ABC | Volume 114, Nº3, March 2020

Original Article Pivatto Júnior et al. Risk scores for surgery in endocarditis Arq Bras Cardiol. 2020; 114(3):518-524 with definite IE based on the modified Duke criteria 13 were enrolled. Patients were identified through surgical schedules and a keyword search of the HCPA electronic medical records system. The present study was approved by the HCPA Research Ethics Committee (protocol 16-0632). Preoperative risk of death was estimated through the mean logistic EuroSCORE 8 and EuroSCORE II 9 , in addition to the IE-specific scores STS-IE, 2 PALSUSE, 10 AEPEI, 11 EndoSCORE 7 and RISK-E 12 (Table 1). Death during hospitalization, regardless of length of stay, was defined as hospital mortality. Creatinine clearance (CC) was estimated through the Cockcroft-Gault formula. 14 Acute renal insufficiency was defined as any of the following: increase in creatinine by ≥ 0.3 mg/dL within 48 hours or to ≥ 1.5 times baseline, which is known or presumed to have occurred within the past 7 days; or urinary output < 0.5 mL/kg/h for 6 hours. 15 Critical preoperative state was defined if any one or more of the following occurred preoperatively during the same hospital admission as the operation: ventricular tachycardia/fibrillation or aborted sudden death; cardiac massage; ventilation before arrival at the anesthesia suite; administration of inotropes; intra-aortic balloon counterpulsation/ventricular-assist device placement before arrival at the anesthesia suite or acute renal failure (anuria or oliguria < 10 mL/h). 9 Active IE (still on antibiotic treatment for IE at time of surgery), chronic pulmonary disease, extracardiac arteriopathy, poor mobility (severe impairment of mobility secondary to musculoskeletal or neurological dysfunction), recent myocardial infarction ( ≤ 90 days), severe pulmonary arterial hypertension (systolic pulmonary artery pressure > 55mmHg), severe renal dysfunction (CC < 50mL/min) and urgency of surgery were also defined as in the EuroSCORE II study. 9 Statistical Analysis Data were collected directly from the patients’ electronic charts and analyzed in IBM SPSS Statistics for Windows, version 21.0; MedCalc, version 12.5; and OpenEpi, version 3.01. 16 Qualitative data were reported as absolute and relative frequency; mean (standard deviation) or median (interquartile range) were used for quantitative variables. The normality of distribution of each variable was evaluated using the Shapiro-Wilk test. Calibration (expressed by the observed/ expected [O/E] mortality ratio, i.e., the standard mortality ratio [SMR]) and discriminant ability (by area under the ROC curve [AUC]) of the scores were evaluated. To calculate the SMR with a 95% confidence interval (CI), we used the mid-P exact test with Miettinen’s modification. Comparison of AUC was performed by the DeLong test. P-values < 0.05 were considered statistically significant. Results During the study period, 107 patients underwent cardiac surgery at the study facility while in the acute phase of IE and were included. Mean age was 58.1 ± 14.5 years and 24.3% were female. Isolated aortic IE was the most prevalent form of IE (43.9%). Patient characteristics and surgical details are described in Table 2. The median vegetation size was 14.0 (9.25-18.0) millimeters. Thirty-one patients (29.0%) experienced at least one embolic event, diagnosed on the basis of symptoms or by incidental detection: 13 (12.1%) to the central nervous system and 11 (10.3%) to the spleen. Twenty-two (20.6%) were on preoperative dialysis: 14 (13.1%) due to chronic kidney disease, 6 (5.6%) due to acute renal failure, and 2 (1.9%) due to acute-on-chronic renal failure. Surgery was performed with a median delay of 12.5 (6.0-22.25) days start of antibiotic therapy. The leading indication for surgery was heart failure (76.6%). The most frequently performed procedure was mechanical aortic valve replacement (n = 26, 24.3%), followed by bioprosthetic aortic valve replacement (n = 22, 20.6%) and bioprosthetic mitral valve replacement (n = 22, 20.6%). Overall hospital mortality was 29.0% (95%CI: 20.4-37.6%). There was a wide variation in expected mortality among the scores, ranging from 10.0% in EndoSCORE to 28.6% in PALSUSE score (Figure 1). The best O/E mortality ratio was achieved by the PALSUSE score (1.01, 95%CI: 0.70-1.42; p = 0.919), followed by the EuroSCORE (1.3, 95%CI: 0.92-1.87; p = 0.123), as seen in Table 3. All other scores significantly underestimated hospital mortality. The logistic EuroSCORE had the highest discriminatory power (AUC 0.77), as seen in Table 3, which was significantly superior to that of EuroSCORE II (p = 0.03), STS-IE (p = 0.03), PALSUSE (p = 0.03), AEPEI (p = 0.03), and RISK-E (p = 0.02), and non-significantly so when compared to EndoSCORE (p = 0.90). All other comparisons were non-significant, except for EndoSCORE versus AEPEI score (p = 0.03). Discussion In this cohort of patients undergoing cardiac surgery for active IE, the best O/E mortality ratio and discriminatory power were achieved by the PALSUSE score (1.01) and the logistic EuroSCORE (AUC 0.77), respectively. The logistic EuroSCORE, which had the second best O/E ratio (1.3), also had significantly better discriminatory power than PALSUSE (AUC 0.68; p=0.03). AUC, also known as the c-statistic or c-index, is a marker of overall diagnostic accuracy 17 and an effective and combined measure of sensitivity and specificity. 18 Discriminative power is thought to be excellent if the AUC is > 0.80, very good if > 0.75, and good (acceptable) if > 0.70. We also evaluated calibration using the O/E mortality ratio. Ideally, this ratio will be 1, i.e., the observed mortality equals expected mortality, denoting a perfectly calibrated predictive model. An O/E value > 1 means the model underestimates mortality, while a value < 1 means the model overestimates mortality. If the 95%CI of the O/E mortality ratio crosses 1, the model is well calibrated. 19 Nevertheless, it is possible for a risk model to have good calibration but poor discrimination, and vice versa. Discrimination is more important than calibration; a model can be recalibrated or adjusted as practice improves, but if the model is built on the wrong risk factors, its discrimination cannot be improved. 20 Although the EndoSCORE did not show significantly worse discriminative power than the logistic EuroSCORE, it did significantly underestimate hospital 519