Despite advances in its medical and surgical treatment, infective endocarditis (IE) remains associated with severe complications and high mortality. The presentation and course of IE are highly variable, depending on host factors (preexisting heart disease, prosthetic valve, implantable cardiac device), the causative organism, and the adequacy of treatment (antibiotics, surgery). The interaction of these factors results in an in-hospital mortality rate for patients with IE ranging from 15 to 30%. ¹ Surgical treatment is required in approximately half of patients with IE due to severe complications. Reasons for considering surgery early in the active phase (while the patient is still receiving antibiotic treatment) are to avoid progressive heart failure and irreversible structural damage caused by severe infection, as well as to prevent systemic embolization. On the other hand, surgical therapy during the active phase of IE is associated with significant risk. ²

Prognostic scores are a reasonable estimate of the risk of death, which is important in clinical decision-making regarding indications for surgery. Estimates are needed to inform patients and their families about surgical risks, and risk stratification allows for a fair comparison of cardiac surgery outcomes, so that surgeons and hospitals treating high-risk patients do not appear worse off than others. ³ The performance of traditional surgical scores (EuroSCORE and EuroSCORE II) and IE-specific scores (STS-IE, PALSUSE, AEPEI, EndoSCORE, RISK-E) in predicting surgical risk in patients undergoing cardiac surgery for IE was recently evaluated in a Brazilian cohort. ⁴ In that study, the best performer in terms of predicting mortality risk was the EuroSCORE when taking both discrimination power and calibration (observed-to-estimated mortality ratio) into account. However, previous work showed conflicting results, with the EuroSCORE underperforming compared to an IE-specific score (STS-IE). ⁵ These differences in the literature support the hypothesis that, to date, no single score has proven to be ideal to identify patients with IE at greater risk for in-hospital mortality.

The SHARPEN score was developed to predict in-hospital mortality in hospitalizations secondary to IE in patients undergoing or not undergoing cardiac surgery. ⁶ Unlike other scores, the SHARPEN score is composed of clinical variables that are readily obtained, which could facilitate its use in clinical practice. Analysis of SHARPEN in a Brazilian cohort in a tertiary care center showed performance similar to that described in the original cohort. ⁷ Within this context, the objective of the present study was to evaluate the SHARPEN score as a predictor of in-hospital mortality in hospitalized patients with IE who underwent cardiac surgery, and to compare it with traditional and IE-specific surgical scores – an analysis not performed in the literature yet.

Retrospective cohort study including all admissions between 2007 and 2016 of patients aged ≥18 years who underwent cardiac surgery due to active IE at Hospital de Clínicas de Porto Alegre (HCPA), a tertiary public teaching hospital in southern Brazil. Only patients with definitive IE, diagnosed according to the modified Duke criteria, ⁸ were included; those whose electronic medical records were unavailable were excluded. Patients were identified from the surgical booking system and by a keyword search in the electronic medical record system of HCPA. The present study was approved by the institutional Research Ethics Committee (protocol no. 16-0632).

The SHARPEN score ⁶ was calculated for each IE admission. Depending on the calculated score, admissions were classified as low- (2 to 6 points), moderate- (7 to 10 points), or high-risk (11 to 20 points). The parameters of interest were:

The performance of the SHARPEN score was compared with that of the logistic EuroSCORE ¹⁰ and EuroSCORE II, ¹¹ as well as the IE-specific STS-IE, ¹² PALSUSE, ¹³ AEPEI, ¹⁴ EndoSCORE, ¹⁵ and RISK-E ¹⁶ scores. Any death during hospitalization, regardless of length of stay, was defined as in-hospital mortality. Creatinine clearance (CrCl) was estimated using the Cockcroft–Gault formula. ¹⁷

Preoperative critical status was defined as the presence of one of the following during the same hospitalization: ventricular tachycardia/fibrillation or aborted sudden death, cardiopulmonary resuscitation, mechanical ventilation before induction of anesthesia, administration of inotropic agents, use of intra-aortic balloon pump/ventricular assist device before induction of anesthesia, or acute kidney injury (anuria or oliguria [urinary output <10 mL/h]). ¹¹ Active IE (still on antibiotics at the time of surgery), chronic lung disease, extracardiac arteriopathy, reduced mobility (severe mobility impairment secondary to neuro-musculoskeletal dysfunction), recent myocardial infarction (≤ 90 days), severe pulmonary artery hypertension (systolic pulmonary artery pressure >55 mmHg), severe kidney injury (CrCl <50 mL/min), and urgency of surgery were also defined according to the EuroSCORE II criteria. ¹¹

We studied 105 hospitalizations of 101 patients (four patients had two hospitalizations each) for active IE, who underwent cardiac surgery, between 2007 and 2016. Of the 107 hospitalizations initially retrieved, two (1.9%) were excluded due to unavailability or inaccessibility to the electronic medical record. Characteristics of the sample are described in Table 1 . Hemodialysis before surgery was performed in 22 hospitalizations (21.0%): 14 (13.3%) for chronic kidney disease (CKD), six (5.7%) for acute kidney injury, and two (1.9%) for acute-on-CKD.

Table 1

Variable	n = 105
Age (years)	57.4 ± 14.6
Male sex	79 (75.2)
Hypertension	58 (55.2)
NYHA functional class III/IV	52 (49.5)
Valve abscess	39 (37.1)
Previous heart surgery	33 (31.4)
Severe PAH (>55mmHg)	30 (28.6)
Prosthetic valve IE	29 (27.6)
Severe renal failure (CrCl <50 mL/min)*	23 (21.9)
Hemodialysis	22 (21.0)
Preoperative critical status	20 (19.0)
Thrombocytopenia	19 (18.1)
LVEF ≤50%	16 (15.2)
IDDM	13 (12.4)
Previous IE	10 (9.5)
Extracardiac arteriopathy	8 (7.6)
Previous MI	8 (7.6)
Reduced mobility	7 (6.7)
Chronic lung disease	6 (5.7)
Recent MI	3 (2.9)
CCS class IV angina	1 (1.0)
Site of IE
Aortic valve	45 (42.9)
Mitral valve	35 (33.3)
Aortic + mitral valves	20 (19.0)
Tricuspid valve	4 (3.8)
Tricuspid + mitral valves	1 (1.0)
Causative pathogen identified	70 (66.7)
Streptococcus viridans	19 (18.1)
Enterococcus sp.	10 (9.5)
Staphylococcus aureus	9 (8.6)
Urgency of procedure
Urgent	96 (91.4)
Emergent	9 (8.6)
CABG	8 (7.6)
On-pump time (min)	83 (63-110)
Ischemic time (min)	64 (51-84)
Logistic EuroSCORE	14.8 (8.1-29.5)
EuroSCORE II	6.9 (3.3-13.6)

CABG: coronary artery bypass grafting; CCS: Canadian Cardiovascular Society; HF: heart failure; IDDM: insulin-dependent diabetes mellitus; IE: infective endocarditis; LVEF: left ventricular ejection fraction; MI: myocardial infarction; min: minutes; NYHA: New York Heart Association; PAH: pulmonary artery hypertension; IV: intravenous. * Patients undergoing hemodialysis preoperatively (n = 22; 21.0%) and those for whom body weight data were not available (n = 11; 10.5%) were excluded which made it impossible to calculate the CrCl. Data are expressed as n (%), mean ± standard deviation, or median (interquartile range).

The leading indication for surgery was heart failure (n = 49; 46.7%); the main procedures performed were mechanical aortic valve replacement (n = 24; 22.9%), biological aortic valve replacement (n = 22; 21.0%), and biological mitral valve replacement (n = 22; 21.0%). The median (IQR) time between prescription of antibiotics and surgery was 13 (6-22) days.

The median (IQR) SHARPEN score was 11 (9-13) points. The prevalence of each score component is described in detail in Table 2 . Overall, 10 admissions (9.5%) were characterized as low-risk, 40 (38.1%) as moderate-risk, and 55 (52.4%) as high-risk.

Table 2

Acronym	Score component	Number (%)
S	SBP <90mmHg (at presentation)	11 (10.5)
H	Heart failure* (during hospitalization)	78 (74.3)
A	Age (years)
	<50 years	29 (27.6)
	50-65 years	41 (39.0)
	>65 years	35 (33.3)
R	Creatinine >2.26 mg/dL (at admission)	23 (21.9)
P	Pneumonia (≥48 hours after admission)	20 (19.0)
E	CRP >200 mg/L (during hospitalization)	36 (34.3)
N	Non-IV drug user	100 (95.2)

IV: intravenous; SBP: systolic blood pressure; CRP: C-reactive protein. *According to Framingham criteria. ⁹

The observed in-hospital mortality was 29.5% (95%CI: 20.8-38.2%); 20.0% in admissions stratified as low or moderate risk and 38.1% in high-risk admissions (P = 0.068). Low- and moderate-risk admissions were pooled for this analysis due to the limited representativeness of the former.

There was no statistically significant difference in observed vs. estimated mortality in the overall analysis, nor in moderate-risk hospitalizations. Conversely, the observed mortality for low-risk hospitalizations was higher than estimated, while that of high-risk hospitalizations was lower than estimated ( Figure 1 ). There was wide variability in estimated mortality among the analyzed scores ( Figure 2 ). Besides the SHARPEN score, only EuroSCORE, PALSUSE, and RISK-E did not present a statistically significant difference between observed and estimated mortality ( Table 3 ).

Figure 1 CI: confidence interval; O/E: observed/estimated; Pts: points.

Figure 2 IE: infective endocarditis

Table 3

Score	O/E Mortality*	95%CI	P	AUC-ROC	95%CI	P
SHARPEN	0.77	0.54-1.09	0.147	0.66	0.54-0.79	0.008
Nonspecific scores
EuroSCORE	1.37	0.94-1.91	0.094	0.77	0.67-0.87	<0.001
EuroSCORE II	2.56	1.77-3.59	<0.001	0.70	0.59-0.81	0.001
IE-specific scores
PALSUSE	1.04	0.72-1.46	0.790	0.71	0.60-0.82	0.001
RISK-E	1.20	0.82-1.71	0.320	0.71	0.61-0.82	0.001
AEPEI	1.76	1.22-2.47	0.004	0.65	0.53-0.77	0.014
STS-IE	2.58	1.79-3.62	<0.001	0.68	0.57-0.80	0.003
EndoSCORE	2.98	2.06-4.18	<0.001	0.77	0.66-0.87	<0.001

AUC-ROC: area under the ROC curve; CI: confidence interval; IE: infective endocarditis; O/E: observed-to-estimated mortality; STS-IE: IE-specific score. *Observed mortality was 29.5%, except for the RISK-E, which was 29.0%: five cases of pulmonary/tricuspid IE were excluded, as they are not included in the analysis of this score

The AUC-ROC for the SHARPEN score was 0.66 (95%CI: 0.54-0.79; P = 0.008) ( Figure 3 ). In comparison with the other scores ( Table 3 ), there was no significant difference in discrimination, with only a trend for discriminative ability to be lower than that of the logistic EuroSCORE (P = 0.098) and EndoSCORE (P = 0.110).

Figure 3

Statistics for the SHARPEN score, and for the other scores analyzed, are shown in the Central Figure. An increase in positive predictive value (PPV) and in accuracy can be observed when adopting a cutoff of >12 points, defined according to Youden’s J statistic, instead of the threshold used in the original study (>10 points). Using the original threshold, mortality was significantly higher (51.7 vs. 21.0%, P = 0.004). However, the CIs of these statistics – as well as those of the negative predictive value (NPV) – overlap, indicating absence of a statistically significant difference between them.

In this cohort of patients, the SHARPEN score was calibrated (P = 0.147), with no statistically significant difference in discriminative power (AUC-ROC = 0.66; P = 0.008) in relation to the other tested scores. Likewise, the PPV, NPV, and accuracy of SHARPEN were not significantly different from those of the other scores included in this analysis. Compared to the original cohort, ⁶ there was a difference in O/E mortality ratio between the low- and high-risk groups, especially in the former (O/E: 10.3; P = 0.003). This difference may be explained by the fact that, while in the original study the proportion of patients undergoing cardiac surgery was 38.3%, our cohort was composed exclusively of surgical patients.

The possibility of using a tool that yields reliable prognostic information based on clinical variables alone is a potential advantage of this new score. In addition, these are objective variables, hence easy to obtain and standardize, which facilitates outcome measurement across different populations. In addition, the SHARPEN score does not include microbiological variables (which may be affected by issues such as difficult confirmation or slow-growing pathogens), unlike other STS-IEs, such as the STS-IE, ¹² PALSUSE, ¹³ EndoSCORE, ¹⁵ and RISK-E. ¹⁶ However, as it incorporates two variables that can only be identified late in the admission (nosocomial pneumonia and peak C-reactive protein >200 mg/L), it may have limited value in early risk estimation.

In the original study cohort including 233 patients with an in-hospital mortality of 23.2%, of which only 51 (21.9%) underwent cardiac surgery, the score showed an AUC-ROC of 0.86 (95%CI: 0.80-0.91). ⁶ In the only study that validated the SHARPEN score in Brazil, ⁷ including 179 hospitalizations for IE in a public hospital, with an in-hospital mortality rate of 22.3% and cardiac surgery performed in 68 patients (38.0%), the score showed an AUC-ROC of 0.76 (95%CI: 0.67-0.85); 0.77 when only clinical treatment was performed and 0.72 in those requiring cardiac surgery. In this study, there was no analysis of specific mortality in the high-risk group (>10 points).

A key strength of the present analysis was the evaluation of the score performance in a cohort of exclusively hospitalizations in which cardiac surgery was performed, unlike the original cohort ⁶ and the subsequent Brazilian study that also evaluated the performance of the score. ⁷ Furthermore, while in the original study only 26.2% of admissions were categorized as high risk, in the present series they represented 52.4% of the total. On the other hand, while the feared S. aureus was the main etiologic agent in the original cohort (48.1%), in this cohort it was the cause of only 8.6% of hospitalizations. In the present study, we observed a nonsignificant increase in score accuracy when we adopted a higher cutoff point, raising the hypothesis that, perhaps, for high-risk surgical patients, the cutoff point should be higher; this will need to be confirmed in future cohorts.

Unlike previous studies and comparative studies, we chose to analyze the score statistics starting from a defined cutoff point, as we believe this information is more useful in daily clinical practice, and then define the best threshold through the optimal sensitivity and specificity ratio (Youden’s index). Using the AUC-ROC allowed us to determine the overall accuracy of the score based on the analysis of all points in the score. Given a specific score, and the score of a given patient, patients can be placed in low- or high-risk groups, yielding an estimate (within a 95%CI) of the associated mortality.

The present study has limitations. The sample was relatively small and restricted to a single tertiary care center. The long period of analysis means that both clinical and surgical management of these patients may have changed over time. Finally, retrospective data collection can compromise the quality of the data obtained.

Several research groups have been searched for the optimal prognostic score for risk stratification in cardiac surgery in IE, although no one score has proven superior to others for this purpose. ¹⁸ The findings obtained in the present cohort corroborate this statement, since none of the tested scores could be considered ideal, with AUC-ROCs considered reasonable at best (0.7 < AUC-ROC ≤ 0.8), with maximum accuracy of 75.2%. Specifically in relation to the SHARPEN score, its properties did not differ significantly from those of the other analyzed scores; despite the easy obtainment of its few variables, it has limited applicability in clinical practice, like other existing scores.