The ACTA PORT-score for predicting perioperative risk of blood transfusion for adult cardiac surgery

Abstract

Background. A simple and accurate scoring system to predict risk of transfusion for patients undergoing cardiac surgery is lacking.

We conducted a retrospective analysis of data collected from the ACTA National Audit. For the derivation dataset, we included data from 20 036 patients, which we then externally validated using a further group of 1047 patients.

Methods. We identified independent risk factors associated with transfusion by performing univariate analysis, followed by logistic regression. We then simplified the score to an integer-based system and tested it using the area under the receiver operator characteristic (AUC) statistic with a Hosmer-Lemeshow goodness-of-fit test. Finally, the scoring system was applied to the external validation dataset and the same statistical methods applied to test the accuracy of the ACTA-PORT score.

Results. Several factors were independently associated with risk of transfusion, including age, sex, body surface area, logistic EuroSCORE, preoperative haemoglobin and creatinine, and type of surgery. In our primary dataset, the score accurately predicted risk of perioperative transfusion in cardiac surgery patients with an AUC of 0.76. The external validation confirmed accuracy of the scoring method with an AUC of 0.84 and good agreement across all scores, with a minor tendency to under-estimate transfusion risk in very high-risk patients.

Conclusions. The ACTA-PORT score is a reliable, validated tool for predicting risk of transfusion for patients undergoing cardiac surgery. This and other scores can be used in research studies for risk adjustment when assessing outcomes, and might also be incorporated into a Patient Blood Management programme.

Editor’s key points

• The authors performed a retrospective analysis of data from 20 036 patients undergoing cardiac surgery in the UK to derive and validate a simple scoring system for risk of transfusion.

• The ACTA PORT-score accurately predicted risk of perioperative blood transfusion in cardiac surgery patients.

• This tool will be useful in risk assessment and preoperative optimization approaches.

Cardiac surgery is associated with comparatively high rates of blood product transfusion. Blood products are a limited resource and are both expensive and resource-intensive; cardiac surgery consumes a significant proportion of global blood resources. There is conflicting evidence supporting the relative merits of restrictive¹ or liberal² transfusion triggers, but there is a significant body of evidence that any perioperative transfusion is associated with higher risk of mortality in both the short-³ and long-term.⁴

Several factors have been shown to be independently associated with transfusion in cardiac surgery patients. These include age,⁵ gender, preoperative haemoglobin concentration (Hb), elevated plasma creatinine³ and low body weight.³ Various attempts have been made to synthesise these predisposing factors into a predictive scoring system,^6–8 but as yet none have become widely established.

Patient blood management (PBM) is an increasingly important concept in perioperative medicine. As with any risk-reduction strategy, the first step is to predict individual risk, followed by targeted strategies to mitigate this risk.⁹ This allows for appropriate and focussed use of PBM strategies, which can be expensive or result in harmful side-effects, and should therefore be reserved for those at higher risk of transfusion. Scoring systems to predict general mortality and morbidity are widely used in cardiac surgery and critical care, such as the EuroSCORE¹⁰ and the recently published ARCTIC score¹¹ and Surgical Outcome Risk Tool (SORT).¹² A similar scoring system to guide effective perioperative PBM could have a major impact on resource allocation and potentially on perioperative morbidity. We therefore decided to design the Association of Cardiothoracic Anaesthetists (ACTA) perioperative risk of blood transfusion score – the ACTA-PORT score, using a large national database collected by members of ACTA.

Methods

This study comprises a national service audit of National Health Service (NHS) cardiac surgery centres that collected relevant patient data as part of routine institutional practice. The Research and Ethics Committee of the London School of Hygiene and Tropical Medicine approved the study, and individual patient consent was not required. Between 1st January 2010 and 31st July 2013, data were collected from 10 cardiac surgery centres in the UK during the first ACTA national audit; an analysis of the effect of anaemia has already been published.¹¹ After the analysis was complete, a further centre provided data from the same study period – this was analysed as the external validation dataset.

Baseline data collected included age, gender, preoperative haemoglobin (Hb) creatinine, weight, height, logistic EuroSCORE, diabetes, hypertension, type of surgery proposed and previous cardiac surgery. BMI and body surface area (BSA) were derived from weight and height. These variables were chosen because we a priori expected them to be associated with outcome. Outcomes recorded included number of units of blood transfused, duration of ICU and hospital stay, and death.

As our goal was to produce a simple-to-use integer risk score, the continuous variables age, preoperative haemoglobin, creatinine, logistic EuroSCORE, BMI and BSA were all categorised using clinical judgement where available or otherwise following graphical inspections and taking into account the distribution of the outcome. We used logistic EuroSCORE as EuroSCORE-2 was not in routine use in the NHS during the study period. Although EuroSCORE was designed to calculate the risk of mortality (as opposed to transfusion), we included it as a separate variable to aid in the calculation of risk of transfusion. Operation type was grouped into three categories: isolated coronary artery bypass graft (CABG) or valve surgery, combination surgeries (CABG and valve, or valve and valve), and other (including operations on the aorta).

The univariate association between each of the baseline variables and the outcome of blood transfusion was assessed using logistic regression. Forward and backward stepwise model building approaches were used in developing a final multivariable logistic regression model using a threshold for inclusion or exclusion of P<0.05. Both approaches yielded identical final models. A restricted set of pre-specified potential interactions were investigated using likelihood ratio tests.

As our goal was to produce a risk score that is generalisable beyond the centres involved in this audit, centre was not included as a fixed effect in the model. We compared multivariable logistic regression models omitting centre completely and multivariable mixed effects logistic regression models including centre as a random effect. These two approaches produced almost identical results in terms of the estimate odds ratios and the overall model performance, and we therefore decided to proceed with the former.

Adjusted odds ratios and 95% confidence intervals for the final multivariable model are presented along with P-values from a likelihood ratio test. For each variable in the model, we set the lowest risk category as the reference group so that the risk score would only involve the addition of points. The logistic odds ratio for each category was converted into an integer by dividing by 0.2 and rounding to nearest the nearest whole number. The total integer risk score for each patient was then calculated by summing the points associated with their combination of baseline risk factors.

The discriminatory performance of the risk score before and after simplification was assessed using the area under the receiver operating characteristic curve (AUC) statistic. The goodness of fit of the models, (i.e. how closely predicted risk matched observed risk), was assessed using the Hosmer-Lemeshow goodness-of-fit test.

The predicted risk of transfusion associated with each value of the total integer risk score was calculated and presented in a table and figure. We grouped the risk score into six equally spaced categories (0-4, 5-9, 10-14, 15-19, 20-24 and 25-30) and plotted the observed vs predicted proportion of patients transfused in each category.

We assessed the sensitivity of our results to the influence of missing data using multiple imputation. Multiple imputation with chained equations was used to generate 20 completed datasets. The selected model was then fitted to each of the 20 completed datasets and the estimated coefficients were combined according to Rubin’s rules.

An external validation of the integer risk score was carried out using data from a further cardiac surgical centre. The integer risk score was calculated for each patient in the external dataset and the performance of the risk score was assessed using the AUC and the Hosmer-Lemeshow goodness-of-fit test. We grouped the risk score for the validation patients into the same categories as described above for the derivation data and plotted the observed vs predicted risk of transfusion. We also used our validation dataset to compute the TRACK score and compared our model with TRACK using the DeLong method. We were unable to calculate any other published risk scores as we did not collect the required variables.

The analysis was carried out using Stata 14 (StataCorp, College Station, TX, USA).

Patient involvement

Patients/service users/lay people were not involved in the design of this study.

Results

We analysed data from 20 036 patients, whose baseline characteristics are shown in Table 1. A total of 8635 (43%) patients were transfused.

Table 1

Baseline characteristics. Values are mean (sd), number (proportion or median (IQR (range)). *Indicates isolated CABG or single valve surgery

	All n=20 036	Not transfused n=11 041	Transfused n=8638	P-value
Age; yr	67.1 (11.9)	65.2 (12.0)	69.7 (11.3)	<0.001
Sex; men	14 303 (71.4%)	9093 (79.8%)	5210 (60.3%)	<0.001
Preoperative Hb; g L−1	132 (17)	138 (15)	125 (17)	<0.001
Missing data	3237 (16.2%)	2077 (18.2%)	1160 (13.4%)
Body surface area; m2	1.9 (0.2)	2.0 (0.2)	1.9 (0.2)	<0.001
BMI; kg m2	28.4 (5.1)	29.0 (5.0)	27.6 (5.0)	<0.001
EuroSCORE	4.3 (2.1–8.7 (0.4–98.4))	3.2 (1.7–6.6 (0.4–98.4))	6.0 (3.1–11.7 (0.4–97.9))
Missing data	393 (2%)	238 (2.1%)	155 (1.8%)
Creatinine; µmol L−1	88 (71–106 (9–1547))	88 (71–97 (9–1547))	88 (71–106 (9–1450))	<0.001
Missing data	2172 (10.8%)	1254 (11%)	918 (10.6%)
Diabetes mellitus	3916 (22.0%)	2114 (20.7%)	1802 (23.8%)	<0.001
Missing data	2267 (11.3%)	1208 (10.6%)	1059 (12.3%)
Hypertension	13 325 (67.8%)	7511 (67.2%)	5814 (68.6%)	0.04
Missing data	384 (1.9%)	224 (2.0%)	160 (1.9%)
Operation type
CABG or valve*	14 575 (73%)	8778 (77%)	5797 (67%)
Double procedure	2858 (14%)	1008 (9%)	1850 (21%)
Other	2594 (13%)	1608 (14%)	986 (11%)	<0.001
Missing data	9 (<0.1%)	7 (0.1%)	2 (<0.1%)

Table 1 shows the baseline characteristics of the patients overall and by the outcome of blood transfusion. The mean age of patients in this audit was 67 yr [range 18, 111], and 71% were male. Mean preoperative haemoglobin was 132 g L⁻¹; 31% of patients were anaemic (<130/<120 g L⁻¹ for males/females, respectively). Haemoglobin was not available for 16% of patients. Of the 20 036 patients 8635 (43%) received a blood transfusion perioperatively.

With the exception of a known history of hypertension, all baseline variables were strongly associated with risk of blood transfusion (all P<0.001) in the univariate analysis. Age, EuroSCORE, female gender, diabetes mellitus and elevated creatinine were positively associated with risk of transfusion. Haemoglobin, BMI and BSA were negatively associated with risk of transfusion. Patients undergoing combined surgery were more likely to be transfused. There were marked differences in transfusion rates among the 10 centres, which ranged from 31% to 56% (Table 2).

Table 2

De-identified centres. The difference in transfusion rates between centres was statistically significant (P<0.001)

Centre	All n=20 036	Not transfused n=11 401	Transfused n=8638	Transfusion rate
A	2559 (13%)	1268 (11%)	1291 (15%)	50%
B	732 (3%)	425 (4%)	307 (4%)	42%
C	2058 (10%)	1410 (12%)	648 (8%)	31%
D	2371 (12%)	1233 (11%)	1138 (13%)	48%
E	5371 (27%)	3283 (29%)	2088 (24%)	39%
F	500 (3%)	292 (3%)	208 (2%)	42%
G	960 (5%)	423 (4%)	537 (6%)	56%
H	1986 (10%)	1029 (9%)	957 (11%)	49%
I	1099 (6%)	618 (5%)	481 (6%)	44%
J	2400 (12%)	1420 (12.5%)	980 (11%)	41%

Table 3 shows the adjusted odds ratios, 95% CIs and P-values for the 7 variables included in the final multivariable risk score. During the model building process, it was found that BSA was a stronger predictor of transfusion than BMI. Neither history of hypertension nor diabetes mellitus were found to be independently associated with risk of transfusion. Table 2 also shows the log-odds ratio, their standard errors and the integer points associated with each category. Other than age (P=0.02), all variables in the multivariable risk score were strongly associated with the outcome (P<0.001). No statistically significant interactions were found. The strongest predictor of transfusion was baseline Hb, followed by BSA and EuroSCORE. The AUC for the integer risk score model was 0.760 (95% CI 0.752, 0.768), and the Hosmer-Lemeshow goodness-of-fit test provided no evidence of a poor fit (P=0.23). The AUC from the non-integer risk model (i.e. using the log-odds ratios) was 0.762 indicating that little predictive power had been lost through the simplification process.

Table 3

Multivariable Risk Score outlining corresponding odds ratios, log odds ratios and how ACTA-PORT score was constructed, showing the number of score-points that were attributed to each group

Characteristic	Category	Odds ratio (95%)	P-value	Log odds ratio (SE)	Points
Age; yr	<70	Ref.			+0
	70+	1.11 (1.01, 1.21)	0.02	0.10 (0.04)	+1
Sex	Male	Ref.			+0
	Female	1.27 (1.15, 1.40)	<0.001	0.24 (0.05)	+1
Haemoglobin	<110	6.36 (5.38, 7.52)		1.85 (0.09)	+9
g L−1	110-	4.60 (3.93, 5.38)		1.53 (0.08)	+8
	120-	3.19 (2.79, 3.65)		1.16 (0.07)	+6
	130-	1.93 (1.70, 2.20)		0.66 (0.07)	+3
	140-	1.55 (1.37, 1.77)		0.44 (0.07)	+2
	150+	Ref.	<0.001		+0
Body surface area	<1.7	3.62 (2.97, 4.42)		1.29 (0.10)	+6
m2	1.7-	2.21 (1.85, 2.64)		0.79 (0.09)	+4
	1.9-	1.56 (1.31, 1.85)		0.44 (0.09)	+2
	2.1-	1.24 (1.04, 1.49)		0.22 (0.09)	+1
	2.3+	Ref.	<0.001		+0
EuroSCORE	<1	Ref.			+0
	1-	1.36 (1.10, 1.70)		0.31 (0.11)	+2
	2-	1.73 (1.39, 2.15)		0.55 (0.11)	+3
	3-	2.16 (1.75, 2.68)		0.77 (0.11)	+4
	9+	2.76 (2.20, 3.46)	<0.001	1.01 (0.12)	+5
Creatinine	<88	Ref.			+0
µmol L−1	88-	1.33 (1.23, 1.44)		0.29 (0.04)	+1
	177-	1.93 (1.54, 2.42)	<0.001	0.66 (0.12)	+3
Type of Operation	CABG/Valve	1.38 (1.22, 1.55)		0.32 (0.06)	+2
	Combination	2.84 (2.46, 3.29)		1.05 (0.07)	+5
	Other	Ref.	<0.001		+0
Intercept		NA		−3.00	(0.15)

The risk score for any patient is simply calculated by adding the points associated with their baseline characteristics. For example, a 65 yr old (+0 points) male (+0 points), with baseline Hb of 135 g L⁻¹ (+3 points), BSA of 2.0 (+2 points), logistic EuroScore of 1.5 (+2 points), creatinine of 1.5 (+1 point) and undergoing CABG surgery (+2 points) would have a total risk score of 10 points. A 75 yr old (+1 point) female (+1 point), with baseline Hb of 125 g L⁻¹ (+6 points), BSA of 1.8 (+4 points), logistic EuroScore of 2 (+3 points), creatinine of 2.5 (+3 points) and undergoing valve surgery (+2 points) would have a total risk score of 20 points.

Table 4 and Fig. 1 show the predicted risk of transfusion associated with each value of the risk score. Figure 1 shows the distribution of risk score among patients in the audit. The risk score can in theory take values ranging from 0 to 30, with a higher score associated with a higher risk. For example, the risk of transfusion for a patient with a risk score of 10 is estimated to be 27% compared with an estimated risk of transfusion of 73% for a patient with a risk score of 20. Figure 1 shows that the risk score is fairly normally distributed in this sample of patients with very few patients having a risk score below 5 (1.9%) or above 24 (1.8%). The median risk score was 14, for which the estimated risk of transfusion was 45%.

Table 4

Integer risk score totals and associated predicted risk of transfusion. Low scores have a very low risk of transfusion (i.e. a score of 1 gives a risk of transfusion<5%), whereas a high score of 30 has >95% risk of requiring a transfusion

Integer risk score	Predicted risk of transfusion	Integer risk score	Predicted risk of transfusion
0	0.0470	15	0.500
1	0.0570	16	0.550
2	0.0690	17	0.599
3	0.0830	18	0.646
4	0.1000	19	0.690
5	0.1190	20	0.731
6	0.1420	21	0.769
7	0.1680	22	0.802
8	0.1980	23	0.832
9	0.2310	24	0.858
10	0.2690	25	0.881
11	0.3100	26	0.900
12	0.3540	27	0.917
13	0.4010	28	0.931
14	0.4500	29	0.943
15	0.5000	30	0.953

Fig 1

Distribution of risk scores in the patient population. The distribution follows a relatively normal curve. Superimposed is a line showing the increasing risk of transfusion associated with higher scores.

Figure 2 shows the observed vs predicted risk of transfusion across categories of the risk score. The score performs well in stratifying the transfusion. Among patients with a score below 10, less than 20% were transfused compared with close to 80% of patients with a score of 20 or above, a four-fold higher risk. There is good agreement between the predicted and observed probability of transfusion.

Fig 2

Observed vs predicted transfusion rates in the derivation dataset. There is a close correlation between predicted and observed rates of transfusion across the range of scores in the derivation dataset.

The AUC for the risk score in the external validation dataset was 0.835 (95% CI 0.810, 0.859). However, the score tends to underestimate risk of transfusion, particularly at higher levels of the score, (e.g. 73% and 89% observed vs 60% and 79% predicted in the 15 – 19 and 20 – 24 categories, respectively).

The score was not designed to predict number of units of blood transfused. However, increasing ACTA-PORT score was associated with increased number of units of blood transfused perioperatively: risk score 0-14, median units of blood transfused 0; score 15-19, median 1 unit; score 20-24, median 2 units; and score 25-30, median 3 units.

The results from the sensitivity analysis using multiple imputation did not make substantial changes to the risk score. We also calculated the performance of the ACTA-PORT score at various integer risk score cut-points; the optimum cut-point was 15, with a positive predictive value of 70% and a negative predictive value of 71%, with 70% of values correctly predicted.

Discussion

We developed a simple, integer-based scoring system that accurately predicted the likelihood of transfusion. We also externally validated this, demonstrating its applicability in a real-world scenario.

The concept of a scoring system designed to predict the risk of bleeding or transfusion during cardiac surgery is not new. One of the first efforts in this area was from Papworth Hospital.⁶ This system aimed to measure blood loss exceeding 2mL kg⁻¹ hr⁻¹, requirement for fresh frozen plasma, platelets or cryoprecipitate, or return to theatre after arrival in the ICU. Whilst the negative predictive value of this score was high, only 27% of patients who the score placed in the highest risk category subsequently demonstrated major bleeding. This low positive predictive value was confirmed by a subsequent external validation.⁷

Whilst the Papworth Bleeding Risk Score sought to identify those patients at risk of excessive blood loss in the ICU after cardiac surgery, subsequent scoring systems have sought to predict the risk of transfusion. A relatively recent example of this is the Transfusion Risk and Clinical Knowledge (TRACK) score.⁸ TRACK aimed to create a simple, easily applied system, based on five predictors of transfusion risk, assigning each variable a proportional risk score based on the clinical condition of the patient. This scoring system was subsequently validated against an external cohort, and proved to be superior to three earlier systems^13–15 with an AUC of 0.70. Like the ACTA-PORT score, TRACK aimed to improve the utility of the scoring system for clinical practice as a result of its relative simplicity, whist at the same time remaining sensitive and specific for predicting transfusion risk. We used our validation dataset to also compute the TRACK score; the AUC of TRACK was 0.781 (ACTA-PORT vs. TRACK P<0.001 using the DeLong method for comparing risk scores). We therefore conclude that the ACTA-PORT score performs significantly better than TRACK.

Recently, Goudie and colleagues¹⁶ published two risk prediction models: one for any red cell transfusion, and another for a requirement for massive transfusion. This is considerably more complex than the simpler TRACK and ACTA-PORT prediction models. When the Goudie model was published, it represented an advance on many existing scoring systems,^10,15,17 with an AUC of 0.77 for any red blood cell transfusion. We were unable to calculate the risk of transfusion from our dataset using the Goudie method as we did not collect all the necessary data.

The use of risk scores for transfusion such as ACTA-PORT might allow clinicians to quantify this risk before surgery, thereby potentially allowing modification of important risk factors during preoperative optimisation. Transfusion has been shown to be associated with increased 30-day mortality,¹⁸ morbidity related to ischaemia,¹⁹ infection,^20–23 renal impairment,²⁴ post-CABG graft occlusion²⁵ and acute lung injury.²⁶ With respect to longer term outcomes, Engoren and colleagues found that blood transfusion during cardiac surgery was associated with a doubling of the risk of death at five yr.²⁷ Yet this clinical intervention, with appropriate preoperative warning and preparation, can potentially be avoided.

In our study, the only realistically modifiable risk factor associated with requirement for blood transfusion was Hb. Patients with an Hb < 130 g L⁻¹ accounted for nearly 50% of all transfusions, despite making up only one-third of the total cohort. Using the risk profile of those patients included in the ACTA-PORT cohort, a PBM program able to increase haemoglobin from 120 to 130 g L⁻¹ would theoretically decrease risk of transfusion during the perioperative period by 40%, with an implied reduction in perioperative morbidity and mortality.

Similar to previous studies that have used the retrospective analysis of large databases to generate a risk score, our study suffers from some limitations. First, the preoperative management of patients presenting for cardiac surgery at the centres involved in the study was not standardised. The possibility that patients at certain centres were exposed to different PBM strategies therefore cannot be excluded, and could potentially confound any subsequent analysis. Such strategies might include differences in the cessation of anti-platelet therapies and use of cell salvage, and transfusion preferences of individual surgeons and centres. All centres administered tranexamic acid routinely, but at different doses depending on institutional preference.

Secondly, despite demonstrating overall reliability in predicting risk of transfusion, the score does slightly underestimate transfusion risk in the higher risk categories. Patients with risk score >20 had a roughly 10% higher observed rate of transfusion relative to predicted risk. This might reflect the nature of the validation cohort, being from a single centre, as opposed to the multi-centre model derivation dataset. Consequently, the transfusion practices in the specific centre might not accurately reflect general transfusion practice. This could be as a result of regional variation in anaemia incidence as described,¹⁷ or a higher incidence of complex cardiac surgery at this specific centre. The decision by the authors to not specifically correct for regional variation was made in order to retain generalisation, enabling the scoring system to be used at centres outside those that participated in the initial cohort. Consequently, if a centre has policies or surgeons that make transfusions more likely (compared with the average in the audit), the score will underestimate risk, as evidenced by the results of the validation cohort. Whilst ACTA-PORT will be useful to stratify risk of transfusion in any patient presenting for cardiac surgery, the system will need to be recalibrated if centres outside of the control cohort wish to use it to predict absolute risk. We plan to design a simple App/online calculator to calculate the ACTA-PORT score when planning surgery or discussing risk with patients. In addition, we were only able to compare the ACTA-PORT score with the TRACK score, and were unable to compute other risk scores because of lack of appropriate data.

Finally, the score makes use of the EuroSCORE¹⁰ as an overall marker of patient mortality. This might limit the applicability of the scoring system beyond health systems that routinely collect this information, particularly centres in China²⁸ and Australia.²⁹ Furthermore, risk prediction models are subject to constant revision,³⁰ potentially further limiting the applicability of derived models that make use of them.

In summary, using a large, multicentre cohort of patients collected from multiple cardiac centres, we derived a robust, simple and accurate system for predicting risk of transfusion for patients undergoing cardiac surgery. Future research will ideally include independent validation against a further external cohort, comparing ACTA-PORT with other bleeding/transfusion risk scores. This and other scores could be used in research studies for risk adjustment of patients when assessing the outcome of an intervention, and could also be incorporated into Patient Blood Management programmes.

Authors’ contributions

Study design/planning: A.A.K., T.R.

Study conduct: A.A.K., N.F., C.E.

Data analysis: T.C.

Writing paper: T.C., J.Y., L.F.M.

Revising paper: all authors

Declaration of interest

All authors have completed the ICMJE uniform disclosure form at http://www.icmje.org/coi_disclosure.pdf and declare: A.A.K. has received funding for research/education and/or honoraria and travel support from Pharmacosmos, Vifor Pharma, CSL Behring and Fisher and Paykel; C.E. has received honoraria and travel support from Pharmacosmos; T.R. has received funding for research/education and/or honoraria and travel support from Pharmacosmos and Vifor Pharma. S.F. is the President of the Association of Cardiothoracic Anaesthesia and Critical Care (ACTACC). The authors received no support from any organisation for the submitted work; and no financial relationships with any organisations that might have an interest in the submitted work in the previous three yr; and no other relationships or activities that could appear to have influenced the submitted work.

Appendix 1

Contributors

R. P. Alston, Royal Infirmary of Edinburgh, Scotland

H. Pauli, Freeman Hospital, Newcastle

A. Vijayan, Castle Hill Hospital, Hull

A. Pai, Essex Cardiothoracic Centre, Basildon, Essex

D. Krahne, Kings College Hospital, London

D. Glasgow, Royal Victoria Hospital, Belfast

P. Fernandez Jimenez, Manchester Royal Infirmary, Manchester)

S. Agarwal, Liverpool Heart and Chest Centre, Liverpool)

A. Kelleher, Royal Brompton Hospital, London

A. Cohen, Bristol Heart Institute, Bristol

N. Balani and G. Hallward, St Thomas Hospital, London, UK

Author notes

*

Handling editor: Hugh C Hemmings Jr