**1. Introduction**

28 Coronary Interventions

[114] Rao SV, Kaul PR, Liao L, et al. Association between bleeding, blood transfusion, and

[115] Kugelmass AD, Cohen DJ, Brown PP, Simon AW, Becker ER, Culler SD. Hospital

[116] Pinto DS, Stone GW, Shi C, et al. Economic evaluation of bivalirudin with or without

[117] Cooper CJ, El-Shiekh RA, Cohen DJ, et al. Effect of transradial access on quality of life

[118] Roussanov O, Wilson SJ, Henley K, et al. Cost-effectiveness of the radial versus

[119] Amoroso G, Sarti M, Bellucci R, et al. Clinical and procedural predictors of nurse

[120] Rinfret S, Kennedy WA, Lachaine J, et al. Economic impact of same-day home

bolus-only abciximab regimen. JACC Cardiovasc Interv 2010;3:1011-9.

systematic radial access. Eur J Cardiovasc Nurs 2005;4:234-41.

from the REPLACE-2 trial. J Am Coll Cardiol 2004;44:1792-800.

coronary interventions. Am J Cardiol 2006;97:322-7.

Am Heart J 2008;155:369-74.

Cardiol 2008;52:1758-68.

1999;138:430-6.

2007;19:349-53.

glycoprotein IIB/IIIA inhibition for percutaneous coronary intervention: results

costs among patients with non-ST-segment elevation acute coronary syndromes.

resources consumed in treating complications associated with percutaneous

glycoprotein IIb/IIIa inhibition versus heparin with routine glycoprotein IIb/IIIa inhibition for early invasive management of acute coronary syndromes. J Am Coll

and cost of cardiac catheterization: A randomized comparison. Am Heart J

femoral artery approach to diagnostic cardiac catheterization. J Invasive Cardiol

workload during and after invasive coronary procedures: the potential benefit of a

discharge after uncomplicated transradial percutaneous coronary intervention and

Coronary artery disease remains highly prevalent in contemporary society and over 1000000 revascularisation procedures by percutaneous coronary intervention (PCI) are performed annually worldwide. PCI has seen significant improvement in clinical outcomes with the current generation of drug eluting stents. The role of PCI in multivessel coronary disease has been expanded with current trial evidence indicating equipoise between PCI and coronary artery by-pass surgery in selected groups.

Increasingly, coronary artery by-pass surgery (CABG) or PCI is being considered as an equivalent revascularisation strategy within the same patient population. Given the options available to clinicians and patients it is important to have robust tools to accurately compare the risk and benefits of selected strategies when making management decisions. Whilst these tools have been available to the cardiac surgeons for some time (Granton & Cheng, 2008), an equivalent tool for the interventional cardiologist has only recently been published.

The CANADA Score is a risk prediction model for determining 30 day mortality risk in patients undergoing elective, urgent and emergent PCI. Its development and validation will be discussed with reference to the established cardiac surgical risk calculators currently available.

### **2. Risk prediction models**

Risk prediction models are statistical models produced from patient databases using a combination of individual risk predication markers and are used by clinicians and patients for making treatment decisions. Model inaccuracy and ineffectiveness can therefore have negative implications on risk measurement and subsequent patient decisions and outcomes. The accuracy of the model is typically summarised in terms of the model's discrimination and calibration (Janes et al, 2008). The applicability of a risk model to a patient population is determined by validation.

#### **2.1 Discrimination**

Discrimination is the ability of the model to correctly classify outcomes (Nathanson & Higgins, 2008). The statistical measures of area under the curve (AUC) or concordance index (C-index) are commonly used to describe how well patients are classified within the model.

Percutaneous Coronary Intervention and 30-Day Mortality: The CANADA Score 31

surgical procedures performed in the United States. In the interval 1997-1999 there were 503 478 CABG-only procedures identified from 495 participating centres. From this, 30 potential risk factors for mortality were identified on univariate screening (Table 1). The over-all 30 day mortality rate was 3.05%. Using multivariate logistic regression an STS Score model was developed that had good discrimination (c-index 0.78) and modest calibration (H-L p =

A European multi-national database was established in 1995 (Nashef et al, 1999) and information on risk factors and mortality was collected for 19 030 consecutive adult patients undergoing cardiac surgery under cardiopulmonary bypass in 128 surgical centres in eight European states (Roques et al, 2003). Data were collected for 68 preoperative and 29 operative risk factors proven or believed to influence mortality. From this a series of objective risk factors (Table 1) were weighted by regression analysis and developed into an additive score (additive EuroSCORE) to predict mortality. Overall, 14 799 patients were divided into a developmental cohort (n= 13 302) and validation cohort (n= 1479). The 30 day mortality was for the entire cohort was 4.7%. The additive EuroSCORE had good discrimination in both the development (c-index 0.79) and validation (c-index 0.76) cohort, as well as good calibration (H-L p value <0.40 & <0.68 respectively) (Table 2). The additive EuroSCORE was further externally validated in a North American population. Despite demographic differences the model performed well with discriminatory c-index 0.75 and

excellent calibration (predicted and observed mortality 4.15%) (Nashef et al, 2002).

Age X X X Gender X X X Renal failure X X X a Critical preoperative state X X X

X X

Chronic lung disease X X PVD / CVD X X

Previous cardiac surgery X X

A limitation of the additive EuroSCORE was underestimation of risk in very high risk populations (Sergeant et al, 2001). A second model was published using the coefficient of the variables in the logistic regression data rather than additive weights to predict mortality. The logistic EuroSCORE had similar discrimination (c-index 0.785) to the additive model but superior accuracy in high risk populations. The models diverged at a predicted mortality of

**STS Score EuroSCORE CANADA Score** 

0.0016) (Table 2) (Shroyer et al, 2003).

**3.2 EuroSCORE** 

8-10% (Michel et al, 2003).

CVA / Neurological

dysfunction

*Patient Factors*

Multiple reoperations X

Patients typically are assigned a positive classification if the model predicts the probability of an outcome as >0.5. Conversely, a negative classification indicates the model predicts the probability of an outcome is <0.5. A patient is therefore correctly classified when an outcome event occurs in a patient with a positive classification, or when no event occurs in a patient with a negative classification. Sensitivity and specificity are derived from the fraction of correctly classified patients. The receiver operating characteristic (ROC) curve is derived from the plot of specificity against (1 – sensitivity), and the area under the ROC curve (AUC) measures the discriminatory ability of the model. An AUC of 0.5 indicates no discriminatory ability and 1.0 indicates perfect discrimination (Cook, 2008). A good model would have AUC 0.7 – 0.9.

#### **2.2 Calibration**

Calibration is determined by comparing the predicted and observed outcomes within subgroups of increasing risk within the dataset and applying the Hosmer-Lemeshow (H-L) statistical test (Hosmer & Lemeshow, 2000) to assess "goodness of fit". The H-L goodness of fit test divides subjects into deciles based on predicted probabilities and calculates a chisquare from observed and expected frequencies. If the H-L statistic has p-value >0.05, it implies there is no significant difference between observed and model-predicted values and therefore the model is well calibrated. Calibration plots can be used to give a graphical representation of model calibration.

#### **2.3 Validation**

Strategies to validate statistical models include (Altman & Royston, 2000):

a. Internal – evaluation from a single dataset

Internal validation refers to the application of a model to the same cohort from which it was derived, often by splitting one dataset into separate training and validation cohorts. This can be problematic as models tend to over fit the data and calibration appears erroneously good (Vickers & Cronin, 2010). Internal validation also does not address real differences that may exist between different cohorts. Methods to improve internal validation include cross validation and bootstrap resampling.


#### **3. Cardiac surgical models**

#### **3.1 STS score**

The Society of Thoracic Surgeons (STS) National Adult Cardiac Database was established in 1989 and currently contains over 4.5 million records. It represents >90% of adult cardiac surgical procedures performed in the United States. In the interval 1997-1999 there were 503 478 CABG-only procedures identified from 495 participating centres. From this, 30 potential risk factors for mortality were identified on univariate screening (Table 1). The over-all 30 day mortality rate was 3.05%. Using multivariate logistic regression an STS Score model was developed that had good discrimination (c-index 0.78) and modest calibration (H-L p = 0.0016) (Table 2) (Shroyer et al, 2003).

#### **3.2 EuroSCORE**

30 Coronary Interventions

Patients typically are assigned a positive classification if the model predicts the probability of an outcome as >0.5. Conversely, a negative classification indicates the model predicts the probability of an outcome is <0.5. A patient is therefore correctly classified when an outcome event occurs in a patient with a positive classification, or when no event occurs in a patient with a negative classification. Sensitivity and specificity are derived from the fraction of correctly classified patients. The receiver operating characteristic (ROC) curve is derived from the plot of specificity against (1 – sensitivity), and the area under the ROC curve (AUC) measures the discriminatory ability of the model. An AUC of 0.5 indicates no discriminatory ability and 1.0 indicates perfect discrimination (Cook, 2008). A good model would have

Calibration is determined by comparing the predicted and observed outcomes within subgroups of increasing risk within the dataset and applying the Hosmer-Lemeshow (H-L) statistical test (Hosmer & Lemeshow, 2000) to assess "goodness of fit". The H-L goodness of fit test divides subjects into deciles based on predicted probabilities and calculates a chisquare from observed and expected frequencies. If the H-L statistic has p-value >0.05, it implies there is no significant difference between observed and model-predicted values and therefore the model is well calibrated. Calibration plots can be used to give a graphical

Internal validation refers to the application of a model to the same cohort from which it was derived, often by splitting one dataset into separate training and validation cohorts. This can be problematic as models tend to over fit the data and calibration appears erroneously good (Vickers & Cronin, 2010). Internal validation also does not address real differences that may exist between different cohorts. Methods to improve internal

Temporal validation involves collecting data from the same sources but at a later time point. It is a prospective evaluation of the original model, but it may take considerable

External validation addresses the generalizability of the statistical model by application to a different population from that used to derive the model. It can be performed retrospectively making it attractive for widespread application and can help address

The Society of Thoracic Surgeons (STS) National Adult Cardiac Database was established in 1989 and currently contains over 4.5 million records. It represents >90% of adult cardiac

Strategies to validate statistical models include (Altman & Royston, 2000):

validation include cross validation and bootstrap resampling. b. Temporal – evaluation of a second dataset after the original cohort

c. External – evaluation of geographically separate cohorts

time to accrue an adequate number of events in the second dataset.

AUC 0.7 – 0.9.

**2.2 Calibration** 

**2.3 Validation** 

representation of model calibration.

a. Internal – evaluation from a single dataset

issues related to sample selection.

**3. Cardiac surgical models** 

**3.1 STS score** 

A European multi-national database was established in 1995 (Nashef et al, 1999) and information on risk factors and mortality was collected for 19 030 consecutive adult patients undergoing cardiac surgery under cardiopulmonary bypass in 128 surgical centres in eight European states (Roques et al, 2003). Data were collected for 68 preoperative and 29 operative risk factors proven or believed to influence mortality. From this a series of objective risk factors (Table 1) were weighted by regression analysis and developed into an additive score (additive EuroSCORE) to predict mortality. Overall, 14 799 patients were divided into a developmental cohort (n= 13 302) and validation cohort (n= 1479). The 30 day mortality was for the entire cohort was 4.7%. The additive EuroSCORE had good discrimination in both the development (c-index 0.79) and validation (c-index 0.76) cohort, as well as good calibration (H-L p value <0.40 & <0.68 respectively) (Table 2). The additive EuroSCORE was further externally validated in a North American population. Despite demographic differences the model performed well with discriminatory c-index 0.75 and excellent calibration (predicted and observed mortality 4.15%) (Nashef et al, 2002).

A limitation of the additive EuroSCORE was underestimation of risk in very high risk populations (Sergeant et al, 2001). A second model was published using the coefficient of the variables in the logistic regression data rather than additive weights to predict mortality. The logistic EuroSCORE had similar discrimination (c-index 0.785) to the additive model but superior accuracy in high risk populations. The models diverged at a predicted mortality of 8-10% (Michel et al, 2003).


Percutaneous Coronary Intervention and 30-Day Mortality: The CANADA Score 33

STS Score development 503478 3.05% 0.78 .0016

Logistic EuroSCORE 14799 4.7% 0.785 N/A

CANADA Score validation 6549 1.4% 0.91 0.12

Table 2. Comparison of Cohort and Model Performance for Various Risk Calculators.

facilitate an appropriate comparison of outcomes between these strategies.

Early models examining risk associated with percutaneous coronary interventions were well validated for the predication of in-hospital mortality ( Moscucci et al, 2001; Qureshi et al, 2003; Resnic et al, 2001; Shaw et al, 2002; Singh et al, 2002; Wu et al, 2006). However, these models had the potential to miss adverse events due to the nature of contemporary PCI where many patients are discharged within 24 hours of admission. As patients may be suitable for revascularization by either CABG or PCI it was important to develop a tool to

Contemporary risk scores were developed from the National Cardiovascular Data Registry (NCDR) (Peterson et al, 2010). The NCDR CathPCI Registry catalogues patient characteristics, angiographic and procedural details and in-hospital outcomes. From this, various risk models to predict in-hospital mortality were derived from pre-procedural and procedural data (full model), as well as a simplified model based on pre-procedure data only. To establish 30-day mortality, the NCDR records for patients aged over 65 were linked to claims data from the national Centers for Medicare and Medicaid Services (CMS). Linked data from 204111 patients observed in-hospital mortality as 1.99% and 30 day mortality as 2.94%. The c-index for predicating 30-day mortality using the NCDR in-patient mortality

The British Columbia Cardiac Registry (BCCR) is a population based database for all invasive cardiac procedures performed in British Columbia, Canada. The registry is used for clinical, administrative and research purposes. The linkage of BCCR data with the death registry of the British Columbia (BC) Vital statistics Agency facilitates outcome research. All procedures were performed at four academic tertiary centres that collectively perform 7500

Additive EuroSCORE

Additive EuroSCORE

CANADA Score training set

NCDR CathPCI Risk Score

**4.1 NCDR CathPCI risk score** 

development

(development)

**4. PCI models** 

model was 0.86.

PCI annually.

**5. The CANADA score** 

validation

subset

**Model Patients 30 day mortality C-Index H-L p-value** 

13302 4.7% 0.783 <0.40

1497 4.7% 0.76 <0.68

26350 1.5% 0.90 0.84

204111 2.94% 0.86 N/A


a, Dialysis; b, Contraindications to left ventricular contrast angiography include significant aortic stenosis (valve area <1.0 cm2); c, Critical pre-procedure state includes the anticipated need for IABP

Table 1. Comparison of Risk Factors Used to Predict 30-Day Mortality in Various Models.



Table 2. Comparison of Cohort and Model Performance for Various Risk Calculators.

### **4. PCI models**

32 Coronary Interventions

Aortic stenosis X X b LV dysfunction (LVEF) X X X NYHA IV X X X IABP X X X c LMS X X Triple vessel disease X X Current ACS X

Urgent status X X X

a, Dialysis; b, Contraindications to left ventricular contrast angiography include significant aortic stenosis (valve area <1.0 cm2); c, Critical pre-procedure state includes

Table 1. Comparison of Risk Factors Used to Predict 30-Day Mortality in Various Models.

Prior MI X X

Active endocarditis X Pulmonary hypertension X

Other than isolated CABG X Surgery on thoracic aorta X Post infarct septal rupture X

PTCA <6 hrs X Mitral insufficiency X

Ethnicity X BSA X Diabetes X Hypercholesterolemia X Hypertension X Immunosuppressive Rx X Smoker X

*Cardiac-Related Factors*

STEMI recurrent / on-

*Operation Related Factors*

the anticipated need for IABP

going

**STS Score EuroSCORE CANADA Score** 

X

Early models examining risk associated with percutaneous coronary interventions were well validated for the predication of in-hospital mortality ( Moscucci et al, 2001; Qureshi et al, 2003; Resnic et al, 2001; Shaw et al, 2002; Singh et al, 2002; Wu et al, 2006). However, these models had the potential to miss adverse events due to the nature of contemporary PCI where many patients are discharged within 24 hours of admission. As patients may be suitable for revascularization by either CABG or PCI it was important to develop a tool to facilitate an appropriate comparison of outcomes between these strategies.

#### **4.1 NCDR CathPCI risk score**

Contemporary risk scores were developed from the National Cardiovascular Data Registry (NCDR) (Peterson et al, 2010). The NCDR CathPCI Registry catalogues patient characteristics, angiographic and procedural details and in-hospital outcomes. From this, various risk models to predict in-hospital mortality were derived from pre-procedural and procedural data (full model), as well as a simplified model based on pre-procedure data only. To establish 30-day mortality, the NCDR records for patients aged over 65 were linked to claims data from the national Centers for Medicare and Medicaid Services (CMS). Linked data from 204111 patients observed in-hospital mortality as 1.99% and 30 day mortality as 2.94%. The c-index for predicating 30-day mortality using the NCDR in-patient mortality model was 0.86.

#### **5. The CANADA score**

The British Columbia Cardiac Registry (BCCR) is a population based database for all invasive cardiac procedures performed in British Columbia, Canada. The registry is used for clinical, administrative and research purposes. The linkage of BCCR data with the death registry of the British Columbia (BC) Vital statistics Agency facilitates outcome research. All procedures were performed at four academic tertiary centres that collectively perform 7500 PCI annually.

Percutaneous Coronary Intervention and 30-Day Mortality: The CANADA Score 35

**c = 0.902** 

Fig. 2. Receiver operator characteristic (ROC) curve for CANADA Score training set.

Fig. 3. Receiver operator characteristic (ROC) curve for CANADA Score validation set.

and Massachusetts, United States of America.

**Sensitivity** 

**Sensitivity** 

was 0.88 (Kurana et al, 2010a).

cohort was 0.87 (Kurana et al, 2010b).

The CANADA Score was further externally validated using data sets from Alberta, Canada,

**c = 0.911** 

**Specificity** 

The Alberta Provincial Project for outcome Assessment in Coronary Heart Disease (APPROACH) began 15 years ago as a cardiac catheterization and cardiac surgery database. The database includes clinical information on 126,500 patients with diagnostic cardiac catheterization and/or revascularization procedures, and 34000 hospital admissions for ACS (www.approach.org). The Canada Score model was evaluated against 9483 PCI performed from April 2005 – March 2008. The 30-day mortality was 1.8%. The c-index for this cohort

The Massachusetts Department of Public Health began collecting patient specific outcome data to evaluate all cardiac surgery and coronary intervention programs in 2002. Data were submitted to the Massachusetts Data Analysis Center (Mass-DAC) with data collection for coronary interventions using the ACC-NCDR Instrument beginning in 2003 (www.massdac.org). During the period Jan 2005 – September 2007, 36 341 PCI procedures were performed. The 30-day mortality was 2.05% and c-index for the Canada Score in this

All patients who had PCI performed in British Columbia (BC) from 2000 – 2005 who were BC residents were included in the study (Hamburger et al, 2009). PCI was defined as any coronary artery procedure that included balloon angioplasty, stent implantation, atherectomy, brachytherapy and thrombectomy. Second or subsequent PCI were not included for further analysis. All cause mortality data was obtained from the BC Vital Statistics Agency.

The study cohort was divided into two groups. Procedures between January 01, 2000 and December 31, 2004 formed the training set that was used to develop the multivariable predictive model for all-cause 30-day mortality. Procedures from 2005 were used to validate the model.

Variables for predicting 30-day mortality post PCI included patient demographics, comorbidities and clinical features such as indication for procedure and disease anatomy. Variables that were significantly associated with 30-day mortality in the univariate analysis (Table 3) or that were considered to be clinically important predictors for 30-day mortality were assessed in a stepwise logistic regression analysis. Only significant predictors (*P* < 0.05) in the multiple logistic regression analysis were kept in the final predictive model (Table 4).

A total of 32 899 procedures were performed. These were divided into 26 350 in the training set and 6549 in the validation set. The overall 30-day mortality was 1.5% (n=500), with mortality in the training set 1.5% (n=406) and validation set 1.4% (n=94) respectively. Of note, overall approximately one third of deaths occurred beyond 7 days (161/500, 32.2%) with similar proportions in the training (121/406, 29.8%) and validation sets (40/94, 42.6%) (Figure 1). The discrimination of the CANADA Score was good with the c-index for the training set 0.90 (Figure 2) and 0.91 for the validation set (Figure 3). The calibration was also good (H-L p values 0.84 & 0.12 respectively) (Table 2).

Fig. 1. Cumulative mortality versus time for all deaths in CANADA Score study population.

All patients who had PCI performed in British Columbia (BC) from 2000 – 2005 who were BC residents were included in the study (Hamburger et al, 2009). PCI was defined as any coronary artery procedure that included balloon angioplasty, stent implantation, atherectomy, brachytherapy and thrombectomy. Second or subsequent PCI were not included for further analysis. All cause mortality data was obtained from the BC Vital

The study cohort was divided into two groups. Procedures between January 01, 2000 and December 31, 2004 formed the training set that was used to develop the multivariable predictive model for all-cause 30-day mortality. Procedures from 2005 were used to validate

Variables for predicting 30-day mortality post PCI included patient demographics, comorbidities and clinical features such as indication for procedure and disease anatomy. Variables that were significantly associated with 30-day mortality in the univariate analysis (Table 3) or that were considered to be clinically important predictors for 30-day mortality were assessed in a stepwise logistic regression analysis. Only significant predictors (*P* < 0.05) in the multiple logistic regression analysis were kept in the final predictive model

A total of 32 899 procedures were performed. These were divided into 26 350 in the training set and 6549 in the validation set. The overall 30-day mortality was 1.5% (n=500), with mortality in the training set 1.5% (n=406) and validation set 1.4% (n=94) respectively. Of note, overall approximately one third of deaths occurred beyond 7 days (161/500, 32.2%) with similar proportions in the training (121/406, 29.8%) and validation sets (40/94, 42.6%) (Figure 1). The discrimination of the CANADA Score was good with the c-index for the training set 0.90 (Figure 2) and 0.91 for the validation set (Figure 3). The calibration was also

Fig. 1. Cumulative mortality versus time for all deaths in CANADA Score study population.

**Days from PCI** 

good (H-L p values 0.84 & 0.12 respectively) (Table 2).

**Cumulative 30-day Death (%)** 

Statistics Agency.

the model.

(Table 4).

Fig. 2. Receiver operator characteristic (ROC) curve for CANADA Score training set.

Fig. 3. Receiver operator characteristic (ROC) curve for CANADA Score validation set.

The CANADA Score was further externally validated using data sets from Alberta, Canada, and Massachusetts, United States of America.

The Alberta Provincial Project for outcome Assessment in Coronary Heart Disease (APPROACH) began 15 years ago as a cardiac catheterization and cardiac surgery database. The database includes clinical information on 126,500 patients with diagnostic cardiac catheterization and/or revascularization procedures, and 34000 hospital admissions for ACS (www.approach.org). The Canada Score model was evaluated against 9483 PCI performed from April 2005 – March 2008. The 30-day mortality was 1.8%. The c-index for this cohort was 0.88 (Kurana et al, 2010a).

The Massachusetts Department of Public Health began collecting patient specific outcome data to evaluate all cardiac surgery and coronary intervention programs in 2002. Data were submitted to the Massachusetts Data Analysis Center (Mass-DAC) with data collection for coronary interventions using the ACC-NCDR Instrument beginning in 2003 (www.massdac.org). During the period Jan 2005 – September 2007, 36 341 PCI procedures were performed. The 30-day mortality was 2.05% and c-index for the Canada Score in this cohort was 0.87 (Kurana et al, 2010b).

Percutaneous Coronary Intervention and 30-Day Mortality: The CANADA Score 37

Yes 167 56 33.53 37.33 26.62–52.35 <0.001

Yes 186 83 44.62 64.64 47.44–88.07 <0.001

Yes 285 82 28.77 32.09 24.28–42.42 <0.001

Yes 474 145 30.59 43.26 34.36–54.47 <0.001

Yes 931 30 3.22 2.22 1.52–3.24 <0.001

Yes 1,865 48 2.57 1.78 1.31–2.42 <0.001

Yes 604 37 6.13 4.57 3.23–6.47 <0.001

Yes 5,924 115 1.94 1.40 1.13–1.75 0.003

Yes 14,226 191 1.34 0.78 0.64–0.95 0.013

Yes 15,345 132 0.86 0.35 0.28–0.43 <0.001

**30-Day mortality rate (%)** 

**Odds ratio** 

**95% Confidence intervals** 

*P*

**Number of deaths**

No 26,168 349 1.33 1.00 Ref.

No 26,149 322 1.23 1.00 Ref.

No 26,065 324 1.24 1.00 Ref.

No 25,876 261 1.01 1.00 Ref.

No 25,419 376 1.48 1.00 Ref.

No 24,485 358 1.46 1.00 Ref.

No 25,641 361 1.41 1.00 Ref.

No 20,324 283 1.39 1.00 Ref.

No 12,022 207 1.72 1.00 Ref.

No 10,903 266 2.44 1.00 Ref.

**Category Number of** 

Hemodynamically unstable prior to procedure

Cardiogenic shock

for IABP

Critical preprocedural

Treated with IIb/IIIa inhibitor preprocedure

Lytic therapy preprocedure

Ongoing dialysis

Hypertension

Diabetes mellitus

Hyperlipidaemia

or serum creatinine >200 μmol/L

state

Anticipated need

**patients** 



**30-Day mortality rate (%)** 

**Odds ratio** 

26,350 1.05 1.04–1.06 <0.001

**95% Confidence intervals** 

*P*

**Number of deaths**

Male 19,283 265 1.37 1.00 Ref.

Non-urgent 7,626 219 0.18 1.00 Ref.

vessel disease 17,420 176 1.01 1.00 Ref.

>50% 14,521 50 0.34 1.00 Ref.

No 7,395 17 0.23 1.00 Ref.

No 25,570 311 1.22 1.00 Ref.

Stable Angina 8,426 18 0.21 1.00 Ref.

Female 7,064 141 2.00 1.46 1.19–1.80 <0.001

 Emergency 2,965 171 7.39 43.36 25.22–74.56 <0.001 Urgent 15,643 14 1.09 6.01 3.48–10.37 <0.001

disease 7,830 164 2.09 2.68 2.03–3.54 <0.001 Left main disease 1,100 66 6.00 7.99 5.70–11.22 <0.001

 <30% 783 48 6.13 18.90 12.63–28.28 <0.001 30%–50% 5,488 82 1.49 4.39 3.08–6.25 <0.001

Contraindicated 3,989 197 4.94 15.04 11.00–20.55 <0.001

Yes 16,364 361 2.21 9.79 6.02–15.94 <0.001

Yes 776 95 12.24 11.33 8.89–14.43 <0.0001

 STEMI Ongoing 2,253 168 7.46 37.64 23.09–61.34 <0.001 STEMI Recurrent 1,216 23 1.89 9.01 4.85–16.74 <0.001 Other ACS 14,440 196 1.36 5.75 3.13–10.55 <0.001

**Category Number of** 

patients 26,350 406

Total number of

Age (continuous) Mean ± SD: 64.3 ±

11.5

Gender

Urgency of procedure

Extent of coronary artery

disease

Single or two

Triple-vessel

Left ventricular ejection fraction

Clinically

CCS class IV angina

NYHA dyspnoea ≥ 3 or congestive heart failure

Indication for procedure

**patients** 

Percutaneous Coronary Intervention and 30-Day Mortality: The CANADA Score 39

Age (per 10 year increase) 0.39 1.48 1.32 1.65 Gender 0.23 1.26 0.98 1.61 Emergency 0.95 2.58 1.87 3.57 Left main disease 1.09 2.98 2.06 4.29 Triple-vessel disease 0.45 1.57 1.22 2.02 LVEF < 30% 1.84 6.27 4.02 9.77 LVEF 30–50% 0.86 2.36 1.63 3.39

Contraindicated 1.55 4.71 3.33 6.66 NYHA ≥ 3/CHF 0.82 2.26 1.65 3.10 Critical preprocedural state 1.97 7.20 5.33 9.74 STEMI Ongoing 2.00 7.40 4.07 13.46 STEMI Recurrent 1.43 4.19 2.08 8.43 Other ACS 1.35 3.87 2.30 6.53

<sup>μ</sup>mol/L 0.76 2.13 1.40 3.23

**6. Comparison of risks – Surgical versus percutaneous revascularisation** 

Increasingly it has become relevant to select the optimal revascularization strategy for patients deemed appropriate for revascularisation by either coronary artery bypass grafting or percutaneous intervention. Published studies of coronary anatomy alone have shown to predict the need for future revascularisation (Serruys et al, 2009; Sianos et al, 2005), but not mortality. The scores derived from an anatomical assessment alone has been shown to have only modest correlation to predicted risk using either surgical (logistic EuroSCORE) or percutaneous (CANADA Score) risk calculators that incorporate clinical and anatomical factors (Hoole & Hamburger, 2011). The same study found comparative risk assessment using either logistic EuroSCORE or CANADA Score has good correlation (R=0.80) and importantly recognised that patients with high predicted risk for surgery may have higher risk for a percutaneous revascularisation strategy (Figure 4). This implies patients declined for surgery should not necessarily default to a PCI treatment strategy. It must be noted that the definition of risk factors in different models may vary and must be considered when

failure; STEMI, ST-elevation myocardial infarction; ACS, acute coronary syndrome; Critical preprocedural state, Hemodynamically unstable prior to procedure or Cardiogenic shock or

Table 4. Predictors of Mortality in the CANADA Score Multivariable Model.

applying multiple models to individual patients (Table 5).

LVEF, Left ventricular ejection fraction; NYHA, New York Heart Association; CHF, Congestive heart

Intercept −9.89

LVEF Clinically

Dialysis/Creatinine > 200

Anticipated need for IABP.

**β Coefficient Adjusted OR 95% Confidence intervals**


Table 3. Baseline Variables in CANADA Score Univariate Analysis.

Yes 2,056 48 2.33 1.63 1.20–2.21 0.002

Yes 1,815 55 3.03 2.20 1.64–2.93 <0.001

Yes 4,542 55 1.21 0.54 0.40–0.73 <0.001

months 11,569 118 1.02 0.45 0.36–0.57 <0.001

Yes 7,804 124 1.59 1.07 0.87–1.33 0.531

Yes 6,674 86 1.29 0.80 0.63–1.02 0.075

Yes 3,132 53 1.69 1.14 0.85–1.52 0.3901

Yes 2,013 50 2.48 1.75 1.30–2.36 <0.001

Yes 1,325 30 2.26 1.55 1.06–2.25 0.023

Yes 1,491 26 1.74 1.16 0.78–1.74 0.460

**30-Day mortality rate (%)** 

**Odds ratio** 

**95% Confidence intervals** 

*P*

**Number of deaths**

No 24,192 350 1.45 1.00 Ref.

No 24,433 343 1.40 1.00 Ref.

No 10,137 225 2.22 1.00 Ref.

No 18,444 274 1.49 1.00 Ref.

No 19,588 313 1.60 1.00 Ref.

No 23,121 345 1.49 1.00 Ref.

No 24,235 348 1.44 1.00 Ref.

No 24,923 368 1.48 1.00 Ref.

No 24,757 372 1.50 1.00 Ref.

Table 3. Baseline Variables in CANADA Score Univariate Analysis.

**Category Number of** 

Peripheral vascular disease

disease

Cerebrovascular

Exsmoker > 3

Previous PCI

Previous CABG

History of chronic pulmonary disease requiring treatment

Potentially lifelimiting

hepatobiliary or gastrointestinal

disease

Diagnosis of malignancy

Previous myocardial infarction

Cigarette smoker

**patients** 


LVEF, Left ventricular ejection fraction; NYHA, New York Heart Association; CHF, Congestive heart failure; STEMI, ST-elevation myocardial infarction; ACS, acute coronary syndrome; Critical preprocedural state, Hemodynamically unstable prior to procedure or Cardiogenic shock or Anticipated need for IABP.

Table 4. Predictors of Mortality in the CANADA Score Multivariable Model.

#### **6. Comparison of risks – Surgical versus percutaneous revascularisation**

Increasingly it has become relevant to select the optimal revascularization strategy for patients deemed appropriate for revascularisation by either coronary artery bypass grafting or percutaneous intervention. Published studies of coronary anatomy alone have shown to predict the need for future revascularisation (Serruys et al, 2009; Sianos et al, 2005), but not mortality. The scores derived from an anatomical assessment alone has been shown to have only modest correlation to predicted risk using either surgical (logistic EuroSCORE) or percutaneous (CANADA Score) risk calculators that incorporate clinical and anatomical factors (Hoole & Hamburger, 2011). The same study found comparative risk assessment using either logistic EuroSCORE or CANADA Score has good correlation (R=0.80) and importantly recognised that patients with high predicted risk for surgery may have higher risk for a percutaneous revascularisation strategy (Figure 4). This implies patients declined for surgery should not necessarily default to a PCI treatment strategy. It must be noted that the definition of risk factors in different models may vary and must be considered when applying multiple models to individual patients (Table 5).

Percutaneous Coronary Intervention and 30-Day Mortality: The CANADA Score 41

**Risk Factor EuroSCORE STS Score CANADA Score** 

Patient has

occlusion

than 24 h

bypass

of surgery

pump, or patient required cardiopulmonary resuscitation within 1 h before the start of the operative procedure

> 200 mmol/l preoperatively

A central neurologic deficit persisting more

Prior cardiac surgical operation(s) with or without the use of cardiopulmonary

Patient currently under antibiotic treatment for endocarditis at the time

Any one or more of the following: sustained ventricular tachycardia or ventricular fibrillation requiring cardioversion and/or IV amiodarone, preoperative inotropic support, preoperative intra-aortic balloon

> 200 umol/l or

Cardiogenic shock (a systolic blood pressure of <90 mmHg for at least 30 min and/or the

dialysis

need for supportive measures to maintain a systolic blood pressure of >90 mmHg, clinical evidence of end-

organ

hypoperfusion) or hemodynamic

Neurological dysfunction disease

Previous cardiac surgery

Serum creatinine / renal

Active endocarditis

Critical preoperative

state

insufficiency

Severely affecting

functioning

pericardium

> 200 mmol/l preoperatively

Patient still under antibiotic treatment for endocarditis at the time of

Any one of more of the

ventricular tachycardia or fibrillation or aborted sudden death, preoperative cardiac massage, preoperative ventilation before arrival

surgery

following:

in the

anaesthetic room, preoperative

aortic balloon counterpulsation or preoperative acute renal

inotropic support, intra-

ambulation or day to day

Requiring opening of the

of the following:

cerebrovascular disease, documented by any one

Unresponsive coma >24 h; CVA (symptoms >72 h after onset); RIND (recovery within 72 h); TIA (recovery within 24 h); or noninvasive carotid test with >75%

Fig. 4. Comparison of predicted PCI (CANADA Score) and CABG (EuroSCORE) mortality relative to tertiles of subject's anatomical risk (SYNTAX Score).


Fig. 4. Comparison of predicted PCI (CANADA Score) and CABG (EuroSCORE) mortality

**Risk Factor EuroSCORE STS Score CANADA Score** 

Sex Female Female Female

Per 5 years or part thereof over 60 years

Patient required pharmacologic therapy for the treatment of chronic pulmonary compromise, or patient has a FEV1 <75% of predicted value

Patient has peripheral vascular disease as indicated by

claudication either with

exertion or rest; amputation for arterial insufficiency; aorto-iliac occlusive disease reconstruction; peripheral vascular bypass surgery, angioplasty or stent; documented AAA, AAA repair, or stent; positive non-invasive testing documented – or –

Per 10 year increase

relative to tertiles of subject's anatomical risk (SYNTAX Score).

Age Per 5 years or part thereof over 60 years

> Long-term use of bronchodilators or steroids for lung disease

Any one or more of the

claudication, carotid occlusion or >50% stenosis, previous or planned intervention on the abdominal aorta, limb arteries or carotids

following:

Chronic pulmonary disease

Extracardiac arteriopathy


Percutaneous Coronary Intervention and 30-Day Mortality: The CANADA Score 43

**Risk Factor EuroSCORE STS Score CANADA Score** 

Major cardiac procedure other than or in addition

For disorder of ascending, arch or descending aorta

grossly elevated left ventricular end diastolic pressure (>30 mmHg).

Table 5. Definition of Risk Factors for Various Risk Calculators.

to CABG

Other than isolated CABG

Surgery on thoracic aorta

Post-infarct septal rupture

Left main disease

**7. Conclusion** 

(either of the following):

circulatory support; or (2) shock without circulatory support. *Salvage: The patient is undergoing CPR en route to the OR or prior to anaesthesia induction*

Any valve procedure in addition to or separate

aneurysm/dissection

Ventricular septal defect

>= 50% compromise of vessel diameter preoperatively

lesion in any three

Greater than 50%

vessels

lesion

(1) shock with

from CABG

Aortic

repair

3 vessel disease Greater than 50%

NYHA ≥ 3

\*LVEF contraindicated = comorbid conditions preclude left ventricular contrast angiography (significant aortic stenosis (valve area <1.0 cm2), presence of aortic valve prosthesis, impaired renal function (serum creatinine >200umol/L), a critical preprocedural clinical state, NYHA IV dyspnoea,

Predicting procedural risk enables the correct treatment decisions to be made and allows valid informed consent and accurate patient counselling. This is particularly important as PCI has become accepted as a viable alternative to established surgical intervention. Early assessment of risk with PCI was limited to short term events that ignored important late events and prevented direct comparison with surgical risk predication tools. The CANADA Score was developed to accurately predict 30 day mortality risk and has been externally validated in large North American cohorts demonstrating broad applicability to varied patient groups. The CANADA Score confirms that both anatomical and clinical data are required to provide accurate and discriminatory 30 day mortality risk prediction and it therefore allows comparison with well validated surgical risk prediction models to guide optimal revascularisation strategy. Application of the CANADA Score to patients with high surgical risk demonstrates the potential for equal or greater risk with a percutaneous



grossly elevated left ventricular end diastolic pressure (>30 mmHg).

Table 5. Definition of Risk Factors for Various Risk Calculators.
