INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION CONCLUSION REFERENCES

Portable, supine chest X-rays (CXRs) are the most commonly used noninvasive tool for identifying the presence, severity, or changes in pulmonary edema in the intensive care unit. However, their role in distinguishing hydrostatic pulmonary edema (HPE) from increased permeability pulmonary edema (PPE) is controversial (1). Inconsistencies in patient positioning (supine versus erect), film-to-target distances, and film quality are often cited as factors that may hinder radiographic analysis (4, 7). Moreover, the mechanical ventilator may alter the appearance of radiographic findings (8). Milne and colleagues have delineated radiologic criteria that help to differentiate among cardiogenic, renal, and injury patterns of edema (5). Clinical applications of Milne's investigation may be limited by the use of clinical diagnoses (rather than data derived from pulmonary artery catheterization) and postural variations in film technique (upright films in HPE patients versus supine films in 82% of PPE patients).

The optimum approach to interpretation of portable CXRs in mechanically ventilated, supine patients with varied diagnoses and hemodynamic parameters remains unclear. Accurate roentgenographic characterization of pulmonary edema assumes added priority with recently heightened concerns regarding the safety of more invasive procedures, such as pulmonary artery catheter placement (9, 10). Therefore, we performed the following investigations to determine whether there are reliable objective and/or subjective radiographic criteria to help the clinician and radiologist differentiate HPE from PPE in the intensive care unit. In addition, we appraised whether the process of objectively measuring multiple radiographic variables might enhance radiographic accuracy of diagnosis.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION CONCLUSION REFERENCES

The current investigation was a blinded comparison of portable CXR parameters with hemodynamic measurements and clinical data in 33 supine adults in the medical, surgical, and coronary care intensive care units of an 830-bed university hospital. The hemodynamic measurements were obtained within 1 h of the CXRs.

Patient Selection

Study patients were identified during another prospective, randomized, and blinded investigation of portable CXRs in 102 intubated, critically ill patients (8) and were included if they had a pulmonary artery catheter in place. Patients were selected without knowledge of their clinical diagnoses, indications for pulmonary artery catheter placement, or hemodynamic measurements. The exact time and date for each CXR were recorded to assess the relationship of CXR findings to hemodynamic data at the time of each film. The CXRs chosen for analysis were the first conventional portable chest radiographs (as opposed to computerized, digital images) taken for the patients with pulmonary artery catheters in place. They were obtained at various points of intervention during the patients' hospital stays.

Radiographic Technique and Interpretation

Conventional 14- × 17-inch screen portable, supine chest radiographs (Dupont Ultravision, Wilmington, DE) were obtained for each patient. The typical radiographic technique involved a 40-inch focal film distance (FFD), 90 kilovoltage peak (kVp), and 1 milliampere-second (mAs) exposure. A pulmonary/critical-care fellow accompanied radiology technologists for each CXR, simultaneously recording mechanical ventilation parameters from the exact breath upon which each CXR was obtained (8). The CXRs were then labeled randomly by number, obscuring any reference to patient name, age, or sex.

These CXRs were evaluated by three experienced chest radiologists, each of whom reviewed the films independently without any clinical information about the patients, their mechanical ventilation parameters, hemodynamic measurements, or prior CXRs. Radiologists recorded their findings on a standardized report form for each patient (Appendix ). Each CXR was determined by the radiologists to be of sufficient quality and appropriate exposure to allow interpretation. The vascular pedicle width (VPW) was measured as described by Milne (11), by dropping a perpendicular line from the point at which the left subclavian artery exits the aortic arch, and measuring across to the point at which the superior vena cava crosses the right mainstem bronchus (Figure 1). The cardiothoracic (CT) ratio was calculated by dividing the widest transverse diameter of the cardiac silhouette by the widest transverse diameter of the thorax above the diaphragm (1, 5, 12, 13).

View larger version (108K): [in this window] [in a new window]   View larger version (117K): [in this window] [in a new window]   Figure 1.   Portable CXR in a patient with HPE (left panel ) who had cardiogenic shock with a pulmonary artery occlusion pressure of 23 mm Hg and cardiac index of 2.3 L/min/m2. The CXR shows a basilar edema pattern, bilateral pleural effusions, CT ratio of 0.62, and VPW of 83 mm. Portable CXR in a patient with PPE (right panel ) who had acute respiratory distress syndrome with pulmonary artery occlusion pressure of 9 mm Hg and cardiac index of 5.4 L/min/m2. The CXR shows diffuse air space disease with a CT ratio of 0.44 and a VPW of 51 mm.

Correlative Clinical Data

Clinical information, including diagnosis and mechanical ventilation parameters, was recorded prospectively at the time the CXR was obtained. Independent of this information or radiologists' interpretations, the medical records of these patients were reviewed retrospectively to collect the following data corresponding to the time of the CXR: hemodynamic measurements (cardiac index, pulmonary artery occlusion pressure), intravenous fluid balance, and discharge diagnoses. At Wake-Forest University School of Medicine, pulmonary artery occlusion pressures are measured and recorded at end-exhalation with standard, electronically calibrated transducers and read from the wedged pulmonary artery waveform using midchest as the "zero" reference point for the transducer. Cardiac index was determined by thermodilution. These parameters were used to distinguish HPE from PPE: (1) pulmonary artery occlusion pressure, (2) cardiac index, and (3) clinical diagnosis. Patients with HPE were defined as having a pulmonary artery occlusion pressure 18 mm Hg and/or cardiac index  3 L/min/m², with a corresponding diagnosis consistent with cardiac disease and/or overhydration. These diagnoses included congestive heart failure, recent coronary artery bypass grafting, or renal failure (creatinine > 2 mg/dl). Patients with PPE were defined as having a pulmonary artery occlusion pressure < 18 mm Hg and/or a cardiac index > 3 L/min/m², with a clinical diagnosis of acute respiratory distress syndrome, sepsis, trauma, or pancreatitis. The primary criteria for classification were hemodynamic data, and they were satisfied in all patients. The clinical criteria were used to confirm the diagnostic category in each case (Table 1).

Statistical Analysis

Radiographic comparisons of subjects with HPE (n = 14) and PPE (n = 19) were based on combined information from 99 CXR interpretations (three radiologists' readings of CXRs on 33 patients). The approach implemented for testing hypotheses recognized that there were three ratings per subject. A contingency table was created from the cross-classification of diagnosis and a new variable created by combining information across raters. This was achieved by first creating a numerical score for each category of the rated variable in question (1 for present and 0 for absent; a variable with three ordinal categories would have scores 0, 1, and 2 for its categories). Then an average score was calculated from the three raters' scores. This new variable was cross-classified with diagnosis. The p values are based on the Wilcoxon Rank Sum Test, which compared two groups (HPE and PPE) with respect to the ordinal variable. Student's t test, however, was applied to average values of continuous variables (e.g., CT ratio and VPW measurement). The chi-square test was implemented to compare the radiologists' accuracy of diagnosing HPE versus PPE. A p < 0.05 was considered to indicate significance. Receiver-operating-characteristic curves were constructed to determine the discriminatory power of various VPW and CT ratios. Both the trapezoidal and nonparametric techniques were used to determine the area under the receiver-operating-characteristic curves as described by Hanley and McNeil (14, 15).

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION CONCLUSION REFERENCES

Of 102 prospectively identified intensive care unit patients with supine, portable CXRs, 33 had pulmonary artery catheters in place for clinical indications independent of this study. Using criteria outlined in METHODS, 14 patients had HPE and 19 had PPE. Details of patient demographics and diagnostic etiologies of respiratory failure are summarized in Table 1. Clinical diagnoses were consistent with the type of edema designated. Two patients designated as having HPE actually had pulmonary artery occlusion pressure values of 16 and 17 mm Hg at the time of their study CXR (see Figures 2 and 3). Each of these individuals had myocardial infarctions, a cardiac index < 2.5 L/min/m², prior pulmonary artery occlusion pressure > 20 mm Hg, and subsequent diuresis. Likewise, one patient classified as having PPE had a pulmonary artery occlusion pressure of 18 at the time of his CXR. This individual had sepsis with a cardiac index of 4.2 L/min/m², prior pulmonary artery occlusion pressure values < 15 mm Hg, and subsequent volume resuscitation. If these three patients are deleted from analyses, all of the statistically significant observations noted below remain. Overall, 12 of 14 HPE patients had furosemide within the previous 12 h and 3 of the 14 were on dobutamine drips (mean dose ± SD of 6 ± 2 µg/kg/min). Five of the 19 PPE patients were on dopamine drips (mean dose ± SD of 7 ± 2 µg/kg/min), and 7 had received furosemide in the past 12 h.

View larger version (9K): [in this window] [in a new window]   Figure 2.   Scatterplot showing the relationship between the pulmonary artery occlusion pressure and the VPW in all 33 patients (black diamond designates patients with HPE, and open square designates patients with PPE). Each data point represents the mean of three readings of each patient's CXR (correlation coefficient = 0.45, p = 0.0076). The horizontal dashed line represents a pulmonary artery occlusion pressure of 18 mm Hg, while the vertical dashed line represents the cut-off value of 63 mm derived from receiver-operating-characteristic analysis.

View larger version (10K): [in this window] [in a new window]   Figure 3.   Scatterplot showing the relationship between the pulmonary artery occlusion pressure and the CT ratio in all 33 patients (black diamond designates patients with HPE, and open square designates those with PPE). Each data point represents the mean of three readings of each patient's CXR (correlation coefficient = 0.52, p = 0.0016). The horizontal dashed line represents a pulmonary artery occlusion pressure of 18 mm Hg, while the vertical dashed line represents the cut-off value of 0.52 derived from receiver-operating-characteristic analysis.

Radiographic measurements and the frequencies of CXR abnormalities are summarized and related to patterns of pulmonary edema in Table 2. A CT ratio of > 0.50 was seen in 86% of patients with HPE, versus 58% of individuals with PPE (p < 0.001). Subjective impressions of cardiomegaly (p < 0.01) and increased VPW (p = 0.02) were associated with HPE. There was a significant relationship between the pulmonary artery occlusion pressure and VPW (p = 0.0076, r = 0.45) and CT ratio (p = 0.0016, r = 0.52), as shown in Figures 2 and 3. The relationship between the VPW and CT ratio was also significant (Figure 4; p = 0.0032, r = 0.49). In addition, pleural effusions were present in 64% of patients with HPE, versus 3% of those with PPE (p = 0.08). Other CXR criteria did not differentiate patients with HPE from those with PPE (Table 2). Fifteen of 19 (79%) of PPE patients showed one or more roentgenographic signs of volume overload.

View larger version (9K): [in this window] [in a new window]   Figure 4.   Scatterplot showing the relationship between the VPW and CT ratio in all 33 patients (black diamond designates patients with HPE, and open square designates patients with PPE). Each data point represents the mean of three readings of each patient's CXR (correlation coefficient = 0.49, p = 0.0032). The horizontal dashed line represents a CT ratio of 0.52, while the vertical dashed line represents a VPW of 63 mm derived from receiver-operating-characteristic analysis.

The radiologists' estimates of pulmonary artery occlusion pressure are shown in Table 3. Their mean diagnostic accuracy in distinguishing HPE and PPE was 41% (Table 4). Because many patients had pulmonary artery occlusion pressures between 15 and 20 mm Hg, we performed a subgroup analysis to examine the accuracy of the radiologists when considering only those patients who had pulmonary artery occlusion pressures of  15 or  20 mm Hg. We found that in these extremes, the sensitivity and specificity were 71% and 50%, respectively. The radiologists' accuracy only rose to 57%, which was not a significant difference. There was a trend towards a higher accuracy of diagnosing HPE than PPE (p = 0.067).

The receiver-operating-characteristic curves delineating "best" cut-off values of VPW, CT ratio, and their combination are shown in Figures 5-7. The receiver-operating-characteristic curve for VPW measurements (Figure 5) demonstrates a quick rise, allowing for optimal distinction between HPE and PPE somewhere between 63 and 70 mm in supine, intubated patients (area under the curve = 0.706). A similar receiver-operating-characteristic curve was extrapolated for the CT ratio (Figure 6), with an optimal cut-off between 0.52 and 0.54 (area under the curve = 0.715). We constructed a third receiver-operating-characteristic curve by combining various VPW and CT ratio measurements (Figure 7); it had the best distinction at a VPW of > 63 mm plus a CT ratio of > 0.52 (p = 0.027) and an area under the curve of 0.807. Test characteristics for specific cut-off values are shown in Table 5. If the "best" choice cut-offs of VPW > 68 mm, CT ratio > 0.54, or a combination of VPW > 63 mm and CT ratio > 0.52 were applied, radiologists' accuracy in predicting the type of pulmonary edema might have been increased to 67%, 67%, and 73%, respectively (Table 5).

View larger version (74K): [in this window] [in a new window]   Figure 5.   Receiver-operating-characteristic curve for VPW. The curve shows the ability of the VPW to differentiate HPE from PPE at different cut-off points. The area under the curve is 0.706.

View larger version (76K): [in this window] [in a new window]   Figure 6.   Receiver-operating-characteristic curve for the CT ratio. The curve shows the ability of the CT ratio to differentiate HPE from PPE at different cut-off points. The area under the curve is 0.715.

View larger version (77K): [in this window] [in a new window]   Figure 7.   Receiver-operating-characteristic curve for combined VPW and CT ratio measurements. The curve shows the ability of the combined measurements to differentiate HPE from PPE at different cut-off points. The area under the curve is 0.807.

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION CONCLUSION REFERENCES

Previously published reports have repeatedly demonstrated that clinical estimates of hemodynamic status in critically ill patients range from 30% to 50% in accuracy (16). In this investigation, radiologists' estimates of both the patients' pulmonary artery occlusion pressure and their type of pulmonary edema showed a similarly low diagnostic accuracy (Tables 3 and 4). In order to improve the clinical assessment and therapy of patients, clinicians commonly use ancillary diagnostic testing such as portable CXRs or pulmonary artery catheters. With recent concerns about the efficacy and safety of using Swan-Ganz catheterization to assess hemodynamic status (9, 10), enhancing the role of simple, noninvasive diagnostic tools such as the portable chest radiograph in the diagnosis of pulmonary edema is imperative.

In many medical centers in the Untied States, approximately half of all CXRs are now performed at the bedside with mobile equipment (23). Although most patients receiving mechanical ventilation are monitored with portable CXRs on a daily basis, uncertainties remain regarding their appropriate use and interpretation (8, 24, 25). Other investigators have attempted to provide distinct, objective data sets that can accurately differentiate among cardiac, renal, and injury edema (1, 4, 5). However, surprisingly few data address the most common scenario in which the population of mechanically ventilated, intensive care unit patients are filmed: i.e., using portable, supine CXRs. This investigation addresses an objective method of improving the radiologic distinction of HPE from PPE in this challenging setting.

Improving Radiologic Interpretations

In our study of supine patients, we found the following parameters on a single, portable CXR to be valuable in this differentiation: (1) objective measurement of a larger cardiothoracic ratio; (2) a subjective impression of cardiomegaly; (3) a subjective impression of a widened vascular pedicle; (4) a trend toward the presence of pleural effusions in HPE. Despite the potential value of these findings, experienced chest radiologists' overall accuracy in diagnosing and distinguishing HPE from PPE using a single portable CXR was low, ranging from 36% to 45%. It is not unexpected that there would be some discordance between the radiographic findings of pulmonary edema, because they typically lag behind changes in hydrostatic pressure, which can occur more rapidly. In addition, there are clinical situations that may cause the pulmonary artery catheter readings to be discordant from patients' underlying diagnoses. For example, patients with HPE may have pulmonary artery occlusion pressures less than 18 after diuresis. Similarly, patients with PPE can have elevated pulmonary artery occlusion pressures due to the effects of volume administration or positive end-expiratory pressure.

Receiver-operating-characteristic curve analysis of the data from these CXRs suggests that radiologists' accuracy might be improved by employing a cut-off of > 63 mm for VPW along with a CT ratio of > 0.52 (p = 0.027; in Figure 7, area under the curve = 0.807). With this approach, the radiologists may have been able to improve their diagnostic accuracy to 73% (Table 5), a considerable improvement over that previously reported in the literature. The optimal cut-offs for CT ratio and VPW measurements may vary depending on the patient population and sample size that is studied, and further investigation is warranted. Future investigations should also compare radiologists' accuracy when reading films in the absence of clinical data with their accuracy in the common clinical scenario, in which radiologists are given some background information about the patient's underlying medical condition and physical examination. Importantly, this investigation represented a selected group of patients in whom pulmonary catheters were placed for varying clinical uncertainties regarding the type of pulmonary edema, potentially introducing bias against radiologists' performance. It is not clear how these results apply to a more general group of intensive-care unit patients.

Although we used a data form (Appendix ) similar to that of Milne and colleagues (5), our observations differ from theirs. This may reflect not only differences in patient selection, but also the influences of posture and mechanical ventilation upon roentgenographic findings (8, 11). The distinguishing characteristics of different etiologies of edema may be altered when all of the films are taken in the supine position and the patients are receiving mechanical ventilation. For example, there was only a 21% incidence of septal lines in HPE patients. The investigation by Milne and colleagues (5) included universally upright films in the HPE patients and predominantly supine films in the PPE patients. In Aberle's study of critically ill patients, only half of the patients with HPE were filmed supine, while 93% of the PPE patients were supine (1). Furthermore, our patients' films were obtained at varying times during their hospitalization and treatment courses, a confounding problem which mimics the daily scenarios confronting radiologists and chest clinicians.

The Cardiothoracic Ratio

In 1919, Danzer undertook one of the earliest studies of the CT ratio. After investigating nearly 500 patients without the aid of hemodynamic parameters, he determined that any measurement over 0.5 was suspicious, and over 0.52 "definitely pathological" (12). A later report by Comeau and White found that 15 to 25% of normal patients had a CT ratio greater than 0.5, and advised using CT ratio prediction tables based upon sex, image technique, and phase of respiration (13). Nevertheless, the most commonly used radiographic definition of cardiomegaly on the standard, upright, posteroanterior CXR is a CT ratio over 0.5 (26). A recent meta-analysis of 29 studies determined that cardiomegaly on the CXR was the best correlate for predicting a reduced ejection fraction, with a sensitivity and specificity of 51% and 79%, respectively (28). The interobserver agreement was only moderate for this observation.

It is important to recognize confounding factors that can lead to false-positive interpretations of cardiomegaly, such as an apical fat pad, a transversely positioned heart, an expiratory film, or a decrease in the thoracic width (13, 28). Portable films taken in the anteroposterior and supine position enlarge the appearance of the cardiac silhouette. An investigation by Milne and colleagues (31) determined that for portable anteroposterior CXRs with a focal film distance of 40 inches (the type used in this investigation), a correction factor of 12.5% for the CT ratio can be used to approximate whether or not cardiomegaly exists. However, this determination was made in healthy patients and was not correlated with hemodynamic parameters. Such correction factors have not been consistently applied in clinical practice. In our investigation, receiver-operating-characteristic curves demonstrated that the optimum cut-off range for the CT ratio when used in isolation to differentiate HPE from PPE was 0.52 to 0.54. However, like the VPW, when used in the "combination" receiver-operating-characteristic curve (Figure 7), the lower cut-off of 0.52 appeared best.

The Vascular Pedicle Width

The present observations confirm the diagnostic value of VPW in appraising patients with pulmonary edema. Our receiver-operating-characteristic curves in patients with pulmonary edema showed that the best values to differentiate HPE from PPE ranged from 63 mm to 70 mm, depending upon whether or not the VPW was used alone or in combination with the CT ratio. Data from a previous investigation of strictly upright CXRs showed that 95% of normal people have a VPW of 48 ± 10 mm (11). Patient posture and rotation influence clinical use of VPW because supine positioning increases the VPW by 20% in normal and cardiac volunteers (11). A 15-degree rotation of the patient to the right will increase the VPW by 6% (likewise, rotation to the left will decrease the VPW). Further study indicates that any increase in alveolar pressure (i.e., positive end-expiratory pressure, chronic obstructive pulmonary disease) may decrease the VPW by varying amounts. Inspiration and expiration were found to cause little to no change in the VPW (11).

Changes in VPW have been associated with altered intravascular volume status, but anecdotal observations suggest that many clinicians are unaware of this relationship. An increase of 1 L in total blood volume will increase the VPW by 5 mm when filmed in the upright position (32). Widening of the vascular pedicle on serial films in burned patients has antedated development of pulmonary edema following fluid resuscitation (33). Relationships of VPW to pulmonary artery occlusion pressure and CT ratio have not been reported previously. Interestingly, in the previously cited meta-analysis investigating the utility of the CXR to diagnose left ventricular dysfunction (either an elevated preload or reduced ejection fraction) (28), VPW was not addressed.

The "Classic" Features of Hydrostatic Pulmonary Edema and Permeability Pulmonary Edema

Radiologists tended to be more accurate in their diagnosis of HPE than PPE (Table 4, p = 0.067). To better understand this trend, it is necessary to consider the number of patients who tested positive for all of our statistically significant criteria (CT ratio > 0.5, subjective cardiomegaly, subjective widened VPW, and pleural effusions). Half of our patients with HPE were positive for all four (Table 2). This finding suggests that HPE patients often have multiple signs of volume overload, perhaps increasing the likelihood that it will be recognized. Surprisingly, less than 25% of patients with PPE were negative for all four criteria. This observation suggests that a portable CXR often shows evidence of one or two signs of volume overload in patients with PPE, thus confounding the radiologists' accuracy.

The "classic" interstitial and alveolar signs of edema have been found with varying frequencies by different investigators. The low frequencies of correct diagnoses by experienced chest radiologists suggests that there are obstacles in processing or weighing these roentgen signs as an overall interpretation of the film is generated. Milne concluded that septal lines and peribronchial cuffing were seen more often in renal and cardiogenic edema, whereas air bronchograms were seen less often than in edema due to lung injury, i.e., acute respiratory distress syndrome (5). We found that our supine patients with either HPE or PPE had similar frequencies of all of these radiographic findings. Vascular redistribution, although recently determined to be the best indicator of increased preload (sensitivity 65%, specificity 67%) (28), was also not helpful in differentiating HPE from PPE (p = 0.33). As previously documented (34), marked gravitational changes occur in these vessels, which may explain why they are less helpful in supine patients. In some circumstances, the lack of an association may reflect the small sample size (i.e., a type II error).

Format for Chest X-Ray Interpretations

The design of our investigation might have altered our radiologists' performance. We did not allow readers to review clinical information related to the patients because we wished to minimize bias and because radiologists are not consistently provided this information in clinical practice. In a study of pneumonia and acute respiratory distress syndrome, Winer-Muram found that neither a review of serial radiographs nor knowledge of clinical data improved radiologists' accuracy, and in some cases actually worsened their interpretations (35). Controversy exists over whether formal CXR reading forms could skew analysis by interfering with the thought processes of the radiologist. However, guidelines have been proposed for formal incorporation of a standardized format for all radiographs of mechanically ventilated patients (4). We also believed that the only way to maximize objectivity and to approach logical interpretation was to follow a uniform set of data. Nevertheless, despite this objective approach, radiologists' accuracy was poor. We did not address effects related to the timing of the CXR within evolution of the patients illness. It is possible that more accurate differentiation of HPE from PPE at earlier phases, before therapeutic interventions, could be achieved.

	CONCLUSIONS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION CONCLUSION REFERENCES

Although previous investigations have shown certain radiographic characteristics to be valuable in a mixture of supine and upright films, this examination solely studied portable CXRs in supine, mechanically ventilated patients. We determined few characteristics to be useful in differentiating HPE from PPE. These included a CT ratio > 0.50, a subjective impression of both cardiomegaly and an enlarged VPW, and a trend toward pleural effusions. There was a significant correlation between the level of pulmonary artery occlusion pressure and both the VPW and CT ratios. In addition, receiver-operating-characteristic curves suggested that the best criteria for distinguishing HPE from PPE could be found using a combined VPW of > 63 mm and CT ratio > 0.52. This approach may have increased the radiologists' accuracy from 41% to 73%. Newer, digitized, computer radiographs printed at various magnifications could limit the usefulness of these observations and introduce considerable challenges in the application of these data. Because the evaluation of patients with pulmonary edema is usually a dynamic process that considers their response to the prescribed therapy, further prospective investigation possibly using serial, computerized CXRs in a larger group of patients with diverse etiologies of pulmonary edema is needed to validate these findings and to define better the role of the portable CXR in supine intensive care unit patients.