- Email: [email protected]
Predicting Short-term Mortality in Patients With Pulmonary Embolism: A Simple Model
CLINICAL INVESTIGATION
Predicting Short-term Mortality in Patients With Pulmonary Embolism: A Simple Model Andriana I. Papaioannou, MD, PhD, Alcibiades Kastanis, PhD, Foteini Malli, MD, Epameinontas Zakynthinos, MD, PhD, Markos Minas, MD, PhD,* Mariana Mpakarozi, MD, Stamatina Moka, MD, Elias Zintzaras, MSc, PhD, Konstantinos I. Gourgoulianis, MD, PhD and Zoe Daniil, MD, PhD
Background: The aim of the present study was to develop a simple prognostic rule that could classify patients with pulmonary embolism (PE) into categories of increased risk of 30-day mortality. Methods: One hundred patients with PE were enrolled. Clinical and laboratory findings were recorded on admission for each patient. Differences between groups’ survival and death were tested, and the association with the 30day mortality was determined. Results: Three variables had a significant effect on survival: age, Charlson index and the alveolar to arterial (A-a) gradient. A receiver operating characteristic analysis was performed, and the cut-off points used for the comparison of survival were 67 years of age, A-a gradient over 52.8 mm Hg and Charlson index over 2. By combining these variables, a score was established for distinguishing patients with PE who are at high risk. This score was also validated in a group of 30 consecutive patients admitted to the hospital for PE Additionally, a tree method was applied and showed that for patients with a history of diabetes and Charlson index .3, the expected outcome is death. Conclusions: The results of this study suggest that patients with PE could be stratified into categories of increasing risk of 30-day mortality using a simple score based only on routinely available variables. Future studies are needed to validate our prognostic model in a large cohort of patients with PE. Key Indexing Terms: Comorbidities; Congestive heart failure; Diabetes mellitus; Pulmonary embolism; Survival. [Am J Med Sci 2013;345 (6):462–469.]
P
ulmonary embolism (PE) is a major health burden worldwide. A recent epidemiological model derived from 6 European countries with a total population of 310.4 million yielded a PE incidence rate of 98 cases per 100,000 person-years.1 The short-term mortality of the disease varies widely, ranging from less than 2% in patients with nonmassive PE to more than 95% in patients who experience cardiorespiratory arrest.2,3 Although most patients with acute PE remain hospitalized during the initial therapy, it has been suggested that patients presenting without hemodynamic instability and without elevated biomarker levels or imaging findings indicating right ventricular (RV) dysfunction or myocardial injury may constitute a lowrisk group and accordingly it was suggested that these patients may be considered for early discharge and home treatment.4 From the Department of Respiratory Medicine (AIP, FM, MMi, MMp, SM, KIG, ZD); Departments of Biomathematics (AK, EZi) and Critical Care Medicine (EZa), School of Medicine, University of Thessaly, Larissa, Greece; and Institute for Clinical Research and Health Policy Studies (EZi), Tufts Medical Center, Tufts University School of Medicine, Boston, Massachusetts. Submitted February 12, 2012; accepted in revised form June 13, 2012. The authors have no financial or other conflicts of interest to disclose. *Author is deceased. Correspondence: Elias Zintzaras, MSc, PhD, Department of Biomathematics, University of Thessaly School of Medicine, 2 Panepistimiou Str., Biopolis, Larissa 4110, Greece (E-mail: [email protected]).
462
Because a small but significant proportion of patients with acute PE will die or have severe complications during the initial therapy period, reliable prognostic information at presentation would help clinicians to estimate which patients are at low risk and could be discharged early or managed entirely as outpatients and which patients are at high risk and may benefit from a more intensive surveillance.5 Previous models of risk stratification for PE tried to develop a clinical prediction rule for patients with PE that could quantify the risk of mortality across the full spectrum of risk. A prognostic model previously developed by Aujesky et al6 [pulmonary embolism severity index (PESI)] comprises 11 routinely available clinical parameters and accurately identifies patients with PE who are at low risk of fatal or nonfatal outcomes.6,7 Recently, the simplified PESI score was described to provide a more simple model for the identification of patients with PE who are at high risk of death.8 A second model, the “Geneva” prediction rule, created by Wicki et al9 identifies 6 independent predictors, including a venous ultrasound; however, recently published study provided evidence that the Geneva risk stratification does not identify patients with acute PE that have negligible risk for 30-day mortality.9,10 The aim of this study was the evaluation of 30-day mortality in patients with PE confirmed by spiral computed tomography (CT) pulmonary angiography and to develop a fast descriptive rule, relying only on a few readily available clinical parameters that could classify patients into categories of increased risk of mortality and be helpful in guiding initial intensity of treatment.
MATERIALS AND METHODS Patients We identified 100 consecutive patients with PE, who were treated in the Respiratory Medicine Department of the University Hospital of Larissa, Greece, during February 2003 to January 2008. In all patients, the diagnosis of PE was established with spiral CT pulmonary angiography. Patients in a state of hemodynamic instability and patients requiring resuscitation are usually admitted in the intensive care unit (ICU) without a definite diagnosis. Although some of them might have suffered a PE, they were not included in the study. For each patient, all clinical manifestations (symptoms and signs) along with the presence of risk factors11 for venous thromboembolism were recorded on admission to the emergency department. To derive our prediction rule, we used clinical variables routinely available at presentation that were previously shown to be associated with mortality in patients with PE. These variables included demographics and physical examination findings.5,9,12 Measurements of arterial blood gasses and plasma levels of D-dimers were also performed.
The American Journal of the Medical Sciences
Volume 345, Number 6, June 2013
Prediction of Pulmonary Embolism Mortality
The study protocol was approved by the local ethics committee and informed consent was obtained from each participant before the inclusion in the study. Laboratory Tests Venous and arterial blood samples were collected from each patient upon arrival on the emergency department. The arterial blood gases were measured immediately using a commercially available blood gas analyzer (model 1630; Instrumentation Laboratories, Milan Italy). The partial pressure of oxygen in arterial blood (PaO2) and the partial pressure of carbon dioxide in arterial blood (PCO2) were recorded. As some patients received oxygen during the first assessment, we used alveolar to arterial (A-a) gradient for the assessment of the patients’ oxygenation, by the formula: A-a gradient 5 ð713 3 FiO2 2 1:25 PCO2 Þ 2 PaO2 : For patients receiving oxygen through nasal cannulas, FiO2 was calculated as13: FiO2 5 20 þ ð4 3 FlowÞ: Venous blood samples were used for standard laboratory measurements (complete blood count, routine biochemical examinations). Furthermore, plasma D-dimmers were measured with a commercial available kit (nicoCard D-dimer single test; Axis Shield PoC AS, Oslo, Norway). A value more than 300 ng/L was accepted as positive. Comorbidity A full history of the patients was taken and recorded on admission. Patients’ comorbid conditions were recorded and evaluated by the previously validated Charlson comorbidity index.14 The index encompasses 19 medical conditions weighted 1 to 6, with total scores ranging from 0 to 37. From the weighted conditions, a sum score can be tallied to yield the total comorbidity score.14 The parameters included in the calculation of the Charlson index of comorbidity are shown in Table 1. Diagnosis Confirmation In all patients, the diagnosis was confirmed with the use of Spiral CT pulmonary angiography of the chest. The CT scans were performed using either a Somaton HiQ or a Somaton Plus scanner (Siemens, Erlangen, Germany). All films were evaluated by 2 experienced radiologists, and the findings consistent with PE were recorded. The size and side of the embolism were also recorded.
TABLE 1. Weighted index of comorbidity Condition Assigned weight Myocardial infraction Congestive heart failure Peripheral vascular disease Cerebrovascular disease Dementia Chronic pulmonary disease Connective tissue disease Ulcer disease Liver disease, mild Diabetes
Ó 2012 Lippincott Williams & Wilkins
1 1 1 1 1 1 1 1 1 1
Management Strategy and Clinical Outcome The clinicians caring for each patient made all decisions concerning the diagnostic workup and treatment. All patients were treated according to the British Thoracic Society Guidelines.15 The study purpose was to assess the differences in 30-days mortality according to the patients’ characteristics and to establish a score which will be associated with death from PE. Statistical Analysis Differences between groups’ survival and death were tested using the Mann-Whitney’s test for continuous variables and Fishers’ exact test for binary variables. First, each of the statistically significant variables was independently evaluated to determine its association with the 30-day mortality due to PE using a cox regression model. Then, for determining the values of the significant continuous variables over which the risk of death increases, receiver operating characteristic (ROC) curves were generated by plotting the sensitivity against 1-specificity. Recorded Variables We determined the following variables in all study subjects: age, gender, smoking habit, pack-years, alveolar to arterial (A-a) gradient and Charlson index to determine the degree of comorbidity and the presence of the following comorbid conditions: myocardial infract, congestive heart failure (CHF), peripheral vascular disease, cerebrovascular disease, dementia, diabetes mellitus, liver disease (any) and malignancy. Validation of Our Score in a Small Population of Patients To perform validation of our score, we used a population of 30 consecutive patients admitted to the hospital for acute PE. After having calculated our score in this study group, an ROC curve was created again and the sensitivity and specificity of the same cut-off point of the score for the prevention of death in the new study group were tested. Finally, a forward growing classification tree method16,17 was used to derive a decision tree for predicting the outcome (dead or alive) based on the significant variables identified by the survival analysis. In this method, a node is split into 2 subnodes in such a way that each time the subnodes are purer than the parent node. A node is maximally impure when all the clones are equally mixed in it and it is pure when the node contains only one clone. The performance of the decision tree was estimated by apparent misclassification rate. Analyses were performed using SPSS 16 statistical package (SPSS, Chicago, IL). P value , 0.05 was considered as statistically significant.
Condition
Assigned weight
Hemiplegia Renal disease, moderate or severe Diabetes with end-organ damage Any malignancy Leukemia Malignant lymphoma Liver disease, moderate or severe Metastatic solid malignancy AIDS
2 2 2 2 2 2 3 6 6
463
Papaioannou et al
TABLE 2. Comparison of demographical and clinical characteristics of the study subjects according to their 30-day mortality from pulmonary embolism Characteristic Survived (n 5 90) Died (n 5 10) P Age Gender (female), n (%) Smoking habit (current smokers), n (%) Pack-years Alveolar to arterial (A-a) gradient Charlson index History of myocardial infract Obesity Congestive heart failure Chronic pulmonary disease Peripheral vascular disease History of cerebrovascular disease Dementia Diabetes Liver disease (any) Malignancy Tachycardia on admission Tachypnea on admission PESI index Simplified PESI index
68.0 55 24 0 46.2 1.0
(56.0–75.0) (49.5%) (21.6%) (0–5) (34.1–54.8) (0.0–2.0) 11 29 7 13 10 9 3 8 4 7 56/90 61/90 119 (97, 142) 2 (1.0, 3.0)
74.5 5 1 0 71.4 4.0
(71.5–88.25) (50%) (10%) (0–30) (60.9–93.2) (2.7–5.5) 1 1 3 2 1 2 1 3 0 0 7/10 8/10 170 (146, 231) 3 (2.7, 4.2)
0.013 NS NS NS ,0.001 ,0.001 NS NS 0.023 NS NS NS NS 0.03 NS NS NS NS ,0.001 ,0.001
Obesity was determined as body mass index .30 kg/m2.33 Scores of Charlson index can range from 0 to 37, with higher scores indicating more coexisting conditions. Data are presented as median (interquartile range) unless otherwise indicated. NS, nonsignificant; PESI, pulmonary embolism severity index.
RESULTS Table 2 shows the demographical and clinical characteristics on admission of the 2 study groups (survivors and nonsurvivors at the end of the 30 days of evaluation after PE) and the significance of their comparison. Significant difference (P , 0.05) between the 2 groups was shown for age, alveolar to arterial (A-a) gradient, Charlson index of comorbidity, and the presence of CHF and diabetes mellitus. Subsequently, the statistically significant characteristics were further evaluated to determine its association with the 30-day mortality due to PE using the cox regression model. Three characteristics were found to have a significant effect on survival, HR [(95% confidence interval (CI)]: the subjects’ age [1.12 (1.05–1.21)], the Charlson index of comorbidity [2.26 (1.46–3.33)] and the alveolar to arterial (A-a) gradient at the time of presentation [1.02 (1.01–1.03)] (Table 3). These continuous characteristics with a significant effect on mortality were then analyzed further to determine cut-off points that could be used for predicting survival. Those cut-off points were determined by the application of ROC analysis. For all 3 aforementioned characteristics, the ROC analysis produced a significant area under the curve (AUC) indicating the validity of each characteristic in determining 30-day mortality from PE. By applying
TABLE 3. Risk of death from pulmonary embolism Hazard ratio (95% Variable confidence interval) Age Charlson index Alveolar to arterial (A-a) gradient
464
1.12 (1.05–1.21) 2.26 (1.46–3.33) 1.02 (1.01–1.03)
P 0.002 ,0.001 ,0.001
ROC analysis, the characteristics age, alveolar to arterial (A-a) gradient and Charlson index of comorbidity showed significant discrimination accuracy [AUC (95% CI): 0.741 (0.596–0.886), 0.872 (0.779–0.965) and 0.872 (0.780–0.964), respectively]. The cut-off point for each characteristic (ie, the value in which the combination of sensitivity and specificity was the maximum) was calculated. Those cut-off points were used for the comparison of survival in the study population and were 67 years of age, the alveolar to arterial (A-a) gradient of 52.8 mm Hg and a Charlson comorbidity index of 2. On further exploration of the impact of these cut-off points in survival, the survival analysis showed that patients older than 67 years and patients with an alveolar to arterial (A-a) gradient more than 52.8 mm Hg at the time of their admission had a poorer survival compared with patients younger than 67 years and patients with an alveolar to arterial (A-a) gradient less than 52.8 mm Hg (P 5 0.031 and P , 0.001, respectively). Patients with Charlson comorbidity index $2 also had a poorer overall survival compared with patients with Charlson index ,2 (P , 0.001). Finally, patients with CHF and diabetes mellitus had a poorer survival compared with patients without those comorbidities (P 5 0.02 and P 5 0.03, respectively) (Figures 1–3). To predict the difference in survival by combining the aforementioned factors, we created a score that is calculated according to the presence or absence of those factors (age 67 years and above, alveolar to arterial (A-a) gradient more than 52.8, Charlson index more than 2, and presence of CHF and diabetes mellitus). The presence of each one of these conditions is graded with 1 point. The total score is calculated by adding all points for each variable and ranges between 0 and 5 points. The variables and point values used for the computation of our score are shown in Table 4. Thus, patients with an index value of 3 or more had a poorer survival compared with patients with an index value less than 3 (P , 0.001) (Figure 4). Volume 345, Number 6, June 2013
Prediction of Pulmonary Embolism Mortality
Table 5. Cox regression analysis for the probability of death using each of the three scores (our score, PESI index, simplified PESI index) Hazard ratio (95% confidence PScore interval) value Our score PESI index Simplified PESI index
3.448 (1.902–6.250) 1.030 (1.017–1.042) 3.599 (1.996–6.250)
,0.001 ,0.001 ,0.001
PESI score. AUC (95% CI) values for our score, PESI and simplified PESI are 0.908 (0.841–0.975), 0.857 (0.739–0.975) and 0.860 (0.759–0.962), respectively. Furthermore, the 3 ROC curves do not differ significantly from each other; PESI versus our score P 5 0.548, PESI versus simplified PESI P 5 0.956, our score versus simplified PESI P 5 0.589. Additionally, Cox regression analysis was performed and the HR (95% CI) for the probability of death was estimated for each score [our score 3.448 (1.902–6.250), the PESI score 1.030 (1.017– 1.042) and the simplified PESI score 3.599 (1.996–6.490)]. The above results are shown in Table 5. In a population of 30 consecutive patients admitted to the hospital for acute PE, a value of our score $3 revealed a 100% sensitivity and 78.57% specificity for the prediction of death. The ROC curve that corresponds in this group is shown in Figure 6. The median value of the score between the patients who died from PE (n 5 2) was significantly higher compared with the patients who survived (n 5 28) [median (interquartile range) of 4 (3–4) versus 1.5 (0–2); P 5 0.002]. Finally, the tree method was applied to the 5 aforementioned characteristics and a tree with apparent misclassification rate of 6% was derived, indicating their good classification performance. The tree used only 4 of the 5 characteristics: diabetes mellitus, Charlson index, alveolar to arterial (A-a) gradient, and CHF. According to our findings, for patients without a history of diabetes and CHF and with Charlson index #3, the expected
FIGURE 1. Kaplan-Meier survival curves for patients (A) older than 67 years and younger than 67 years (P 5 0.031). (B) Alveolar to arterial (A-a) gradient ,52.8 mm Hg and $52.8 mm Hg (P , 0.001).
To compare the prognostic value of our score with the already known prognostic scores for PE (ie, PESI6 and simplified PESI8 score), ROC curves were performed in our study population. As shown in Figure 5, in our study population, the validity of our score is similar to that of the PESI and the simplified TABLE 4. Variables and point values for the computation of the index in patients with pulmonary embolism Points on index Variable Age (yr) Alveolar to arterial (A-a) gradient Charlson index Congestive heart failure Diabetes mellitus Ó 2012 Lippincott Williams & Wilkins
0
1
,67 ,52.8 0–1 Absent Absent
$67 $52.8 $2 Present Present
FIGURE 2. Kaplan-Meier survival curves for patients with Charlson index less than 2 and $2 (P , 0.001).
465
Papaioannou et al
FIGURE 4. Kaplan-Meier survival curves for patients with different scores in our rule (P , 0.001).
provide the practicing clinician with useful prognostic information. According to the study results, patients with PE aged older than 67 years and those with an alveolar to arterial (A-a) gradient more than 52.8 have higher 30-day mortality compared with patients younger than 67 years and with a less impaired alveolar to arterial (A-a) gradient. We have also shown that patients with PE and more or more severe comorbid diseases (Charlson index .2) have poorer 30-day survival due to PE compared with patients with fewer or less severe comorbid
FIGURE 3. Kaplan-Meier survival curves for patients (A) with and without CHF (P 5 0.02) and (B) with and without diabetes mellitus (P 5 0.03).
outcome is survival from PE. In contrast, for patients with a history of diabetes and Charlson index .3, the expected outcome is a definite death (Figure 7).
DISCUSSION In this study, we have proposed 2 different approaches for predicting 30-day mortality from PE. The first approach was based on results from survival and ROC analysis, and the second approach was set on a classification tree method. The cut-off points are expected to be not quite similar in the 2 approaches because their philosophy is absolutely different. However, both approaches produce significant results and can be used for the recognition of patients who are at increased risk of death from PE. We identified 5 variables, the age, the alveolar to arterial (A-a) gradient, the Charlson index and the presence of CHF and diabetes mellitus, that classify patients into 2 groups: a low risk of death group and one of high risk. All these variables can be quickly evaluated in common practice in most cases and can
466
FIGURE 5. Receiver operating characteristic (ROC) curves for the prediction of death for our score (score) (dotted line), PESI (continuous line) and simplified PESI (discontinuous line). AUC (95% confidence interval) values for our score, PESI and simplified PESI are 0.908 (0.841–0.975), 0.857 (0.739–0.975) and 0.860 (0.759– 0.962), respectively. There is no statistically significant difference between the 3 ROC curves. PESI versus our score, P 5 0.548; PESI versus simplified PESI, P 5 0.956; our score versus simplified PESI, P 5 0.589. PESI, pulmonary embolism severity index. Volume 345, Number 6, June 2013
Prediction of Pulmonary Embolism Mortality
FIGURE 6. Receiver operating characteristic curve for the prediction of death in the validation study group of 30 patients with pulmonary embolism. AUC (95% confidence interval): 0.937 (0.785–0991). In this group, a value of our score $3 predicts death from pulmonary embolism with a sensitivity of 100% and a specificity of 78.57%.
diseases. The presence of CHF and diabetes mellitus were also found to be associated with higher 30-day mortality from PE. Our rule accurately distinguishes patients with PE who are at a low risk of death after 30 days from patients who are at high risk using a simple score. At the same time, the use of our tree method can predict the outcome according to the presence or absence of diabetes mellitus, CHF and comorbid diseases. Recent evidence suggests that many patients with nonmassive PE can be safely treated entirely as outpatients using
low-molecular-weight heparins or discharged early.18–20 On the basis of this evidence, the European Society of Cardiology recommends early discharge and/or outpatient treatment for clinically stable patients with low-risk PE.4 Thus, our rule could provide clinicians with a simple tool for identifying low-risk patients with PE who might be potential candidates for outpatient treatment or early hospital discharge. Previous studies have shown that age is associated with early mortality from PE.6,21 In one series, appropriately treated PE had markedly lower mortality (2.5%) in a group of younger patients than in an older cohort (18%).22 One possible explanation could be that increased age is also associated with more frequent comorbid conditions and with a decrease in pulmonary function. However, in our study, subjects’ age remained an independent risk factor for death even after controlling for comorbid conditions. The worse outcome seen in older patients probably reflects the less overall cardiopulmonary fitness and reserve of these patients. There are several mechanisms that cause a widened alveolar to arterial (A-a) gradient in patients with PE, including a reduction of cardiac output and V/Q mismatch. In a previous study, it has been shown that the alveolar to arterial (A-a) gradient was correlated with the severity of the PE as assessed by the pulmonary artery mean pressure and by the number of mismatched vascular perfusion defects on the ventilation/ perfusion lung scan. Patients with less severe PE were more likely to have a normal alveolar to arterial (A-a) gradient than those with more severe PE.23 An elevated alveolar to arterial (A-a) gradient could also be associated with the coexistence of a lung disease other than PE. Several studies have shown that chronic pulmonary disease increases the risk of death in patients with PE.6,7,21 For our study population, no difference in survival from PE between patients with and without chronic pulmonary disease was found. However, we have to keep in mind that usually the main chronic pulmonary diseases, such as asthma and COPD, are underdiagnosed in the general population.24 This fact cannot exclude the possibility that the alveolar to arterial (A-a) gradient impairment could be in part the result
FIGURE 7. Classification tree for the prediction of outcome (dead or alive) based on 4 of the 5 significant variables [diabetes mellitus, congestive heart failure, Charlson index of comorbidity $3, and alveolar to arterial (A-a) gradient .63 mm Hg].
Ó 2012 Lippincott Williams & Wilkins
467
Papaioannou et al
of the coexistence of a comorbid chronic pulmonary disease. However, the smoking history of the study subjects did not seem to be related to survival from PE. Charlson index was developed to classify comorbid conditions that might alter the risk of mortality.25 It is known that the increase in Charlson index is associated with the increase of the risk of 1-year mortality.25 In our study, we have shown that Charlson index is also associated with an increase in the 30-day mortality in patients with PE. This observation can lead to the hypothesis that when PE is present in a patient with severe comorbid conditions, this patient could probably benefit from more intensive surveillance, early admittance to an ICU and the prompt communication with a specialist. In our study, the presence of CHF and diabetes mellitus was associated with poorer survival from PE. Heart failure has also been shown to be associated with poorer survival from PE in preceding studies.6,21 Diabetes mellitus is also associated with an increased risk of death due to cardiovascular causes and for that reason could be associated with the presence of cardiovascular disease26 in our study subjects. However, in a recent study, diabetes mellitus has been found to be an independent risk factor associated with mortality in patients with postoperative venous thromboembolism.27 The reason for the association of increased mortality from PE in patients with diabetes mellitus is not clear. In the past, several rules have been used to predict the risk of death in patients with PE, such as the PESI6,8 and the Geneva Risk score.21 Our rule has several strengths compared with previous prognostic models of mortality due to PE. First, it considers only mortality due to PE and not due to comorbid conditions, although the latter seem to contribute to this outcome. Second, it consists of clearly defined routinely available predictors and can be easily calculated in the emergency department without the need of radiologic procedures, such as CT or echocardiography.28,29 Furthermore, our rule does not include the measurement of biomarkers such as BNP or troponin whose measurement is expensive and primarily not always available.30 However, under no circumstances do we claim that our model is superior to previously documented laboratory or imaging parameters (biomarkers or echocardiography findings), which have been shown in numerous studies to predict mortality from PE. We did not proceed in such analysis as it was not the purpose of this study. However, our rule is a very quick, easy, economic and always available way that can be used in the first assessment of the patient in the emergency to distinguish highrisk patients from the low-risk ones. However, our study has several limitations. First, the main limitation is that our sample size (n 5 100) and number of events (n 5 10) might be considered relatively low. This fact probably makes our model rather unstable and for that reason future accumulation of more data, derived from a multicenter study, will enable us to further explore the validity of our prognostic model. Second, it must be emphasized that our population consisted of rather stable “nonsevere PE” as patients with shock or requiring resuscitation at emergency—for any reason, probably including PE—were not included in the study. However, the aim of our study was to diagnose rather stable PE and probably to discriminate patients for outpatient treatment or early discharge. It is well known that patients in shock from PE have a high mortality rate, more than 50% in most studies, and therefore must be confronted in the ICU.2,5,31 Therefore, the accuracy of our prognostic model for PE has been tested only on a low-risk population with less severe PE. However, considering that mortality in the acute phase of PE vary widely covering a range between 1% and well over 50%,5,25,32,33 the
468
mortality rate of 10% in our study is in accordance to the aforementioned rates. Third, none of our study subjects underwent thrombolysis, although we cannot estimate whether thrombolysis could possibly alter disease outcome in some patients. The reason for not using thrombolysis was either because of the absence of clear indications (patients with shock or needing resuscitation were not included in the study, as stated) or because of the presence of contraindications for its use.15 However, the benefit of thrombolytic treatment and its role on prognosis in these patients remains unclear.34 Fourth, no imaging or laboratory markers of right ventricular dysfunction (as BNP, troponin or echocardiography) were tested in comparison or in addition to the developed score. Finally, validation of our score has been performed in a small group of patients. Although in that validation group the cut-off value that we have suggested (ie, score $3) gives a satisfactory sensitivity and specificity, this result has been based in only 2 deaths. A larger study is required to explore the validation of our risk stratification score in an independent cohort. In conclusion, we developed a simple score based on variables commonly available and easily measured in the emergency ward that could be helpful in guiding the initial intensity of treatment. According to the results of our study, patients older than 67 years with severe comorbidities, who have impaired oxygenation at the time of their admission or have a medical history of CHF or diabetes mellitus are at increased risk of death when experiencing PE. Patients with a low score in our rule could possibly be candidates for outpatient treatment or early discharge to decrease the cost of health care services. Although further studies are clearly needed to validate our prognostic model in a large cohort of patients, this would have significant implication for the care of patients with PE.
ACKNOWLEDGMENTS The authors thank Anna Laluvein (Bachelor of Arts in English with Contemporary Cultural Studies, Wolverhampton University) for improvements in the quality of written English and her assistance in editing the manuscript. REFERENCES 1. Cohen AT, Agnelli G, Anderson FA, et al. Venous thromboembolism (VTE) in Europe. The number of VTE events and associated morbidity and mortality. Thromb Haemost 2007;98:756–64. 2. Kurkciyan I, Meron G, Sterz F, et al. Pulmonary embolism as a cause of cardiac arrest: presentation and outcome. Arch Intern Med 2000;160: 1529–35. 3. Simonneau G, Sors H, Charbonnier B, et al. A comparison of lowmolecular-weight heparin with unfractionated heparin for acute pulmonary embolism. The THESEE Study Group. Tinzaparine ou Heparine Standard: evaluations dans l’Embolie Pulmonaire. N Engl J Med 1997; 337:663–9. 4. Torbicki A, Perrier A, Konstantinides S, et al. Guidelines on the diagnosis and management of acute pulmonary embolism: the Task Force for the Diagnosis and Management of Acute Pulmonary Embolism of the European Society of Cardiology (ESC). Eur Heart J 2008;29: 2276–315. 5. Goldhaber SZ, Visani L, De Rosa M. Acute pulmonary embolism: clinical outcomes in the International Cooperative Pulmonary Embolism Registry (ICOPER). Lancet 1999;353:1386–9. 6. Aujesky D, Obrosky DS, Stone RA, et al. Derivation and validation of a prognostic model for pulmonary embolism. Am J Respir Crit Care Med 2005;172:1041–6.
Volume 345, Number 6, June 2013
Prediction of Pulmonary Embolism Mortality
7. Aujesky D, Perrier A, Roy PM, et al. Validation of a clinical prognostic model to identify low-risk patients with pulmonary embolism. J Intern Med 2007;261:597–604.
21. Nendaz MR, Bandelier P, Aujesky D, et al. Validation of a risk score identifying patients with acute pulmonary embolism, who are at low risk of clinical adverse outcome. Thromb Haemost 2004;91:1232–6.
8. Jimenez D, Aujesky D, Moores L, et al. Simplification of the pulmonary embolism severity index for prognostication in patients with acute symptomatic pulmonary embolism. Arch Intern Med 2010;170:1383–9.
22. Green RM, Meyer TJ, Dunn M, et al. Pulmonary embolism in younger adults. Chest 1992;101:1507–11.
9. Wicki J, Perrier A, Perneger TV, et al. Predicting adverse outcome in patients with acute pulmonary embolism: a risk score. Thromb Haemost 2000;84:548–52. 10. Jimenez D, Yusen RD, Otero R, et al. Prognostic models for selecting patients with acute pulmonary embolism for initial outpatient therapy. Chest 2007;132:24–30. 11. Fedullo PF, Tapson VF. Clinical practice. The evaluation of suspected pulmonary embolism. N Engl J Med 2003;349:1247–56. 12. Grifoni S, Olivotto I, Cecchini P, et al. Short-term clinical outcome of patients with acute pulmonary embolism, normal blood pressure, and echocardiographic right ventricular dysfunction. Circulation 2000;101:2817–22. 13. Keane MP, Strieter RM, Belperio JA. (20060 9Macrophage Inflammatory Proteins (CCL3, CCL4, CCL9/10, CCL15, CCL18, CCL19, CCL20, CXCL2)9 In: Laurent GJ, and Shapiro SD (eds). Encyclopedia of Respiratory Medicine. Oxford: Elsevier Limited. 14. Hall WH, Ramachandran R, Narayan S, et al. An electronic application for rapidly calculating Charlson comorbidity score. BMC Cancer 2004;4:94. 15. British Thoracic Society guidelines for the management of suspected acute pulmonary embolism. Thorax 2003;58:470–84. 16. Zintzaras E, Bai M, Douligeris C, et al. A tree-based decision rule for identifying profile groups of cases without predefined classes: application in diffuse large B-cell lymphomas. Comput Biol Med 2007;37:637–41. 17. Zintzaras E, Kowald A. Forest classification trees and forest support vector machines algorithms: demonstration using microarray data. Comput Biol Med 2010;40:519–24. 18. Wells PS, Anderson DR, Rodger MA, et al. A randomized trial comparing 2 low-molecular-weight heparins for the outpatient treatment of deep vein thrombosis and pulmonary embolism. Arch Intern Med 2005;165:733–8. 19. Lim AY, Parr DG, Stableforth DE, et al. Early discharge and home supervision of patients with pulmonary embolism treated with lowmolecular weight heparin. Eur J Intern Med 2003;14:89–93. 20. Beer JH, Burger M, Gretener S, et al. Outpatient treatment of pulmonary embolism is feasible and safe in a substantial proportion of patients. J Thromb Haemost 2003;1:186–7.
Ó 2012 Lippincott Williams & Wilkins
23. Stein PD, Goldhaber SZ, Henry JW. Alveolar-arterial oxygen gradient in the assessment of acute pulmonary embolism. Chest 1995;107:139–43. 24. Bednarek M, Maciejewski J, Wozniak M, et al. Prevalence, severity and underdiagnosis of COPD in the primary care setting. Thorax 2008; 63:402–7. 25. Charlson ME, Pompei P, Ales KL, et al. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J Chronic Dis 1987;40:373–83. 26. Cannon CP. Cardiovascular disease and modifiable cardiometabolic risk factors. Clin Cornerstone 2007;8:11–28. 27. Gangireddy C, Rectenwald JR, Upchurch GR, et al. Risk factors and clinical impact of postoperative symptomatic venous thromboembolism. J Vasc Surg 2007;45:335–41; discussion 341–2. 28. ten Wolde M, Sohne M, Quak E, et al. Prognostic value of echocardiographically assessed right ventricular dysfunction in patients with pulmonary embolism. Arch Intern Med 2004;164:1685–9. 29. Gibson NS, Sohne M, Buller HR. Prognostic value of echocardiography and spiral computed tomography in patients with pulmonary embolism. Curr Opin Pulm Med 2005;11:380–4. 30. Jimenez D, Yusen RD. Prognostic models for selecting patients with acute pulmonary embolism for initial outpatient therapy. Curr Opin Pulm Med 2008;14:414–21. 31. Silverstein MD, Heit JA, Mohr DN, et al. Trends in the incidence of deep vein thrombosis and pulmonary embolism: a 25-year populationbased study. Arch Intern Med 1998;158:585–93. 32. Kasper W, Konstantinides S, Geibel A, et al. Management strategies and determinants of outcome in acute major pulmonary embolism: results of a multicenter registry. J Am Coll Cardiol 1997;30: 1165–71. 33. Lankeit M, Konstantinides S. Is it time for home treatment of pulmonary embolism? Eur Respir J. 2012;40:742–49. 34. Kearon C, Kahn SR, Agnelli G, et al. Antithrombotic therapy for venous thromboembolic disease: American College of Chest Physicians Evidence-Based Clinical Practice Guidelines (8th Edition). Chest 2008; 133:454S–545S.
469
Predicting Short-term Mortality in Patients With Pulmonary Embolism: A Simple Model Andriana I. Papaioannou, MD, PhD, Alcibiades Kastanis, PhD, Foteini Malli, MD, Epameinontas Zakynthinos, MD, PhD, Markos Minas, MD, PhD,* Mariana Mpakarozi, MD, Stamatina Moka, MD, Elias Zintzaras, MSc, PhD, Konstantinos I. Gourgoulianis, MD, PhD and Zoe Daniil, MD, PhD
Background: The aim of the present study was to develop a simple prognostic rule that could classify patients with pulmonary embolism (PE) into categories of increased risk of 30-day mortality. Methods: One hundred patients with PE were enrolled. Clinical and laboratory findings were recorded on admission for each patient. Differences between groups’ survival and death were tested, and the association with the 30day mortality was determined. Results: Three variables had a significant effect on survival: age, Charlson index and the alveolar to arterial (A-a) gradient. A receiver operating characteristic analysis was performed, and the cut-off points used for the comparison of survival were 67 years of age, A-a gradient over 52.8 mm Hg and Charlson index over 2. By combining these variables, a score was established for distinguishing patients with PE who are at high risk. This score was also validated in a group of 30 consecutive patients admitted to the hospital for PE Additionally, a tree method was applied and showed that for patients with a history of diabetes and Charlson index .3, the expected outcome is death. Conclusions: The results of this study suggest that patients with PE could be stratified into categories of increasing risk of 30-day mortality using a simple score based only on routinely available variables. Future studies are needed to validate our prognostic model in a large cohort of patients with PE. Key Indexing Terms: Comorbidities; Congestive heart failure; Diabetes mellitus; Pulmonary embolism; Survival. [Am J Med Sci 2013;345 (6):462–469.]
P
ulmonary embolism (PE) is a major health burden worldwide. A recent epidemiological model derived from 6 European countries with a total population of 310.4 million yielded a PE incidence rate of 98 cases per 100,000 person-years.1 The short-term mortality of the disease varies widely, ranging from less than 2% in patients with nonmassive PE to more than 95% in patients who experience cardiorespiratory arrest.2,3 Although most patients with acute PE remain hospitalized during the initial therapy, it has been suggested that patients presenting without hemodynamic instability and without elevated biomarker levels or imaging findings indicating right ventricular (RV) dysfunction or myocardial injury may constitute a lowrisk group and accordingly it was suggested that these patients may be considered for early discharge and home treatment.4 From the Department of Respiratory Medicine (AIP, FM, MMi, MMp, SM, KIG, ZD); Departments of Biomathematics (AK, EZi) and Critical Care Medicine (EZa), School of Medicine, University of Thessaly, Larissa, Greece; and Institute for Clinical Research and Health Policy Studies (EZi), Tufts Medical Center, Tufts University School of Medicine, Boston, Massachusetts. Submitted February 12, 2012; accepted in revised form June 13, 2012. The authors have no financial or other conflicts of interest to disclose. *Author is deceased. Correspondence: Elias Zintzaras, MSc, PhD, Department of Biomathematics, University of Thessaly School of Medicine, 2 Panepistimiou Str., Biopolis, Larissa 4110, Greece (E-mail: [email protected]).
462
Because a small but significant proportion of patients with acute PE will die or have severe complications during the initial therapy period, reliable prognostic information at presentation would help clinicians to estimate which patients are at low risk and could be discharged early or managed entirely as outpatients and which patients are at high risk and may benefit from a more intensive surveillance.5 Previous models of risk stratification for PE tried to develop a clinical prediction rule for patients with PE that could quantify the risk of mortality across the full spectrum of risk. A prognostic model previously developed by Aujesky et al6 [pulmonary embolism severity index (PESI)] comprises 11 routinely available clinical parameters and accurately identifies patients with PE who are at low risk of fatal or nonfatal outcomes.6,7 Recently, the simplified PESI score was described to provide a more simple model for the identification of patients with PE who are at high risk of death.8 A second model, the “Geneva” prediction rule, created by Wicki et al9 identifies 6 independent predictors, including a venous ultrasound; however, recently published study provided evidence that the Geneva risk stratification does not identify patients with acute PE that have negligible risk for 30-day mortality.9,10 The aim of this study was the evaluation of 30-day mortality in patients with PE confirmed by spiral computed tomography (CT) pulmonary angiography and to develop a fast descriptive rule, relying only on a few readily available clinical parameters that could classify patients into categories of increased risk of mortality and be helpful in guiding initial intensity of treatment.
MATERIALS AND METHODS Patients We identified 100 consecutive patients with PE, who were treated in the Respiratory Medicine Department of the University Hospital of Larissa, Greece, during February 2003 to January 2008. In all patients, the diagnosis of PE was established with spiral CT pulmonary angiography. Patients in a state of hemodynamic instability and patients requiring resuscitation are usually admitted in the intensive care unit (ICU) without a definite diagnosis. Although some of them might have suffered a PE, they were not included in the study. For each patient, all clinical manifestations (symptoms and signs) along with the presence of risk factors11 for venous thromboembolism were recorded on admission to the emergency department. To derive our prediction rule, we used clinical variables routinely available at presentation that were previously shown to be associated with mortality in patients with PE. These variables included demographics and physical examination findings.5,9,12 Measurements of arterial blood gasses and plasma levels of D-dimers were also performed.
The American Journal of the Medical Sciences
Volume 345, Number 6, June 2013
Prediction of Pulmonary Embolism Mortality
The study protocol was approved by the local ethics committee and informed consent was obtained from each participant before the inclusion in the study. Laboratory Tests Venous and arterial blood samples were collected from each patient upon arrival on the emergency department. The arterial blood gases were measured immediately using a commercially available blood gas analyzer (model 1630; Instrumentation Laboratories, Milan Italy). The partial pressure of oxygen in arterial blood (PaO2) and the partial pressure of carbon dioxide in arterial blood (PCO2) were recorded. As some patients received oxygen during the first assessment, we used alveolar to arterial (A-a) gradient for the assessment of the patients’ oxygenation, by the formula: A-a gradient 5 ð713 3 FiO2 2 1:25 PCO2 Þ 2 PaO2 : For patients receiving oxygen through nasal cannulas, FiO2 was calculated as13: FiO2 5 20 þ ð4 3 FlowÞ: Venous blood samples were used for standard laboratory measurements (complete blood count, routine biochemical examinations). Furthermore, plasma D-dimmers were measured with a commercial available kit (nicoCard D-dimer single test; Axis Shield PoC AS, Oslo, Norway). A value more than 300 ng/L was accepted as positive. Comorbidity A full history of the patients was taken and recorded on admission. Patients’ comorbid conditions were recorded and evaluated by the previously validated Charlson comorbidity index.14 The index encompasses 19 medical conditions weighted 1 to 6, with total scores ranging from 0 to 37. From the weighted conditions, a sum score can be tallied to yield the total comorbidity score.14 The parameters included in the calculation of the Charlson index of comorbidity are shown in Table 1. Diagnosis Confirmation In all patients, the diagnosis was confirmed with the use of Spiral CT pulmonary angiography of the chest. The CT scans were performed using either a Somaton HiQ or a Somaton Plus scanner (Siemens, Erlangen, Germany). All films were evaluated by 2 experienced radiologists, and the findings consistent with PE were recorded. The size and side of the embolism were also recorded.
TABLE 1. Weighted index of comorbidity Condition Assigned weight Myocardial infraction Congestive heart failure Peripheral vascular disease Cerebrovascular disease Dementia Chronic pulmonary disease Connective tissue disease Ulcer disease Liver disease, mild Diabetes
Ó 2012 Lippincott Williams & Wilkins
1 1 1 1 1 1 1 1 1 1
Management Strategy and Clinical Outcome The clinicians caring for each patient made all decisions concerning the diagnostic workup and treatment. All patients were treated according to the British Thoracic Society Guidelines.15 The study purpose was to assess the differences in 30-days mortality according to the patients’ characteristics and to establish a score which will be associated with death from PE. Statistical Analysis Differences between groups’ survival and death were tested using the Mann-Whitney’s test for continuous variables and Fishers’ exact test for binary variables. First, each of the statistically significant variables was independently evaluated to determine its association with the 30-day mortality due to PE using a cox regression model. Then, for determining the values of the significant continuous variables over which the risk of death increases, receiver operating characteristic (ROC) curves were generated by plotting the sensitivity against 1-specificity. Recorded Variables We determined the following variables in all study subjects: age, gender, smoking habit, pack-years, alveolar to arterial (A-a) gradient and Charlson index to determine the degree of comorbidity and the presence of the following comorbid conditions: myocardial infract, congestive heart failure (CHF), peripheral vascular disease, cerebrovascular disease, dementia, diabetes mellitus, liver disease (any) and malignancy. Validation of Our Score in a Small Population of Patients To perform validation of our score, we used a population of 30 consecutive patients admitted to the hospital for acute PE. After having calculated our score in this study group, an ROC curve was created again and the sensitivity and specificity of the same cut-off point of the score for the prevention of death in the new study group were tested. Finally, a forward growing classification tree method16,17 was used to derive a decision tree for predicting the outcome (dead or alive) based on the significant variables identified by the survival analysis. In this method, a node is split into 2 subnodes in such a way that each time the subnodes are purer than the parent node. A node is maximally impure when all the clones are equally mixed in it and it is pure when the node contains only one clone. The performance of the decision tree was estimated by apparent misclassification rate. Analyses were performed using SPSS 16 statistical package (SPSS, Chicago, IL). P value , 0.05 was considered as statistically significant.
Condition
Assigned weight
Hemiplegia Renal disease, moderate or severe Diabetes with end-organ damage Any malignancy Leukemia Malignant lymphoma Liver disease, moderate or severe Metastatic solid malignancy AIDS
2 2 2 2 2 2 3 6 6
463
Papaioannou et al
TABLE 2. Comparison of demographical and clinical characteristics of the study subjects according to their 30-day mortality from pulmonary embolism Characteristic Survived (n 5 90) Died (n 5 10) P Age Gender (female), n (%) Smoking habit (current smokers), n (%) Pack-years Alveolar to arterial (A-a) gradient Charlson index History of myocardial infract Obesity Congestive heart failure Chronic pulmonary disease Peripheral vascular disease History of cerebrovascular disease Dementia Diabetes Liver disease (any) Malignancy Tachycardia on admission Tachypnea on admission PESI index Simplified PESI index
68.0 55 24 0 46.2 1.0
(56.0–75.0) (49.5%) (21.6%) (0–5) (34.1–54.8) (0.0–2.0) 11 29 7 13 10 9 3 8 4 7 56/90 61/90 119 (97, 142) 2 (1.0, 3.0)
74.5 5 1 0 71.4 4.0
(71.5–88.25) (50%) (10%) (0–30) (60.9–93.2) (2.7–5.5) 1 1 3 2 1 2 1 3 0 0 7/10 8/10 170 (146, 231) 3 (2.7, 4.2)
0.013 NS NS NS ,0.001 ,0.001 NS NS 0.023 NS NS NS NS 0.03 NS NS NS NS ,0.001 ,0.001
Obesity was determined as body mass index .30 kg/m2.33 Scores of Charlson index can range from 0 to 37, with higher scores indicating more coexisting conditions. Data are presented as median (interquartile range) unless otherwise indicated. NS, nonsignificant; PESI, pulmonary embolism severity index.
RESULTS Table 2 shows the demographical and clinical characteristics on admission of the 2 study groups (survivors and nonsurvivors at the end of the 30 days of evaluation after PE) and the significance of their comparison. Significant difference (P , 0.05) between the 2 groups was shown for age, alveolar to arterial (A-a) gradient, Charlson index of comorbidity, and the presence of CHF and diabetes mellitus. Subsequently, the statistically significant characteristics were further evaluated to determine its association with the 30-day mortality due to PE using the cox regression model. Three characteristics were found to have a significant effect on survival, HR [(95% confidence interval (CI)]: the subjects’ age [1.12 (1.05–1.21)], the Charlson index of comorbidity [2.26 (1.46–3.33)] and the alveolar to arterial (A-a) gradient at the time of presentation [1.02 (1.01–1.03)] (Table 3). These continuous characteristics with a significant effect on mortality were then analyzed further to determine cut-off points that could be used for predicting survival. Those cut-off points were determined by the application of ROC analysis. For all 3 aforementioned characteristics, the ROC analysis produced a significant area under the curve (AUC) indicating the validity of each characteristic in determining 30-day mortality from PE. By applying
TABLE 3. Risk of death from pulmonary embolism Hazard ratio (95% Variable confidence interval) Age Charlson index Alveolar to arterial (A-a) gradient
464
1.12 (1.05–1.21) 2.26 (1.46–3.33) 1.02 (1.01–1.03)
P 0.002 ,0.001 ,0.001
ROC analysis, the characteristics age, alveolar to arterial (A-a) gradient and Charlson index of comorbidity showed significant discrimination accuracy [AUC (95% CI): 0.741 (0.596–0.886), 0.872 (0.779–0.965) and 0.872 (0.780–0.964), respectively]. The cut-off point for each characteristic (ie, the value in which the combination of sensitivity and specificity was the maximum) was calculated. Those cut-off points were used for the comparison of survival in the study population and were 67 years of age, the alveolar to arterial (A-a) gradient of 52.8 mm Hg and a Charlson comorbidity index of 2. On further exploration of the impact of these cut-off points in survival, the survival analysis showed that patients older than 67 years and patients with an alveolar to arterial (A-a) gradient more than 52.8 mm Hg at the time of their admission had a poorer survival compared with patients younger than 67 years and patients with an alveolar to arterial (A-a) gradient less than 52.8 mm Hg (P 5 0.031 and P , 0.001, respectively). Patients with Charlson comorbidity index $2 also had a poorer overall survival compared with patients with Charlson index ,2 (P , 0.001). Finally, patients with CHF and diabetes mellitus had a poorer survival compared with patients without those comorbidities (P 5 0.02 and P 5 0.03, respectively) (Figures 1–3). To predict the difference in survival by combining the aforementioned factors, we created a score that is calculated according to the presence or absence of those factors (age 67 years and above, alveolar to arterial (A-a) gradient more than 52.8, Charlson index more than 2, and presence of CHF and diabetes mellitus). The presence of each one of these conditions is graded with 1 point. The total score is calculated by adding all points for each variable and ranges between 0 and 5 points. The variables and point values used for the computation of our score are shown in Table 4. Thus, patients with an index value of 3 or more had a poorer survival compared with patients with an index value less than 3 (P , 0.001) (Figure 4). Volume 345, Number 6, June 2013
Prediction of Pulmonary Embolism Mortality
Table 5. Cox regression analysis for the probability of death using each of the three scores (our score, PESI index, simplified PESI index) Hazard ratio (95% confidence PScore interval) value Our score PESI index Simplified PESI index
3.448 (1.902–6.250) 1.030 (1.017–1.042) 3.599 (1.996–6.250)
,0.001 ,0.001 ,0.001
PESI score. AUC (95% CI) values for our score, PESI and simplified PESI are 0.908 (0.841–0.975), 0.857 (0.739–0.975) and 0.860 (0.759–0.962), respectively. Furthermore, the 3 ROC curves do not differ significantly from each other; PESI versus our score P 5 0.548, PESI versus simplified PESI P 5 0.956, our score versus simplified PESI P 5 0.589. Additionally, Cox regression analysis was performed and the HR (95% CI) for the probability of death was estimated for each score [our score 3.448 (1.902–6.250), the PESI score 1.030 (1.017– 1.042) and the simplified PESI score 3.599 (1.996–6.490)]. The above results are shown in Table 5. In a population of 30 consecutive patients admitted to the hospital for acute PE, a value of our score $3 revealed a 100% sensitivity and 78.57% specificity for the prediction of death. The ROC curve that corresponds in this group is shown in Figure 6. The median value of the score between the patients who died from PE (n 5 2) was significantly higher compared with the patients who survived (n 5 28) [median (interquartile range) of 4 (3–4) versus 1.5 (0–2); P 5 0.002]. Finally, the tree method was applied to the 5 aforementioned characteristics and a tree with apparent misclassification rate of 6% was derived, indicating their good classification performance. The tree used only 4 of the 5 characteristics: diabetes mellitus, Charlson index, alveolar to arterial (A-a) gradient, and CHF. According to our findings, for patients without a history of diabetes and CHF and with Charlson index #3, the expected
FIGURE 1. Kaplan-Meier survival curves for patients (A) older than 67 years and younger than 67 years (P 5 0.031). (B) Alveolar to arterial (A-a) gradient ,52.8 mm Hg and $52.8 mm Hg (P , 0.001).
To compare the prognostic value of our score with the already known prognostic scores for PE (ie, PESI6 and simplified PESI8 score), ROC curves were performed in our study population. As shown in Figure 5, in our study population, the validity of our score is similar to that of the PESI and the simplified TABLE 4. Variables and point values for the computation of the index in patients with pulmonary embolism Points on index Variable Age (yr) Alveolar to arterial (A-a) gradient Charlson index Congestive heart failure Diabetes mellitus Ó 2012 Lippincott Williams & Wilkins
0
1
,67 ,52.8 0–1 Absent Absent
$67 $52.8 $2 Present Present
FIGURE 2. Kaplan-Meier survival curves for patients with Charlson index less than 2 and $2 (P , 0.001).
465
Papaioannou et al
FIGURE 4. Kaplan-Meier survival curves for patients with different scores in our rule (P , 0.001).
provide the practicing clinician with useful prognostic information. According to the study results, patients with PE aged older than 67 years and those with an alveolar to arterial (A-a) gradient more than 52.8 have higher 30-day mortality compared with patients younger than 67 years and with a less impaired alveolar to arterial (A-a) gradient. We have also shown that patients with PE and more or more severe comorbid diseases (Charlson index .2) have poorer 30-day survival due to PE compared with patients with fewer or less severe comorbid
FIGURE 3. Kaplan-Meier survival curves for patients (A) with and without CHF (P 5 0.02) and (B) with and without diabetes mellitus (P 5 0.03).
outcome is survival from PE. In contrast, for patients with a history of diabetes and Charlson index .3, the expected outcome is a definite death (Figure 7).
DISCUSSION In this study, we have proposed 2 different approaches for predicting 30-day mortality from PE. The first approach was based on results from survival and ROC analysis, and the second approach was set on a classification tree method. The cut-off points are expected to be not quite similar in the 2 approaches because their philosophy is absolutely different. However, both approaches produce significant results and can be used for the recognition of patients who are at increased risk of death from PE. We identified 5 variables, the age, the alveolar to arterial (A-a) gradient, the Charlson index and the presence of CHF and diabetes mellitus, that classify patients into 2 groups: a low risk of death group and one of high risk. All these variables can be quickly evaluated in common practice in most cases and can
466
FIGURE 5. Receiver operating characteristic (ROC) curves for the prediction of death for our score (score) (dotted line), PESI (continuous line) and simplified PESI (discontinuous line). AUC (95% confidence interval) values for our score, PESI and simplified PESI are 0.908 (0.841–0.975), 0.857 (0.739–0.975) and 0.860 (0.759– 0.962), respectively. There is no statistically significant difference between the 3 ROC curves. PESI versus our score, P 5 0.548; PESI versus simplified PESI, P 5 0.956; our score versus simplified PESI, P 5 0.589. PESI, pulmonary embolism severity index. Volume 345, Number 6, June 2013
Prediction of Pulmonary Embolism Mortality
FIGURE 6. Receiver operating characteristic curve for the prediction of death in the validation study group of 30 patients with pulmonary embolism. AUC (95% confidence interval): 0.937 (0.785–0991). In this group, a value of our score $3 predicts death from pulmonary embolism with a sensitivity of 100% and a specificity of 78.57%.
diseases. The presence of CHF and diabetes mellitus were also found to be associated with higher 30-day mortality from PE. Our rule accurately distinguishes patients with PE who are at a low risk of death after 30 days from patients who are at high risk using a simple score. At the same time, the use of our tree method can predict the outcome according to the presence or absence of diabetes mellitus, CHF and comorbid diseases. Recent evidence suggests that many patients with nonmassive PE can be safely treated entirely as outpatients using
low-molecular-weight heparins or discharged early.18–20 On the basis of this evidence, the European Society of Cardiology recommends early discharge and/or outpatient treatment for clinically stable patients with low-risk PE.4 Thus, our rule could provide clinicians with a simple tool for identifying low-risk patients with PE who might be potential candidates for outpatient treatment or early hospital discharge. Previous studies have shown that age is associated with early mortality from PE.6,21 In one series, appropriately treated PE had markedly lower mortality (2.5%) in a group of younger patients than in an older cohort (18%).22 One possible explanation could be that increased age is also associated with more frequent comorbid conditions and with a decrease in pulmonary function. However, in our study, subjects’ age remained an independent risk factor for death even after controlling for comorbid conditions. The worse outcome seen in older patients probably reflects the less overall cardiopulmonary fitness and reserve of these patients. There are several mechanisms that cause a widened alveolar to arterial (A-a) gradient in patients with PE, including a reduction of cardiac output and V/Q mismatch. In a previous study, it has been shown that the alveolar to arterial (A-a) gradient was correlated with the severity of the PE as assessed by the pulmonary artery mean pressure and by the number of mismatched vascular perfusion defects on the ventilation/ perfusion lung scan. Patients with less severe PE were more likely to have a normal alveolar to arterial (A-a) gradient than those with more severe PE.23 An elevated alveolar to arterial (A-a) gradient could also be associated with the coexistence of a lung disease other than PE. Several studies have shown that chronic pulmonary disease increases the risk of death in patients with PE.6,7,21 For our study population, no difference in survival from PE between patients with and without chronic pulmonary disease was found. However, we have to keep in mind that usually the main chronic pulmonary diseases, such as asthma and COPD, are underdiagnosed in the general population.24 This fact cannot exclude the possibility that the alveolar to arterial (A-a) gradient impairment could be in part the result
FIGURE 7. Classification tree for the prediction of outcome (dead or alive) based on 4 of the 5 significant variables [diabetes mellitus, congestive heart failure, Charlson index of comorbidity $3, and alveolar to arterial (A-a) gradient .63 mm Hg].
Ó 2012 Lippincott Williams & Wilkins
467
Papaioannou et al
of the coexistence of a comorbid chronic pulmonary disease. However, the smoking history of the study subjects did not seem to be related to survival from PE. Charlson index was developed to classify comorbid conditions that might alter the risk of mortality.25 It is known that the increase in Charlson index is associated with the increase of the risk of 1-year mortality.25 In our study, we have shown that Charlson index is also associated with an increase in the 30-day mortality in patients with PE. This observation can lead to the hypothesis that when PE is present in a patient with severe comorbid conditions, this patient could probably benefit from more intensive surveillance, early admittance to an ICU and the prompt communication with a specialist. In our study, the presence of CHF and diabetes mellitus was associated with poorer survival from PE. Heart failure has also been shown to be associated with poorer survival from PE in preceding studies.6,21 Diabetes mellitus is also associated with an increased risk of death due to cardiovascular causes and for that reason could be associated with the presence of cardiovascular disease26 in our study subjects. However, in a recent study, diabetes mellitus has been found to be an independent risk factor associated with mortality in patients with postoperative venous thromboembolism.27 The reason for the association of increased mortality from PE in patients with diabetes mellitus is not clear. In the past, several rules have been used to predict the risk of death in patients with PE, such as the PESI6,8 and the Geneva Risk score.21 Our rule has several strengths compared with previous prognostic models of mortality due to PE. First, it considers only mortality due to PE and not due to comorbid conditions, although the latter seem to contribute to this outcome. Second, it consists of clearly defined routinely available predictors and can be easily calculated in the emergency department without the need of radiologic procedures, such as CT or echocardiography.28,29 Furthermore, our rule does not include the measurement of biomarkers such as BNP or troponin whose measurement is expensive and primarily not always available.30 However, under no circumstances do we claim that our model is superior to previously documented laboratory or imaging parameters (biomarkers or echocardiography findings), which have been shown in numerous studies to predict mortality from PE. We did not proceed in such analysis as it was not the purpose of this study. However, our rule is a very quick, easy, economic and always available way that can be used in the first assessment of the patient in the emergency to distinguish highrisk patients from the low-risk ones. However, our study has several limitations. First, the main limitation is that our sample size (n 5 100) and number of events (n 5 10) might be considered relatively low. This fact probably makes our model rather unstable and for that reason future accumulation of more data, derived from a multicenter study, will enable us to further explore the validity of our prognostic model. Second, it must be emphasized that our population consisted of rather stable “nonsevere PE” as patients with shock or requiring resuscitation at emergency—for any reason, probably including PE—were not included in the study. However, the aim of our study was to diagnose rather stable PE and probably to discriminate patients for outpatient treatment or early discharge. It is well known that patients in shock from PE have a high mortality rate, more than 50% in most studies, and therefore must be confronted in the ICU.2,5,31 Therefore, the accuracy of our prognostic model for PE has been tested only on a low-risk population with less severe PE. However, considering that mortality in the acute phase of PE vary widely covering a range between 1% and well over 50%,5,25,32,33 the
468
mortality rate of 10% in our study is in accordance to the aforementioned rates. Third, none of our study subjects underwent thrombolysis, although we cannot estimate whether thrombolysis could possibly alter disease outcome in some patients. The reason for not using thrombolysis was either because of the absence of clear indications (patients with shock or needing resuscitation were not included in the study, as stated) or because of the presence of contraindications for its use.15 However, the benefit of thrombolytic treatment and its role on prognosis in these patients remains unclear.34 Fourth, no imaging or laboratory markers of right ventricular dysfunction (as BNP, troponin or echocardiography) were tested in comparison or in addition to the developed score. Finally, validation of our score has been performed in a small group of patients. Although in that validation group the cut-off value that we have suggested (ie, score $3) gives a satisfactory sensitivity and specificity, this result has been based in only 2 deaths. A larger study is required to explore the validation of our risk stratification score in an independent cohort. In conclusion, we developed a simple score based on variables commonly available and easily measured in the emergency ward that could be helpful in guiding the initial intensity of treatment. According to the results of our study, patients older than 67 years with severe comorbidities, who have impaired oxygenation at the time of their admission or have a medical history of CHF or diabetes mellitus are at increased risk of death when experiencing PE. Patients with a low score in our rule could possibly be candidates for outpatient treatment or early discharge to decrease the cost of health care services. Although further studies are clearly needed to validate our prognostic model in a large cohort of patients, this would have significant implication for the care of patients with PE.
ACKNOWLEDGMENTS The authors thank Anna Laluvein (Bachelor of Arts in English with Contemporary Cultural Studies, Wolverhampton University) for improvements in the quality of written English and her assistance in editing the manuscript. REFERENCES 1. Cohen AT, Agnelli G, Anderson FA, et al. Venous thromboembolism (VTE) in Europe. The number of VTE events and associated morbidity and mortality. Thromb Haemost 2007;98:756–64. 2. Kurkciyan I, Meron G, Sterz F, et al. Pulmonary embolism as a cause of cardiac arrest: presentation and outcome. Arch Intern Med 2000;160: 1529–35. 3. Simonneau G, Sors H, Charbonnier B, et al. A comparison of lowmolecular-weight heparin with unfractionated heparin for acute pulmonary embolism. The THESEE Study Group. Tinzaparine ou Heparine Standard: evaluations dans l’Embolie Pulmonaire. N Engl J Med 1997; 337:663–9. 4. Torbicki A, Perrier A, Konstantinides S, et al. Guidelines on the diagnosis and management of acute pulmonary embolism: the Task Force for the Diagnosis and Management of Acute Pulmonary Embolism of the European Society of Cardiology (ESC). Eur Heart J 2008;29: 2276–315. 5. Goldhaber SZ, Visani L, De Rosa M. Acute pulmonary embolism: clinical outcomes in the International Cooperative Pulmonary Embolism Registry (ICOPER). Lancet 1999;353:1386–9. 6. Aujesky D, Obrosky DS, Stone RA, et al. Derivation and validation of a prognostic model for pulmonary embolism. Am J Respir Crit Care Med 2005;172:1041–6.
Volume 345, Number 6, June 2013
Prediction of Pulmonary Embolism Mortality
7. Aujesky D, Perrier A, Roy PM, et al. Validation of a clinical prognostic model to identify low-risk patients with pulmonary embolism. J Intern Med 2007;261:597–604.
21. Nendaz MR, Bandelier P, Aujesky D, et al. Validation of a risk score identifying patients with acute pulmonary embolism, who are at low risk of clinical adverse outcome. Thromb Haemost 2004;91:1232–6.
8. Jimenez D, Aujesky D, Moores L, et al. Simplification of the pulmonary embolism severity index for prognostication in patients with acute symptomatic pulmonary embolism. Arch Intern Med 2010;170:1383–9.
22. Green RM, Meyer TJ, Dunn M, et al. Pulmonary embolism in younger adults. Chest 1992;101:1507–11.
9. Wicki J, Perrier A, Perneger TV, et al. Predicting adverse outcome in patients with acute pulmonary embolism: a risk score. Thromb Haemost 2000;84:548–52. 10. Jimenez D, Yusen RD, Otero R, et al. Prognostic models for selecting patients with acute pulmonary embolism for initial outpatient therapy. Chest 2007;132:24–30. 11. Fedullo PF, Tapson VF. Clinical practice. The evaluation of suspected pulmonary embolism. N Engl J Med 2003;349:1247–56. 12. Grifoni S, Olivotto I, Cecchini P, et al. Short-term clinical outcome of patients with acute pulmonary embolism, normal blood pressure, and echocardiographic right ventricular dysfunction. Circulation 2000;101:2817–22. 13. Keane MP, Strieter RM, Belperio JA. (20060 9Macrophage Inflammatory Proteins (CCL3, CCL4, CCL9/10, CCL15, CCL18, CCL19, CCL20, CXCL2)9 In: Laurent GJ, and Shapiro SD (eds). Encyclopedia of Respiratory Medicine. Oxford: Elsevier Limited. 14. Hall WH, Ramachandran R, Narayan S, et al. An electronic application for rapidly calculating Charlson comorbidity score. BMC Cancer 2004;4:94. 15. British Thoracic Society guidelines for the management of suspected acute pulmonary embolism. Thorax 2003;58:470–84. 16. Zintzaras E, Bai M, Douligeris C, et al. A tree-based decision rule for identifying profile groups of cases without predefined classes: application in diffuse large B-cell lymphomas. Comput Biol Med 2007;37:637–41. 17. Zintzaras E, Kowald A. Forest classification trees and forest support vector machines algorithms: demonstration using microarray data. Comput Biol Med 2010;40:519–24. 18. Wells PS, Anderson DR, Rodger MA, et al. A randomized trial comparing 2 low-molecular-weight heparins for the outpatient treatment of deep vein thrombosis and pulmonary embolism. Arch Intern Med 2005;165:733–8. 19. Lim AY, Parr DG, Stableforth DE, et al. Early discharge and home supervision of patients with pulmonary embolism treated with lowmolecular weight heparin. Eur J Intern Med 2003;14:89–93. 20. Beer JH, Burger M, Gretener S, et al. Outpatient treatment of pulmonary embolism is feasible and safe in a substantial proportion of patients. J Thromb Haemost 2003;1:186–7.
Ó 2012 Lippincott Williams & Wilkins
23. Stein PD, Goldhaber SZ, Henry JW. Alveolar-arterial oxygen gradient in the assessment of acute pulmonary embolism. Chest 1995;107:139–43. 24. Bednarek M, Maciejewski J, Wozniak M, et al. Prevalence, severity and underdiagnosis of COPD in the primary care setting. Thorax 2008; 63:402–7. 25. Charlson ME, Pompei P, Ales KL, et al. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J Chronic Dis 1987;40:373–83. 26. Cannon CP. Cardiovascular disease and modifiable cardiometabolic risk factors. Clin Cornerstone 2007;8:11–28. 27. Gangireddy C, Rectenwald JR, Upchurch GR, et al. Risk factors and clinical impact of postoperative symptomatic venous thromboembolism. J Vasc Surg 2007;45:335–41; discussion 341–2. 28. ten Wolde M, Sohne M, Quak E, et al. Prognostic value of echocardiographically assessed right ventricular dysfunction in patients with pulmonary embolism. Arch Intern Med 2004;164:1685–9. 29. Gibson NS, Sohne M, Buller HR. Prognostic value of echocardiography and spiral computed tomography in patients with pulmonary embolism. Curr Opin Pulm Med 2005;11:380–4. 30. Jimenez D, Yusen RD. Prognostic models for selecting patients with acute pulmonary embolism for initial outpatient therapy. Curr Opin Pulm Med 2008;14:414–21. 31. Silverstein MD, Heit JA, Mohr DN, et al. Trends in the incidence of deep vein thrombosis and pulmonary embolism: a 25-year populationbased study. Arch Intern Med 1998;158:585–93. 32. Kasper W, Konstantinides S, Geibel A, et al. Management strategies and determinants of outcome in acute major pulmonary embolism: results of a multicenter registry. J Am Coll Cardiol 1997;30: 1165–71. 33. Lankeit M, Konstantinides S. Is it time for home treatment of pulmonary embolism? Eur Respir J. 2012;40:742–49. 34. Kearon C, Kahn SR, Agnelli G, et al. Antithrombotic therapy for venous thromboembolic disease: American College of Chest Physicians Evidence-Based Clinical Practice Guidelines (8th Edition). Chest 2008; 133:454S–545S.
469