The association between perioperative acute risk change (ARC) and longterm survival after cardiac surgeryThis is the author manuscript accepted for publication and has undergone full peer
review but has not been through the copyediting, typesetting, pagination and
proofreading process, which may lead to differences between this version and the
Version of Record. Please cite this article as doi: 10.1111/anae.13967
This article is protected by copyright. All rights reserved
Article Type: Original Article
T. G. Coulson, 1 M. Bailey,
2 C. M. Reid,
2, 8 L. Tran,
3 D. V. Mullany,
7 J. A. Smith
1 Consultant, 2 Professor, 3 Scientist, Department of Epidemiology and Preventive
Medicine, School of Public Health and Preventive Medicine, 6 Professor, Department
of Surgery, Monash University, Melbourne, Australia
4 Professor, Department of Intensive Care, The Alfred Hospital, Melbourne, Victoria,
Australia
5 Professor, ANZICS Centre for Outcome and Resource Evaluation, Levers Terrace,
Carlton, Melbourne, Australia
7 Consultant, Critical Care Research Group, University of Queensland, Brisbane,
Australia
8 Professor, School of Public Health, Curtin University, Perth, Western Australia
Keywords: quality measures, patient care; cardiac surgery; survival; outcomes A u
th o
r M
a n
2
Short title: Peri-operative acute risk change and long-term survival after cardiac
surgery
Summary
Acute risk change has been described as the difference in calculated mortality risk
between the pre-operative and postoperative periods of cardiac surgery. We aimed
to assess whether this was associated with long-term survival after cardiac surgery.
We retrospectively analysed 22,570 cardiac surgical patients, with minimum and
maximum follow up of 1 and 6.7 years. Acute risk change was calculated as the
arithmetic difference between pre- and postoperative mortality risk. ‘Rising risk’
represented an increase in risk from pre- to postoperative phase. The primary
outcome was one-year mortality. Secondary outcomes included mortality at three
and five years, and time to death. Univariable and multivariable analyses were
undertaken to examine the relationship between acute risk change and outcomes.
Rising risk was associated with higher mortality (5.6% vs. 3.5%, p<0.001). After
adjusting for baseline risk, rising risk was independently associated with increased
one-year mortality (OR 2.6 (95%CI 2.2-3.0), p<0.001). The association of rising risk
with long-term survival was greatest in patients with highest baseline risk. Cox
regression confirmed rising risk was associated with shorter time to death (HR 1.86,
1.68-2.05, p<0.001). Acute risk change may represent peri-operative clinical events
in combination with unmeasured patient risk and noise. Measuring risk change could
potentially identify patterns of events that may be amenable to investigation and
intervention. Further work with case review, and risk scoring with shared variables
may identify mechanisms, including the interaction between miscalibration of risk
and true differences in peri-operative care.
A u
th o
r M
a n
t
3
Introduction
Long-term survival with good functional status is the ultimate aim of all surgical
procedures, including cardiac surgery. Outcomes after cardiac surgery are recorded
in most countries [1, 2]. Operative mortality, commonly defined as death in-hospital
or within 30 days of surgery, [3] is low, with reported rates of approximately 2%, 5%
and 7% for coronary artery bypass grafting (CABG), valve surgery and combined
CABG and valve surgery, respectively. Long-term survival is lower, with 90%, 80%
and 70% of these patients reported alive at five years, respectively [1]. The cardiac
surgical processes and performance of the peri-operative team not only impact on
short term mortality and the occurrence of complications [4-6], but may also
influence long-term survival. Peri-operative complications such as stroke, new renal
failure, deep sternal wound infection, sepsis and gastro-intestinal complications
have all been shown to be associated with reduced long-term survival [7]. Since long-
term survival is important to patients, carers, clinicians and funders [8,9], it is vital to
be able to measure how this is affected by peri-operative events.
Currently, while there are measures assessing individual
outcomes, structures or processes, there is no single marker in common usage that
represents events during the overall peri-operative process, nor one that has been
validated for both short and long-term outcomes. In addition, when assessing peri-
operative events, both procedure type and baseline risk must be taken into account.
For instance, worse long-term survival is associated with baseline risks such as
increased age, female sex, extremes of BMI, low ejection fraction, endocarditis,
urgent surgery, left main coronary artery disease, diabetes, renal dysfunction,
respiratory disease and previous cardiac surgery [1, 10, 11].
The authors have developed acute risk change (ARC), a measure based on the
difference in risk from the pre-operative to postoperative phase. The principle of
ARC is that patients presenting to the intensive care unit (ICU) after cardiac surgery
should have the same risk of death as they had before surgery. A rising ARC (a
A u
th o
r M
a n
t
4
proportion more than zero) represents a higher risk of death postoperatively, while a
falling ARC (proportion less than zero) represents a lower risk of death. A rising or
falling ARC therefore may represent a change in patient status during peri-operative
care and may, thus, be associated with ‘negative’ or ‘positive’ occurrences during the
peri-operative period, respectively. In previous studies, ARC was shown to be
associated with intra-operative events and morbidity at an individual and unit level
[12, 13]. Although further validation is required, it is plausible that ARC may be
useful to identify peri-operative team misadventure in the surgical period. In this
study, we describe further evaluation of ARC. Our aim was to assess whether an
increase in peri-operative risk after arrival in the ICU following cardiac surgery was
independently associated with long-term mortality, after adjustment for baseline
patient characteristics.
Methods
This study was approved by the Alfred Hospital Research Ethics Committee. We
retrospectively analysed the combined Australian and New Zealand Society of
Cardiac and Thoracic Surgeons Cardiac Surgery Database (ANZSCTS-CSD) and
Australian and New Zealand Intensive Care Society Adult Patient Database (ANZICS-
APD). The ANZSCTS-CSD contains patient data on all cardiac surgical procedures at
participating hospitals including baseline characteristics, pre-, intra- and
postoperative data. The ANZICS-APD contains demographic, diagnostic, and
physiological data from the first 24 hours of ICU admission for calculation of severity
of illness scores such as the acute physiological and chronic health evaluation
(APACHE)-3 score. Both databases are regularly audited to assess reliability of
submitted data [14, 15]. Linkage of the two databases was carried out by
probabilistic matching and has been described previously [13].
The combined database included 24,046 patients undergoing CABG and/or
valvular surgery between January 2008 and October 2013. Patients with missing
AusScore (required to calculate preoperative risk of death, n=1253), APACHE-3 score
(required to calculate postoperative risk of death, n=155) or hospital mortality data
(n=68) were excluded from the database, leaving 22,570 patients. Long-term
A u
th o
r M
a n
t
5
mortality was provided by linkage with the Australian Institute of Health and
Welfare’s National Death Index. The censor date was 7 th
The ARC for each patient was calculated by subtracting pre-operative risk
from postoperative risk of death, representing an absolute difference in risk. A rising
ARC value thus indicated an increase in mortality risk from the pre-operative to
postoperative periods.
October 2014; this
corresponded to a minimum and maximum follow-up time of 1 and 6.7 years,
respectively.
The pre-operative risk of hospital mortality was generated using logistic
regression from the raw AusScore [2], a pre-operative score indicating risk of death
using baseline patient risk factors (age, comorbidities etc.) and procedural factors
(type of surgery, urgency], analogous to EuroSCORE-2 [16]. Similarly, a logistic
regression model was used to generate postoperative risk of death from the surgical
operative diagnosis and the APACHE-3 score, a severity of illness score routinely
calculated following admission to Australian and New Zealand ICUs. Calibration and
discrimination was assessed using area under the receiver operating curve,
calibration plots and Hosmer-Lemeshow X 2
Our primary outcome was mortality at one year. To assess whether ARC was
independently associated with mortality at one year, patients were split into two
groups: those with an increase in risk of death in the postoperative compared to the
pre-operative period (i.e. rising ARC > 0, representing potential adverse occurrences
during the cardiac surgical process); and those with reduced risk of death
postoperatively (i.e. falling ARC <= 0, representing a potential ‘good’ peri-operative
experience).
. Univariable analysis was undertaken to
compare the demographics, characteristics and outcomes of those alive at one year
to those who had died. All patients were included irrespective of whether they had
died in hospital or after discharge.
Variables associated with one-year mortality (p<0.1) on univariable analysis
were included in a multivariable logistic regression model to identify factors
independently associated with survival. We analysed ARC group and variables
previously validated for short or long term mortality in other published studies [10,
17]. These included: age group (<50, 50-60, 60-70, 70-80, >80 yrs); BMI (<25, >= 25
A u
th o
r M
a n
t
6
kg.m -2
Time to death was analysed as a secondary outcome. For each of two groups,
rising ARC and falling ARC, Kaplan-Meier survival plots were constructed and then
stratified by initial baseline pre-operative risk of in-hospital death (low, medium and
high risk groups in tertiles). Comparisons were performed using the log-rank test for
equality across strata.
); sex; urgency of surgery (elective or non-elective); previous cardiac surgery;
shortness of breath using New York Heart Association category (> 2); ejection
fraction (<30%, 30-45%, >45%); renal disease requiring dialysis;
hypercholesterolaemia; peripheral arterial disease; cerebrovascular disease;
haemodynamic shock; respiratory disease; diabetes; infective endocarditis; and type
of procedure (CABG, valve or combined CABG/valve procedure). To determine
whether the effect of ARC on one-year survival varied according to baseline risk, an
interaction term was fitted between ARC and baseline risk (AusScore). A sensitivity
analysis using ARC as a continuous outcome was also carried out.
To identify factors independently associated with survival, multivariable Cox
regression analysis was performed using the same variables used in the primary
outcome analysis. To determine if the relationship between ARC and survival time
differed according to baseline risk, an interaction between ARC and baseline risk was
fitted. The proportionality assumption was assessed using graphical methods. Model
fit was assessed using Harrell’s C-statistic.
To further account for potential differences between those with rising or
falling ARC values, a matched propensity analysis was performed. Using all available
data and the software ‘MatchIt’ [18], a logistic regression model was used to
determine the probability of ‘allocation’ to the rising ARC group, conditional on the
covariates. Nearest neighbor matching with replacement was used to match each
‘rising ARC’ patient to a ‘falling ARC’ patient [19,20]. Replacement allowed each
rising ARC unit to be matched to more than one falling ARC unit. A weighting was
given to each falling ARC unit reflecting the frequency with which it was used.
Variables were transformed to dichotomous values prior to matching. Covariate
balance was assessed after matching using difference in means. Cox regression was
then repeated using the weighted propensity-matched database.
A u
th o
r M
a n
t
7
Multivariable logistic regression was undertaken to assess the association
between an increase in peri-operative mortality risk, measured by a rising ARC and
mortality at three years and five years after adjusting for baseline pre-operative risk
of death (including only patients with sufficient three- and five-year data,
respectively). To ensure that observed results were not driven by early deaths, an
additional sensitivity analysis was performed after removal of patients who died
within 30 days of surgery.
All statistical analyses except for propensity score matching were carried out using
Stata 12.1 (StataCorp, College Station, Texas, USA) [21]. ‘R’ was used for propensity
score matching [22]. A p-value of < 0.05 was considered significant.
Results
Complete baseline risk data was available for 98% of the population (Table 1). Non-
survivors at one year were older and had a greater incidence of baseline risk factors.
Non-survivors had higher baseline pre-operative AusScore scores and higher
APACHE-3 scores after admission to the ICU. Discrimination and calibration for
predicted hospital mortality using AusScore and APACHE-3 were good (Table 1 and
online supporting information Fig. A1).
An increase in peri-operative mortality risk, as evidenced by a rising ARC, was
more common in non-survivors at one year (48% vs. 35%, p<0.001). Non-survivors
had higher ARC values than survivors, median (IQR [range]) -0.08 (IQR -1.6 to 2.9
[range -55 to 94]) vs median -0.2 (IQR -0.9 to 0.2 [range -49 to 90])). At all three
levels of baseline risk (as defined by AusScore), ARC values were higher in those who
died.
The primary outcome, mortality at one year, was higher in the rising ARC
group (5.6% vs. 3.5%, p<0.001). After adjusting for baseline risk factors, a rising ARC
was independently associated with an increased one-year mortality (OR 2.6 (95%CI
2.2-3.0), p<0.001) in multivariable analysis (Table 2). A sensitivity analysis
considering ARC as a continuous variable demonstrated higher ARC values were also
independently associated with increased mortality (online supporting information
Table A1).
A u
th o
r M
a n
t
8
The nature of the relationship between ARC and mortality differed according
to baseline risk (interaction p=0.001). Figure 1 shows one-year mortality stratified by
baseline pre-operative risk profile. In the highest baseline risk group, a rising ARC
was associated with three-fold one-year mortality, compared with two-fold in the
lowest risk group. Similar differences in one-year mortality were seen when ARC was
further subdivided into sextiles (online supporting information, Fig. A2).
Figure 2a shows Kaplan-Meier plots after dividing the cohort into two groups, falling
ARC vs. rising ARC. Those with a rising ARC (i.e. an increase in peri-operative risk) had
reduced survival (p<0.001). Patients with a rising ARC had a one-year survival of 94%,
compared with those with a falling ARC who had a one-year survival of 97%. Survival
at three and five years was 90% and 82%, respectively, in the rising ARC group,
compared with 92% and 87% in the falling ARC group.
Figure 2b shows survival plots stratified by baseline pre-operative risk. The
rising ARC group had reduced survival in all three baseline risk groups (p<0.001).
Differences were more pronounced in patients with higher initial baseline pre-
operative risk. In the highest baseline risk group, patients who had a rising ARC had a
five-year survival of only 60% compared with those in the falling ARC group, who had
a five-year survival of 77%, an absolute difference of 17% (p<0.001). In contrast,
amongst the lowest baseline risk group, the absolute difference in five-year survival
was 3% between the rising ARC group (five-year survival 91%) and the falling ARC
group (five-year survival 94%) (p=0.02).
After adjusting for baseline pre-operative risk factors, a rising ARC remained
associated with a reduced survival (mortality hazard ratio 1.86, 95%CI 1.68-2.05,
p<0.001). The Cox proportional hazards model is shown in Table 3. There was also a
significant interaction between baseline risk (AusScore) and ARC group (p=0.009),
supporting the fact that ARC had a differential effect depending on the level of
baseline risk. The C-statistic for the model was 0.78. A sensitivity analysis using ARC
as a continuous variable showed similar findings and is presented in online
supporting information Table A2.
Propensity matching resulted in a well-balanced database of 8637 patients in
the rising ARC group and 11,109 patients in the falling ARC group. Full covariate
balancing statistics are shown in the online data supplement (online supporting
A u
th o
r M
a n
t
9
information Table A3). An increase in peri-operative risk, indicated by a rising ARC,
remained significantly associated with reduced survival time (mortality hazard ratio
1.73, 95%CI 1.56-1.91, p<0.001).
After adjusting for confounding factors, a rising ARC was independently
associated with increased mortality at three and five years (OR 1.88 (1.62-2.19) and
1.76 (1.40-2.20), respectively, p<0.001). This relationship persisted when those who
died before 30 days were removed from the dataset (1.49 (1.26-1.77) and 1.54
(1.21-1.96), respectively, p<0.001).
Discussion:
Our study showed that higher mortality risk following arrival in the ICU after cardiac
surgery, as measured by a rising ARC, was independently associated with long-term
mortality, not only at one year but up to five years after the procedure. These
findings were consistent, irrespective of the type of analysis used.
In an ideal setting, cardiac surgery patients would have the same risk after
surgery as they do before surgery (ARC = 0). There are a few important reasons for
ARC not to be equal to zero. These are: systematic variation in predicted risk of
death and other sources of noise resulting in the generation of ARC; the presence of
inaccurate risk scores (for example due to unmeasured patient risk factors or lack
thereof); and the occurrence of adverse events in the immediate peri-operative
period. For example, events such as inadequate myocardial protection or major
haemorrhage could lead to a patient presenting to the ICU with a higher risk of
death.
We investigated the first possibility using calibration plots, and by plotting
the relationship between risk scores and predicted risk of death. We were unable to
identify systematic variation in mortality estimates that could have confounded the
relationship between ARC and long-term survival. Nonetheless, it is still possible that
noise influences the generation of ARC. The extent of the effect of noise remains
difficult to quantify.
The second possibility is that there are unmeasured patient or surgical risk
factors, not included in AusScore, that become apparent on risk assessment using
A u
th o
r M
a n
t
10
the physiological variables in APACHE-3. The value of ARC could therefore represent
the ‘true’ patient risk rather than be temporally related to intra-operative and
immediate postoperative events. While we have made attempts to mitigate the
possibility of both measurable and immeasurable patient risk affecting ARC by using
multivariate analysis and propensity scores; these methods are not infallible. It is
likely that in some cases, changes in risk (ARC) are associated with inaccurate risk
scoring rather than true differences in peri-operative events. It follows that patients
with higher baseline risk (albeit unmeasured) would have higher long-term mortality.
The third possibility is that a rising ARC is associated with peri-operative
events. In this case, the mechanism by which a rising ARC may influence long-term
survival deserves explanation. A rising ARC (ARC more than zero) refers to
postoperative risk being higher after surgery than before, and thus perhaps that
immediate peri-operative care did not go according to plan. The converse would be
true for falling ARC. We can conceptualise this, albeit simplistically, into ‘optimal’ or
‘non-optimal’ peri-operative care. As such, non-optimal care is associated with
higher acute and long-term mortality. Our previous studies suggest that rising ARC is
associated with higher morbidity, these complications may then be associated with
reduced long term survival [12, 13]. In patients with higher baseline pre-operative
risk profiles, a rising ARC was a more important finding, being associated with the
highest long-term mortality rates. This may suggest that intra-operative events have
the greatest effect on those who are the most fragile, the high-risk patients.
The occurrence of surgical, anaesthetic or other mishaps, for example
accidental vascular injury, failure to adequately revascularise or failed valve repairs,
may lead to longer cardiopulmonary bypass times and physiological derangement,
which could result in higher ARC. More subtle problems could include variation in
processes, such as: peri-operative team performance; unexpected intra-operative
findings; surgical skill [23]; and quality of pre-operative patient work-up [24,25]. Not
all of these complications may be documented in the notes. Acute risk change may,
therefore, be a quantitative method of assessing the complex process of peri-
operative care. The identification of high ARC could highlight patterns of peri-
operative events amenable to practice change or intervention (for example,
myocardial protection or supervision). At present, these suggestions are speculative
A u
th o
r M
a n
t
11
and the possibility of association of ARC with specific events requires further
investigation to be confirmed.
In reality, it is likely that noise, inaccurate risk scoring and adverse events are
responsible for ARC. The relative contribution of each is unclear, but may be
determined in future studies linking changes in ARC back to patients and subsequent
case note review. A new and more comprehensive risk scoring system, with shared
variables between the pre-operative and postoperative risk calculation, could help
eliminate the influence of any possible systematic variation in risk estimation. It
would be unlikely, however, to eliminate the influence of unmeasured risk. This
situation is not unlike the current situation in quality monitoring where it must be
determined whether risk-adjusted outcome variation (e.g. mortality) occurs as a
result of altered quality of care or altered case-mix.
Strengths of this study are that, to our knowledge, there is no other study
that has looked specifically at the effect of early changes in peri-operative risk on
long-term mortality. Our findings suggest that the patient peri-operative experience
may have a long-lasting effect on their survival. Unlike other factors associated with
long-term mortality, the majority of which represent baseline pre-operative risks,
ARC may be modifiable. We used two high quality databases to generate our data.
Sensitivity analyses showed our findings were robust to examination using different
statistical techniques. In addition, many of the other risk factors found to be
associated with long-term mortality were similar to those reported in other studies,
and of similar magnitude [10], implying external validity of our findings. This suggests
that our results may be applicable outside Australia and New Zealand.
Additional limitations of the study should be addressed. APACHE-3 has been
criticised for its ability to predict outcomes in cardiac patients. However, when
recalibrated for procedure type, we found it to be both a discriminative and well-
calibrated test [13]. Other ICU risk estimation scores have since been developed
specifically for cardiac surgery and could be used in future [26]. Acute risk change
requires the collection of two risk scores, currently not collected in any single
database. However, many hospitals collect both datasets and a regular database
match or the collection of additional data between databases would be a feasible
option. Although the AusScore has been derived solely for use in Australia and New
A u
th o
r M
a n
t
12
Zealand, it uses similar prognostic variables to other cardiac surgery risk tools such
as EuroSCORE-2 (online supporting information Table A4). It is therefore possible
that the principles underlying ARC would be applicable in other regions. This study
required probabilistic matching of two databases, and it is possible that matched
patients were not representative of the general population. However, we achieved
approximately 95% database match and characteristics were similar between
groups. Patients with the very lowest ARC values had higher mortality than those
with moderately low ARC values. This may represent the initial high baseline risk
required to generate a low ARC and may limit the use of ARC to discriminate outlier
units with different patient risk profiles. Acute risk change is only relevant to those
patients who are admitted to ICU after cardiac surgery, and has not been
investigated in patients who are already in the ICU and undergo cardiac surgery or in
patients who are not admitted to ICU after cardiac surgery. Although the utility and
value of ARC appear to be robust across multiple analyses, it is a new concept and
requires external validation and peer review.
Acute risk change is a new concept of acute mortality risk change between
the pre- and postoperative periods. A rising ARC is independently associated with
long-term mortality. This may represent a combination of noise, adverse peri-
operative occurrences (avoidable and unavoidable) and unmeasured patient risk.
Further work should focus on the development of risk estimation using shared
variables between preoperative and postoperative estimation and characterisation
of the intra-operative and other factors that lead to high ARC values. This may
identify new targets for improvement in care pathways.
Acknowledgements
TC was funded by an Alfred-Monash Research Fellowship. CR is funded by a National
Health and Medical Research Council Research Fellowship. ANZSCTS-CSD is funded
by the Department of Health Victoria, the Health Administration Corporation
(GMCT) and the Clinical Excellence Commission (CEC) NSW, and funding from
individual units. ANZICS-APD is managed by the ANZICS Centre for Outcome and
Resource Evaluation (CORE). ANZICS CORE is funded by the state and territory health
A u
th o
r M
a n
t
13
departments of Australia and Ministry of Health, New Zealand. No other external
funding or competing interests declared.
References:
1. The Society for Cardiothoracic Surgery in Great Britain & Ireland. Sixth
National Adult Cardiac Surgical Database Report 2008.
http://www.scts.org/_userfiles/resources/SixthNACSDreport2008withco
vers.pdf (accessed 16/05/2017).
2. Billah B, Reid CM, Shardey GC, Smith JA. A preoperative risk prediction
model for 30-day mortality following cardiac surgery in an Australian
cohort. European Journal of Cardiothoracic Surgery. 2010 May; 37:
1086–92.
3. Jacobs JP, Mavroudis C, Jacobs ML et al. What is operative mortality?
Defining death in a surgical registry database: a report of the STS
Congenital Database Taskforce and the Joint EACTS-STS Congenital
Database Committee. Annals of Thoracic Surgery 2006; 81: 1937–41.
4. Papachristofi O, Sharples LD, Mackay JH, Nashef SAM, Fletcher SN, Klein
AA. The contribution of the anaesthetist to risk-adjusted mortality after
cardiac surgery. Anaesthesia 2016; 71: 138-46.
5. Barbeito A, Lau WT, Weitzel N, Abernathy JH, Wahr J, Mark JB. FOCUS:
the Society of Cardiovascular Anesthesiologists' initiative to improve
quality and safety in the cardiovascular operating room. Anesthesia and
Analgesia 2014; 119: 777–83.
6. Glance LG, Hannan EL, Fleisher LA, et al. Feasibility of report ards for
A u
th o
r M
a n
t
14
Analgesia 2016; 122 :1603–13.
7. Rahmanian PB, Adams DH, Castillo JG, Carpentier A, Filsoufi F. Predicting
hospital mortality and analysis of long-term survival after major non-
cardiac complications in cardiac surgery patients. Annals of Thoracic
Surgery 2010; 90: 1221–9.
8. Myles PS. Meaningful outcome measures in cardiac surgery. Journal of
Extracorporeal Technology 2014; 46: 23–7.
9. Stiefel M, Nolan K. A guide to measuring the triple aim: population
health, experience of care, and per capita cost. Institute for Healthcare
Improvement, 2012. www.IHI.org (accessed 16/05/2017)
10. Wu C, Camacho FT, Wechsler AS, et al. Risk score for predicting long-
term mortality after coronary artery bypass graft surgery. Circulation
2012; 125: 2423–30.
11. Shahian DM, O'Brien SM, Sheng S, et al. Predictors of long-Term survival
after coronary artery bypass grafting surgery: results from the Society of
Thoracic Surgeons adult cardiac surgery database (The ASCERT Study).
Circulation 2012; 125: 1491–500.
12. Coulson TG, Bailey M, Reid CM, et al. Acute risk change (ARC) identifies
outlier institutions in perioperative cardiac surgical care when the
standardized mortality ratio cannot. British Journal of Anaesthesia 2016;
117: 164–71.
13. Coulson TG, Bailey M, Reid CM, et al. Acute risk change for
cardiothoracic admissions to intensive care: A new measure of quality in
cardiac surgery. Journal of Thoracic and Cardiovascular Surgery 2014;
16: 1–8.
14. Reid CM, Rockell M, Skillington PD, et al. Initial twelve months
A u
th o
r M
a n
t
15
experience and analysis for 2001-2002 from the Australasian Society of
Cardiac and Thoracic Surgeons--Victorian database project. Heart, Lung
and Circulation 2004 ;13: 291–7.
15. Stow PJ, Hart GK, Higlett T, et al. Development and implementation of a
high-quality clinical database: the Australian and New Zealand Intensive
Care Society adult patient database. Journal of Critical Care 2006; 13:
133–41.
16. Nashef SAM, Roques F, Sharples LD, et al. EuroSCORE II. European
Journal of Cardiothoracic Surgery 2012; 41:734–44.
17. Huq M, Billah B, Reid C. Long term mortality risk prediction following
isolated aortic valve replacement surgery for the Australian population.
Heart, Lung and Circulation 2013; 22: S238.
18. Ho DE, Imai K, King G, Stuart EA. MatchIt: nonparametric preprocessing
for parametric causal inference (version 2.211) [software]. Journal of
Statistical Software 2009; 42: 1-27.
19. Rubin DB, Thomas N. Matching using estimated propensity scores:
relating theory to practice. Biometrics 1996; 52: 249–64.
20. Caliendo M, Kopeinig S, Zukunft der Arbeit zur F. Some practical
guidance for the implementation of propensity score matching. Journal
of Economic Surveys 2008; 22: 31-72.
21. StataCorp. 2011. Stata Statistical Software: Release 12. College Station,
TX: StataCorp LP.
22. R Core Team 2013. Foundation for Statistical Computing, Vienna,
Austria.
23. Burt BM, El Bardissi AW, Huckman RS, et al. Influence of experience and
the surgical learning curve on long-term patient outcomes in cardiac
surgery. Journal of Thoracic and Cardiovascular Surgery 2015; 150:
A u
th o
r M
a n
t
16
1061–3.
24. Martinez EA, Shore A, Colantuoni E, et al. Cardiac surgery errors: results
from the UK National Reporting and Learning System. International
Journal for Quality in Health Care 2011; 23: 151–8.
25. Merry AF, Weller J, Mitchell SJ. Improving the quality and safety of
patient care in cardiac anesthesia. Journal of Cardiothoracic and
Vascular Anesthesia 2014; 28: 1341–51.
26. Shahin J, Ferrando-Vivas P, Power GS, et al. The Assessment of Risk in
Cardiothoracic Intensive Care (ARCtIC): prediction of hospital mortality
after admission to cardiothoracic critical care. Anaesthesia 2016; 71:
1410–6.
Figure legends
Figure 1 One-year mortality stratified by baseline risk of hospital mortality (low=0-
1%, n=10605; medium=1-2%, n=4666; and high=>2%, n=7299). Checkered bars
represent falling acute risk change (ARC), while paler solid bars represent rising ARC.
Figure 2a Kaplan-Meier survival plot comparing rising acute risk change (ARC) group
(dotted line) with falling ARC group (solid line) (log-rank p<0.001).
Figure 2b Kaplan-Meier survival plot comparing rising (dotted line) with falling (solid
line) acute risk change (ARC) in three baseline-risk groups. Top = low mortality risk
(0-1%); middle = moderate mortality risk (1-2%); bottom = high mortality risk (>2%)
(all plots log-rank p<0.001)
A u
th o
r M
a n
t
17
Table 1 Univariable statistics and demographics for one-year mortality, including
acute risk change (ARC) and baseline risk factors. AusScore score is a pre-operative
risk score. Values are number (proportion), mean (SD) or median (IQR [range])
Baseline characteristics Alive at 1-year
n=21,612
Valve surgery 5011 (23%) 279 (29%) <0.001
Combined 2711 (13%) 240 (25%)
Previous cardiac surgery 1504 (7%) 118 (12%) <0.001
Non-elective surgery 5876 (27) 357 (37%) <0.001
Pre-operative patient baseline demographics
NYHA category >2 5424 (25%) 452 (47%) <0.001
Ejection fraction category
30-45% 2781 (13%) 205 (22%) <0.001
>45% 17722 (83%) 620 (66%)
BMI >25 kg.m 16,205 (75%) -2
623 (65%)
146 (128) <0.001
Peripheral arterial disease 2060 (9.5%) 186 (19%) <0.001
Cerebrovascular disease 2333 (11%) 189 (20%) <0.001
Shock 310 (1.4%) 79 (8.3%) <0.001
Respiratory disease 2820 (13%) 237 (25%) <0.001
Diabetes 6921 (32%) 405 (42%) <0.001
Infective endocarditis 513 (2.4%) 69 (7.2%) <0.001
Pre-operative baseline severity of Illness and scores
AusScore* 5 (2-8 [-3 – 31]) 10 (6-13 [-3 – 31]) <0.001
AusScore predicted % mortality risk 1.2 (0.7-2 [0.3-62]) 3 (1.4-5.2 [0.3-62]) <0.001
Low mortality risk (<1%) 10,447 (48%) 158 (16%)
<0.001
Postoperative severity of illness scores
APACHE-3 score 45 (35-55 [0-162]) # 62 (48-77 [12-176]) <0.001
APACHE-3 predicted mortality risk 0.8 (0.4-1.6 [0.06-91]) % 2.8 (1.1-7.1 [0.1-97]) % <0.001
A u
th o
r M
a n
t
18
ARC
ARC -0.2 (-0.9 to 0.2 [-49-
90]) %
94]) %
<0.001
ARC in low baseline risk group (<1%) -.03 (-0.3 to 0.3 [-0.9-
75]) %
91]) %
<0.001
ARC in moderate baseline risk group (1-2%) -0.5 (-0.9 to 0.2 [-1.6-
90]) %
80]) %
<0.001
ARC in high baseline risk group (>2%) -1.5 (-2.7 to -0.4 [-49-
88])%
94])%
<0.001
*AusScore - area under ROC 0.78 (0.76-0.80), X 2 , 8.5; p = 0.59
# APACHE-3 - area under ROC 0.84 (0.82-0.86); X
2 , 8.1; p = 0.62
CABG, coronary artery bypass graft; APACHE, acute physiology and chronic health
evaluation; NYHA, New York Heart Association; BMI, body mass index.
Table 2 Logistic regression model for one-year mortality.
Predictors OR (95%CI) p value
Rising ARC (> 0) 2.60 (2.23-3.03) <0.001
Procedure type
combined 1.69 (1.41-2.03) <0.001
Previous cardiac surgery 1.40 (1.12-1.76) 0.003
Sex; male 0.69 (0.59-0.80) <0.001
Age; years
t
19
<0.001
<0.001
Cerebrovascular disease 1.42 (1.19-1.70) <0.001
Pre-operative shock 2.59 (1.92-3.49) <0.001
Respiratory disease 1.64 (1.39-1.93) <0.001
Diabetes 1.39 (1.20-1.62) <0.001
Infective endocarditis 2.09 (1.52-2.88) <0.001
ARC, acute risk change; CABG, coronary artery bypass graft; NYHA, New York Heart
Association; BMI, body mass index.
Predictors of time to death Hazard ratio (95%CI) p value
Rising ARC 1.86 (1.68-2.05) <0.001
Sex; male 0.83 (0.76-0.92) <0.001
Non-elective surgery 1.40 (1.26-1.55) <0.001
Previous cardiac surgery 1.12 (0.96-1.3) 0.157
A u
th o
r M
a n
t
20
Ejection fraction >45% 1.0 (reference)
30-45% 1.76 (1.58-1.97) <0.001
<30% 3.11 (2.66-3.6) <0.001
BMI > 25; kg.m 0.77 (0.7-0.8) -2
<0.001
Procedure type CABG 1.0 (reference)
Valve surgery 1.45 (1,28-1.64) <0.001
Combined 1.54 (1.37-1.74) <0.001
A u
th o
r M
a n
t
21
ARC; acute risk change; CABG, coronary artery bypass graft; NYHA, New York
Heart Association; BMI, body mass index.
A u
th o
r M
a n
t
) anae_13967_f1.pdf
This article is protected by copyright. All rights reservedA u th
o r
anae_13967_f2a.pdf
A u th
anae_13967_f2b.pdf
A u th
M a n u s c ri p t
Minerva Access is the Institutional Repository of The University of Melbourne
Author/s: Coulson, TG;Bailey, M;Reid, CM;Tran, L;Mullany, DV;Smith, JA;Pilcher, D
Title: The association between peri-operative acute risk change (ARC) and long-term survival after cardiac surgery
Date: 2017-12-01
Citation: Coulson, T. G., Bailey, M., Reid, C. M., Tran, L., Mullany, D. V., Smith, J. A. & Pilcher, D. (2017). The association between peri-operative acute risk change (ARC) and long- term survival after cardiac surgery. ANAESTHESIA, 72 (12), pp.1467-1475. https:// doi.org/10.1111/anae.13967.
Persistent Link: http://hdl.handle.net/11343/293182