We made a cohort of CO poisoning and included all consecutive patients with CO exposure visiting the ED in Chungnam National University Hospital, prospectively. The diagnosis of CO poisoning was done with the presence of compatible clinical history of CO exposure and an elevated serum carboxy-hemoglobin level at the time of admission. All patients followed the same protocol to find out the presence of LVSD. Baseline echocardiography was scheduled within the first 6 hours in the emergency department. The mean interval from ED visit to taking echocardiography was 5.2±4.2 hours. Their clinical variables were obtained from patients' interviews and their medical records. If there was a patient with LVSD, the patient was admitted to the intensive care unit to observe clinical course and managed with standard medical care for LVSD including angiotensin converting enzyme inhibitor/angiotensin receptor blocker, and beta-blocker. Diuretics were used in cases with pulmonary edema. The presence of recovery of LVSD were checked by the follow-up echocardiography daily for 3 to 5 days. If there was no recovery of LVSD or presence of regional wall motion abnormality, we recommended taking the coronary angiography to find out significant coronary arterial disease. The study protocol has been reviewed and approved by the institutional review board of Chungnam National University Hospital (IRB no 2013-12-001).

Venous blood samples were taken and the plasma was assayed for N-terminal pro-B type natriuretic peptide (NT pro-BNP) using an established radioimmunoassay. Creatinine kinase (CK), CK-MB, troponin-I, and lactic acid levels were also analyzed together. The carboxyhemoglobin (CO-Hb) level was measured by a CO-oxymeter. An electrocardiography was performed in all patients at the time of arrival with commercially available electrocardiography machines. An expert radiologist (S.S. Kim) who was unaware of other clinical features of the participants confirmed the presence of pulmonary edema on a chest X-ray.

Standard 2 dimensional echocardiographic examinations with Doppler studies were performed by expert echocardiographers using Vivid 7 or E9 echocardiographic machines with an M4S probe (GE Vingmed, Horten, Norway) and stored digitally. The echocardiograms were systematically reviewed and measured by 2 experienced cardiologists (J-H. Park and B.J. Sun). The 2- and 4-chamber images were used to calculate LV end-diastolic and end-systolic volumes, and LV ejection fraction (LVEF) using modified Simpson's biplane method.8) The presence of LVSD was defined by LVEF less than 50% assessed by biplane Simpson's method. Regional wall motion abnormality (RWMA) was assessed visually by two well-trained cardiologists.

Categorical variables are presented as a frequency and percentage. Continuous variables are presented as mean±standard deviation (SD) or median with interquartile range. Categorical variables were compared with χ² test and continuous variables were compared with Student's t-test. Numerical variables were made binary by the use of cut-points with the goal of defining a simple, easily implemented scoring method. Thresholds for categorization of these numeric variables were specified using generally accepted criteria or best cut-off values from receiver operator characteristic (ROC) curves. A binary logistic regression was used in the prediction of LVSD and a multivariable analysis was performed on parameters that were significant on the univariate analysis. We performed a multivariate analysis with bootstrapping to prevent overfitting of the model.

The clinical scoring with the lowest regression coefficient among 5 variables in the model was assigned a numeric value of 2, and other 4 variables were assigned scores based on the values of their regression coefficients, relative to that of the lowest value. Thus, each of the binary clinical parameters was assigned a score based on the regression coefficient in the multivariable model. This clinical score was derived by summation of the assigned numeric score. The optimal cutoff value of the clinical score for predicting LVSD was determined by the ROC curve analysis. Comparison of areas under the curve (AUC) of the clinical score and other single variables was done with the method suggested by Hanley and McNeil.9) A two-tailed value of p<0.05 was considered significant. All statistical analyses were performed using SPSS 20.0 (IBM Inc., Chicago, IL, USA), MedCalc (version 12.3.0.0, MedCalc Software, Mariakerke, Belgium) and with the assistance of a professional statistician (I.S. Kwon) from our institution.

We enrolled a total of 81 patients (48 men, 47±19 years old) with CO exposure who were admitted to our ED from May 2009 to June 2015. The mean carboxy-hemoglobin level was 23.8±14.8%. The etiology of intoxication was intentional in 41 patients (51%), and 36 patients (44.4%) showed altered mentality. Their baseline characteristics are presented in Table 1.

Twenty five of them (31%) showed LVSD and 9 had RWMA. Comparison of clinical parameters according to the presence of LVSD was summarized in Table 2. In the LVSD group, initial heart rate, troponin-I, NT pro-BNP, and serum lactate level were significantly higher. Also, pulmonary edema was more prevalent in the LVSD group. Due to the fact that patients without LVSD were discharged in the ED after receiving emergent management, the hospital duration was significantly longer in the LVSD group (5.1±3.6 vs. 1.3±1.9 days, p<0.001). A follow-up echocardiographic examination was completed for 21 patients, but were unavailable for 4 patients (1 died and 3 refused admission). Of the 21 patients with a completed echocardiographic examination, 18 patients showed complete recovery of LV systolic function in about 3 days (mean 2.8±1.7 days). There were 9 patients with RWMA; Six patients demonstrated complete recovery within 7 days. The remaining 3 patients did not show recovery of RWMA and had to undergo coronary angiography; 2 of them had significant coronary arterial stenosis, which was treated with the coronary artery stenting. The clinical outcomes of this study population are summarized in the Fig. 1.

In the LVSD group, 4 patients also had rhabdomyolysis upon admission. One of them was temporarily treated with continuous renal replacement therapy. Also, there were 7 patients with delayed onset encephalopathy associated with CO poisoning.

We selected clinical variables based on univariate analysis according to the presence of LVSD (Table 2). Pulmonary edema on an initial chest X-ray showed statistical significance (HR=10.50, 95% confidence interval [CI]=1.997-53.20, p=0.005). The optimal cut-off values of continuous variables were calculated with ROC curve analyses; initial heart rate >100/min (sensitivity=80%, specificity=61%, AUC=0.785, 95% CI=0.680-0.869, p<0.001), NT pro-BNP >100 pg/mL (sensitivity=80%, specificity=68%, AUC=0.789, 95% CI=0.684-0.872, p<0.001), troponin-I >0.1 ng/mL(sensitivity=84%, specificity=75%, AUC=0.867, 95% CI=0.773-0.932, p<0.001) and serum lactic acid >4.0 mg/dL(sensitivity=75%, specificity=71%, AUC=0.740, 95% CI=0.550-0.883, p=0.001). After the multivariate analysis in the prediction of LVSD with 5 clinical parameters, each was assigned a numeric value based on its relative effect in 5 variables with statistical significance. Accordingly, heart rate (>100/min) was assigned a numeric value of 2; lactic acid >4.0 mg/dL, troponin-I >0.1 ng/mL, and NT pro-BNP >100 pg/mL were assigned a value of 3 each; presence of pulmonary edema, a value of 5 (Table 3). A clinical score was constructed by adding the numeric values of the factors identified in each patient and ranged from 0 to 16. Using ROC curve analysis for the prediction of LVSD, a total score of ≥8 showed optimal sensitivity (84%) and specificity (91%) (Fig. 2). The area under the curve of our clinical score was significantly higher than those of other clinical variables (Table 4).

There were several limitations in our study. First, a limited number of patients were included only from one hospital. Moreover, there were only 25 patients (31%) with LVSD. This number was insufficient for the multivariate analysis. We performed a multivariate analysis with bootstrapping to prevent overfitting of the model.

Second, most of our study population were transferred to our hospital using the emergency medical system. Use of high concentrations of oxygen during their transport could have influenced measurements o f C O-Hb l evels. A lso, t here w as a possibility that some patients with severe cardiotoxicity after CO exposure might have expired as consequences of CO poisoning. So, we could have a selection bias in our study cohort. However, our cohort could also represent a real clinical practice. Finally, we could not evaluate the long term clinical outcome of CO-induced cardiotoxicity. To overcome these limitations, large multicenter studies, including more comprehensive laboratory, imaging and other assessments are needed.

We found that about 31% of patients showed LVSD associated with CO poisoning, and most of them showed complete recovery within about 3 days. The presence of LVSD was associated with initial heart rate >100/min, presence of pulmonary edema on chest X-ray, NT pro-BNP >100 pg/mL, troponin-I >0.1 ng/mL, and lactic acid >4.0 mg/dL. We made a clinical scoring with these variables to predict LVSD, which may be helpful in detecting LVSD in CO-poisoned patients, especially in small ED without echocardiographic support. Patients with a higher clinical score (≥8) might have LVSD, which can be detected with an echocardiograph.