EuroSCORE Predicts Intensive Care Unit Stay and Costs of Open Heart Surgery

EuroSCORE Predicts Intensive Care Unit Stay and Costs of Open Heart Surgery Johan Nilsson, MD, Lars Algotsson, MD, PhD, Peter Höglund, MD, PhD, Carsten Lührs, MD, and Johan Brandt, MD, PhD Departments of Cardiothoracic Surgery and Cardiothoracic Anesthesiology, Heart and Lung Center, and Department of Clinical Pharmacology, University Hospital, Lund, Sweden Background. This study aimed to determine whether the preoperative risk stratification model EuroSCORE predicts the different components of resource utilization in open heart surgery. Methods. Data for all adult patients undergoing heart surgery at the University Hospital of Lund, Sweden, between 1999 and 2002 were prospectively collected. Costs were calculated for the surgery and intensive care and ward stay for each patient (excluding transplant cases and patients who died intraoperatively). Regression analysis was applied to evaluate the correlation between EuroSCORE and costs. The predictive accuracy for prolonged postoperative intensive care unit (ICU) stay was assessed by the Hosmer-Lemeshow goodnessof-fit test. The discriminatory power was evaluated by calculating the areas under receiver operating characteristics curves. Results. The study included 3,404 patients. The mean cost for the surgery was $7,300, in the ICU $3,746, and in the ward $3,500. Total cost was significantly correlated with EuroSCORE, with a correlation coefficient of 0.47 (p < 0.0001); the correlation coefficient was 0.31 for the surgery cost, 0.46 for the ICU cost, and 0.11 for the ward cost. The Hosmer-Lemeshow p value for EuroSCORE prediction of more than 2 days stay in the ICU was 0.40, indicating good accuracy. The area under the receiver operating characteristics curve was 0.78. The probability of an ICU stay exceeding 2 days was more than 50% at a EuroSCORE of 14 or more. Conclusions. In this single-institution study, the additive EuroSCORE algorithm could be used to predict ICU cost and also an ICU stay of more than 2 days after open heart surgery. (Ann Thorac Surg 2004;78:1528 35) 2004 by The Society of Thoracic Surgeons Quality control is an established feature of contemporary medicine, and required as an instrument for improving the standard of care. It is known that variations in outcome may be influenced by excess in workload [1]. As a result of an increasing cost awareness, and a relative scarcity of resources, it has become important to optimize and quality-assure medical interventions. Operative mortality is widely used as an indicator of the quality of cardiac surgery. To make an accurate comparison between different institutions or surgeons, mortality data must be adjusted to the risk profiles of the patients. During the last decades several models to calculate mortality risk before surgery have been developed. However, a preoperative risk algorithm specifically designed to predict the need for hospital care resources in cardiac surgery is lacking. As a result, a number of authors have evaluated different risk algorithms designed to prognosticate outcomes such as morbidity and mortality, to predict the need of resources after heart surgery. Most studies have focused on the total resource Accepted for publication April 20, 2004. Address reprint requests to Dr Nilsson, Department of Cardiothoracic Surgery, Heart and Lung Center, University Hospital, SE 221 85 Lund, Sweden; e-mail: johan.nilsson@thorax.lu.se. requirement for coronary artery bypass grafting only patients [2, 3]. The predictive value of risk scoring on the different components of resource utilization in heart surgery (surgery, intensive care unit [ICU], and ward) has previously not been studied. The purpose of the present study was to evaluate whether the preoperative risk stratification model European System for Cardiac Operative Risk Evaluation (EuroSCORE) [4] predicts the different components of resource utilization in cardiac surgery, by applying this algorithm to a local, large Swedish adult cardiac surgery database. Material and Methods The study was approved by the Ethical Committee of the Medical Faculty, Lund University. Data Risk factors for all adult patients undergoing heart surgery at the University Hospital of Lund between October 1,1999, and December 31, 2002, were prospectively collected when the patients were admitted to the Department of Cardiothoracic Surgery. The patient record form contained a total of 248 variables (preoperative, intraoperative, and postoperative) based on the Higgins, Parsonnet, The Society of Thoracic Surgeons, and EuroSCORE 2004 by The Society of Thoracic Surgeons 0003-4975/04/$30.00 Published by Elsevier Inc doi:10.1016/j.athoracsur.2004.04.060

Ann Thorac Surg NILSSON ET AL 2004;78:1528 35 EUROSCORE AND COSTS Table 1. Patient Characteristics and Weights (Score) of the Risk Factors in European System for Cardiac Operative Risk Evaluation (EuroSCORE) in 3,413 Open Heart Operations Variables Mean SD, or % Score Age (years) 67.5 10.5 1 a Female sex 27.8% 1 Chronic pulmonary disease 10.4% 1 Extracardiac arteriopathy 15.8% 2 Neurological dysfunction 5.9% 2 Previous cardiac surgery 4.5% 3 Serum creatinine 200 mol/l 2.4% 2 (2.27 mg/dl) Active endocarditis 1.3% 3 Critical preoperative state b 7.0% 3 Unstable angina (requiring 13.6% 2 intravenous nitrates) Left ventricular dysfunction EF 0.30 0.50 35.5% 1 EF 0.30 9.0% 3 Preoperative myocardial infarction 34.4% 2 90 days Pulmonary hypertension (systolic 4.3% 2 PAP 60 mm Hg) Emergency 10.1% 2 Other than isolated CABG 27.1% 2 Surgery on thoracic aorta 3.5% 3 Postinfarction ventricular septal 0.3% 4 rupture closure EuroSCORE 6.1 3.8 a One score per 5 years (or part thereof) over 60 years. b Any one or more of the following: ventricular tachycardia or fibrillation or aborted sudden death, preoperative cardiac massage, preoperative ventilation before arrival in the anesthetic room, preoperative inotropic support, intraaortic balloon counterpulsation or preoperative acute renal failure (anuria or oliguria 10 ml/h). CABG coronary artery bypass grafting; EF ejection fraction; PAP pulmonary arterial pressure; SD standard deviation. patient record forms. The data were stored in a local adult cardiac surgery database. Eighteen variables (Table 1) were imported into the statistical software package Intercooled Stata (version 8.2, 2003; Stata Corporation, USA). The duration of anesthesia (minutes), the Lund ICU workload score, the length of stay (LOS) in the ICU and in the ward, and the total in-hospital stay were collected. Patients who underwent transplantation, died intraoperatively, or in whom any of the preoperative, intraoperative, or postoperative data were missing were excluded from the study. The Lund ICU workload score is a modification of a nursing care recording system [5], by which each patient in the ICU gets a score three times a day, depending on the resources needed for his or her condition (eg, medication, volume therapy, transfusions, need of ventilator assistance, need of further technical support such as dialysis or cardiac assist device, nurse workload). Scoring points are directly related to the cost of the specific resource used. The total number of points is computed daily for each patient and entered into the database. Table 2. Principles for Calculation of Costs of Care Costs Equation Surgery Starting Cost Op (Anesthesia duration Constant Op ) Implantable material ICU Starting Cost ICU (Total Lund ICU workload score Constant ICU ) Ward Starting Cost Ward (LOS ward Constant Ward ) ICU intensive care unit; LOS length of stay; Op operating room. EuroSCORE is an additive risk algorithm developed on patients operated on in 128 surgical centers in eight European countries in 1995. Risk factors (97 variables) and mortality data from 19,030 consecutive adult patients undergoing all types of cardiac surgery were collected and stored in the EuroSCORE multinational database [6]. The risk algorithm was constructed by using logistic regression analysis. The included variables and their scores are listed in Table 1. In the present study, the total risk score for every patient was calculated according to the EuroSCORE additive algorithm [4] (http://www.euroscore.org), and the individual cost was calculated according to a formula used by the hospital accounting system. Principles for calculations of cost of care are shown in Table 2. The hospital economy department established all starting costs and constants yearly. The 30-day mortality was obtained from the Population and Welfare Statistics Sweden, Statistiska Centralbyrån, Stockholm, Sweden. Statistics Values are given as mean standard deviation, median, and range. Univariate linear regression analysis was used to test the correlation between the EuroSCORE and either cost or LOS. Multivariate linear regression analysis was used to test which combination of the individual risk factors in the EuroSCORE model were significantly correlated to total cost. The cost and LOS were normalized by logarithmically transforming the data [7]. Analyses were performed using both individual patient data and patients grouped into six risk cohorts (Table 3) [3, 7]. The EuroSCORE was used as a univariate predictor in developing a logistic regression model from the present data set. Predicted ICU stay more than 1 day and more Table 3. Patients Cohorts Based on European System for Cardiac Operative Risk Evaluation (EuroSCORE) Risk Stratification Cohort EuroSCORE Risk No. of operations I 0 2 612 II 3 4 672 III 5 6 700 IV 7 8 614 V 9 10 423 VI 10 392 1529 CARDIOVASCULAR

1530 NILSSON ET AL Ann Thorac Surg EUROSCORE AND COSTS 2004;78:1528 35 Fig 1. Graph of costs (mean standard deviation) for each risk score. (USD US dollars). than 2 days was calculated using these models. Predictive accuracy was assessed by comparing the observed and the expected more than 1 day LOS and more than 2 days LOS at the ICU for equal-sized quantiles of risk, by applying the Hosmer-Lemeshow goodness-of-fit test, after dividing the study patients into 10 different ordered risk groups [8]. The discriminatory power of the logistic regression model was evaluated by calculating the areas under the receiver operating characteristics (ROC) curves [9, 10]. These areas are presented with 95% confidence intervals. An area of 1.0 under the ROC curve indicates perfect discrimination, whereas an area of 0.50 indicates complete absence of discrimination. Any intermediate value is a quantitative measure of the ability of the risk predictor model to distinguish between a shorter and longer LOS at the ICU. One-way analysis of variance was used to compare the difference between predicted and observed number of patients with an ICU stay of more than 2 days for each risk cohort. Graphs and statistical analyses were performed with Intercooled Stata. Results Between October 1, 1999, and December 31, 2002, 3,819 patients underwent cardiac surgery at our institution. Patients who underwent transplantation (n 66), died intraoperatively (n 37), or in whom any of the preoperative, intraoperative, or postoperative data were missing (n 312) were excluded from the study. Thus, 3,404 patients, undergoing 3,413 operations, were included in the analysis. There was accurate documentation of data including 30-day mortality in all cases, and no patient was lost to follow-up. Preoperative characteristics are listed in Table 1. The average age was 67.5 10.5 years (range, 18 to 89 years). The majority of patients were men (72%). A coronary artery bypass grafting only operation was performed in 2,487 cases (73%), 710 (21%) cases had a valve procedure with or without coronary artery bypass grafting surgery, and 216 (6%) were miscellaneous procedures (postinfarction septal rupture, aortic aneurysm or dissection, and so forth). The actual 30-day postoperative mortality was 2.5% (95% confidence interval, 2.0% to 3.0%). The mean cost for the surgery was $7,300 $2,120 (median, $6,613; range, $2,563 to $25,988), in the ICU $3,746 $6,032 (median, $2,182; range, $632 to $134,263), and in the ward $3,500 $2,605 (median, $2,999; range, $0 to $41,626), and the mean total cost was $14,546 $7,658 (median, $12,546; range, $6,995 to $157,912). The mean costs ( standard deviation) were calculated for the EuroSCORE risk groups 0 to 24 for the surgery, the ICU, and the ward, with the results shown in Figure 1. The log-transformed cost for the individual patients was significantly correlated to EuroSCORE. The strongest correlation was between the EuroSCORE and log-transformed ICU costs, with a correlation coefficient (r) of 0.46 (p 0.0001; Table 4). When patients were grouped into cohorts of similar predicted EuroSCORE risk (Table 3), the correlation between log-transformed mean costs was improved. The mean total cost was significantly correlated to mean EuroSCORE risk for each risk cohort, with an r of 0.99 (p 0.00005); r was 0.99 for the mean surgery cost, 0.98 for the mean ICU cost, and 0.94 for the mean ward cost (Fig 2). In the multivariate linear regression analysis with the 18 EuroSCORE risk factors as regressor variables and log-transformed cost as the dependent variable, 15 Euro- SCORE variables were found to be significantly (p 0.05) associated with the log-transformed cost (Table 5), with an r of 0.63 (p 0.0001). The mean LOS in the ICU was 1.76 2.39 days

Ann Thorac Surg NILSSON ET AL 2004;78:1528 35 EUROSCORE AND COSTS Table 4. Regression Analysis Results (n 3,413) Comparison Equation Coefficient 95% CI r p Value r 2 Cost versus EuroSCORE Surgery $6,182 1.021 EuroSCORE 1.019 1.024 0.31 0.0001 0.10 ICU $1,752 1.076 EuroSCORE 1.071 1.081 0.46 0.0001 0.21 Ward $2,873 1.014 EuroSCORE 1.010 1.019 0.11 0.0001 0.01 Total $10,688 1.040 EuroSCORE 1.038 1.043 0.47 0.0001 0.22 LOS versus EuroSCORE ICU 0.88 days 1.071 EuroSCORE 1.066 1.076 0.45 0.0001 0.21 Ward 6.05 days 1.015 EuroSCORE 1.010 1.020 0.11 0.0001 0.01 Total 7.14 days 1.028 EuroSCORE 1.024 1.032 0.24 0.0001 0.06 Probability of ICU stay 1 day Probability of ICU stay 2 days a Pseudo r 2. 1531 3.0 0.28 EuroSCORE) e 0.25 0.30 0.0001 0.16 a 1 e ( 3.0 0.28 EuroSCORE) 4.0 0.29 EuroSCORE) e 0.26 0.32 0.0001 0.18 a 1 e ( 4.0 0.29 EuroSCORE) CARDIOVASCULAR CI confidence interval; ICU intensive care unit; LOS length of stay. (median, 1 day; range, 1 to 41 days; Table 6). Logtransformed LOS at the ICU was significantly correlated to EuroSCORE with an r of 0.45 (p 0.0001; Table 4). The results from the logistic regression analysis are presented in Table 4. Of all patients, 25.4% had an ICU stay of more than 1 day, and 13.7% had an ICU stay of more than 2 days. Hosmer-Lemeshow test gave a p value of 0.24 for the EuroSCORE to predict an ICU stay of more than 1 day and a p value of 0.40 to predict an ICU stay of more than 2 days, which indicates a good accuracy. The area under the ROC curve for an ICU stay of more than 1 day was smaller compared with the ROC area for an ICU stay of more than 2 days (0.76; 95% confidence interval, 0.74 to 0.78; and 0.78; 95% confidence interval, 0.76 to 0.81, respectively; Fig 3). The probability of an ICU stay exceeding 2 days was more than 50% at a EuroSCORE of 14 or more (Fig 4). The sensitivity and specificity for this cutoff point were 21% and 98%, respectively. During the entire study period (169 weeks), the mean Fig 2. Linear regression analysis of mean EuroSCORE versus logtransformed mean costs in each cohort for surgery (circle), in the intensive care unit (square), in the ward (cross), and total (diamond). Cohorts are defined in Table 3. (USD US dollars.) weekly number of patients entering the ICU was 20 7.13 (median, 22; range, 5 to 35). During this period, the EuroSCORE algorithm predicted the number of patients with an ICU stay more than 2 days exactly in 51 weeks (30%), and within 1 patient in 127 weeks (75%). The predictive accuracy was independent of the EuroSCORE risk cohort (p 0.65). Table 5. Regression Analysis: European System for Cardiac Operative Risk Evaluation (EuroSCORE) Variables Independently Associated With Total Costs (n 3,413) Variables % Cost Increase 95% CI p Value Age a 0.28 0.20 0.37 0.0001 Female sex 1.1 3.0 0.89 0.283 Chronic pulmonary disease 2.9 0.0 5.8 0.047 Extracardiac arteriopathy 3.4 1.0 6.0 0.006 Neurologic dysfunction 6.9 3.1 10.9 0.0001 Previous cardiac surgery 12.0 7.5 16.8 0.0001 Serum creatinine 200 mol/l (2.27 mg/dl) 16.4 10.0 23.2 0.0001 Active endocarditis 11.4 3.1 20.4 0.007 Critical preoperative state b 19.9 14.9 25.1 0.0001 Unstable angina (requiring 2.2 0.7 5.3 0.132 intravenous nitrates) Left ventricular dysfunction EF 0.30 0.50 3.3 1.4 5.3 0.001 EF 0.30 12.1 8.6 15.8 0.0001 Preoperative myocardial infarction 90 days 2.7 0.6 4.7 0.010 Pulmonary hypertension 15.5 10.5 20.8 0.0001 (systolic PAP 60 mm Hg) Emergency 9.3 5.3 13.5 0.0001 Other than isolated CABG 37.5 34.4 40.6 0.0001 Surgery on thoracic aorta 24.4 18.2 30.8 0.0001 Postinfarction ventricular septal rupture closure 12.0 4.9 31.9 0.174 a % increase per year of age. b See Table 1 for definition. CI confidence interval; EF ejection fraction; CABG coronary artery bypass grafting; PAP pulmonary arterial pressure.

1532 NILSSON ET AL Ann Thorac Surg EUROSCORE AND COSTS 2004;78:1528 35 Table 6. Intraoperative and Postoperative Data in 3,413 Open Heart Operations Variables Mean SD, or % Median Range Anesthesia duration (minutes) 280 70 267 75 1,120 Total Lund ICU workload score 109 183 60 14 3,802 ICU LOS (days) 1.76 2.4 1 1 41 Ward LOS (days) 7.53 5.9 6 0 105 Total LOS (days) 9.28 6.6 8 1 105 30-day mortality 2.5% ICU intensive care unit; LOS length of stay; SD standard deviation. Comment The purpose of the present study was to evaluate whether the preoperative risk stratification model Euro- SCORE predicts the different cost components in open heart surgery. The results show that the EuroSCORE algorithm correlates to both costs and length of ICU stay. Fifteen of the 18 variables included in the risk algorithm were significantly correlated to the total cost of open heart surgery. The closest correlation was between EuroSCORE and cost at the ICU. The EuroSCORE algorithm also showed a good predictive power (accuracy) of an ICU stay more than 1 day, and even better for an ICU stay more than 2 days. In our experience, patients staying in the ICU more than 2 days are likely to remain there for prolonged periods. As other authors [11 13], we therefore chose to focus the additional analyses on a more than 2-day ICU stay as being clinically more important. The results from our study also suggest that accuracy and discrimination will be better for a stay of more than 2 days. At higher risk scores the EuroSCORE predictive performance was less than at lower scores, as previously reported in mortality studies [14]. The lower numbers of patients in these high-risk score groups might contribute to this finding, but it may also reflect a weakness of the Fig 3. The receiver operating characteristic curves for sensitivity and 1 minus specificity for the EuroSCORE prediction of an intensive care unit stay of more than 1 day (diamonds) and an intensive care unit stay of more than 2 days (circles). The solid line represents no discrimination. Fig 4. Percentage of patients with an intensive care unit (ICU) stay more than 2 days (left y axis) for each EuroSCORE risk group (x axis). Predicted intensive care unit stay more than 2 days (dotted line) with 95% confidence interval (shaded area); observed intensive care unit stay more than 2 days (diamond) with 95% confidence interval (bars). The histogram shows the number of patients (right y axis) in each risk group. risk score algorithm. Indeed, the EuroSCORE was developed to predict mortality and not cost or duration of ICU stay. In our study, as in others [3, 15], the predictive value was limited for individual patients, but an excellent correlation was seen if the patients were grouped in risk cohorts. The predictive accuracy of the total number of patients with an ICU stay more than 2 days was independent of which risk cohort the patients belonged to. This indicates that preoperative prediction of ICU stay using EuroSCORE is possible for groups of patients operated on during a certain period of time (e.g., 1 week), and may therefore be used in clinical practice. In the present study the strongest correlation was between EuroSCORE and resource requirements in the ICU. A patient with a high preoperative risk score is expected to need more ICU care and support; therefore, the cost of this may be better measured by total ICU workload than by LOS alone. The resource utilization in the ICU is primarily dependent on the patient s medical condition, whereas numerous other factors may influence the time to discharge from the regular ward. Earlier studies have indicated that preoperative risk variables can be used to predict costs of cardiac surgery [16, 17]. An increasing Cleveland risk score has been shown to be associated with an increase in total cost and longer postoperative LOS [2], and similar results have been shown with the CABDEAL risk algorithm [18]. Riordan and colleagues [3] found that grouping patients in risk cohorts resulted in a correlation between The Society of Thoracic Surgeons risk algorithm and total cost, but for individual patients the prediction was poor. In these studies the relationships found were between the risk algorithm and total cost for coronary artery bypass grafting only patients. In a study of the different components of heart surgery cost, Ferraris and associates

Ann Thorac Surg NILSSON ET AL 2004;78:1528 35 EUROSCORE AND COSTS [19] showed that the greatest expenses were generated by anesthesia and the surgery, but any relationship between the risk and cost was not evaluated. The additive EuroSCORE model has been shown to work well to predict 30-day mortality in many European countries [20] and in the United States [21], and compares favorably with The Society of Thoracic Surgeons risk stratification algorithm [22]. Use of the EuroSCORE risk algorithm to predict the need for different resources therefore appears logical. Pintor and coworkers [7] and Sokolovic and associates [23] recently demonstrated a correlation between EuroSCORE and cost for open heart surgery. Both these studies were rather small (488 and 201 patients, respectively), and their focus was on total cost and not on the different components of heart surgery resource utilization. Several studies have tried to identify preoperative risk variables that predict LOS in the ICU after cardiac surgery. The Parsonnet risk algorithm seems able to preoperatively identify patients who are likely to spend more than 24 hours in the ICU [24] with an ROC area of 0.70, which is somewhat less than the ROC area for an ICU stay of more than 1 day in the present study (0.76). A strength of the EuroSCORE is that it has been developed fairly recently (1995) on a large multiinstitutional patient material. The Parsonnet score, on the other hand, is developed on a single-institution material collected between 1982 and 1987. Three additional studies [11 13], comparing different risk algorithms, have found a correlation between Euro- SCORE and ICU stay of more than 2 days. Tu and Guerriere [25] used a neural network as a predictive instrument, finding both advantages and disadvantages as compared with other statistical techniques. Our study is based on patients treated in a single, European institution, but the EuroSCORE and the evaluation were applied to a relatively large patient material, in which the data were prospectively collected on a regular basis in the daily clinical work. We chose to exclude cases dying intraoperatively as they did not require any postoperative resources. The exclusion of these mainly high-risk patients probably reduces the actual predictive power of the analysis, but the difference will be minor because the number of patients who died intraoperatively was small (1.1% during the period in question). Improved preoperative prediction of care requirements in cardiac surgery could give patients better access to treatment, and could also guide the selection of patients for surgery versus alternative, nonsurgical therapies. A preoperative risk algorithm specifically designed to predict the need for hospital care resources in cardiac surgery could help in decision-making to realistically estimate the need for resources and plan the care for high-risk patients more efficiently. By making a weekly operation schedule that is adjusted for expected resource utilization, the workload on the ICU could be more equal over time, which may improve patient management. This could also have positive effects on the outcome of surgery [1]. The EuroSCORE algorithm appears valuable in this process. References 1533 1. Tarnow-Mordi WO, Hau C, Warden A, Shearer AJ. Hospital mortality in relation to staff workload: a 4-year study in an adult intensive-care unit. Lancet 2000;356:185 9. 2. Kurki TS, Hakkinen U, Lauharanta J, Ramo J, Leijala M. Evaluation of the relationship between preoperative risk scores, postoperative and total length of stays and hospital costs in coronary bypass surgery. Eur J Cardiothorac Surg 2001;20:1183 7. 3. Riordan CJ, Engoren M, Zacharias A, et al. Resource utilization in coronary artery bypass operation: does surgical risk predict cost? Ann Thorac Surg 2000;69:1092 7. 4. Nashef SA, Roques F, Michel P, Gauducheau E, Lemeshow S, Salamon R. European system for cardiac operative risk evaluation (EuroSCORE). Eur J Cardiothorac Surg 1999;16: 9 13. 5. Hjortso E, Buch T, Ryding J, et al. The nursing care recording system. A preliminary study of a system for assessment of nursing care demands in the ICU. Acta Anaesthesiol Scand 1992;36:610 4. 6. Roques F, Nashef SA, Michel P, et al. Risk factors and outcome in European cardiac surgery: analysis of the Euro- SCORE multinational database of 19030 patients. Eur J Cardiothorac Surg 1999;15:816 23. 7. Pintor PP, Bobbio M, Colangelo S, Veglia F, Marras R, Diena M. Can EuroSCORE predict direct costs of cardiac surgery? Eur J Cardiothorac Surg 2003;23:595 8. 8. Hosmer DW, Lemeshow S. Applied logistic regression. New York: John Wiley & Sons, 2000:147 56. 9. Hanley JA, McNeil BJ. The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology 1982;143:29 36. 10. Swets JA. Measuring the accuracy of diagnostic systems. Science 1988;240:1285 93. 11. Stoica SC, Sharples LD, Ahmed I, Roques F, Large SR, Nashef SA. Preoperative risk prediction and intraoperative events in cardiac surgery. Eur J Cardiothorac Surg 2002;21: 41 6. 12. Huijskes RVHP, Rosseel PMJ, Tijssen JGP. Outcome prediction in coronary artery bypass grafting and valve surgery in the Netherlands: development of the Amphiascore and its comparison with the Euroscore. Eur J Cardiothorac Surg 2003;24:741 9. 13. Pitkanen O, Niskanen M, Rehnberg S, Hippelainen M, Hynynen M. Intra-institutional prediction of outcome after cardiac surgery: comparison between a locally derived model and the EuroSCORE. Eur J Cardiothorac Surg 2000; 18:703 10. 14. Pintor PP, Colangelo S, Bobbio M. Evolution of case-mix in heart surgery: from mortality risk to complication risk. Eur J Cardiothorac Surg 2002;22:927 33. 15. Williams TE Jr, Fanning WJ, Benton WC, et al. What is the marginal cost for marginal risk in cardiac surgery? Ann Thorac Surg 1998;66:1969 71. 16. MaWhinney S, Brown ER, Malcolm J, et al. Identification of risk factors for increased cost, charges, and length of stay for cardiac patients. Ann Thorac Surg 2000;70:702 10. 17. Smith PK, Smith LR, Muhlbaier LH. Risk stratification for adverse economic outcomes in cardiac surgery. Ann Thorac Surg 1997;64(Suppl):S61-3 S80-2. 18. Kurki TS, Kataja MJ, Reich DL. Validation of a preoperative risk index as a predictor of perioperative morbidity and hospital costs in coronary artery bypass graft surgery. J Cardiothorac Vasc Anesth 2002;16:401 4. 19. Ferraris VA, Ferraris SP, Singh A. Operative outcome and hospital cost. J Thorac Cardiovasc Surg 1998;115:593 603. 20. Roques F, Nashef SA, Michel P, et al. Does EuroSCORE CARDIOVASCULAR

1534 NILSSON ET AL Ann Thorac Surg EUROSCORE AND COSTS 2004;78:1528 35 work in individual European countries? Eur J Cardiothorac Surg 2000;18:27 30. 21. Nashef SA, Roques F, Hammill BG, et al. Validation of European system for cardiac operative risk evaluation (EuroSCORE) in North American cardiac surgery. Eur J Cardiothorac Surg 2002;22:101 5. 22. Nilsson J, Algotsson L, Höglund P, Lührs C, Brandt J. Early mortality in coronary bypass surgery: the EuroSCORE versus the STS risk algorithm. Ann Thorac Surg 2004;77:1235 40. 23. Sokolovic E, Schmidlin D, Schmid ER, et al. Determinants of costs and resource utilization associated with open heart surgery. Eur Heart J 2002;23:574 8. 24. Lawrence DR, Valencia O, Smith EE, Murday A, Treasure T. Parsonnet score is a good predictor of the duration of intensive care unit stay following cardiac surgery. Heart 2000;83:429 32. 25. Tu JV, Guerriere MR. Use of a neural network as a predictive instrument for length of stay in the intensive care unit following cardiac surgery. Comput Biomed Res 1993;26: 220 9. INVITED COMMENTARY The article by Nilsson and coworkers claims that the EuroSCORE (a popular mortality risk scoring system used extensively in Europe) is also useful for predicting resource utilization in cardiac surgical patients. This work deserves comment in three areas. First, there are some quirks about the authors measurements that need to be pointed out. The authors use log transformation of the dependent variable (hospital cost) in order to measure the relationship of EuroSCORE to hospital costs. Their Figure 2 shows a very linear regression between cost and EuroSCORE, only after the cost data are log transformed. The advantage of log transformation is that the influence of outliers is minimized and the regression coefficient looks better. The disadvantage is that the resulting regression line using log transformed data are always more linear than nonlog transformed data and gives an inflated sense of dependence of the outcome variable (cost) on the predictor variable (EuroSCORE). It would be wishful thinking to suppose that the EuroSCORE predicts hospital cost without significant variability. In fact, the authors acknowledge that the EuroSCORE is not a useful measure of hospital costs for individual patients because of this variability. Other measurement issues that are a little quirky include the way that hospital costs were measured by the authors. This method of cost calculation is unlikely to be duplicated by any other center in the world. The cost calculation uses factors like the Lund ICU workload score and the Starting Cost Op, which is a factor estimated by the hospital accounting system (see their Table 2). This idiosyncratic estimate of cost is unlikely to be duplicated by other institutions that may wish to copy the authors methods. In addition, the authors chose to exclude operative deaths from their analysis. We found that the most costly patients who undergo cardiac operations are those who die from the operation [1]. It may be that exclusion of the intraoperative deaths omits important information about determinants of hospital costs. For these reasons, the prediction model of hospital costs by Nilsson and coworkers is neither transferable nor robust. Second, each of the reviewers of this article asked a similar question (ie, What is the clinical usefulness of the cost prediction model based on the EuroSCORE for the individual patient?). The short answer to this question is that prediction of costs for individual patients by this system is inaccurate and the EuroSCORE should not be used for this purpose. The authors acknowledge this shortcoming and recognize that other investigators have not been able to apply models of risk-adjusted cost to individual patients, only to larger patient cohorts [2, 3]. They did find some correlation of the EuroSCORE with weekly costs in certain patient cohorts. They suggest a novel use of the EuroSCORE, namely, predicting weekly intensive care unit costs in certain patient cohorts. This may allow week-to-week assessment of resource needs in the intensive care unit or operating room based on the risk profile of patients being operated upon during a given week. This is an intriguing possibility and should be tested in larger cohorts. Third, if the EuroSCORE can not be used to predict individual patient costs and the authors means of measuring hospital costs are quirky, then what is the value of this analysis? The answer lies in the assessment of what a health care payer must pay to treat patients. In the United States, the Leapfrog Group (Washington, D.C.) and other consortiums of health care payers are concerned about the spiraling increase in health care costs and ultimately the cost that corporations must pay to provide health care to their beneficiaries. By default, and because there is not really a good measure of hospital (and individual provider) quality that is readily understood by corporate executives and beneficiaries, these consortiums have chosen provider volume as a surrogate for quality. These same consortiums assume (without much justification) that high quality will translate into lower costs. The problem with using hospital volume as an indicator of quality and ultimately lower health care costs is that volume alone is a poor indicator of both quality and cost [4]. Some high-volume providers have poor outcomes or high costs and some low-volume providers have good outcomes or low costs. It seems so obvious that volume is a poor indicator of quality, but the consortiums that worry about health care costs are at a loss to find a better indicator of quality than hospital (or provider) volume. Herein lies the potential of studies like that of Nilsson and coworkers. The EuroSCORE is a far better predictor of cost than hospital volume. The score is easy to apply and provides a more accurate way to compare providers than simple hospital volume. Is the EuroSCORE the best indicator of hospital cost? It probably is not. However, studies like that of Nilsson and coworkers are the first steps in predicting the costs and the value of services provided by hospitals and individual practitioners. We can only hope that corporate exec- 2004 by The Society of Thoracic Surgeons 0003-4975/04/$30.00 Published by Elsevier Inc doi:10.1016/j.athoracsur.2004.05.078