SinoSCORE: a logistically derived additive prediction model for post-coronary artery bypass grafting in-hospital mortality in a Chinese population

Abstract

This study aims to construct a logistically derived additive score for predicting in-hospital mortality risk in Chinese patients undergoing coronary artery bypass surgery (CABG). Data from 9839 consecutive CABG patients in 43 Chinese centers were collected between 2007 and 2008 from the Chinese Coronary Artery Bypass Grafting Registry. This database was randomly divided into developmental and validation subsets (9:1). The data in the developmental dataset were used to develop the model using logistic regression. Calibration and discrimination characteristics were assessed using the validation dataset. Thresholds were defined for each model to distinguish different risk groups. After excluding 275 patients with incomplete information, the overall mortality rate of the remaining 9564 patients was 2.5%. The SinoSCORE model was constructed based on 11 variables: age, preoperative NYHA stage III or IV, chronic renal failure, extracardiac arteriopathy, chronic obstructive pulmonary disease, preoperative atrial fibrillation or flutter (within 2βweeks), left ventricular ejection fraction, other elective surgery, combined valve procedures, preoperative critical state, and BMI. In the developmental dataset, calibration using a Hosmer-Lemeshow (HL) test was at P =β0.44 and discrimination based on the area under the receiver operating characteristic curve (ROC) was 0.80. In the validation dataset, the HL test was at P =β0.34 and the area under the ROC (AUC) was 0.78. A logistically derived additive model for predicting in-hospital mortality among Chinese patients undergoing CABG was developed based on the most up-to-date multi-center data from China.

Keywords

coronary artery bypass grafting / risk stratification / in-hospital mortality

Introduction

Risk stratification is of great importance for cardiac surgery [1]. Since the 1980s, several risk models have been established worldwide. Most of these models are logistic or Bayesian, which require complex calculations, and some are proprietary [2-4]. Among these models, EuroSCORE [5,6] was proven successful because of its simplicity and openness. However, several recent studies have found that this widely adopted model overestimates mortality [7-9]. This result may be attributed to differences between Chinese and Western populations and the improved operative outcomes over the decade years. In recent years, the developed off-pump technique, myocardial protection technology, and the improved off-pump technique significantly reduce in-hospital mortality.

The incidence of coronary artery disease (CAD), known as the “epidemiological transition,” is greatly increasing in China [10]. Accordingly, the number of patients who need revascularization via coronary artery bypass surgery (CABG) dramatically increased [11]. This increase is due to the unavailability of a published operative risk model specifically developed for the increasing number of patients in China undergoing CABG. Creating a model is especially important because recent literature reported that EuroSCORE overestimates in-hospital mortality for this cohort of patients [12]. Consequently, we were unable to determine the actual risk of the patients using the EuroSCORE model, which will waste medical resources. This study aims to develop a logistically derived additive model for estimating post-CABG in-hospital mortality based on a large, multi-center Chinese database that accurately assesses the risk among Chinese bypass patients and guides clinical work.

Methods

The Chinese CABG Registry [13] is a national multi-center study that primarily aims to construct a new risk stratification system for Chinese patients undergoing CABG. Its steering group includes a number of Chinese cardiac surgeons and epidemiologists with interest in predicting surgical outcomes in the Fuwai Hospital. All patients (nβ=β9839) who underwent CABG at 43 participating institutions all over China between January 2007 and December 2008 were enrolled. Detailed information on patient demographics, preoperative risk factors, operative risk factors, and in-hospital mortality was collected. The data collected from the registry included 67 preoperative factors and 30 operative parameters (Table 1) that were selected based on our clinical experience and on prior published CABG risk-prediction models.

The Chinese CABG Registry database was carefully examined and patients<18βyears old and patients with incomplete data records were excluded. Among the patients, 1 was excluded for being<18βyears old and 274 were excluded because of incomplete data. Of the patients with incomplete data, 5 had no age, 70 had no body weight, 52 had no height, 205 had no ejection fractions, 1 did not have information on unstable angina, 2 had no information on cardiogenic shock, and 80 had no NYHA staging. Some overlaps in the data were also observed. A total of 275 patients (6 patients died) were excluded from the study, and 9564 patients were analyzed. The 9564 patients were randomly divided into two subsets with a 9:1 ratio according to method used to develop the EuroSCORE model [6]. These subsets were the developmental dataset (nβ=β8602), which was used to construct the risk model, and the validation subset (nβ=β962), which was used to test and validate the model. SPSS 13.0 (SPSS Inc., Chicago, Illinois) was used to perform the statistical analysis. This study was approved by the local IRB/EC (IRB/EC of Fuwai Hospital; approval number: 219).

In the developmental dataset, all the possible risk factors associated with early mortality were screened using chi-square tests for categorical covariates and t-tests or Wilcoxon rank sum tests for continuous covariates. Variables with P values>0.2 were excluded. A stepwise multivariate logistic regression was performed using the backward method (criteria for entry and removal of variables were 0.05 and 0.10) with one variable eliminated at a time. The stability of the model was checked every time a variable was eliminated. When all statistically non-significant variables were eliminated, a Hosmer-Lemeshow (HL) chi-square test was used to evaluate the calibration. The area under the ROC curve (AUC) was utilized to assess the discriminatory power. In the final score (that is, the SinoSCORE), the weights attributed to each variable were obtained from the logistic regression b coefficients. The multicollinearity of the models was checked using tolerance and inflation statistics. The calibration and discriminatory power were then tested in the validation dataset. This logistically derived additive model was then applied to the entire population (developmental and validation cohorts) divided into three risk levels.

Results

A total of 9839 patients from 43 surgical centers in 16 provinces in China participated in the project. After error checking and quality control procedures, 9564 patients were analyzed. Elective procedures were 97.1%, and isolated CABG accounted for 87.7%. Overall in-hospital mortality was 2.5%.

Patient characteristics

The mean age was 62.1βyears old, with a standard deviation of 9.2 and 77.3% were male. The mean BMI was 25.1β±β3.3 with 62.6%>24 and 1.1%<18. A history of hypertension occurred in 37.0% of patients. Diabetes was present in 29.8% of patients, with 5.7% on diet control, 17.5% on oral therapy, and 6.6% on insulin.

Construction of the SinoSCORE

Risk factor identification

Multivariate analysis identified 11 risk factors that are related to in-hospital mortality. These factors were classified into three groups, namely, patient related, cardiac, and operation related (Table 2). Multicollinearity diagnostics with tolerance statistics ruled out the multicollinearity of the 11 factors.

Risk model construction

Table 3 and Fig. 1 list the statistical features of the developmental and validation datasets. The validation analysis proved that the calibration of the model was satisfactory and the discrimination power was characterized by an AUC of 0.80. This system, which is similar to the EuroSCORE, is additive.

Risk model application

The SinoSCORE was applied to the whole database. The patients were divided into three risk groups according to the risk scores (Table 4). The observed and predicted mortalities overlapped in all risk groups.

Discussion

Necessity of developing a Chinese model

The incidence of CAD in China is increasing dramatically, with an estimated CAD incidence rate as high as 98 per 100 000 [10]. This rate might increase by>50% from 2010 to 2030 [14]. Although this incidence rate is lower compared with that in the US, considering the large population of China (more than 1300 million), the number of new CAD patients and patients who need revascularization via CABG in China is considerably high and rapidly increasing. The last decade has witnessed immense advances in surgical, anesthetic, perfusion procedures, and postoperative intensive care worldwide [11]. Considering most of the existing risk prediction models were developed more than a decade ago and are based on large populations in Europe and the US, developing a risk stratification system applicable for contemporary Chinese patients is highly appropriate. This necessity is further supported by three reasons. First, some of the existing risk models, such as the proprietary model of the Society of Thoracic Surgeons (STS) from the USA, are proprietary and unavailable to Chinese clinicians [15]. Second, open models such as EuroSCORE, which is based on European populations, has been proven unsuitable for Chinese CABG patients [12]. Third, most of the previous models are based on Western populations and the difference between Chinese and Western populations highlights the need to construct a model for Chinese patients. Comparison of the demographic characteristics of the study population based on the EuroSCORE multinational database revealed that 43.6% had hypertension and 16.7% had diabetes [16], whereas the SinoSCORE database indicated 37.0% had hypertension and 29.8% had diabetes.

This study, based on the largest most up-to-date multi-center database in China, is the first attempt to develop a risk stratification system for Chinese CABG patients. To our knowledge, this is the first risk stratification model for CABG patients in Asia, the largest continent and home to more than two-thirds of the world’s population. This model is worth validating in other Asian countries or among migrants of Asian origin in Western countries.

Logistically derived model

The SinoSCORE model incorporates 11 variables, which were mostly included in previous models [3,4,6]. However, our model differs with other existing models in at least two aspects. First, re-operation (“previous cardiac surgery” or “previous CABG”) was included for analysis (Table 1), but was not as a major risk factor, in the Chinese model. However, this factor was given a heavy weight in non-Chinese models. This result might be due to population differences and the fact that patients with previous cardiac surgery accounted for a very small proportion in our cohort. Previous cardiac surgery accounted for 1.1%, whereas previous CABG accounted for 0.3%. These low re-operation rates reflect the current situation in China (perhaps as well as in other developing countries where CAD is becoming increasingly prominent). The second difference is the high percentage of off-pump procedures (52.4%) in our cohort. Many of the existing models were constructed based on data more than a decade ago [3,6] during which the off-pump technique was uncommon. These models do not reflect the advances in surgical techniques during the past 10βyears, especially the introduction of off-pump surgery. Moreover, the percentage of off-pump CABG was much lower in Western populations. The 1998-2009 Nationwide Inpatient Sample databases from the USA identified 3 046 709 patients who underwent CABG procedures, 744 636 (24.4%) of whom were off-pump CABG [16]. Thus, our study constructed a model from a population with a majority of patients undergoing off-pump CABG, which contributes to existing literature.

The effects of off-pump procedures on operative mortality have been debated. One prospective randomized trial [17] demonstrated that off-pump CABG does not differ significantly from on-pump CABG in terms of 30-day composite outcome (death and major complications). The latest study “CORONARY,” the largest trial yet conducted, compared the performance of off-pump CABG with on-pump CABG. The study reported that off-pump CABG has a better safety and efficacy [18]. Our study indicated that off-pump CABG was not a risk factor for post-CABG in-hospital mortality.

Limitations

This study utilized in-hospital mortality as the major endpoint, whereas the EuroSCORE model used 30-day mortality as the endpoint [6]. The difference between the different models determined using this study might be due to differences in the endpoints. However, the use of in-hospital death as the endpoint is justified by the following reasons. First, other studies also used in-hospital mortality [19,20]. Second, the length of postoperative hospital stay is much longer in China than the US and Europe: the mean postoperative length of stay (PLOS) in our study population was 13.5βdays, whereas a study that utilized data (496 797 procedures from 1997 to 2001) from the STS database had a mean PLOS of 6.9βdays [21]. Third, out of hospital postoperative death in China within 30βdays is very unusual because of the lack of extended care and ventilator facilities compared with the US and Europe.

The results of this study were not validated using external data; thus, the SinoSCORE model should be applied to non-Chinese populations with great caution. Similarly, using models developed in the US and European populations may not be as accurate when applied to a contemporary Chinese population. Thus, our model may not be applicable to other populations and requires further validation.

This model is a logistically derived additive model. Given that categorization of continuous data might cause information loss, a logistic model that incorporates continuous variables without dichotomization would be more accurate. However, a considerable number of hospitals in China do not have access to the sophisticated information technology and need to utilize a logistic model. Thus, a logistically derived additive model that is easy to use for most Chinese doctors is preferred. This approach exhibits good discrimination and calibration of the current additive model. Considering the rapid development of the Chinese economy and the exponential growth of IT technology, we anticipated that a logistic model with better accuracy would be available in most Chinese centers in the future.

Conclusions

SinoSCORE, an 11-variable logistically derived additive model, was developed to predict in-hospital mortality among Chinese patients undergoing CABG. This model was based on the most recent multi-center data in China.

References

[1] HannanEL, SiuAL, KumarD, RaczM, PryorDB, ChassinMR. Assessment of coronary artery bypass graft surgery performance in New York.Is there a bias against taking high-risk patients? Med Care1997; 35(1): 49–56

[2] EdwardsFH, GroverFL, ShroyerAL, SchwartzM, BeroJ. The Society of Thoracic Surgeons National Cardiac Surgery Database: current risk assessment. Ann Thorac Surg1997; 63(3): 903–908

[3] ParsonnetV, DeanD, BernsteinAD. A method of uniform stratification of risk for evaluating the results of surgery in acquired adult heart disease. Circulation1989; 79(6 Pt 2 suppl I): I3–I12

[4] EagleKA, GuytonRA, DavidoffR, EwyGA, FongerJ, GardnerTJ, GottJP, HerrmannHC, MarlowRA, NugentW, O’ConnorGT, OrszulakTA, RieselbachRE, WintersWL, YusufS, GibbonsRJ, AlpertJS, GarsonA Jr, GregoratosG, RussellRO, RyanTJ, SmithSC Jr. ACC/AHA guidelines for coronary artery bypass graft surgery: executive summary and recommendations : A report of the American College of Cardiology/American Heart Association Task Force on Practice Guidelines (Committee to revise the 1991 guidelines for coronary artery bypass graft surgery). Circulation1999; 100(13): 1464–1480

[5] RoquesF, NashefSAM, MichelP, GauducheauE, de VincentiisC, BaudetE, CortinaJ, DavidM, FaichneyA, GabrielleF, GamsE, HarjulaA, JonesMT, PintorPP, SalamonR, ThulinL. Risk factors and outcome in European cardiac surgery: analysis of the EuroSCORE multinational database of 19030 patients. Eur J Cardiothorac Surg1999; 15(6): 816–822, discussion 822–823

[6] NashefSA, RoquesF, MichelP, GauducheauE, LemeshowS, SalamonR. European system for cardiac operative risk evaluation (EuroSCORE). Eur J Cardiothorac Surg1999; 16(1): 9–13

[7] MikkelsenMM, JohnsenSP, NielsenPH, JakobsenCJ. The EuroSCORE in western Denmark: a population-based study. J Cardiothorac Vasc Anesth2012; 26(2): 258–264

[8] AkarAR, KurtcepheM, SenerE, AlhanC, DurduS, KuntAG, GüvenirHA; Group for the Turkish Society of Cardiovascular Surgery and Turkish Ministry of Health. Validation of the EuroSCORE risk models in Turkish adult cardiac surgical population. Eur J Cardiothorac Surg2011; 40(3): 730–735

[9] ParolariA, PesceLL, TrezziM, LoardiC, KassemS, BrambillascaC, MiguelB, TremoliE, BiglioliP, AlamanniF. Performance of EuroSCORE in CABG and off-pump coronary artery bypass grafting: single institution experience and meta-analysis. Eur Heart J2009; 30(3): 297–304

[10] OkrainecK, BanerjeeDK, EisenbergMJ. Coronary artery disease in the developing world. Am Heart J2004; 148(1): 7–15

[11] HuSS. Coronary heart diseases. In: HuSS, KongLZ, editors. Annual report 2005 on cardiovascular diseases in China (English edition). Beijing: Encyclopedia of China Publishing House. 2007: 45–62

[12] ZhengZ, LiY, ZhangS, HuS; Chinese CABG Registry Study. The Chinese coronary artery bypass grafting registry study: how well does the EuroSCORE predict operative risk for Chinese population? Eur J Cardiothorac Surg2009; 35(1): 54–58

[13] LiY, ZhengZ, HuS; Chinese Coronary Artery Bypass Grafting Registry Study. The Chinese coronary artery bypass grafting registry study: analysis of the national multicentre database of 9248 patients. Heart2009; 95(14): 1140–1144

[14] MoranA, GuD, ZhaoD, CoxsonP, WangYC, ChenCS, LiuJ, ChengJ, Bibbins-DomingoK, ShenYM, HeJ, GoldmanL. Future cardiovascular disease in china: Markov model and risk factor scenario projections from the coronary heart disease policy model-china. Circ Cardiovasc Qual Outcomes2010; 3(3): 243–252

[15] EdwardsFH, GroverFL, ShroyerAL, SchwartzM, BeroJ. The Society of Thoracic Surgeons National Cardiac Surgery Database: current risk assessment. Ann Thorac Surg1997; 63(3): 903–908

[16] GopaldasRR, ChuD, CornwellLD, DaoTK, LeMaireSA, CoselliJS, BakaeenFG. Cirrhosis as a Moderator of Outcomes in CABG and OPCAB Operations: A 12-Year Population-Based Study. STS/AATS Tech-Con 2013 and STS 49th Annual Meeting. abstract

[17] ShroyerAL, GroverFL, HattlerB, CollinsJF, McDonaldGO, KozoraE, LuckeJC, BaltzJH, NovitzkyD; Veterans Affairs Randomized On/Off Bypass (ROOBY) Study Group. On-pump versus off-pump coronary-artery bypass surgery. N Engl J Med2009; 361(19): 1827–1837

[18] LamyA, DevereauxPJ, PrabhakaranD, HuS, PiegasLS, StrakaZ, PaolassoE, TaggartD, LanasF, AkarAR, JainA, NoiseuxN, OuY, ChrolaviciusS, NgJ, YusufS. Rationale and design of the coronary artery bypass grafting surgery off or on pump revascularization study: a large international randomized trial in cardiac surgery. Am Heart J2012; 163(1): 1–6

[19] HannanEL, WuC, BennettEV, CarlsonRE, CullifordAT, GoldJP, HigginsRS, IsomOW, SmithCR, JonesRH. Risk stratification of in-hospital mortality for coronary artery bypass graft surgery. J Am Coll Cardiol2006; 47(3): 661–668

[20] CarosellaVC, NaviaJL, Al-RuzzehS, GrancelliH, RodriguezW, CardenasC, BilbaoJ, NojekC. The first Latin-American risk stratification system for cardiac surgery: can be used as a graphic pocket-card score. Interact Cardiovasc Thorac Surg2009; 9(2): 203–208

[21] PetersonED, CoombsLP, FergusonTB, ShroyerAL, DeLongER, GroverFL, EdwardsFH. Hospital variability in length of stay after coronary artery bypass surgery: results from the Society of Thoracic Surgeon’s National Cardiac Database. Ann Thorac Surg2002; 74(2): 464–473

RIGHTS & PERMISSIONS

Higher Education Press and Springer-Verlag Berlin Heidelberg