1Service dAnesthésie-Réanimation chirurgicale, 2Antenne de Biostatistiques et dInformatique médicale and 3Laboratoire des Explorations Fonctionnelles, Hôpital Tenon, 4 rue de la Chine, F-75970 Paris Cedex 20, France*Corresponding author
This article is accompanied by Editorial I.
Accepted for publication: November 13, 2000
![]() |
Abstract |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
Br J Anaesth 2001; 86: 75462
Keywords: blood volume; intensive care; measurement techniques, Bayes theorem; measurement techniques, radioactive labelled albumin; complications, hypovolaemia; clinical trials
![]() |
Introduction |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
Clinical evaluation of CBV could encounter two potential problems. First, knowing absolute CBV is less important than the effective CBV in determining pre-load responsiveness or in defining the determinants of a particular haemodynamic state. For the physician, hypovolaemia means low pre-load, not just low CBV. Circulating blood volume is one of the most important factors that affect heart pre-load; changes in the compliance of veins maintain the correct relationship between vascular space and CBV.2 9 Second, the clinical signs used are really aimed at measuring total body extracellular volume which includes intravascular volume and interstitial volume, and therefore, one would not expect them to closely predict intravascular volume.
The aim of this study was to evaluate the ability of the routine clinical parameters used to estimate fluid volume status to predict CBV, and to compare with measurements of CBV using the [125I]albumin technique. The study was divided into two parts: first, clinical variables that best discriminated between patients with and without low CBV were determined; and second, a scoring system which combines elements of Bayess theorem with those of logistic regression was constructed and prospectively assessed in another group of patients.
![]() |
Patients and methods |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
The patient population consisted of subjects admitted to our ICU during a 2-yr period between January 1996 and December 1997, in whom clinical assessment of CBV status had to be confirmed by laboratory measurement, because, after several days in the ICU, the physician was uncertain about the CBV status. Bedside CBV determination is readily available at our institution. The initial study population consisted of 68 patients and represented the model development sample from which the prediction system was derived. Another 30 patients were then included to constitute the test data set. The main reasons for admission to ICU included 39 post-operative cases (orthopaedic, thoracic, vascular and abdominal surgery); 45 cases of sepsis as previously defined,10 six cases of gastrointestinal haemorrhage and eight miscellaneous causes.
The following information was recorded: age, sex, height and usual weight, new Simplified Acute Physiology Score (SAPS II),11 duration of ICU stay, time of CBV determination after ICU admission, need for mechanical ventilation or vasoactive drugs, and mortality in ICU.
Clinical assessment of extracellular fluid volume
Each patient underwent two independent physical examinations by two attending physicians before CBV determination. Results of these examinations were subsequently compared. When a disagreement occurred, a third physician was required to make final decision. A chest x-ray was available for all patients. According to previous reports and guidelines, 6 12 13 clinical examination paid special attention to seven signs readily available at the bedside: (1) presence or absence of fluid losses since ICU admission (chest and abdominal drainage, aspiration of gastric contents); (2) fluid balance in the last 24 h by recording intake and output (positive fluid balance if above 400 ml); (3) skin mottling; (4) presence of pulmonary congestion based on detection of pulmonary rales and crackles on physical examination, and/or alveolar oedema and pulmonary vasculature redistribution on chest x-ray; (5) presence of congestive heart failure supported by a past medical history, cardiac enlargement and pulmonary oedema on physical and chest x-ray examination, and gallop rhythm; (6) peripheral oedema; and (7) detection of an enlarged third space: ascites (bulging flanks, fluid wave, shifting dullness), and pleural effusion (dullness on physical examination, compatible chest x-ray). The following variables were also recorded: systolic and diastolic arterial pressure, heart rate, temperature, urine output over the last 24 h, plasma and urinary sodium concentrations, serum total proteins, haemoglobin and haematocrit values.
Central venous pressure (CVP) was measured with a pressure transducer (PVB, Kirchsceon, Germany). Pressures were obtained after calibration, zeroing to atmospheric pressure and using the mid-chest level as reference. Transducers were connected to bedside amplifiers (HP M10469102B, Hewlett Packard). Central venous pressure was recorded at end-expiration.
Circulating blood volume determination
Measurement of CBV was performed with [125I]human serum albumin (SERALB-125®; CIS bio international, Gyf sur Yvette, France) and CBV equipment (Volumetron®; AMES Co, Div. Miles Lab. Inc.; Elkhart, IN, USA), which automatically calculated the volume from the radioactivity injected and from the radioactivity of a post-injection whole blood sample, as previously described.5
SERALB-125® was supplied as a sterile solution of human serum albumin labelled with iodine 125I, made isotonic with sodium chloride. The radioactive concentration was 185 kBq ml1 (5 µCi ml1) at the calibration date; SERALB-125® contains 9 mg of human serum albumin ml1. Not less than 97% of the total radioactivity was bound to human serum albumin. After treatment any acute episode of severe hypotension, 3.5 µCi of SERALB-125® were injected i.v. (# 0.7 ml of solution). The radioactivity injected was assumed to be the difference of the activities contained in the syringe before and after injection. Blood was withdrawn from a vein of the contralateral arm after 10 min, as previously recommended.5 The concentration of test substance in the sample was obtained from simultaneous measurements of fixed volumes of pre-mixing and post-mixing blood samples.
Plasma volume and cell volume were calculated from CBV and peripheral haematocrit corrected for the body-to-venous haematocrit ratio (0.91). Interpretation of results was based on comparison between observed values and expected values in healthy subjects of the same sex, height, weight and age. The expected values for healthy subjects have been previously reported.14 15 Precision of CBV determination is ±5% (Dr F. Paillard, personal communication). Hypovolaemia was defined as a CBV at least 10% lower than the predicted mean normal CBV.
Statistical analysis
Hypovolaemic and non-hypovolaemic patients were compared using the chi-squared test or Fishers exact test for categorical variables, and Students t- or MannWitney U-test for continuous variables. A two-sided formulation with a P value <0.05 was required for statistical significance. Results are expressed as mean (SD) for continuous variables, and as per cent for categorical variables.
Agreement between physicians on the presence or absence of a clinical sign was assessed by the kappa measurement of agreement.16 Kappa values exceeding 0.75 represent excellent agreement, values between 0.4 and 0.75 indicate fair to good agreement, and values less than 0.4 indicate poor agreement.16
Standard formulas were used to calculate sensitivity (true positives/[true positives+false negatives]), specificity (true negatives/[true negatives+false positives]), positive predictive value (true positives/[true positive+false positives]), and negative predictive value (true negative/[true negatives+ false negatives]) for each clinical parameter.17
To decide the optimum cut off point for CVP, a Receiver Operating Characteristics (ROC) curve was constructed, which plots true- and false-positives rates (sensitivity and 1 minus specificity, respectively) for a series of cut off points.17
Calculation of SpiegelhalterKnill-Jones weightings
Many statistical methods have been developed to improve the physicians judgment. Over the past few years, Spiegelhalter and Knill-Jones have proposed a simple scoring system which adds precision to risk assessment in individual patients.18 19 This statistical method combines elements of Bayes theorem with those of logistic regression. The result is a system that neatly sidesteps some of the main disadvantages of the two original techniques. For example, it does not assume that all risk factors are acting independently within each outcome class (the independence Bayes assumption, which is central to many bayesian analyses) because an adjustment is made, while at the same time predictions are presented in a form which is less mathematical and much more clinically relevant than the output of a conventional logistic regression analysis. Another reason why this method may be preferred over simple logistic regression is the ability to integrate zero for missing data. Finally, the weights of evidence may be viewed as tools for refining risk prediction by taking into account key clinical parameters in the individual patient. Once the SpiegelhalterKnill-Jones weightings have been derived they can be used by the non-mathematician, but the process of derivation from the training data set requires statistical skill (see Appendix). The SpiegelhalterKnill-Jones scoring system, applied to the evaluation of CBV could, therefore, improve the physicians clinical judgment.
Validation of the clinical scoring system
The predictive accuracy of the Spiegelhalter and Knill-Jones weighting system was measured in the 30 patients constituting the test data set. The predictive probabilities have been grouped into four categories and the observed number of hypovolaemic patients (O) in each category is noted. The four categories are defined according to the range of probabilities of expected hypovolaemia: category 1, 010%; category 2, 1149%; category 3, 5070%; and category 4, 71100%.
The expected number of hypovolaemic patients (E), assuming the prediction is completely reliable, is also shown. According to the null hypothesis that the predictions are reliable, O would be approximately distributed with the mean E and a standard error (SE) equal to [E(1E/n)]1/2. Hence, Z=(OE)/(SE) will be approximately a standard normal statistic according to the null hypothesis of perfect reliability. Values of Z greater than 2 suggest that the probability attributed to hypovolaemia is too low, while values of Z less than 2 suggest that the probabilities are too high.19
![]() |
Results |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
|
|
The accuracy of each of the seven clinical signs in the model development sample is shown in Table 3. Sensitivity was 0.94 for absence of congestive heart failure, which was marginally higher than the sensitivity of absence of pulmonary oedema or absence of third spacing (both equal to 0.92). Specificity was highest for the presence of fluid losses and presence of skin mottling (both equal to 0.78), and lowest for absence of third spacing (0.19). Positive predictive value was highest for the absence of peripheral oedema (0.62). Negative predictive value was highest for the presence of skin mottling (0.49), closely followed by the presence of fluid losses (0.48).
|
|
Two steps are required to predict the risk of hypovolaemia in an individual patient. First, by referring to Table 4, the starting score is added to the appropriate adjusted weights of evidence for that patient. Then, by referring to Figure 1, the total score obtained is converted into a probability of hypovolaemia. These steps are illustrated in the following example. A patient presented with septic shock after cholangitis. Three days after ICU admission, physical examination showed peripheral oedema, fluid losses, and a negative fluid balance. No signs of pulmonary congestion, congestive heart failure, skin mottling, or increased third space was found. Central venous pressure could not be measured. The adjusted weights of evidence for pulmonary congestion (+20), peripheral oedema (29), skin mottling (10), congestive heart failure (+11), third spacing (+27), fluid losses (+14), fluid balance (+41), and CVP (0) were added to the starting score (5 in every case) (Table 4). The total score was +74. A total score of +74 corresponds to a predicted probability of hypovolaemia of 70% (Fig. 1).
|
Figure 2 gives the reliability of adjusted predictions. The percentage of hypovolaemia observed in patients in the test data set is plotted against the percentage of hypovolaemia expected from the calculated predictions. The line produced is close to the line of identity. The results indicate that the predicted probabilities are reliable, especially for patients with a high probabilitythat is, patients predicted to have, say 60% chance of hypovolaemia will actually have hypovolaemia about 60% of the time.
|
![]() |
Discussion |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
We chose the radioactive-labelled albumin method for quantitative CBV determination as the standard. However, using this method in ICU patients raises several problems.2 First, measurement of CBV depends on adequacy of mixing of [125I]albumin throughout the vascular system and rapid whole body disappearance of the tracer.20 In the present study, the measurement of CBV 10 min after tracer injection corresponded to the best compromise between complete mixing of the tracer with plasma and its disappearance.5 21 22 However, despite a close correlation with other techniques,23 this method could lead to a slight overestimation of CBV.20 23 Second, septic patients have altered capillary permeability.24 However, [125I]albumin disappearance rate has been reported to be almost unmodified after endotoxin administration21 and determination of CBV has been shown to be accurate in septic patients.25 Third, CBV measurements are commonly compared with predicted CBV values derived from measurements in a normal population. However, the normal values of CBV could be different in critically ill patients. Unfortunately, to our knowledge, there are no published data concerning normal CBV values in critically ill patients. Despite these concerns, direct measurement of CBV by a radioactive tracer in ICU patients has been shown to be accurate and reproducible.5 25 26
We have defined hypovolaemia in term of intravascular volume because CBV is one of the most important factors affecting pre-load. In most clinical studies, the term hypovolaemia refers to either volume depletion or dehydration.8 One could argue that measurement of CBV only refers to intravascular volume and ignores interstitial and intracellular volumes, but it has been demonstrated that CBV correlates well with extracellular fluid volume.27 Both CBV and vascular capacitance determine the effective CBV. It could, therefore, be hypothesized that septic patients could have an unchanged CBV despite a decreased pre-load. However, Rothe and colleagues reported that, in an endotoxin shock model, CBV decreased by nearly 30% after endotoxin infusion.21 It has also been shown that effective vascular compliance is already decreased in patients under mechanical ventilation with sepsis syndrome, or when sympathetic nervous system tone is elicited.9 It could also be argued that the division into hypovolaemia and non-hypovolaemia on the basis of a greater than 10% reduction in the measured CBV could be too stringent. However, the mean CBV deficit was 948 ml in hypovolaemic patients and only two non-hypovolaemic patients overlapped with this threshold.
Previous studies have indicated that clinical estimates of a patients fluid volume status are not reliably perceived by physical examination alone, with an accuracy range from 30 to 50%.58 2831 One possible explanation is that physicians do not optimally use clinical information when assessing fluid volume status. 32 In our series, hypovolaemia was present in about one half of ICU patients, as reported by Shoemaker and colleagues using the same method in a comparable patient population.26 Ourselves and others25 33 have noted that the use of routine haemodynamic variables are of little value to differentiate hypovolaemic and non-hypovolaemic patients. Likewise, spot urinary sodium concentration is not useful in this setting in contrast with its value in stable, non-critically ill patients.6 Renal losses of sodium and water are often a result of an osmotic diuresis. One common culprit is glycosuria, but other causes (e.g. post-obstructive diuresis, profuse diuresis during recovery from acute renal failure) may account for such differences. The value of haemoglobin concentration has several limitations. During rapid and severe bleeding, the patients may exsanguinate before transcapillary migration of interstitial fluid can significantly reduce the haemoglobin value. On the other hand, haemoglobin changes become almost uninterpretable when patients have been given large volumes of packed red cells and fluids.
The clinical signs that were used are really aimed at measuring total body extracellular volume and are not directly indicative of intravascular volume. However, there is a relationship between clinical features such as oedema and aspiratory crackles and CBV.27 On the other hand, the appearance of reduced CBV in the face of expanded extracellular fluid has been described.26 As a consequence, based on the values of the likelihood ratio and as shown in Table 3, none of these findings is particularly helpful when present in isolation, as previously reported.8 On the other hand, combinations of physical signs appear to be more helpful.8 Inspection of the range of adjusted weights given in Table 4 allows comparison of the value of the symptoms. The wider the range of an indicator, the greater its potential value as a diagnostic item. However, it is possible to have a very powerful symptom that is so rarely present that it would not normally be ascertained in clinical evaluation. Central venous pressure was the last variable taken into account in the score, as CVP is commonly measured in critically ill ICU patients.32 However, we found that only a value <2 mm Hg provided evidence in favour of hypovolaemia. CVP may vary considerably in critically ill patients, because of heartlung interactions, especially in mechanically ventilated patients,34 and therefore only extreme values seem to have any clinical significance.
Regarding the SpiegelhalterKnill-Jones system, three points must be stressed. First, the predictive accuracy of the system appears reliable. However, the system seems to be less reliable for patients with a low probability of hypovolaemia (Figure 2). Second, some cardiopulmonary signs have been suggested to be of questionable reliability.35 36 However, most of the signs used to construct the system frequently occur in the ICU setting, and unreliability of the signs seems to be linked to their low frequency of appearance.35 36 Finally, if the criteria of hypovolaemia change, the sensitivity and specificity of each of the signs may be different,8 affecting the prediction of the model.
Following recent concerns and doubts about the efficacy and safety of using Swan-Ganz catheterization to assess haemodynamic status,37 there are no guidelines to help the physician to decide when to use or to withhold invasive monitoring in individual patients.29 We believe the application of such a method could lead to a more judicious selection of diagnostic tests. In conclusion, physicians working in ICU are daily called upon to predict the patients fluid volume status on the basis of existing symptoms and signs, physical findings, and laboratory results. As illustrated by the SpiegelhalterKnill-Jones method, a science of clinical prediction has been developed, and it is now possible to make quantitative predictions by using statistical models and to more rigorously assess the accuracy of these predictions.
![]() |
Appendix |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
(1) Calculation of the likelihood ratio (LR)
To calculate the LR for a given sign, the patients were divided into hypovolaemic and non-hypovolaemic patients. A standard 2x2 table allowed calculation of sensitivity, specificity, and LR for the presence or absence of a given sign:
LR for the presence of a particular sign=sensitivity/(1specificity)
LR for the absence of a particular sign=(1sensitivity)/specificity.
(2) Bayes theorem for calculation of the post-test probability of disease
The independence Bayes equation may be expressed as:
Posterior odds=prior oddsxLR of sign 1xLR of sign 2 . . .xLR of sign N (1)
where posterior odds is the predicted odds of hypovolaemia in an individual, and prior odds is the odds of hypovolaemia in the study population.
(3) Converting LR into scores and weights
To convert LR into a simple score that can be added up, we took the natural logarithm of this LR, as the logarithm can be added rather than multiplied, thus simplifying the process. For further simplicity, this value was multiplied by 100 and rounded off so that crude weights could be expressed as whole numbers.
Equation (1) therefore becomes:
100 Ln posterior odds=100 Ln prior odds+100 Ln LR of sign 1+100 Ln LR of sign 2 . . .+100 Ln LR of sign N (2).
By using the terminology of the Spiegelhalter and Knill-Jones method, equation (2) becomes:
Total score (T)=starting score+crude weights of sign 1+crude weights of sign 2 . . .+crude weights of sign N (3).
The starting score reflects the prior probability of hypovolaemia and could be different according to the specificity of the ICU.
(4) Adjustment of crude weights
The use of Bayes theorem could lead to considerable problems in overestimating the probabilities of hypovolaemia, as the assumption of independence of different signs is rarely satisfied in practice. A statistical method of making an adjustment must therefore be defined. The Spiegelhalter and Knill-Jones method calculates adjusted weights of evidence, which are obtained by entering the value of the crude weights as independent variables in a logistic regression equation. Data were computerized (Compaq prolinea 575E) and analysed using BMDP statistical packages (BMDP Statistical Software, 7.0 software release 1992; Inc. Los Angeles, CA, USA). Goodness of fit was assessed by the Hosmer and Lemeshow chi-squared test. The resulting regression coefficients a0, a1, a2, an are displayed, together with their standard error (SE). After multiplication of crude weights by their respective regression coefficient, equation (3) becomes:
Total score (T)=a0+adjusted weights of sign 1+adjusted weights of sign 2 . . .+adjusted weights of sign N (4).
(5) Converting scores back to probability of disease
Because T=100 (Ln posterior odds) and because odds=(probability of event)/(1probability of event), it may be calculated that probability of hypovolaemia (in %)=(eT/100/1+eT/100 )x100=1/(eT/100+1)x100. This can be performed more rapidly by using a simple graph.
![]() |
References |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
2 Jones JG, Wardrop CAJ. Measurement of blood volume in surgical and intensive care practice. Br J Anaesth 2000; 84: 22635
3 Shoemaker WC, Montgomery ES, Kaplan E, Elwyn DH. Physiologic patterns in surviving and nonsurviving shock patients. Arch Surg 1973; 106: 6306[ISI][Medline]
4 Mitchell JP, Schuller D, Calandrino FS, Schuster DP. Improved outcome based on fluid management in critically ill patients requiring pulmonary artery catheterization. Am Rev Resp Dis 1992; 145: 9908[ISI][Medline]
5 Williams JA, Fine J. Measurement of blood volume with a new apparatus. N Engl J Med 1961; 264: 8428[ISI]
6 Chung H-M, Kluge R, Schrier RW, Anderson RJ. Clinical assessment of extracellular fluid volume in hyponatremia. Am J Med 1987; 83: 9058[ISI][Medline]
7 Cook DJ. Clinical assessment of central venous pressure in the critically ill. Am J Med Sci 1990; 229: 1758
8 McGee S, Abernethy WB, Simel DL. Is this patient hypovolemic? JAMA 1999; 281: 10229
9 Stéphan F, Novara A, Tournier B, et al. Determination of total effective vascular compliance in patients with sepsis syndrome. Am J Respir Crit Care Med 1998; 157: 506
10 American College of Chest Physicians/Society of Critical Care Medicine Consensus Committee. Definitions for sepsis and organ failures and guidelines for the use of innovative therapies in sepsis. Crit Care Med 1992; 20: 86474[ISI][Medline]
11 Le Gall J-R, Lemeshow S, Saulnier F. A new simplified acute physiology score (SAPS II) based on a European/North American multicenter study. JAMA 1993; 270: 295763[Abstract]
12 Brezis M, Rosen S, Epstein KH. Acute renal failure. In: Brenner BM, Rector Jr FC eds. The Kidney, 4th edn. Philadelphia: WB Saunders Company, 1991; 9931061
13 Greco BA, Jacobson HR. Fluid and electrolyte problems with surgery, trauma, and burns. In: Koko JP, Tannen RL, eds. Fluids and Electrolytes, 3rd edn. Philadelphia: WB Saunders Company, 1996; 72958
14 Wennesland R, Brown E, Hopper J, et al. Red cell, plasma and blood volume in healthy men measured by radiochromium cell tagging and hematocrit: influence of age, somatotype and habits of physical activity on the variance after regression of volumes to height and weight combined. J Clin Invest 1959; 38: 106577[ISI]
15 Brown E, Hopper J, Hodges JL, Bradley B, Wennesland R, Yamauchi H. Red cell, plasma, and blood volume in healthy women measured by radiochromium cell-labelling and hematocrit. J Clin Invest 1962; 41: 218290[ISI]
16 Armitage P, Berry G. Statistical Methods in Medical Research, 6th edn. Oxford: Blackwell Scientific Publications, 1994
17 Griner PF, Mayewski RJ, Mushlin AI, Greenland P. Selection and interpretation of diagnostic tests and procedures. Principles and applications. Ann Intern Med 1981; 94: 553600[ISI]
18 Spiegelhalter DJ, Knill-Jones RP. Statistical and knowledge-based approaches to clinical decision-support system, with an application in gastroenterology. J R Statist Soc A 1984; 147: 3577[ISI]
19 Spiegelhalter DJ. Probabilistic prediction in patient management and clinical trials. Stat Med 1986; 5: 42133[ISI][Medline]
20 Bent-Hansen I. Initial plasma disappearance and distribution volume of 131I-albumin and 125I-fibrinogen in man. Acta Physiol Scand 1989; 136: 45561[ISI][Medline]
21 Rothe CF, Murray RH, Bennett TD. Actively circulating blood volume in endotoxin shock measured by indicator dilution. Am J Physiol 1979; 236: H291H300
22 Payen J-F, Vuillez J-P, Geoffray B, et al. Effects of preoperative intentional hemodilution on the extravasation rate of albumin and fluid. Crit Care Med 1997; 25: 2438[ISI][Medline]
23 Thomsen JK, Fogh-Andersen N, Bülow K, Devantier A. Blood and plasma volumes determined by carbon monoxide gas, 99mTc-labelled erythrocytes, 125I-albumin and the T1824 technique. Scand J Clin Lab Invest 1991; 51: 18590[ISI][Medline]
24 Ellman H. Capillary permeability in septic patients. Crit Care Med 1984; 12: 62933[ISI][Medline]
25 Shippy CR, Appel PL, Shoemaker WC. Reliability of clinical monitoring to assess blood volume in critically ill patients. Crit Care Med 1984; 12: 10712[ISI][Medline]
26 Shoemaker WC, Bryan-Brown CW, Quigley L, Stahr L, Elwyn DH, Kark AE. Body fluid shifts in depletion and poststress states and their correction with adequate nutrition. Surg Gynecol Obstet 1973; 136: 3714[ISI][Medline]
27 Bogaard HJ, de Vries JPPM, de Vries PMJM. Assessment of refill and hypovolaemia by continuous surveillance of blood volume and extracellular fluid volume. Nephrol Dial Transplant 1994; 9: 12837[Abstract]
28 Eisenberg PR, Jaffe AS, Schuster DP. Clinical evaluation compared to pulmonary artery catheterization in the hemodynamic assessment of critically ill patients. Crit Care Med 1984; 12: 549553[ISI][Medline]
29 Connors AF Jr, Dawson NV, Shaw PK, Montenegro HD, Nara AR, Martin L. Hemodynamic status in critically ill patients with and without acute heart disease. Chest 1990; 98: 12006[Abstract]
30 Steingrub JS, Celoria G, Vickers-Lahti M, Teres D, Bria W. Therapeutic impact of pulmonary artery catheterization in a medical/surgical ICU. Chest 1991; 99: 14515[Abstract]
31 Mimoz O, Rauss A, Rekik N, Brun-Buisson C, Lemaire F, Brochard L. Pulmonary artery catheterization in critically ill patients: a prospective analysis of outcome changes associated with catheter-prompted changes in therapy. Crit Care Med 1994; 22: 5739[ISI][Medline]
32 Connors AF, Dawson NV, McCaffree DR, Gray BA, Siciliano CJ. Assessing hemodynamic status in critically ill patients: do physicians use clinical information optimally? J Crit Care 1987; 2: 17480[ISI]
33 Lazrove S, Waxman K, Shippy C, Shoemaker WC. Hemodynamic, blood volume, and oxygen transport responses to albumin and hydroxyethyl starch infusions in critically ill postoperative patients. Crit Care Med 1980; 8: 3026[ISI][Medline]
34 Pinsky MR. The hemodynamic consequences of mechanical ventilation: An evolving story. Intensive Care Med 1997; 23: 493503[ISI][Medline]
35 Koran LM. The reliability of clinical methods, data and judgments (first of two parts). N Engl J Med 1975; 293: 6426[ISI][Medline]
36 Spiteri MA, Cook DG, Clarke SW. Reliability of eliciting physical signs in examination of the chest. Lancet 1988; i: 8735
37 Connors A, Speroff T, Dawson N, Thomas C, Knauss W. The effectiveness of right heart catheterization in the initial care of critically ill patients. JAMA 1996; 276: 88997[Abstract]