient Safety and Service Quality, St Mary's Campus (Mr Almoudaris
and Prof Vincent), Imperial College London, England.
Importance Gastroesophageal cancer resections are associated with significant
reintervention and perioperative mortality rates.
Objective To compare outcomes following operative and nonoperative reinterventions
between high- and low-mortality gastroesophageal cancer surgical units
Design All elective esophageal and gastric resections for cancer between
2000 and 2010 in English public hospitals were identified from a national
administrative database. Units were divided into low- and high-mortality
units (LMUs and HMUs, respectively) using a threshold of 5% or less
for 30-day adjusted mortality. The groups were compared for reoperations
and nonoperative reinterventions following complications.
Setting Both LMUs and HMUs.
Participants Patients who underwent esophageal and gastric resections for
Exposure Elective esophageal and gastric resections for cancer, with
reoperations and nonoperative reinterventions following complications.
Main Outcomes and Measures Failure to rescue is defined as the death of a patient following
a complication; failure to rescue–surgical is defined as the
death of a patient following reoperation for a surgical complication.
Results There were 14 955 esophagectomies and 10 671 gastrectomies
performed in 141 units. For gastroesophageal resections combined,
adjusted mortality rates were 3.0% and 8.3% (P < .001) for LMUs and HMUs, respectively. Complications
rates preceding reoperation were similar (5.4% for LMUs vs 4.9% for
HMUs; P = .11). The failure to
rescue–surgical rates were lower in LMUs than in HMUs (15.3%
vs 24.1%; P < .001). The LMUs
performed more nonoperative reinterventions than the HMUs did (6.7%
vs 4.7%; P < .001), with more
patients surviving in LMUs than in HMUs (failure to rescue rate, 7.0%
vs 12.5%; P < .001). Overall,
LMUs reintervened more than HMUs did (12.2% vs 9.6%; P < .001), and LMUs had lower failure to rescue
rates following reintervention than HMUs did (9.0% vs 18.3%; P = .001). All P values stated refer to 2-sided values.
Conclusions and Relevance Overall, LMUs were more likely to reintervene and rescue patients
following gastroesophageal cancer resections in England. Patients
were more likely to survive following both reoperations and nonsurgical
interventions in LMUs.
In 2011, there were 462 000 new cases of esophageal cancer1 and 988 000 new cases of gastric cancer
diagnosed worldwide.2 Surgery is
the mainstay for a cure. Esophageal and gastric cancer resections,
however, are associated with significant postoperative mortality,
with notable variability among centers. The volume-outcome relationship
has been demonstrated in upper gastrointestinal cancer surgery,3-6 which has led the drive to centralization of services in England.
Another major determinant of outcome is the quality of postoperative
complication management. The relationship between a hospital's complication
rate and its mortality rate is not a linear one. It has been shown
that hospitals that rank as the worst for mortality are not necessarily
those that rank as the worst for the number of complications accrued,
which suggests that the complication rate is not the only determinant
of outcome following postoperative complications.7 In colorectal surgery, it has been shown
that, despite similar surgical complication rates, variability in
outcome following the management of these complications occurs between
high- and low-mortality hospitals; this is termed failure to rescue–surgical, which is defined as the
death of a patient following reoperation for a surgical complication.8 This metric is thought to represent a marker
of how well postoperative surgical complications are managed. It may
explain to a degree why such mortality differences and variability
occur among colorectal cancer surgical units. In gastroesophageal
cancer surgery, such a relationship between reinterventions and postoperative
outcomes has not been examined.
Our hypothesis is that units with lower overall postoperative
mortality reintervene more often and are subsequently more successful
than units with high mortality. Our study aims to ascertain if differences
in patient survival following reoperations for complications occur
between high- and low-mortality hospitals undertaking gastroesophageal
cancer resections in England. Furthermore, our study also aims to
assess the use of and the subsequent outcome following nonoperative
interventions, such as the use of endoscopic and interventional radiological
therapy between high- and low-mortality gastroesophageal cancer surgical
units in England.
Our study was a retrospective national cohort study of all patients
who received a diagnosis of gastric or esophageal cancer and who underwent
an elective primary major surgical resection between April 2000 and
March 2010 inclusive in England. Data were obtained from the Hospital
Episodes Statistics (HES) database. We obtained approval under Section
251 granted by the National Information Governance Board for Health
and Social Care (formerly Section 60 by the Patient Information Advisory
Group). We obtained approval from the South East Research Ethics Committee.
The HES database is an administrative database covering all
English public hospitals known as National Health Service (NHS) hospitals
and treatment centers from 1989, with more than 15 million new records
added each year, and has been described previously.9,10 Demographic, diagnostic,
and procedural interventions are reported routinely for each patient
episode. Each patient record now contains a primary diagnosis field
and up to 19 secondary diagnoses fields, as well as up to 24 procedural
(including operative) fields. These are reported using International Statistical Classification of Diseases, 10th Revision (ICD-10) and Operating
and Procedural Codes version 4 (OPCS-4) coding, respectively. Anonymized patient level outcome measures
can be derived, and this data set has been extensively used for such
purposes previously.11-13 Linkage
of HES data with the UK Office of National Statistics was performed
to identify patients who had died within 30 days of admission to the
All patients who underwent a primary resection for esophageal
or gastric cancer in England were included in the analysis. The ICD-10 codes were used to identify these patients.
The OPCS-4 procedure codes used to identify
esophagectomy and gastrectomy were G01, G02, G03, G27, and G28.
Units were stratified according to overall risk-adjusted mortality.
A cutoff of 5% was chosen as an arbitrary threshold for combined mortality
of a unit's esophageal and gastric 30-day mortality rate. This figure
was chosen because high-volume centers worldwide reportedly reproduce
such rates for patients undergoing elective resections for cancer.14-17 Units with an adjusted mortality of 5% or lower were termed low-mortality units (LMUs), and those units with
a higher adjusted mortality (>5%) were termed high-mortality
Reinterventions were only counted if specific surgical pathologies
were found. The reoperative procedures were selected as those that
most likely reflect serious surgical complications rather than planned
“re-looks.” Thus, reoperations in this context can be
thought of as a surrogate for serious surgical complications for patients
who underwent an intervention. The reoperations broadly fall under
3 categories: thoracotomy, laparotomy, and laparoscopy. Examples of
indications for reoperation include bleeding, organ space infections,
and bowel obstruction. Common procedures included under the heading “laparotomy”
were drainage of abscess/collection, bowel resection and/or stoma
formation, and washout of abdominal cavity. Patients who unexpectedly
underwent 1 or more operative procedures subsequent to the index procedure
during the primary admission were deemed to have undergone a reoperation.
Any patient who did not undergo a reoperation but who underwent
a radiologically guided percutaneous drainage procedure or an upper
gastrointestinal endoscopic procedure postoperatively were identified
and were deemed to have undergone a nonoperative reintervention. Procedures
that were performed before midnight on the day of the index operation
were excluded because these procedures may have been planned procedures
that were undertaken as a part of the primary resection. Reasons for
undertaking these reinterventions are not discernible from the database.
However, radiologically guided drainages and endoscopies that were
performed after the day of surgery are likely to reflect a surgeon's
desire/threshold to investigate any deviations from a “normal”
recovery. Thoroscopic interventions were also considered, but none
were identified as reinterventions.
The outcome measures were (1) mortality following reoperations
and (2) mortality following nonoperative reintervention (endoscopic
The validated modified Clavien-Dindo classification of surgical
complications stratifies surgical complications according to 7 grades,
I to V, with 2 subgrades for grades III and IV (subgrades IIIa and
IIIb and subgrades IVa and IVb) (eTable 1).18 For our study, we were specifically interested
in subgrade IIIa and subgrade IIIb complications. Our primary outcome
measure was failure to rescue a patient after nonoperative reintervention;
it would represent outcome following Clavien-Dindo grade IIIa complications,
and FTR-S would represent outcome following Clavien-Dindo grade IIIb
complications (eTable 1). Of note, the original
failure to rescue definition would have considered deaths following
Clavien-Dindo grades I through IVb complications together.
Risk adjustment was performed by creating a multiple regression
model to predict the likelihood of binary outcomes with the covariates
of sex, patient age (considered in bands of <60,60-70, 71-80, >80
years), Carstairs index,19 Charlson
comorbidity score (grouped as those with scores of <2 and those
with scores of ≥2, where the latter group indicates more comorbid
patients), and type of procedures performed (esophagectomy or gastrectomy).
Factors with a significance level of 0.1 or less on bivariate analysis
were included in the multiple regression analyses. Unit-level adjusted
death rates were obtained for each hospital by dividing the hospital's
observed deaths by its model-predicted deaths and multiplying by the
national crude death rate. Statistical analyses were performed with
SPSS version 18.0 (SPSS Inc). All P values
stated refer to 2-sided values.
A total of 25 626 patients were electively admitted to
141 NHS units in England over the 10-year period and were included
in our study. The demographic characteristics of the patients are
shown in eTable 2, as are the odds ratios
of 30-day mortality as independently predicted by the considered covariates.
Patients older than 80 years of age, more socially deprived patients
(ie, those with a higher Carstairs index), and patients with more
comorbidities (ie, those with a Charlson score of >2) were more likely
to die in the hospital within 30 days (with odds ratios of 5.00, 1.36,
and 1.84, respectively, and P < .05
for all 3 groups of patients) (eTable 2).
There were 1348 deaths in the whole cohort, giving an overall
crude death rate of 5.3% for esophageal and gastric resections combined.
The crude death rates were 5.1% (758 of 14955) for patients who underwent
an esophagectomy and 5.5% (590 of 10671) for those who underwent a
When the units are stratified using a risk-adjusted mortality
threshold of 5%, 65 units with 11 803 patients (46.1%) appear
in the LMU group (≤5% adjusted mortality), and 76 units with 13 823
patients (53.9%) appear in the HMU group (>5% adjusted mortality).
The demographic characteristics of the 2 groups are shown in eTable
3. There were no statistical differences between the 2 groups
in terms of the type of resection performed (esophageal or gastric; P = .20), patients' presenting ages (P = .77), or patients' sex (P = .34). There were relatively more
socially deprived patients (indicated by a higher Carstairs index)
in the HMUs than in the LMUs (P < .001).
There were also more comorbid patients (indicated by a higher Charlson
score) in the LMUs than in the HMUs (P < .001).
More patients in LMUs than HMUs underwent procedures with a minimally
invasive approach (985 vs 792 patients; P < .001). Teaching hospital status by mortality group
was not statistically different between LMUs and HMUs (10 vs 14 units; P = .40).
The LMUs and the HMUs had equivalent reoperation rates for surgical
complications (thoracotomy, laparotomy, and laparoscopy combined)
(5.4% vs 4.9%; P = .11). The LMUs
and the HMUs had significantly different nonoperative reintervention
rates, with the LMUs performing more nonoperative reinterventions
(6.7% vs 4.7%; P < .001).
Patients were significantly more likely to die after reoperations
in HMUs than in LMUs (ie, the failure to rescue–surgical rate,
24.1% vs 15.3%; P < .001). Patients
were more likely to die after a nonoperative reintervention in the
HMUs (12.5% vs 7.0%; P < .001).
Eighty-two patients underwent both nonoperative and operative
reinterventions that were coded under the operative category. These
were equally distributed between the mortality groups (P = .18).
Overall failure to rescue rates did not appear to be related
to total unit case volume (Figure). Statistical outliers (>2 SDs from the population mean) are seen
in both high- and low-volume centers.
Using English national administrative data, we have shown that,
when surgical units undertaking esophageal and gastric cancer resections
are grouped according to adjusted mortality, index admission reoperation
rates for complications are equivalent between LMUs and HMUs. However,
the LMUs are more likely to nonoperatively intervene and are more
likely to rescue patients from subsequent death after both reoperative
and nonreoperative reinterventions.
What our study reiterates is that variability in outcome following
management of serious complications does also occur in patients undergoing
an upper gastrointestinal resection. This is in keeping with the current
literature in other specialties.20 Our study also demonstrates that failure to rescue–surgical
rates differ by mortality grouping, as in lower gastrointestinal surgery.8 Although the concept that LMUs rescue patients
after reoperations more frequently than HMUs do may not be novel,
the present study is the first to demonstrate this concept in patients'
undergoing upper gastrointestinal cancer surgery. Theoretically, if
all the patients undergoing an operation in an HMU were undergoing
the operation in an LMU, approximately 50 more patients per year would
have been alive. Through analysis of nonoperative reinterventions,
and subsequent outcome, we are able to reflect on those aspects of
surgical care that are encountered when surgeons are faced with serious
surgical complications in HMUs and LMUs. In those circumstances, surgeons
have the option of watchful waiting and/or medical therapy, nonoperative
reintervention (radiological drainages/endoscopic therapy), or reoperation.
What determines the treatment path taken is dependent on many factors,
including the patient's comorbidity and physiological reserve and
the available resources and experience. Ultimately, surgeons bear
the responsibility on the final decision taken.
The timely diagnosis of a major surgical complication and the
performance and reliability of surgical teams are of vital importance.
Perhaps a more rapid identification of actual or potential complications
underlies some of the observed differences. Supportive care such as
intensive therapy units, physiotherapy, and dietetic input may contribute
toward the successful rescue of a patient. A greater understanding
of the quality and capabilities of such services between HMUs and
LMUs may uncover some of the observed outcome differences. Similarly,
structural factors such as out-of-hours radiology services and nurse-patient
ratios may vary between LMUs and HMUs. Although the hospital volume-mortality
relationship has been shown previously in esophageal surgery and,
to a lesser extent, gastric surgery, no relationship has been investigated
with regard to the hospital volume–failure to rescue–surgical
rate relationship in gastroesophageal cancer surgery. We found failure
to rescue outliers at both high and low volumes, which suggests that
unit volume does not predict ability to rescue patients following
reintervention. Future studies are needed to examine the relationship
between failure to rescue rates in different surgical units and the
aforementioned factors that determine the treatment path chosen and
the outcomes encountered.
We specifically chose to analyze the modified Clavien-Dindo
group III types of complications for several reasons. First, certain
postoperative complications may be present on admission, such as pneumonia
or deep vein thrombosis, which are not necessarily discernible from
administrative databases and are subsequently discovered in the postoperative
period. This would influence failure to rescue rates, and, although
important, if they cannot be adjusted for, this would not faithfully
reflect the actual care given by a unit. The interventions that we
have chosen are not open to variability of definition. Certain postoperative
complications, such as respiratory tract infections and wound infections,
may be underreported owing to differences in definitions between LMUs
and HMUs. It must be noted that the absolute complication rate is
not represented here. Nor can the total complication rate be assumed
from the reintervention rate presented. There are likely to have been
some serious surgical complications and interventions, such as the
failure of a cervical anastomosis (where the opening up of a neck
wound on the ward would have been the treatment) or the insertion
of chest drains by the bedside. Such “rescue” would not
have been discerned from our study. Although underestimating the true
serious surgical complication rate by using reoperations as a proxy,
of the patients who underwent a reoperation, differences are still
observed among units with different mortality statuses. Our study
has clearly defined complications in that patients were either returned
to the operating theater or not, and in that patients either underwent
postoperative endoscopies or drainages or they did not. Finally, Clavien-Dindo
grade III complications are arguably those that most likely represent
surgical technical quality and anastomotic complications.21 In gastroesophageal cancer surgery, these
are factors that have important implications perioperatively for survival.22,23
Our study is based on an administrative data source. Any coding
errors may potentially influence the results. Systemic underreporting
of reinterventions by disparate organizations within the same mortality
grouping over the study period could, in theory, lead to reporting
bias. However, this is unlikely. A more general coding error could
have influenced the results. Again, however, this is unlikely given
the proven accuracy of this data set in recording diagnostic fields.24,25 Data on site of anastomosis
and histology are not available from the database and may influence
the risk adjustment and, hence, subsequent outcome. However, we have
no evidence to suggest that geographically disparate hospitals grouped
as HMUs or LMUs would collectively perform more procedures of one
type than another, or perform operations for patients with similar
histological features rather than for patients with different histological
features. Unfortunately, the HES database does not capture cancer
stage (although the Charlson score includes metastases, which data
were included in the adjustment model), and this may well have an
influence on perioperative outcome, as will a case mix not fully adjusted
for using the available data. However, it is unlikely that patients
with stage IV disease would be undergoing major surgery, unlike in
colorectal cancer for which some patients with obstructing colon cancer
are stented as a bridge to elective (albeit expedited) surgery.
A strength of our study is that it is not subject to reporting
bias. Another strength is that, over similar time periods, the HES
database has been shown to record more deaths than a voluntarily recorded
clinical registry.10 Given the sample
size and the number of years considered, our study truly reflects
the national outcome from gastroesophageal cancer surgery.
Reoperation following gastroesophageal cancer surgery has been
repeatedly associated with poor perioperative outcomes in the literature.
Such findings may influence contemporary decision making by surgeons
faced with the difficulties of managing postoperative surgical complications.
The implications of our study raise the question of whether more aggressive
and appropriate reinterventions can in fact confer better outcomes.
Surgeons should interpret these findings within the context and limitations
of their own units. With centralization of services, commissioners
should ensure the availability of interventional endoscopy and radiology
services to complement the surgical service. Improved outcome from
gastroesophageal cancer surgery is more complex than just the volume-outcome
relationship. Complication rates have repeatedly been shown to be
equivalent between HMUs and LMUs in different specialties. Complication
management is becoming more widely recognized as an important discriminator
of surgical outcome. Surgeons should be supported with all the facilities
and expertise necessary to ensure that all facets of the care that
they deliver are optimal. The focus and priority of future research
should be to identify whether quality, expertise, or familiarity with
reintervention is the causal factor in the association with LMUs.
In conclusion, there is variability in outcome following serious
surgical complications necessitating reoperations among gastroesophageal
cancer surgical units in England. Units with lower overall mortality
reintervene more often and are subsequently more successful than units
with higher overall mortality. Future work should focus on why such
variability occurs and should identify methods for mitigating this
Correspondence: George B. Hanna,
PhD, Department of Surgery and Cancer, Faculty of Medicine, St Mary's
Hospital, Imperial College London, Praed St, London W2 1NY, England
Accepted for Publication: July 5, 2012.
Author Contributions:Study concept and design: Almoudaris, Aylin, Vincent, Faiz,
and Hanna. Acquisition of data: Bottle and
Aylin. Analysis and interpretation of data: Almoudaris, Mamidanna, Bottle, Aylin, Vincent, and Hanna. Drafting of the manuscript: Almoudaris, Mamidanna,
Faiz, and Hanna. Critical revision of the manuscript
for important intellectual content: Almoudaris, Bottle, Aylin,
Vincent, and Hanna. Statistical analysis:
Almoudaris and Mamidanna. Obtained funding: Aylin and Vincent. Administrative, technical,
and material support: Almoudaris, Aylin, and Vincent. Study supervision: Bottle, Vincent, Faiz, and Hanna.
Conflict of Interest Disclosures: None
Funding/Support: This study was supported
by funding from the National Institute for Health Research for research
into patient safety (to Mr Almoudaris). The researchers had complete
independence from Mr Almoudaris' funders. Drs Bottle and Aylin work
within the Dr Foster Unit at the Imperial College London, which is
largely funded by a research grant from Dr Foster Intelligence (an
independent health service research organization, which also has no
influence on any part of publication). The Dr Foster Unit at Imperial
College London is affiliated with the Imperial Centre for Patient
Safety and Service Quality at Imperial College Healthcare NHS Trust,
which is funded by the National Institute of Health Research. The
Department of Primary Care and Social Medicine is supported by the
National Institute for Health Research Biomedical Research Centre
Role of the Sponsor: The National Institute
for Health Research did not participate in the design and conduct
of the study; in the collection, analysis, and interpretation of the
data; or in the preparation, review, or approval of the manuscript.