Development of an Iranian CPB Based Risk Stratification Score Model, an Iranian Risk Stratification Model

1. Background

Despite improvements in technology and surgeons’ experience, open-heart surgery still portends a risk of mortality and morbidity (1), therefore investigators have tried to decrease complications by appropriate selection of patients.

Operative mortality represents an indicator of cardiac surgery quality (2). Interest to estimate operative mortality, has led to designation of several predictive models (3). Most of the models are multifactorial including preoperative information, operation data and 30-days post outcomes. One of the first scoring algorithms formulated by Parsonnet in 1989 (4) and thereafter more risk score calculators developed.

Difference between institutions and geographic areas necessitates local risk models (1). The most populated models include Euro SCORE, Parsonnet, 2000 Bernstein-Parsonnet (BP) estimation score, the Society of Thoracic Surgeons (STS) algorithm and United Kingdom score models (5-8).

On the other hands, off-pump coronary artery bypass grafting (OPCAB) is gaining world wide acceptance and the number of OPCAB procedures is increasing (9). Several studies have evaluated the validity of populated mortality score models in OPCAB patients (9, 10). Cardiopulmonary bypass (CPB) technique is a paramount factor that can affect operative mortality but no model has inserted this risk factor in multivariate analysis. Conversion of OPCAB to On-pump coronary artery bypass grafting (ONCAB) deserves attention indeed.

Respect to wide spread usage of OPCAB technique, the aim of this study is to designate a mortality score model including CPB technique and conversion to ONCAB as a new factor and its comparison with Parsonnet and Euro SCORE model because of their popularity.

2. Methods

2.1. Patients’ Populations

The study has been performed in the cardiothoracic surgery department of Razavi hospital. 1920 consecutive patients who underwent emergent and elective cardiac surgery from April 2009 to March 2011 entered. Patients’ data were collected while in hospital according to a comprehensive database including 445 variables and stored in a Razavi adult cardiac surgery database.

2.2. Data Collection

The most relevant 81 patients related variables are depicted in Table 1 combined with 39 items derived from Parsonnet, and Euro SCORE variables were imported into the statistical software (Table 1). Before calculation missing data were excluded from database.

2.3. CPB Dependent Models

Our model named as Iranian model was designed by entering variable relevant to CBP into standard logistic Euro SCORE and Parsonnet model. Mean post operative and 30 days mortality was calculated by each model and the actual mortality was compared with each other.

2.4. Statistical Analysis

Since the utilized models are very complicated and large in size and in order to get better results, the R software and relative packages were applied. To check the ability of entering variables, we used the cross tabulation tables, Chisquare and Fisher’s exact test. If any significant correlation between mortality and the explanatory variables exist at the level of two percent (0.2), variable was inserted. The best model to fit the data is Binomial Logistic Regression. After finding the best model, we checked the adequacy of the final models.

Discrimination can be assessed by the area under the receiver operative characteristic curve (ROC). The ROC area can be interpreted as the probability that a patient who died had a higher risk score than a patient who survived. Thus the area under the curve is the percentage of randomly drawn pairs for which this is true. This is a fairly subjective measure and values greater than 0.8 usually indicate potentially useful discrimination. A value of 0.5 indicates random predictions.

The AIC statistic (2(log-likelihood) + 2(number of parameters in the model)) increases with an increasing number of coefficients but decreases when a better adaptability to data is achieved. It represents the measure of how much a specific model is suitable to describe the study phenomenon and is a function of the model’s residual variance (prediction error): the less the variance the more the accuracy. According to Akaike, the model exhibiting the smallest AIC value is the model providing the most information on the study sample (11).

3. Results

3.1. Patients’ Variables

The mean age of the 1920 patients was 59 ± 12.5 years, 69.1% were men.

Coronary artery bypass grafting (CABG) consisted 90.7% of surgeries along with 2.8 ± 1.6 grafts in each patient, compared to 9.9 % valve surgery and 3.8 % aortic surgery respectively. 78.7 % of operations were using off – pump technique and in 15 (0.8%) the operation was converted to on- pump technique.

Coronary patients suffered from 1.96% overall and 0.44% off pump mortality (Table 1).

3.2. Euro SCORE and Parsonnet Variables

After excluding missing data, 936 patients entered. Additive and logistic scores were evaluated (Table 2). EuroSCORE estimated mortality 8.4 ± 10.8 by logistic model. Parsonnet additive model estimated mortality 6.2 ± 9.98. These two models overestimate mortality in comparison to 2.3 % in our patients.

3.3. Euro SCORE Model (with or without CPB)

CPB as a variable was inserted to this model and analyzed again. The P value of these two models were < 2e-16. Results manifested that CPB decreased overestimation of the model and better estimated mortality rate. Mortality of Euro SCORE including CPB or not were 0.6267039 and 0.5056874 (Table 3).

3.4. Parsonnet Model (with or without CPB)

Parsonnet model was recalibrated by entering CPB and analyzed again. The p value of these two models were < 2e-16. Overestimation of mortality of Standard Parsonnet and CPB dependent Parsonnet models were 0.6483348 and 0.5271963 (Table 4).

3.5. CPB Dependent Risk Prediction Model (Iranian Model)

Firstly all variables of 1920 patients were analyzed by SPSS 19 software and the items with P < 0.2 were selected and entered while others were excluded. New items include using prophylactic or therapeutic drugs like ASA, Digoxin and oral hypoglycemic drugs, laboratory data like INR and Potassium, number of coronary grafts, pericardial effusion and utilization of cardiopulmonary bypass.

Then significant variables were entered into R software to heighten sensitivity and multivariable analysis was performed. P value of model was < 2e-16 and over estimation was considered 0.4375621 (Table 5).

3.6. Comparison Between Models

Analysis revealed that P value of all models were < 2e-16, however Iranian model possessed the highest accuracy for mortality evaluation and the lowest overestimation of mortality (0.4375621%), followed by Euro SCORE including CPB (0.5056874%), Parsonnet with CPB (0.5271963%), Euro SCORE (0.6267039%) and Parsonnet (0.6483348%) (Table 6).

Results showed that new version of Parsonnet model with new calibration and β-coefficients of variables has the least number of mistakes in estimation of mortality and the lowest Akaike information criterion (AIC) score (-1246.7561), followed by Parsonnet with CPB (-1122.1585), Euro SCORE (-978.0662), Iranian model (-917.1253) and Euro SCORE with CPB (-889.2239).

Finally ROC curve of models showed better area under curve (AUC) for CPB dependent Parsonnet models (0.9841) in comparison with standard Parsonnet (0.9551), Iranian model (0.9537), CPB dependent Euro SCORE (0.9465) and Euro SCORE (0.8659) (Table 6).

Iranian model had the lowest overestimation in predicting mortality and recalibrated Parsonnet model had the highest AUC overall. The point is by entering CPB variable, AUC increased and overestimation of mortality decreased. For example in Euro SCORE AUC and overestimation of mortality changed from 0.8659 to 0.9465 and 0.6267 to 0.5056 respectively.

4. Discussion

Our main purpose was to evaluate cardiopulmonary bypass effect on 30-day postoperative mortality besides designing a new model for better estimation of mortality in OPCAB and ONCAB patients. Until now, no model with CPB variable is available.

OPCAB technique is gaining more popularity worldwide and surgeons’ experiences are increasing 11. However a question still exists. Which one is superior? On pump or off pump? (12). Some investigators tried some models for better estimation of mortality in these two groups. Hirose et al in 2010 assessed mortality in CABG group by Euro SCORE model (13). They concluded that Euro SCORE was not an appropriate risk stratification model for off pump patients and should be modified.

Parolari et al (14) in 2009 estimated postoperative mortality in 1140 OPCAB and 3440 ONCAB patients by additive and logistic Euro SCORE models and finally reported no significant difference between these two groups. ROCs of additive Euro SCORE were 0.808 and 0.779 in ONCAB and OPCAB whereas ROCs of Logistic Euro SCORE included 0.813 and 0.773 in ONCAB and OPCAB, respectively. Mortality overestimation was noticed in both models. Farrokhyar et al in 2007 estimated a good prediction of mortality in on and off pump by using society of thoracic surgeons (STS) and Euro SCORE models although CPB had not been evaluated (15). ROC curve of STS for off-pump and on-pump was 0.81 and 0.82 and by Euro SCORE was 0.79 and 0.81 respectively. Similarly Toumpoulis et al in 2004 evaluated Euro SCORE model in CABG patients and reported logistic and standard Euro SCORE model were strong predictor models in CABG patients (16).

Two clinical trials reported difference between ONCAB and OPCAB mortality and both study showed lower mortality and morbidity in OPCAB group. Calafiore et al in 2001 reported CPB as an independent risk factor for higher morbidity and mortality (17). Al-Ruzzeh et al in 2003 used mortality prediction model reported by the Society of Cardiothoracic Surgeons of Great Britain and Ireland (SCTS) and reported OPCAB group had a lower mortality in UK national database (10). Our findings accommodate with the last studies mentioned. OPCAB group mortality was lower than ONCAB patients (0.44% (3 of 674) Vs. 8.69 (10 of 115)). Possible explanation of higher ONCAB mortality rate is that patients are operated using OPCAB technique except those who could not tolerate and converted to ONCAB or another operation than CABG should be performed such as valve surgery plus CABG.

Because of demographic differences among countries specified prediction models should be applied. Euro SCORE model is a good mortality predictor in Europe and North America (14) however, it may overestimate postoperative risk and require recalibration in different countries. Youn et al. reported overestimation of prediction in Korea (observed mortality 1.3% Vs. Logistic and standard Euro SCORE prediction 4.5% and 5% respectively) (9). Yap et al in 2006 reported Euro SCORE as an inappropriate model in Australia and should be recalibrated (Observed 3.2% Vs. additive and logistic 5.31% and 8.76% respectively) (18), the same as in Denmark and Italy (19, 20). Parsonnet score is a simple prediction model but like Euro SCORE model it would not be suitable for many populations and should be recalibrated. Varennes et al in 2007 used Parsonnet score for prediction of mortality in Canadian patients and results showed overall mortality was 6.4 vs. model estimation which was 18.8 ± 13.7 (21).

Accordingly, our results depicted overall mortality was 2.3% and estimation for logistic Euro SCORE was 8.4 ± 10.86 and for Parsonnet score was 6.2 ± 9.98. Therefore recalibration was performed (Tables 3 and 4). Overestimation changed to 0.6267039 and 0.6483348 after modification of Euro Score and Parsonnet models respectively.

After considering CPB as a variable, results indicated significant decrease in overestimation (0.6267039 to 0.5056874 for Euro Score and 0.6483348 to 0.5271963 for Parsonnet). So CPB significantly accentuated the accuracy of mortality prediction.

Comparing other models, Iranian model includes the lowest overestimation (0.4375621). Regarding new variables, aortic surgery encompassing Bental operation (P value 3.41 × 10^-6), aortic dissection (P value 0.000362), aortic aneurysm (P value 0.002585) signifies higher early mortality. Pericardial effusion (P value 0.000108) along with CPB especially in emergent situations (P value 0.001224) significantly augments postoperative mortality. Conversely, consuming drugs preoperatively lessens early mortality irrespective of time duration.

Although our study was single centered and limited in respect to the number of patients, we developed a new risk prediction model of postoperative mortality named as Iranian model. By inserting CPB and other variables into existing models we claim that our model accommodates better with mortality rate. It has been justified with our demographic characteristics and has reduced overestimation of mortality comparing to Euro SCORE and Parsonnet.

In fact, this study suggests inserting CPB as a determinant variable in predicting mortality. In case of application of popular predictor models recalibration considering demographic characteristics seems necessary.