Predictive Nomogram of the in-Hospital Mortality of Patients with Cardiogenic Shock

transcutaneous pacing.

4.66), and renal replacement therapy (OR 3.08) were associated with higher in-hospital mortality. These factors were used to develop a risk-predictive nomogram. A receiver operating characteristic (ROC) curve was drawn using scores from 0 to 198.5 (area under the curve [AUC] = 0.92), and the cut-off score of 65 was identi ed. The model accurately predicted mortality (AUC = 0.81) in a validation set of 68 patients, with a sensitivity of 81% and a speci city of 74%.

Conclusion
Factors were used to develop a highly accurate risk-predictive nomogram for the in-hospital mortality of patients with CS, which may aid physicians in making prognoses.

Background
Cardiogenic shock (CS) is the most severe form of acute heart failure. 1 It can be described as a state of ineffective cardiac output, which results in clinical and biochemical manifestations of inadequate tissue perfusion. 1, 2 CS is a leading cause of mortality after myocardial infarction (MI) and complicates up to 10% of cases of acute myocardial infarction (AMI). 3 It is more common in patients with ST-elevation myocardial infarctions (STEMI) than in those with non-ST-elevation myocardial infarctions (NSTEMI). 3,4 Despite advances in treatment options, CS mortality remains high at approximately 35 to 50 percent. 1,3,5 Several risk scores that help predict short-term mortality have been established. The 'SHould we emergently revascularize Occluded coronaries for Cardiogenic shocK?" (SHOCK) trial score had a large registry (n = 1217) and emphasized the importance of early coronary revascularization in MI patients with shock. 6 The SHOCK measure employs a two-stage process to identify variables that can predict short-term mortality, including age, shock on admission, evidence of end-organ hypoperfusion, anoxic brain damage, systolic blood pressure, prior coronary artery bypass grafting (CABG), noninferior MI, creatinine level, stroke work, and left ventricular ejection fraction (LVEF) of less than 28%. 6 The model had an area under the curve (AUC) of 0.74 but has not been validated using an independent data set. 6 The Intra-aortic Balloon Pump in Cardiogenic Shock II (IABP-SHOCK II) trial is the largest randomized trial to date (n = 480); however, only patients with MI were included. 7 Similar to the SHOCK trial, age and creatinine were included in the risk score, among other factors, such as prior stroke, glucose level, blood lactate level, and thrombolysis in myocardial infarction (TIMI) ow grading. 7 The model had a fair predictive performance (AUC = 0.79). 7 Furthermore, internal validation using data from the IABP-SHOCK II registry patients (AUC = 0.79) and external validation based on the CardShock population (AUC = 0.73) were performed. 7 The CardShock study is by far the largest prospective, observational study (n = 219) to include patients with all etiologies of CS. 8 Age, previous MI or CABG, reduced LVEF, high lactate level, and confusion at presentation (presence of systemic hypoperfusion) were also used as predictors of mortality, as were acute coronary syndrome (ACS) etiology and estimated glomerular ltration rate (eGFR). 8 This model had a good predictive power (AUC = 0.85) and the IABP-SHOCK II trial patients were used for external validation (AUC = 0.71). 8 A comparative study of these risk scores revealed modest prognostic accuracy, which indicated the need for novel methods of risk prediction. 9 Furthermore, these scores were largely based on Western populations, and their applicability to different populations had not been tested. 9 Zhang et al. developed a predictive model of the risk of developing CS in Chinese patients; however, the primary outcome was not mortality and the population was limited to patients with AMI, instead of those with the entire spectrum of related etiologies. 10 The predictive factors included age, sex, body mass index (BMI), Killip class, mean arterial pressure (MAP), heart rate, previous coronary disease, the outcome of thrombolytic therapy, and MI location. 10 The patient characteristics included in the aforementioned models were mostly from the medical history and biochemistry results. The variance in these risk factors illustrates the heterogeneity of CS presentations. Furthermore, a patient's clinical course is often nonlinear, and thus factors throughout their hospitalization should be explored. 11 Therefore, this study aimed to develop a risk-predictive model of inhospital mortality for patients with CS based on their medical history, examination results, and interventions throughout hospitalization to aid physicians in their clinical management and to improve patients' prognoses.

Study population
This retrospective study was approved by the Institutional Review Board of National Taiwan University Hospital (NTUH). This study was conducted in the NTUH, a 2500-bed tertiary hospital with 110,000 annual emergency room (ER) visits. NTUH is located in Taipei City, Taiwan, which has a population density of approximately 10,000 people/km 2 . 12 This study enrolled 509 nontraumatic adult patients (≥ 20 years old) who received inotropic support at the ER, and had subsequent admission to the cardiac care unit (CCU) from January 2017 to December 2019. After the exclusion of 105 patients who had out-of-hospital-cardiac-arrest (OHCA), 48 who received inotropic support for bradycardia and conduction system disorders, and 110 patients with noncardiogenic shock (obstructive, hypovolemic, distributive shock), 246 patients were included in this study to develop the predictive model (Fig. 1).
Another set of 163 patients with the same inclusion criteria who visited the ER from January to August 2020 were enrolled as the validation set (sFigure 1). Patients who presented with OHCA (n = 35), conduction disorders (n = 18), and non-cardiogenic shock (n = 42) were excluded. A validation set of 68 patients was thus used to determine the model's diagnostic accuracy.

Data collection and variable de nitions
The primary outcome of this study was to predict the in-hospital mortality of patients with cardiogenic shock. The following information were collected from the individual medical records: age, sex, body mass index (BMI), smoking history, alcohol intake, preexisting comorbidities, laboratory exams, electrocardiogram (ECG) characteristics, chest X-ray ndings, echocardiogram ndings, medications administered, clinical management, discharge diagnosis, and length of hospital stay. The imaging and laboratory exams nearest to the shock time at the ER were recorded.
Medications and clinical management provided at the ER and throughout the hospitalization were recorded. Inotropic use was de ned as the administration of dopamine, norepinephrine bitartrate, epinephrine, vasopressin, or dobutamine to achieve hemodynamic stability. Cases were recorded either as using only a single inotrope or multiple inotropic agents (two or more). Patients who did not require any supplemental oxygen, nasal cannula, oxygen mask, or nonrebreathing mask were classi ed as requiring low respiratory support; whereas patients who required bilevel positive airway pressure (BiPAP), high ow oxygen therapy, or endotracheal intubation were classi ed as having high respiratory support.
Transcutaneous pacing (TCP) refers to the noninvasive mode of temporary pacing by applying pads to the chest, whereas pacemaker implantation refers to the invasive method of inserting either temporary or permanent pacing devices. Fluid challenge is the infusion of more than 250 mL of a crystalloid to the patient when shock was noted, before administration of any inotropic agent or other interventions. Component transfusion during hospitalization involved the administration of either packed red blood cells, platelets, fresh frozen plasma, or cryoprecipitate. The use of mechanical circulatory support devices include extracorporeal membrane oxygenation (ECMO) or intra-aortic balloon pump (IABP) insertion.
Cardiopulmonary resuscitation (CPR) was performed on patients with cardiac arrest at the ER or during hospitalization until spontaneous circulation (ROSC) was achieved or an ECMO device was attached.
Emergent coronary angiography (CAG) was performed within 24 hours of shock. CABG and renal replacement therapy (RRT) at the ER or during hospitalization were documented. RRT encompassed patients who received dialysis, sustained low-e ciency dialysis (SLED), or continuous veno-venous hemo ltration (CVVH) after manifesting symptoms of uid overload, respiratory distress, or severe electrolyte imbalance.
Discharge diagnoses were recorded based on the judgement of attending physicians. Cases of STEMI, NSTEMI, and post-MI complications were classi ed as ACS; heart failure, valvular heart conditions, and myocarditis were classi ed as acute decompensated heart failure (ADHF); different types of cardiomyopathy (stress, restrictive, hypertrophic, dilated, ischemic) were classi ed as cardiomyopathy; tachyarrhythmia and bradyarrhythmia were classi ed as arrhythmia; CS with sepsis or pneumonia were classi ed as CS with septic complications; and cardiac tamponade, aortic dissection, and pulmonary hypertension were considered as other causes of CS.

Statistical analyses
Results were presented using frequencies for categorical variables and medians with ranges for continuous variables. Group comparisons of outcomes were performed using Fisher's exact or Pearson's chi-squared test for categorical variables and the Mann-Whitney U test for continuous variables. From the training set, 31 independent variables with signi cant associations (p < 0.05) in the univariate analysis were included in the forward multiple logistic regression analysis (p < 0.05 for inclusion and > 0.1 for elimination) to identify the predictors of in-hospital mortality. The selected variables were further used to develop a risk-prediction nomogram to predict in-hospital mortality. A receiver operating characteristic (ROC) curve was drawn using the total scores from the training set and the Youden Index was used to determine the cut-off value for predicting mortality in patients with CS. Patients from the validation set were scored using the nomogram and to test the model's diagnostic accuracy. The predicted mortality was then compared with the observed mortality.
All statistical analyses were performed using IBM SPSS Statistics for Windows, version 16 (IBM Corp, Armonk, NY, USA) and R statistical software version 4.0.2 was used to construct the nomogram.

Clinical characteristics
A comparison of the characteristics of patients in the training set is presented in Table 1. The median age was 70 years and the majority of patients were men (60.16%). In-hospital mortality for patients with CS was at 26.83%. Coronary artery disease, heart failure, cardiomyopathy, and renal disease were determined to be higher in the patients who did not survive to discharge, whereas dyslipidemia was more frequent in the patients who survived. Patients who died during hospitalization were observed to have higher levels of lactic acid, hemoglobin, platelet, international normalized ratio (INR), creatinine, and pro B-type natriuretic peptide (pro-BNP). A higher heart rate and longer QRS duration on ECG ndings were also observed in patients who did not survive to discharge. Chest X-rays revealing lung edema were more frequent in patients who survived, whereas pleural effusion was observed more frequently in those who did not survive. Echocardiograms with LVEF lower than 40%, pericardial effusion, and valvular heart conditions were also observed more frequently in patients who failed to survive. Patients who did not survive had received albumin infusion, multiple inotropic agents, and heparin more frequently. These patients also more frequently required high respiratory support, pacemaker implantation, mechanical circulatory support devices, CABG surgery, component transfusion, and RRT.
Among patients with cardiac arrest following CS, resuscitation efforts were longer for patients who did not survive than for those who did.
The discharge diagnosis classi cation indicated that ACS was perceived as the main etiology of CS, followed by ADHF, arrhythmia, cardiomyopathy, CS with septic complication, and other causes. Patients with CS complicated by arrhythmia and septic complications had higher rates of survival than those who did not. Patients who did not survive had signi cantly longer CCU stays (Table 2).  60-13.60, p = 0.005), and renal replacement therapy (OR 3.08, 95% CI 1.19-8.02, p = 0.021) remain associated with higher in-hospital mortality, and a risk-prediction nomogram was developed accordingly (Fig. 2). A ROC curve was drawn using scores ranging from 0 to 198.5 and the cut-off value to predict mortality was determined to be 65. The nomogram yielded a sensitivity of 92.42%, a speci city of 72.78%, and an AUC of 0.92 (95% CI 0.88-0.96) for the training set. Using the total scores from the validation set, with scores less than 65 deemed as survival to hospital discharge and scores more than 65 deemed as inhospital mortality, the predicted mortalities were compared with the observed mortalities, which revealed a sensitivity of 81.25%, a speci city of 74%, and an AUC of 0.81 (95% CI 0.71-0.92; Fig. 3).

Discussion
In this retrospective observational study that used a training set of 246 patients, six factors (history of CAD, multiple inotropic use, ejection fraction of less than 40%, longer CPR duration, albumin infusion, and RRT) were associated with higher in-hospital mortality. A risk-predictive nomogram was developed and validated using another set of 68 patients. This model is based on simple parameters that account for not only characteristics that can be observed during the initial presentation but also clinical management decisions that can occur throughout the hospitalization period. This model exhibited a more favorable performance than the previously published models.
The SHOCK trial score identi ed a prior CABG among MI patients and the CardShock risk score identi ed prior MI or CABG and ACS etiologies as risk factors in the general population. 6,8 Our ndings did not indicate that a history of MI was a predictor of in-hospital mortality, whereas a history of CAD was.
Furthermore, Wehner et al. 13 reported that LVEF may have a U-shaped relationship with mortality.
However, most studies demonstrated a relatively linear relationship, with lower LVEF being associated with a higher risk of death. 14,15 Our risk-predictive nomogram, along with the SHOCK trial score and CardShock risk score, identi ed reduced ejection fraction as a predictor of mortality. 6,8 CS can manifest signs of end-organ hypoperfusion; therefore, the prevalence of acute kidney injury (AKI) is high among patients with CS and is a strong predictor of 90-day mortality. 16 Patients who require RRT had higher in-hospital mortality than those with AKI who did not undergo dialysis. 17,18 The SHOCK trial score and CardShock risk score included creatinine levels as a predictor of mortality while our multivariate analysis indicated that the need for RRT during hospitalization was a risk factor. 6,8 Predictors identi ed in this study that were not a part of the other scoring systems were multiple inotropic use, longer CPR duration, and albumin infusion. Pharmacologic management with inotropic agents and vasopressors has become a cornerstone of hemodynamic support for patients with shock. 19 Early diagnosis of CS and rapid initiation of therapy can greatly improve patient prognosis. 19 However, inotropes and vasopressors are recommended for short-term treatment at minimum doses when deemed necessary in the absence of alternatives. 20 Studies have presented con icting results regarding their safety and overall bene ts. Tarvasmäki et al. 19 revealed that a combination of vasopressors such as levosimendan with noradrenaline was associated with a more favorable prognosis; however, adrenaline was associated with an increased mortality, whether administered alone or in combination. Other studies have reported that inotropic therapy is not associated with differences in mortality, and the additive effect does not reduce mortality in patients with heart failure. 20,21 The simultaneous administration of more than one inotropic agent to maintain hemodynamic stability was strongly associated with in-hospital mortality in our study.
Severe cases of CS often progress to cardiac arrest, thus requiring resuscitation efforts. 22 Vallabhajosyula et al. 23 reported that among STEMI patients, CS with cardiac arrest was associated with higher rates of noncardiac organ failure and subsequent in-hospital mortality compared with CS or arrest alone. In patients with cardiac arrest, a prolonged CPR duration is associated with a poorer prognosis, as demonstrated in our study. 22,24 The infusion of albumin as a volume expander or as an alternative to using crystalloids for volume replacement is performed in patients with shock and severe sepsis; however, evidence that its use in critically ill patients improves survival is limited. 25,26 Although improvements in hemodynamic instability can be achieved with the administration of albumin, long-term bene ts require further investigation. 25 Among patients with heart failure or ACS, hypoalbuminemia is correlated with higher overall illness severity and low levels of albumin can act as a frailty biomarker; therefore, it is consistently associated with higher mortality rates. 27 Our ndings also indicated that patients who received albumin infusion were at risk of a less favorable prognosis.

Conclusion
The risk factors of a history of CAD, multiple inotropic use, ejection fraction less than 40%, longer CPR duration, albumin infusion, and RRT are associated with higher in-hospital mortality. A risk-predictive nomogram with high diagnostic accuracy was developed accordingly.

Limitations
Several limitations of this study should be considered. The retrospective nature of the study caused unavoidable selection bias toward patients with more severe conditions requiring more medical interventions. The small sample size may have resulted in nonsigni cant differences between groups in some variables. Unrecognized confounding factors may also be present. This study is largely based on an Asian demographic and thus may be more applicable for similar populations. External validation was not performed for the predictive model. The availability of IABP and ECMO without Impella support may differ from the Western practices. As renal replacement therapy is usually not applied on the rst day of hospitalization, this new score may not be useful for the initial evaluation of patients with CS; rather, it will be up to the physician's discretion as to the appropriate time to apply this score for prognostication purposes. Moreover, the primary outcome of this study was in-hospital mortality, and thus the prediction of long-term prognosis after discharge requires future investigation through well-designed studies.

Declarations Ethics Approval and Consent to Participate
Not applicable

Consent for Publication
Not applicable

Availability of Data and Materials
The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.

Competing Interests
The authors declare that there is no con ict of interest, including the relevant nancial interests, activities, relationships, and a liations.  Flowchart of patient enrolment in the training set. CCU, cardiac care unit; ER, emergency room; OHCA, outof-hospital cardiac arrest.

Figure 2
Predictive nomogram of in-hospital mortality in patients with cardiogenic shock. CAD, coronary artery disease; CPR, cardiopulmonary resuscitation; EF, left ventricular ejection fraction.