Reclassification tables have never been used to compare concentrations of cardiac troponin I (cTnI) with predictive models of risk in the perioperative setting. The current study aimed to evaluate the prognostic value of pre- and/or postoperative serum cTnI when combined with The European System for Cardiac Operative Risk Evaluation (EuroSCORE) in predicting adverse outcome after cardiac surgery.
Nine hundred five consecutive patients were included. Standard EuroSCORE as well as preoperative and 24-h postoperative cTnI were measured in all patients. Major adverse cardiac events and in-hospital mortality were chosen as study endpoints. The performance of EuroSCORE with and without pre- and/or postoperative cTnI were assessed by means of receiver operating characteristic curves, net reclassification index, and integrated discrimination improvement analyses. Data are expressed as ±SD.
Death occurred in 28 of 905 (3%) patients and major adverse cardiac events in 202/905 (22%) patients. Models including EuroSCORE alone were characterized by a low discriminative power (c-index = 0.60 ± 0.05) in predicting major adverse cardiac events. The c-index increased to 0.61 ± 0.05 (P = 0.46), 0.70 ± 0.04 (P < 0.001), and 0.71 ± 0.04 (P < 0.001) when preoperative, postoperative, and pre/postoperative cTnI were included, respectively. The better predictive ability was confirmed by net reclassification index (0.41 ± 0.08, P < 0.001; 0.67 ± 0.08, P < 0.001; and 0.68 ± 0.08, P < 0.001, respectively) and integrated discrimination improvement (0.003 ± 0.002, P = 0.12; 0.099 ± 0.015, P < 0.001; and 0.094 ± 0.016, P < 0.001, respectively). Similar results were observed for in-hospital mortality.
The combination of EuroSCORE and postoperative cTnI provides the best discriminative power and performance in predicting adverse outcome after cardiac surgery and is suggested as being an effective model that improves early identification of high-risk patients.
What We Already Know about This Topic
Both a clinical assessment (EuroSCORE) and postoperative cardiac troponin I (cTnI) predict postoperative cardiac morbidity and mortality
What This Article Tells Us That Is New
Using reclassification table methods in 905 patients, the combination of postoperative EuroSCORE and cTnI predicted adverse outcomes better than either measure alone.
THE strong independent prognostic value of early postoperative cardiac troponin I (cTnI) release after cardiac surgery has been previously reported in studies from our group1,2and others.3–5A recent meta-analysis provided further evidence for a significant association between postoperative cTnI release and short- and long-term all-cause mortality after adult cardiac surgery.6However, the prognostic value of preoperative cTnI has only rarely been considered in the cardiac surgical setting.7,8The general criteria for clinically useful risk markers and the hierarchical steps in demonstrating the clinical interest of a biomarker have been recently published9,10and have served as a reminder that we should compare the diagnostic properties of the biomarker with existing predictive models of risk stratification and validate an incremental prognostic value before recommending wider use for routine practice. As far as we are aware, only a single retrospective study11has compared the prognostic value of cTnI with different clinical predictive risk models after cardiac surgery. This showed the superiority of the combination of postoperative cTnI with the European System for Cardiac Operative Risk Evaluation (EuroSCORE), recognized as being a robust clinical model of risk prediction in cardiac surgery,12in predicting in-hospital death, compared with EuroSCORE alone. Indeed, the authors found an improvement in the area (AUC) under the receiver operating characteristic (ROC) curve, comparing ROCAUCusing models with and without postoperative cTnI.11ROC curves might, however, be considered as insensitive in the evaluation of new biomarkers because it does not provide information about the pretest risk or the proportion of participants who have high- or low-risk values.10Thus, to complete the results obtained by ROC curves, new approaches have been proposed. Among them, reclassification tables appear to be the most interesting because they better focus on the key purpose of risk prediction, namely to classify individuals into clinically relevant risk categories.10Pencina et al .13have recently purposed two ways of assessing improvement in model performance using reclassification tables: the net reclassification index (NRI) and integrated discrimination improvement (IDI). Interestingly, this new approach has never been used to compare cTnI with clinical predictive models of risk in the perioperative setting, although it is now recognized as a more powerful tool than standard methods, such as ROC curves.10
The objective of the current study conducted in adult patients undergoing conventional cardiac surgery with cardiopulmonary bypass (CPB) was to evaluate the additional prognostic value of preoperative and/or postoperative serum cTnI when combined with standard EuroSCORE in predicting both overall in-hospital mortality and nonfatal cardiac morbidity. The hypothesis tested was therefore that the simultaneous use of EuroSCORE and cTnI would improve the performance of the predictive model.
Materials and Methods
The study was conducted in accordance with the Statements for Reporting Studies of Diagnostic Accuracy (the STARD initiative).14We used a comprehensive, prospectively recorded database describing the clinical and surgical characteristics of 905 consecutive adult patients undergoing cardiac surgery with CPB at the Saint-Martin Hospital (Caen, France) from January 2006 to December 2007. An anesthesiologist (J.L.F.) entered the data, and a systematic audit by a trained research technician who participated in previous studies1,2allowed verification of the accuracy in coding data. Missing data were coded as absent. The study was approved by an institutional review board (Comité pour la Protection des Personnes, Pitié-Salpêtrière, Paris, France). Because data were collected during routine care of patients that conformed to standard procedures currently used in this institution, authorization was granted to waive written informed consent. Inclusion criteria were elective conventional cardiac surgery with CPB: coronary artery bypass grafting, aortic valve and/or mitral valve replacement surgery, and combined surgery (coronary artery bypass grafting plus aortic valve and/or mitral valve replacement). Patients undergoing emergency surgery (less than 24 h), off-pump coronary artery bypass grafting, and complex, unusual procedures were excluded from the study.
Perioperative Management
All patients were premedicated with oral lorazepam (2.5 mg the evening before surgery and on the morning of surgery). β-blocking agents and statins were given until the morning of surgery in chronically treated patients. Oral antiplatelet agents were stopped within 7–10 days before surgery and replaced by oral flurbiprofene (50 mg twice) until the day before surgery. An additive EuroSCORE was systematically established by the cardiac surgeon at the bedside before surgery. Standardized total intravenous anesthesia (i .e ., target control propofol infusion, remifentanil, and pancuronium bromide) and monitoring techniques (i .e ., five-lead electrocardiogram with computerized analysis of repolarization, invasive arterial blood pressure, and central venous pressure) were used in all patients and complied with routine practice at the Saint-Martin Hospital.1,2Antifibrinolytic therapy with tranexamic acid (15 mg/kg twice) was routinely administered. CPB was performed under normothermia (more than 35.5°C), and myocardial protection was achieved by intermittent anterograde or combined (anterograde plus retrograde) warm blood cardioplegia, as previously described.1,2Boluses of ephedrine and/or phenylephrine were given intraoperatively as necessary to maintain mean arterial pressure between 50 and 80 mmHg. The heart was defibrillated after aortic unclamping, if sinus rhythm did not resume spontaneously. After the termination of cardiopulmonary bypass, catecholamines were used, when necessary, at the discretion of the attending anesthesiologist.15All patients were admitted postoperatively into the intensive care unit for at least 48 h and were assessed for tracheal extubation within 1–8 h of arrival in the intensive care unit. Standard postoperative care included blood glucose control less than 8 mm,16intravenous heparin (200 U/kg) in patients with valve disease, and aspirin (300 mg, oral or intravenous) and a low molecular weight heparin (nadroparin 2,850 U Anti-Xa, subcutaneous; Fraxiparine®; Sanofi-Synthelabo, Paris, France) in patients with coronary artery disease, beginning 6 h after surgery in the absence of significant mediastinal bleeding (more than 50 ml/h). β-blocking agents and statins were given as soon as possible postoperatively in chronically treated patients.17Postoperative care was delivered by anesthesiologists in the intensive care unit and by cardiac surgeons in the ward.
Measurements of cTnI Concentration
Blood samples were collected the day before surgery and at 24 h after the end of surgery. This single postoperative time point was chosen in accordance with previous studies from our group and others showing that a single 24-h cTnI value is an independent predictor of short- and long-term adverse outcome in cardiac surgical patients.1,4,18A technician who was unaware of the clinical and electrocardiogram data performed the assays. cTnI was analyzed with a sensitive and highly specific immunoenzymometric assay (AxSYM Troponin-I ADV assay; Abbott Laboratories, Rungis, France) that detects both free and complex bound troponin. The assay allows the detection of cTnI within the range of 0.02–23 ng/ml with appropriate dilutions. Values greater than 0.04 ng/ml were considered abnormal. The within-run coefficient of variation was 6% and the between-run coefficient of variation was 11%.
Clinical Outcome and Endpoints
The duration of hospitalization, the length of stay in the intensive care unit, the in-hospital mortality, and the nonfatal major adverse cardiac events (MACEs) were recorded. MACEs included malignant ventricular arrhythmia, defined as sustained ventricular arrhythmias requiring treatment, postoperative myocardial infarction, and the need for inotropic support for at least 24 h, as previously described.15Daily 12-lead electrocardiogram recordings and postoperative two-dimensional echocardiography (systematically performed within 5 days after surgery, according to standard procedures currently used in this institution) were assessed by two experienced physicians blinded to the clinical and biochemical information. Diagnostic criteria for myocardial infarction were the appearance of new Q waves of more than 0.04 s and 1 mm deep or a reduction in R waves of more than 25% in at least two continuous leads of the same vascular territory and/or occurrence of postoperative severe wall motion abnormalities in the same area, as previously described.18In-hospital death was defined as death of any cause occurring at any time during the stay in hospital. Overall in-hospital death and MACEs were chosen as study endpoints.
Statistical Analysis
Data are expressed as mean ± SD, or median (25–75th interquartile) for nonnormally distributed variables (d'Agostino–Pearson omnibus test), or number (percentage), as appropriate. Comparison of two means was performed using the unpaired Student t test, whereas comparison of two medians was performed using the Mann–Whitney U test, and comparison of proportions was performed using Fisher exact method.
To predict postoperative outcome, embedded models were constructed, one with EuroSCORE and the others with EuroSCORE and preoperative and/or postoperative cTnI. The discriminative power of these models was quantified by measurement of ROCAUC(c-index), which is the usual global measure of the performance of a prognostic test, and represents the probability of assigning a greater risk to present the outcome of interest to a randomly selected patient who died compared with a randomly selected patient who survived.19Model calibrations were assessed using the Hosmer–Lemeshow statistics.20Internal validation of our models was done using 10-fold cross-validation.21This method randomly assigns the patients to 10 equally sized partitions. Subsequently, nine partitions were used as a learning set and one as a testing set. This operation was repeated 10 times until each partition was used as a testing set.
The models were then compared using a nonparametric method for comparison of paired ROC curves.22ROC curves were obtained by averaging 1,000 populations bootstrapped from the original study population. This method limits the impact of outliers and allows the provision of more robust representations. CIs of the average ROC curves were depicted using box plots. Presented AUCROCwere the average of the 1,000 populations (more robust estimations of the true values). As the comparison of ROC curves was recognized to be potentially insensitive, two complementary methods (NRI and IDI) were used, as recently described.13NRI requires predefined strata of risk, whereas IDI does not and can be seen as a continuous version of NRI with a probability of endpoint differences used instead of predefined strata.10This approach allowed us to assess the improvement of predictive abilities given by simultaneous use of EuroSCORE and pre- and/or postoperative cTnI. Because NRI and IDI are powerful statistical tools, significant results might only have a poor clinical impact. In order to illustrate the improvement given by pre- and/or postoperative cTnI, we provide reclassification tables, that enabled us to quantify the benefit of cTnI in terms of number of patients correctly reclassified by adding cTnI to the predictive models. Further, the reclassification tables of patients presenting or not the endpoint offer a practical representation of both the relationship between false positive and false negative and the magnitude of the gain of predictability in quantitative terms (number of patients).
A P value of less than 0.05 was considered significant, and all P values were two-tailed. All statistical analyses were conducted using R software23and specific packages.**
Results
The perioperative characteristics of the whole cohort of patients are shown in table 1. The median EuroSCORE was 5 (4–7) (extremes 0–13), whereas the median preoperative cTnI was 0 (0–0) (extremes 0–137) and the median postoperative cTnI was 6.1 (2.9–11.5) (extremes 0.1–510.0). In-hospital death occurred in 28 of 905 (3%) patients, and MACEs occurred in 202 of 905 (22%) patients.
Comparison of Estimations of Postoperative Risk
The univariate discriminative power of EuroSCORE, preoperative cTnI, and postoperative cTnI in predicting both in-hospital mortality and MACEs are shown in table 2. Although all three variables had a significant predictive value for MACE and death, the global discriminative power remained low for both endpoints, even for EuroSCORE (table 2).
Models to Predict MACE or In-hospital Mortality
Embedded models were considered to assess prediction of endpoints. Confirming the univariate ROC analysis, models including only EuroSCORE, although well calibrated (P values associated with Hosmer–Lemeshow statistics were 0.93 for MACEs and 0.29 for in-hospital mortality), were characterized by a low discriminative power (c-index = 0.60 and 0.61, respectively). CPB and aortic cross-clamping times as well as postoperative mediastinal bleeding were also found to be predictors of MACEs and in-hospital mortality. Inclusion of these intra- and early postoperative variables significantly increased the discrimination of EuroSCORE in predicting MACEs (c-index = 0.71; P < 0.001) and in-hospital mortality (c-index = 0.71; P = 0.04). The calibration of these models remained correct (P values associated with Hosmer–Lemeshow statistics were 0.62 for MACEs and 0.30 for in-hospital mortality). After 10-fold cross-validation, the maximum difference in c-index between derivation and validation cohorts was 0.01, suggesting that the presented models were robust, and that their predictive values were not only related to several patients.
Impact of Pre- and/or Postoperative cTnI in Predicting MACE or In-hospital Mortality
The AUCROCof the model based on EuroSCORE to predict MACEs was 0.60 ± 0.05 versus 0.61 ± 0.05 (P = 0.46) in the model that also included preoperative cTnI, 0.70 ± 0.04 (P < 0.001) with postoperative cTnI, and 0.71 ± 0.04 (P < 0.001) with pre- and postoperative cTnI (fig. 1A–C). The better predictive ability of the model including pre- and/or postoperative cTnI was confirmed by the NRI (0.41 ± 0.08, P < 0.001; 0.67 ± 0.08, P < 0.001; and 0.68 ± 0.08, P < 0.001, respectively) and IDI (0.003 ± 0.002, P = 0.12; 0.099 ± 0.015, P < 0.001; and 0.094 ± 0.016, P < 0.001, respectively). Table 3shows a reclassification table for the use of postoperative cTnI to predict MACEs. Patients with MACEs were more frequently classified in a stratum of higher risk in the model including cTnI, compared with EuroSCORE alone, whereas patients without MACEs were more frequently classified in a stratum of lower risk (table 3). Comparisons of models including pre- and postoperative cTnI versus postoperative cTnI (fig. 1C) were not significant (NRI: 0.15 ± 0.08, P = 0.07 and IDI: 0.005 ± 0.003, P = 0.11), implying that preoperative cTnI did not significantly increase the predictive value of a model including EuroSCORE and postoperative cTnI in combination. A reclassification table for the use of preoperative cTnI to predict MACEs when EuroSCORE and postoperative cTnI were taken into consideration is presented in table 4.
The AUCROCof the model based on EuroSCORE to predict in-hospital mortality was 0.61 ± 0.14, compared with 0.62 ± 0.15 (P = 0.64) in the model that also included preoperative cTnI; 0.71 ± 0.13 (P = 0.07) with postoperative cTnI; and 0.71 ± 0.12 (P = 0.05) with pre- and postoperative cTnI (fig. 1D–F). The predictive ability of the models including pre- and/or postoperative cTnI was assessed by the NRI (0.04 ± 0.23, P = 0.86; 0.56 ± 0.23, P = 0.01; and 0.65 ± 0.22, P = 0.004, respectively) and IDI (0.012 ± 0.011, P = 0.26; 0.042 ± 0.003, P = 0.14; and 0.10 ± 0.05, P = 0.06, respectively). This demonstrated that preoperative cTnI did not significantly increase the value of EuroSCORE in predicting in-hospital mortality, and that postoperative cTnI might have a clinical value. Nevertheless, the low rate of mortality in our cohort limited the power of analysis and interpretation. Comparison of models including pre- and postoperative cTnI versus postoperative cTnI were not significant (NRI: 0.09 ± 0.23, P = 0.70 and IDI: 0.058 ± 0.049, P = 0.23). This suggested that preoperative cTnI did not significantly increase the predictive value of a model including EuroSCORE and postoperative cTnI in combination to predict in-hospital mortality.
When intra- and postoperative variables were considered to predict MACEs (fig. 2A–C), preoperative cTnI only slightly increased the predictive value of the model (AUCROC= 0.71 ± 0.04, P = 0.63; NRI = 0.54 ± 0.08, P < 0.001; and IDI = 0.006 ± 0.002, P = 0.01), whereas postoperative cTnI (AUCROC= 0.75 ± 0.04, P < 0.001; NRI = 0.49 ± 0.08, P < 0.001; and IDI = 0.06 ± 0.01, P < 0.001) and the association of pre- and postoperative cTnI (AUCROC= 0.76 ± 0.04, P < 0.001; NRI = 0.49 ± 0.08, P < 0.001; and IDI = 0.07 ± 0.01, P < 0.001) markedly increased the discriminative power of the models. Further, although no significant difference was observed between AUCROC(fig. 2C), preoperative cTnI remained slightly predictive of MACE when postoperative cTnI was already included in the model (NRI = 0.23 ± 0.08, P = 0.004 and IDI = 0.008 ± 0.003, P = 0.02). Table 5presents the magnitude of this improvement in terms of reclassification.
When intra- and postoperative variables were considered to predict in-hospital mortality (fig. 2D–F), preoperative cTnI did not remain predictive of in-hospital mortality (AUCROC= 0.71 ± 0.04, P = 0.67; NRI = 0.32 ± 0.22, P = 0.16; and IDI = 0.03 ± 0.02, P = 0.10), whereas postoperative cTnI (AUCROC= 0.75 ± 0.04, P = 0.08; NRI = 0.48 ± 0.23, P = 0.03; and IDI = 0.02 ± 0.02, P = 0.23) and the association of pre- and postoperative cTnI (AUCROC= 0.76 ± 0.04, P = 0.13; NRI = 0.49 ± 0.23, P = 0.03; and IDI = 0.09 ± 0.05, P = 0.08) appeared to slightly increase the predictive value of the models. Nevertheless, against MACEs, neither AUCROCcomparison (fig. 2F) nor NRI and IDI analyses presented a significant difference in the predictive abilities of preoperative cTnI when postoperative cTnI was already included in the model (NRI = 0.32 ± 0.23, P = 0.16 and IDI = 0.06 ± 0.05, P = 0.18). A reclassification table using empirical stratification of in-hospital mortality risk is shown in table 6.
Discussion
The main findings of the current study are the following: (1) the discriminative power of EuroSCORE alone in predicting both in-hospital mortality and nonfatal severe cardiac morbidity is moderate and inferior to postoperative serum cTnI concentration alone; (2) the diagnostic value of preoperative cTnI is limited and probably useless for routine cardiac surgical practice; and (3) whatever the selected endpoint, EuroSCORE and postoperative serum cTnI when used in combination significantly improve early identification of high-risk patients after conventional cardiac surgery with CPB.
The EuroSCORE has been validated as one of the most robust multifactor risk scores for cardiac surgical patients,12and princeps studies showed good discriminative power in predicting in-hospital mortality.24,25Less convincing results, however, have been reported concerning length of stay and nonfatal cardiac morbidity.26In the current study, we found a quite moderate discrimination of EuroSCORE alone in predicting both in-hospital mortality and severe cardiac morbidity. An explanation could be that our cohort of patients was possibly different from those initially described. Standards of care at the current time are also somewhat different. Moreover, the profile of cardiac surgical patients has changed over the past 10 yr, and preoperative evaluation has markedly progressed with the use of major chronic therapies, such as statins and/or oral antiplatelet agents.17,27Subsequently, the role of intraoperative events in the occurrence of postoperative complications and adverse outcome might be reinforced. It is noteworthy that our models, taking into account intra- and early postoperative events in addition with EuroSCORE, were more accurate predictors of adverse outcome than EuroSCORE alone. Finally, the composite aspect of severe cardiac morbidity might also explain the disappointing results reported with EuroSCORE.
The prognostic value of preoperative cTnI has been scarcely reported in the cardiac surgical setting.7,8In the current study, the discrimination of preoperative cTnI alone in predicting either mortality or severe cardiac morbidity was of limited value. Most preoperative cTnI concentrations were in the normal range, and, for a given patient, an increased concentration in preoperative cTnI might be associated with a low postoperative cTnI concentration. Indeed, a coronary patient suffering from preoperative unstable angina can be immediately improved by surgery and then experience an uneventful outcome. Moreover, neither the combination of EuroSCORE and preoperative cTnI, compared with EuroSCORE alone, nor the association of pre- and postoperative cTnI combined with EuroSCORE, compared with postoperative cTnI combined with EuroSCORE, dramatically improved the prediction of fatal and nonfatal adverse outcome after cardiac surgery, thereby suggesting that the prognosis value of preoperative serum cTnI in cardiac surgery is controversial and probably useless for routine practice.
Postoperative serum cTnI has been strongly associated with short- and long-term outcome after conventional cardiac surgery in well-designed studies1,4and in a recent meta-analysis.6Moreover, a single postoperative 24-h measurement reliably predicts in-hospital adverse outcome and can be recommended in clinical routine practice as an easy, objective, and cost-effective tool for the detection of high-risk patients.18Unfortunately, the accuracy of postoperative cTnI in predicting poor outcome may be different among cardiac surgical procedures (being less in valve than in coronary surgery), thereby necessitating to consider different thresholds for accurate identification of high-risk patients.2In the current study, we found results in accordance with previous reports. Moreover, the combination of additive EuroSCORE and postoperative serum cTnI significantly improved the ability to detect postoperative high-risk patients. A single retrospective study previously compared the prognostic value of EuroSCORE and postoperative cTnI in combination to predict perioperative death after cardiac surgery and found similar results.11In order to quantify the improvement in model performance introduced by adding a new variable (cTnI) to an existing clinical model (EuroSCORE), we used both nonparametric AUCROCcomparisons and other modern methods (reclassification measures). Although the comparison of AUCROCremains the most popular metric to capture discrimination, it appears that for models containing clinical risk and possessing reasonably good discrimination, very important associations between the biomarker and the endpoint are required to provide significantly different AUCROC.28,29In other words, comparisons of AUCROCmight be considered as being powerless in identifying biomarkers of interest in such situations. To address this problem, new ways of evaluating the usefulness of biomarkers have been described. The methods used in this study are based on the quantification of the reassignment of subjects into adapted risk categories. Thus, NRI and IDI provide a more powerful evaluation of the value of serum cTnI than AUCROCcomparison.
Further studies are, however, needed to assess the possible clinical consequences of the identification of high-risk patients and to test appropriate strategies to improve outcome in prospective interventional trials. Identifying a patient as high risk may either indicate the occurrence of an adverse outcome (i .e ., insufficient cardiac protection, inappropriate graft procedure, and postoperative complication) or only a more severe condition, with these two main hypotheses being not exclusive. Both of them may need therapeutic intensification to improve final outcome. Moreover, it has been recently demonstrated that β-blockers may have either beneficial or deleterious effects,30depending on risk stratification of the patients in whom this therapeutic is used. Therefore, future clinical trials testing therapeutic intensification should probably be designed in a well-stratified risk population.
Some comments are necessary concerning the limitations of the current study. First, the study was conducted in a single center. Although we used a very efficient internal method to validate our model, an external validation using other cohorts provided by other centers is mandatory. Second, we only used an additive model of EuroSCORE, in accordance with clinical routine practice in this institution. A logistic model could have been more discriminating for both mortality and cardiac morbidity, especially in high-risk patients.31However, the logistic model is less easy to calculate at the bedside and its discrimination for a composite operative risk remains questionable.32In the same way, we did not compare cTnI with other existing clinical risk scores. Last, we employed a traditional biologic approach by using cTnI alone. We recently demonstrated that an integrating approach measuring postoperative multiple cardiac biomarkers associating cTnI, B-type natriuretic peptide, and C-reactive protein improved the risk assessment of cardiac adverse outcome within 12 months after elective cardiac surgery.33Further studies should compare pre- and postoperative multiple markers strategies, in addition with clinical risk scores, to better identify high-risk patients on a short- and long-term basis and definitely clarify what cardiac biomarkers may add to existing methods of risk stratification in the cardiac surgical setting.
In conclusion, the use of EuroSCORE alone at the bedside as a clinical predictive model of individual poor outcome after conventional cardiac surgery has proved disappointing at the current time. Preoperative serum cTnI is of limited value in the cardiac surgical setting and probably useless for routine clinical practice. The combination of EuroSCORE and a single postoperative cTnI 24-h measurement provides the best discriminative power and performance in predicting both mortality and severe cardiac morbidity after adult cardiac surgery, and could be proposed as an effective combined model that improves early identification of high-risk patients.
The authors thank David Baker, D.M., F.R.C.A. (Emeritus Consultant Anesthesiologist, Department of Anesthesiology and Critical Care, Centre Hospitalier Universitaire Necker-Enfants Malades, Paris, France), for reviewing the manuscript.