Intensive Care Medicine

, Volume 39, Issue 10, pp 1808–1817

Increased perioperative mortality following aprotinin withdrawal: a real-world analysis of blood management strategies in adult cardiac surgery

Authors

  • Graham J. Walkden
    • Bristol Heart InstituteUniversity of Bristol
  • Veerle Verheyden
    • Department of Cardiac SurgeryUniversity Hospitals Bristol NHS Foundation Trust
    • Department of Cardiovascular Sciences, Clinical Sciences Wing, Glenfield General HospitalUniversity of Leicester
  • Rosalind Goudie
    • School of Social and Community MedicineUniversity of Bristol
    • Bristol Heart InstituteUniversity of Bristol
    • Department of Cardiovascular Sciences, Clinical Sciences Wing, Glenfield General HospitalUniversity of Leicester
Original

DOI: 10.1007/s00134-013-3020-y

Cite this article as:
Walkden, G.J., Verheyden, V., Goudie, R. et al. Intensive Care Med (2013) 39: 1808. doi:10.1007/s00134-013-3020-y

Abstract

Purpose

To evaluate the effect of aprotinin withdrawal in 2008 on patient outcomes, to assess the likely risks and benefits of its re-introduction, and to consider the relevance of existing evidence from clinical trials to ‘real-world’ practice.

Methods

We performed a nested case–control study of two cohorts undergoing adult cardiac surgery in a single tertiary centre. The first group underwent surgery between 1 January 2005 and 30 July 2007 (n = 3,578), prior to aprotinin withdrawal; the second group underwent surgery between 1 January 2009 and 31 December 2010 (n = 3,030), after aprotinin withdrawal. Propensity matching was used to select patients matched for 24 covariates in both groups (n = 3,508). We also estimated the effect of aprotinin withdrawal on a subgroup of high-risk patients (n = 1,002). Results were expressed as adjusted odds ratios (OR) and 95 % confidence intervals (CI) for categorical data and hazard ratios (HR) for time-to-event data.

Results

In propensity-matched cohorts, the withdrawal of aprotinin from clinical use was associated with more bleeding, higher rates of emergency re-sternotomy, OR 2.10 (1.04–4.25), and acute kidney injury (AKI), OR 1.86 (1.53–2.25). In high-risk patients, the increases in bleeding and AKI following aprotinin withdrawal were of a greater magnitude. Aprotinin withdrawal was also associated with a significant increase in 30-day mortality, HR 2.51 (1.00–6.29), in the high-risk group. The results were not altered by sensitivity analyses that adjusted for potential selection bias, time series bias and unmeasured confounders.

Conclusions

Aprotinin withdrawal was associated with increased complication rates and patient deaths following cardiac surgery. These real-world findings are at odds with those of randomised trials and cohort studies that have considered the clinical role of aprotinin.

Keywords

AprotininCardiac surgeryMortalityBlood management

Introduction

Coagulopathic haemorrhage and the subsequent requirement for large-volume blood transfusion are important risk factors for poor clinical outcomes in cardiac surgery [13]. This remains a significant problem; 25 % of all UK cardiac surgery patients audited over a 3-month period in 2010 received four or more red blood cell (RBC) units or non-RBC components for coagulopathic bleeding [4]. This reflects the increasingly elderly patients and complex case mix referred for surgery, as well as the increased emphasis on early revascularisation of patients with unstable ischaemic heart disease who are often receiving multiple anti-thrombotic drugs [5]. Safe and effective blood management is therefore a key component of high-quality care, particularly in high-risk patients [6, 7]. Aprotinin, a serine protease inhibitor with broad anti-inflammatory and pro-haemostatic effects, was initially licensed for use in high-risk coronary artery bypass grafting, where it was used widely and had demonstrable efficacy in terms of reducing bleeding, transfusion and re-sternotomy rates [8]. However, between 2005 and 2008 several large observational studies raised safety concerns relating to the use of aprotinin, reporting associations between aprotinin use and acute kidney injury (AKI), myocardial infarction and stroke [911]. This was most evident when aprotinin was compared to the lysine derivatives tranexamic acid or epsilon-aminocaproic acid (EACA), which can also reduce perioperative haemorrhage in cardiac patients. Following the publication of the blood conservation using antifibrinolytics in a randomised trial (BART) [12] in 2008, which reported an apparent increase in mortality in the aprotinin group when compared to groups receiving tranexamic acid and EACA, the drug was withdrawn from the market and licensure was suspended. However, subsequent revisiting of the BART study data raised questions as to the validity of this safety signal [13], and following a comprehensive review it was re-licensed in Europe and Canada in 2010. The aims of the current study were to evaluate the effect of aprotinin withdrawal in 2008 on patient outcomes in a large tertiary cardiac unit, to assess the likely risks and benefits to patients from its re-introduction and to cast light on the relevance of the existing evidence from clinical trials to ‘real-world’ practice.

Methods

This retrospective observational case–control study was approved by the South West research ethics committee under reference 11/SW/0075. The requirement for written informed consent was waived. Our research objectives and methods were specified prior to execution. A more detailed description of our methodology is available in an eSupplement.

Study population and data sources

The Bristol Royal Infirmary (Bristol, UK) established a database of adult cardiac surgical patients in April 1996 [Patient Analysis and Tracking System (PATS), Dendrite Clinical Systems, London, UK] for which standardised perioperative and postoperative data are routinely and prospectively collected. All patients ≥16 years old in the PATS database whose data could be linked to the hospital’s biochemistry databases (for the determination of AKI) and the blood bank database (of blood products issued and used) between 1 January 2005 and 31 December 2010 were initially included (n = 8,795).

Institutional blood management

Aprotinin was administered according to individual physician preference, in many cases without published international guidelines [6]. Alternatively, tranexamic acid (Cyclokapron®, Pfizer, UK) was administered as a 2-g intravenous bolus, plus an additional 2-g dose administered to the pump prime for procedures utilising cardiopulmonary bypass. The use of intraoperative cell salvage was recorded in the PATS database only for those patients where salvaged blood was processed and autotransfused (Fresenius C.A.T.S.®, Terumo, Surrey, UK). Clinical concern about bleeding was managed by a Thromboelastogram (TEG®, Haemoscope Corporation, USA) guided algorithm, with administration of up to two cycles of four units of fresh frozen plasma (FFP) and one pooled unit of platelets, followed by four units of cryocoprecipitate and then recombinant activated factor VIIa (Novoseven®, Novo Nordisk Inc., NJ, USA).

Definitions of exposures of interest

We defined two cohorts for comparison based on the declining use of aprotinin at the hospital (eFigure 1): the first group was operated on between 1 January 2005 and 30 June 2007 (n = 3,578); the second group was operated on between 1 January 2009 and 31 December 2010 (n = 3,030). To determine the effect of withdrawing aprotinin upon high-risk patients, we also extracted a subgroup defined by their receipt of aprotinin in the first group (n = 756) and compared them with a propensity-matched subgroup who underwent surgery in the second group.

Definitions of outcomes of interest

Three co-primary outcomes were prespecified: (1) the transfusion of allogenic blood components (RBC, FFP, platelets and cryoprecipitate); (2) re-sternotomy due to bleeding and blood lost postoperatively on the intensive care unit (ITU); (3) measures of morbidity including low cardiac output, new onset AKI (defined by the creatinine criteria of the KDIGO classification), pulmonary complications and infective complications. We also examined all-cause mortality and the durations of ventilation, ITU stay and overall hospital stay.

Statistical analysis

We selected 24 preoperative covariates (recorded on the y-axis of Fig. 1) for inclusion in a propensity model [14]. Propensity scores were then estimated for patients for whom all 24 preoperative characteristics were recorded via logistic regression using the methods of Rosenbaum and Rubin [15].
https://static-content.springer.com/image/art%3A10.1007%2Fs00134-013-3020-y/MediaObjects/134_2013_3020_Fig1_HTML.gif
Fig. 1

Standardised differences across all 24 preoperative variables included in the propensity score for a all patients and b high-risk patients. This metric was found to be <5 % in the all-patient analysis and <10 % in the high-risk analysis

Our specification of a one-to-one ‘nearest neighbour’ propensity-matching algorithm with replacement permitted was chosen to minimise selection bias at the expense of variance. A total of 1,754 post-cessation patients were retrospectively matched to 1,754 pre-cessation counterparts. To maximise the number of unique patients in the pre-cessation arm of the high-risk subgroup analysis, 501 pre-cessation patients were prospectively matched to 501 post-cessation equivalents. After matching, t tests for the equality of covariate means in pre- and post-withdrawal groups were found to be non-significant (eTable 1), and standardised differences across all covariates (a metric of covariate balance defined by Rosenbaum and Rubin [16]) were found to be <5 % in the all-patient analysis and <10 % in the high-risk analysis (Fig. 1). We then proceeded to derive adjusted frequencies and odds ratios for our pre-specified outcomes. Length-of-stay and mortality outcomes were considered as time-to-event data. We employed Cox proportional hazards regression to compute hazard ratios across the entire duration of follow-up and also for epochs of follow-up time. Log-rank tests for differences between groups provided associated P values. To account for the correlation introduced by propensity matching, we present cluster confidence intervals for all results.

Given the high likelihood of outcomes changing over time independent of the effect of aprotinin withdrawal, we conducted a sensitivity analysis to investigate the potential bias introduced by the temporal definitions of pre- and post-withdrawal groups. The patient sample was serially decreased in size by excluding 3-month periods from the beginning and end of the time series (eFigure 1), and the outcomes re-quantified. We conducted further analyses to determine the sensitivity of our results to alternative propensity-matching strategies and to simulated unobserved confounders using the methods proposed by Ichino et al. [17]. Finally, we examined the interaction between aprotinin and alternate blood management strategies on outcomes using multivariate logistic regression. All analyses were performed in STATA 12 (Stata Corp., Texas, USA).

Results

Study population

Of the 8,795 cardiac surgical patients whose preoperative characteristics we present in Table 1 and eTable 2, transfusion data were available for 8,377. Blood loss on ITU was recorded for 7,218 subjects. Postoperative complications could be derived for all patients, with the exception of AKI for 8,500 patients. It was possible to calculate ventilator time, ITU stay and hospital stay for 8,178, 8,773 and 8,772 cases, respectively. All-cause mortality was recorded for all patients, 213 (2.4 %) of whom died in hospital. A total of 756 (8.6 %) patients were administered aprotinin. Comparisons of the pre- and post-cessation cohorts indicated that patients’ preoperative Euroscores, procedural complexity and CPB duration were greater in the latter cohort.
Table 1

Characteristics of precessation patients (1 January 2005 to 30 June 2007) and post-cessation patients (1 January 2009 to 31 December 2012)

Characteristic

Pre-cessation (n = 3578)

Post-cessation (n = 3030)

P value

N

Statistic

N

Statistic

Demographics

 Age, median (IQR), years

3,578

68.1 (54, 82.2)

3,030

68.8 (54.1, 83.5)

0.049

 Sex, %

  Male

2,691

75.2

2,225

73.4

0.099

  Female

887

24.8

805

26.6

 

Cardiac disease

 Angina, %

  No angina

944

26.4

1,062

35.2

0.000

  CCS 1–4

2,588

72.3

1,957

64.6

 

 Dyspnoea, %

  No dyspnoea

2

0.1

0

0.0

0.119

  NYHA 1–4

3,527

98.6

3,018

99.6

 

 Previous MI, %

2,247

62.8

2,030

67.0

0.000

 MI within 30 days of surgery, %

432

11.8

575

19.1

0.000

 Catheterisation within 30 days of surgery, %

1,257

35.1

1,144

38.2

0.011

 Left ventricular ejection fraction, %

  ≥50 %

2,597

72.6

2,294

75.7

0.000

  <50 %

867

24.2

666

22.0

 

 Coronary disease severity (>50 % stenosis), %

  No vessels

682

19.0

734

24.2

0.000

  1–3 vessels

2,712

75.8

2,033

67.1

 

 Left main stem disease, %

736

20.5

555

18.3

0.117

 Cardiogenic shock, %

37

1.0

38

1.3

0.401

Co-morbidities

 Diabetes mellitus, %

656

18.3

532

17.6

0.324

 Hypertension, %

2,417

67.6

2,045

67.5

0.516

 Smoking status, %

  Never

1,144

32.0

1,173

38.7

0.000

  Current or ex-smoker

2,376

66.4

1,844

60.9

 

 Renal status, %

  Normal

3,517

98.3

2,956

97.6

0.357

  Abnormal

61

1.7

65

2.1

 

 eGFR, mean (SD), ml/min/1.73 m2

3,482

63.9 (15.5)

2,991

62.6 (16.5)

0.000

 Euroscore, median (IQR)

3,567

4 (0, 8)

2,918

5 (1, 9)

0.000

 Time between catheterisation and surgery, median (IQR), days

3,357

65 (−37, 167)

2,586

38 (−33, 109)

0.000

Operative complexity

 Type of operation, %

  CABG ± valve

2,619

73.1

1,938

64.0

0.000

  Valve only

592

16.5

578

19.1

 

 Operative priority, %

  Elective

2,004

56.0

1,723

56.9

0.665

  Urgent or Emergency

1,563

43.7

1,265

41.7

 

 CPB, %

2,148

60.0

1,806

59.6

0.000

 Cross-clamp, %

2,024

56.6

1,716

56.6

0.000

 Aprotinin, %

756

21.1

0

0.0

0.000

 Tranexamic acid, %

2,350

65.7

2,384

78.7

0.000

 Cell salvage, %

404

11.3

380

12.5

0.106

 rFVIIa, %

25

0.7

30

1.0

0.301

 Lowest core temperature, mean (SD), °C

3,367

33.6 (3.0)

2,361

32.9 (3.5)

0.000

Significant differences between the two groups determined for continuous variables via the non-parametric Wilcoxon rank sum test; categorical variables via the Pearson χ2 test

CABG coronary artery bypass graft, CCS Canadian Cardiovascular Society, CPB cardiopulmonary bypass, eGFR estimated glomerular filtration rate, IQR interquartile range, MI myocardial infarction, NYHA New York Heart Association, Pre-op pre-operative, rFVIIa recombinant factor VIIa, SD standard deviation. An expanded summary of patient characteristics can be found in eTable 1

Transfusion requirements

Patients received more blood product transfusions after aprotinin had been withdrawn (Table 2). Adjusted odds ratios for the associations of aprotinin withdrawal with RBC, platelet, FFP and cryoprecipitate transfusion were 1.24 (CI 1.04–1.49), 2.16 (CI 1.69–2.76), 1.81 (CI 1.33–2.45) and 3.90 (CI 1.32–11.51), respectively. In the high-risk group, adjusted odds ratios for the associations of aprotinin withdrawal for RBC, platelet, FFP and cryoprecipitate transfusions were 1.22 (CI 0.89–1.66), 1.89 (CI 1.34–2.65), 1.63 (CI 1.11–2.39) and 2.64 (CI 0.81–8.65), respectively.
Table 2

Adjusted odds ratios comparing patients after aprotinin withdrawal to those before aprotinin withdrawal

Outcome

All patients

High risk patients

Pre-cessation (n = 1754)

Post-cessation (n = 1754)

Odds ratio (CI)

Pre-cessation (n = 501)

Post-cessation (n = 501)

Odds ratio (CI)

N

Statistic

N

Statistic

N

Statistic

N

Statistic

Transfusion requirements

 RBC, %

706

40.3

800

45.6

1.24 (1.04–1.49)*

278

55.5

302

60.3

1.22 (0.89–1.66)

 RBC ≥4 units, %

156

8.9

248

14.1

1.68 (1.29–2.21)**

99

19.8

103

20.6

1.05 (0.74–1.53)

 Platelets, %

225

12.8

423

24.1

2.16 (1.69–2.76)**

99

19.8

159

31.7

1.89 (1.34–2.65)**

 FFP, %

140

8.0

238

13.6

1.81 (1.33–2.45)**

71

14.2

106

21.2

1.63 (1.11–2.39)*

 Cryoprecipitate, %

7

0.4

27

1.5

3.90 (1.32–11.51)*

5

1.0

13

2.6

2.64 (0.81–8.65)

Morbidity

 Blood loss, med (IQR), ml

1,754

575 (125, 1,025)

1,754

600 (175, 1,025)

 

501

475 (100, 850)

501

600 (150, 1,050)

 

  <1,000 ml, %

1,469

83.8

1,429

81.5

0.85 (0.67–1.08)

437

87.2

399

79.6

0.57 (0.38–0.87)**

  ≥1,000 ml, %

285

16.3

325

18.5

1.17 (0.92–1.49)

64

12.8

102

20.4

1.75 (1.15–2.64)**

 Re-sternotomy for bleeding, %

32

1.8

66

3.8

2.10 (1.04–4.25)*

10

2.0

27

5.4

2.80 (1.25–6.25)**

 Low cardiac output, %

915

52.2

708

40.4

0.62 (0.52–0.74)**

332

66.3

266

53.1

0.58 (0.42–0.79)**

 AKI, %

410

23.4

634

36.2

1.86 (1.53–2.25)**

163

32.5

207

41.3

1.46 (1.06–2.01)*

 AKI stage

  Stage 0, %

1,344

76.6

1,120

63.9

0.54 (0.44–0.65)**

338

67.5

294

58.7

0.68 (0.50–0.94)**

  Stage 1, %

303

17.3

440

25.1

1.60 (1.29–1.99)**

109

21.8

142

28.3

1.42 (0.99–2.05)

  Stage 2, %

63

3.6

116

6.6

1.90 (1.25–2.89)**

27

5.4

38

7.6

1.44 (0.79–2.62)

  Stage 3, %

44

2.5

78

4.5

1.81 (1.11–2.95)*

27

5.4

27

5.4

1.00 (0.53–1.90)

 Pulmonary complications, %

319

18.2

182

10.4

0.52 (0.41–0.66)**

119

23.8

54

10.8

0.38 (0.25–0.59)**

 Infective complications, %

223

12.7

113

6.4

0.47 (0.36–0.62)**

92

18.4

31

6.2

0.29 (0.17–0.50)**

AKI acute kidney injury, CI 95 % confidence interval, FFP fresh frozen plasma, IQR interquartile range, RBC red blood cells. The corresponding unadjusted results are presented in eTable 3

* P < 0.05, ** P < 0.01

Blood loss and re-sternotomy

Table 2 demonstrates that the distribution of blood lost on ITU shifted from lower volumes (<1,000 ml) towards higher volumes (≥1,000 ml) after aprotinin had been withdrawn. This effect was more marked amongst high-risk patients where patients losing ≥1,000 ml of blood increased from 12.8 to 20.4 % [adjusted odds ratio 1.75 (CI 1.15–2.64)]. Overall, the adjusted rate of re-sternotomy due to bleeding increased from 1.8 % pre-cessation to 3.8 % post-cessation, with an associated odds ratio of 2.10 (CI 1.04–4.25). The cessation of use of aprotinin had a greater effect upon re-operation in the high-risk subgroup: re-sternotomy increased from 2.0 to 5.4 %, yielding an adjusted odds ratio of 2.80 (CI 1.25–6.25).

Postoperative morbidity

Postoperative cardiac, pulmonary and infectious morbidity decreased after aprotinin had been withdrawn. This effect was manifested both before and after propensity matching (eTable 3). Adjusted odds ratios for low cardiac output, pulmonary and infective complications were 0.62 (CI 0.52–0.74), 0.52 (CI 0.41–0.66) and 0.47 (CI 0.36–0.62), respectively. In contrast, aprotinin withdrawal conferred a significant increase in postoperative AKI from 23.4 to 36.2 % of all patients [adjusted odds ratio 1.86 (CI 1.53–2.25)]. High-risk patients were similarly affected by the cessation of use of aprotinin, with corresponding adjusted odds ratios of 0.58 (CI 0.42–0.79), 0.38 (CI 0.25–0.59) and 0.29 (CI 0.17–0.50) for low cardiac output, pulmonary and infective complications, respectively. Postoperative AKI increased from 32.5 to 41.3 % of high-risk subjects [adjusted odds ratio 1.46 (CI 1.06–2.01)].

Resource utilisation

Time-to-event analyses showed that aprotinin withdrawal was associated with shorter ITU [HR 0.87 (CI 0.80–0.95)] and hospital [HR 0.91 (CI 0.85–0.99)] stays (eTable 4). In high-risk patients, aprotinin withdrawal was not associated with significant reductions in resource use: ITU stay [HR 0.91 (CI 0.77–1.07)], hospital stay [HR 0.96 (CI 0.82–1.11)], and in fact ventilation time was increased in this group [HR 1.33 (CI 1.14–1.54)].

Mortality

We present Kaplan-Meier curves of cumulative deaths over the first postoperative year and adjusted hazard ratios for the effect of aprotinin withdrawal on mortality in Fig. 2, eFigure 2 and eTable 4. A trend toward increased all-cause mortality was evident in the first postoperative month following the withdrawal of the drug [HR 2.10 (CI 0.94–4.69)]. The relative increase in mortality was greater amongst high-risk patients [HR 2.51 (CI 1.00–6.29)]. Examination of the causes of in-hospital death amongst pre- versus post-aprotinin withdrawal did not demonstrate any clear change in the types of major morbidity preceding death, although this may reflect low statistical power for this analysis (eTable 5).
https://static-content.springer.com/image/art%3A10.1007%2Fs00134-013-3020-y/MediaObjects/134_2013_3020_Fig2_HTML.gif
Fig. 2

Kaplan-Meier failure functions showing the cumulative proportion of patients dying from any cause over the first postoperative year for patients pre-cessation (black) and post-cessation (red) in the high-risk subgroup. Dashed lines depict corresponding 95 % confidence intervals. Hazard ratios were calculated for epochs 0–30 days and 30 days to 1 year. See eFigure 2 for corresponding plots for the all-patient analysis

Sensitivity analyses

Sensitivity analyses performed on our binary outcomes demonstrated that both their magnitudes and directions of effect were highly robust to potentially unobserved confounders, alternate propensity-matching strategies and serially reducing durations of the pre- and post-withdrawal eras (eTables 6, 7 and eFigure 3). For example, our simulated confounders reduced the mean effect of aprotinin withdrawal on AKI in all patients by just 10–17 %. Re-calculating this outcome with alternate matching strategies conferred anything between an 11 % reduction and a 6 % increase. Alternative propensity-matching strategies including ‘greedy matching’ and stipulation of the ‘common support’ condition did not alter our results (eTables 8, 9). Serial restrictions in the duration of the pre- and post-withdrawal eras reduced the odds ratio for the effect of aprotinin withdrawal on AKI in all patients by no more than 21 % (eFigure 3).

Interaction among alternate blood management strategies, aprotinin and outcome

Associations between aprotinin and adverse outcomes may be confounded by alternate blood management strategies and by the administration of allogenic blood components. In this cohort, aprotinin withdrawal was associated with an increased use of tranexamic acid and recombinant factor VIIa (rFVIIa). In an exploratory analysis of the entire cohort, we estimated the associations among aprotinin, tranexamic acid or rFVIIa administration, increased bleeding and transfusion and death using multivariate logistic regression. This demonstrated significant associations between aprotinin, re-sternotomy for bleeding, the administration of allogenic RBC and non-RBC blood components, and 30-day mortality (Table 3). Examination of the contribution of each of these variables to the variance of the outcome measure indicated that RBC transfusion explained 13 % of the variance, whereas aprotinin explained 2 % and the administration of tranexamic acid 0.4 %.
Table 3

Adjusted odds ratios showing associations between alternate blood management strategies and in-hospital mortality

Predictors

Odds ratio

95 % confidence interval

P value

R2

Lower

Upper

Propensity score

1.07

1.05

1.09

0.000

0.07

Aprotinin

1.05

1.04

1.00

0.045

0.02

Tranexamic acid

1.00

0.93

1.07

0.954

0.00

No antifibrinolytic

0.98

0.92

1.03

0.441

0.00

Re-sternotomy for bleeding

1.07

1.04

1.10

0.000

0.01

RBC transfusion

1.34

1.44

1.24

0.000

0.13

Platelet transfusion

1.29

1.22

1.38

0.000

0.06

FFP transfusion

1.35

1.29

1.42

0.000

0.06

rFVIIa

1.02

1.00

1.04

0.060

0.00

R2 is the fraction of the variance in the outcome variables explained by each predictor variable

rFVIIa recombinant factor VIIa

Discussion

The main finding of this observational nested case-control study is that after adjustment for differences in baseline risk factors, the withdrawal of aprotinin from clinical use was associated with more bleeding and transfusions, higher rates of emergency re-sternotomy and AKI, and significant reductions in pulmonary, infectious and cardiac morbidity. In high-risk patients, the increases in bleeding, transfusion and AKI were of a greater magnitude. Aprotinin withdrawal was also associated with a significant increase in 30-day mortality in the high-risk group.

These results are at odds with the findings of systematic reviews of both randomised trials and observational cohort studies that have attempted to define the indications for aprotinin in cardiac surgery [7, 1822]. Early randomised controlled trials (RCT) demonstrated clear clinical benefits attributable to aprotinin when compared to placebo [7]. However, the very high transfusion rates in the placebo arm for what were low-risk procedures by modern standards and the introduction of cheap alternative anti-fibrinolytics, such as the lysine analogues tranexamic acid and EACA, which themselves have proven efficacy, now mean that these findings have limited relevance to contemporary practice. When aprotinin was compared to these lysine analogues in a mixed risk contemporary cardiac surgery population in the BART study [12], no clinically important benefit for aprotinin relative to the other treatments beyond reduced blood loss was identified. More importantly, BART raised the possibility of increased mortality attributable to aprotinin use. This finding has been supported by a subsequent systematic review conducted by several of the BART investigators [18]. However, reviews of similar data by other groups have not supported these findings [18, 19]. Importantly, no contemporary review of the evidence from RCTs has shown a survival benefit from aprotinin, as suggested in our own study. Observational cohort studies have also failed to demonstrate clinically important benefits attributable to aprotinin use, although these types of studies are limited by systematic bias, unmeasured confounders, and a lack of standardisation of analyses and reporting between studies. For example, cohort studies that use propensity matching and/or logistic regression modelling to account for differences in the types of patients that do or do not receive aprotinin often demonstrate strong associations between aprotinin and adverse outcomes in mixed populations where aprotinin is used in selected high-risk patients [711]. Conversely, when this bias is reduced in cohort studies where aprotinin use is non-selective, or where high bleeding risk cohorts are considered in isolation, the safety signal attributable to aprotinin is absent [2326].

The inclusion or omission of important confounders such as, for example, alternative haemostatic agents or other anti-fibrinolytics also produces conflicting results in observational cohort studies [27, 28]. We identified this as a potential source of confounding in the current study: alternative antifibrinolytics, rFVII, and the administration of blood components changed significantly following aprotinin withdrawal and may affect outcomes. In an exploratory analysis we established that the independent association between aprotinin and mortality was diminished following adjustment for volume of RBC transfusion, non-RBC blood components and alternative haemostatic strategies. This may be interpreted as showing that the adverse effects associated with aprotinin withdrawal are due to large-volume blood transfusions. However, this was not the primary aim of this study and we would caution against over-interpretation of this result.

Importantly, and unlike the results of RCTs and cohort studies, our real-world study suggested that the withdrawal of aprotinin was associated with increased complication rates and patient deaths following cardiac surgery. This apparent contradiction mandates a careful examination of the limitations of the current study. We attempted to minimise the bias attributable to the selected use of aprotinin by adopting a ‘before and after’ nested case-control design with propensity matching to select cohorts matched for key covariates and by performing two analyses; one that compared the entire cohort before and after aprotinin withdrawal, as well as a second comparing the high-risk cohort who received aprotinin before 2008 versus a matched cohort who we estimated would have received aprotinin after 2008. Bias will have been reduced by the fact that aprotinin was no longer available after 2008, by the use of a highly restrictive matching strategy that resulted in very similar patient groups in the before and after cohorts, and by the consistency of our observations in both mixed and high-risk populations. The chief limitation of this approach, as with all ‘before and after’ studies, is the high likelihood of time series bias. Mortality following elective coronary artery bypass grafting (CABG) in the UK has fallen from 2.2 to 0.8 % between 2005 and 2012 despite an increasingly high-risk case mix [6] and we have also reported improvements in clinical outcomes in our own unit over this period [29]. Time series bias in this case may have been expected to result in improvements in outcome in the later cohort, and indeed we did observe improvements in pulmonary and cardiac morbidity and resource use in this group consistent with historical trends. Importantly, however, the effect on bleeding and renal morbidity was in the opposite direction, and this result was not altered by our sensitivity analyses that utilised different historical cohorts, adjusted for simulated unmeasured confounders and used alternate matching strategies. The mixed positive and negative effects of aprotinin withdrawal argue against our results simply reflecting the increasingly high-risk and complex case mix in the latter cohort. Our findings mirror those of other ‘before and after’ studies from Europe [30], North America [31] and Asia [32, 33], where aprotinin withdrawal has increased bleeding complications, blood component exposure and perioperative morbidity. Importantly, our results also show an increase in mortality attributable to aprotinin withdrawal, an effect that was greater in the high-risk group. The possibility of time series bias notwithstanding, the mixed positive and negative effects of aprotinin withdrawal merit comment. Other ‘before and after’ studies have reported mixed changes in the frequencies of adverse events after aprotinin withdrawal; Von Heymann and colleagues documented reductions in stroke [30], whereas Martin et al. [34] reported lower rates of myocardial infarction in CABG patients, despite increased bleeding and transfusion post-aprotinin withdrawal. The trend towards increased mortality attributable to aprotinin in the BART trial has not been adequately explained and one possibility is that serine protease inhibition by aprotinin may have mixed adverse and beneficial clinical effects that are as yet poorly understood.

In conclusion, our study highlights the clinical importance of coagulopathic haemorrhage, and the need for improved, evidence-based blood management in cardiac surgery. Our study does not demonstrate a causal relationship between aprotinin withdrawal and adverse outcome. However, it demonstrates the disparity between frequently cited systematic reviews of randomised trials, the results of observational cohort studies that have considered the therapeutic role of aprotinin and our real-world experience of aprotinin withdrawal on patient outcomes. We suggest that this represents sufficient evidence of clinical equipoise to justify further evaluation of aprotinin in appropriately designed and powered multicentre RCTs.

Acknowledgments

This article presents independent research supported by the National Institute for Health Research (NIHR) under its Programme Grants for Applied Research Programme (grant reference no. RP-PG-0407-10384). The views expressed are those of the author(s) and not necessarily those of the NHS, the NIHR or the Department of Health. We thank Dr Neil Davies for his assistance with the statistical analyses.

Conflicts of interest

The authors declare that they have no conflict of interest.

Ethical standards

This retrospective observational case-control study was approved by the South West (UK) research ethics committee under reference 11/SW/0075.

Supplementary material

134_2013_3020_MOESM1_ESM.doc (340 kb)
Supplementary material 1 (DOC 340 kb)
134_2013_3020_MOESM2_ESM.doc (42 kb)
Supplementary material 2 (DOC 42 kb)
134_2013_3020_MOESM3_ESM.doc (286 kb)
Supplementary material 3 (DOC 285 kb)

Copyright information

© Springer-Verlag Berlin Heidelberg and ESICM 2013