(Chest. 2002;121:1916-1920.)
© 2002
American College of Chest Physicians
Multilevel Likelihood Ratios for Identifying Exudative Pleural Effusions*
John E. Heffner, MD, FCCP;
Steven A. Sahn, MD, FCCP and
Lee K. Brown, MD, FCCP
* From the Medical University of South Carolina (Drs. Heffner and Sahn), Charleston, SC; and Lovelace Health Systems (Dr. Brown), University of New Mexico School of Medicine, Albuquerque, NM.
Correspondence to: John E. Heffner, MD, FCCP, Medical University of South Carolina, Division of Pulmonary and Critical Care Medicine-812CSB, 96 Jonathan Lucas St, PO Box 250623, Charleston, SC 29425; e-mail heffnerj{at}musc.edu
 |
Abstract
|
|---|
Study objectives: To determine multilevel likelihood ratios for pleural fluid tests that are commonly used to discriminate between exudative and transudative pleural effusions.
Design: Meta-analysis of patient-level data.
Patient data: Selected studies included patients with diagnoses of exudative or transudative pleural effusions who underwent thoracentesis and laboratory analysis of their pleural fluid.
Measurements and methods: Studies were identified by searching MEDLINE and related bibliographies. Data were obtained for 1,448 patients from seven primary investigators or extracted from dot plots in published reports. Likelihood ratios were calculated from extracted data stratified across ranges of test result values.
Results: Sufficient data were available to calculate multilevel likelihood ratios for the elements of Light’s criteria (pleural fluid lactate dehydrogenase [LDH], ratio of pleural fluid to serum LDH, and ratio of pleural fluid to serum protein), pleural fluid protein, ratio of pleural fluid to serum cholesterol, pleural fluid cholesterol, and gradient of pleural fluid to serum albumin. Each of these tests provided levels of likelihood ratios through the most clinically relevant range (0 to 10).
Conclusion: Multilevel likelihood ratios combined with a clinician’s estimation of the pretest probability of an exudative effusion improve the diagnostic accuracy of discriminating between exudative and transudative pleural effusions. Likelihood ratios avoid the use of confusing terms, such as "pseudoexudates," that derive from the use of single cutoff points for pleural fluid tests.
Key Words: body fluids • pleura • pleural disease • pleural effusion • thoracentesis
|
Introduction
|
|---|
Pleural effusions of uncertain etiology are extremely common diagnostic challenges that warrant thoracentesis and pleural fluid analysis to categorize the effusion as an exudate or transudate. Because exudative effusions present a broad differential diagnosis of various inflammatory and malignant conditions that often require additional diagnostic or therapeutic interventions,1
accurate categorization of an effusion is fundamentally important.
Multiple investigations have examined the discriminative properties of different pleural fluid tests for identifying exudative effusions.2
3
4
5
6
7
8
9
10
These studies usually compare the operating characteristics of individual tests among patients with different etiologies of pleural effusions in order to identify tests with the highest sensitivity and specificity. Established clinical practice has favored diagnostic strategies that combine pleural fluid lactate dehydrogenase (LDH), the ratio of pleural fluid to serum LDH (LDH-R), and the ratio of pleural fluid to serum protein combined in "or" rules (Light’s criteria),2
wherein an exudative effusion is identified if any one of the criteria is fulfilled. More recent studies have examined the diagnostic utility of pleural fluid cholesterol,3
4
5
6
7
bilirubin,8
10
and albumin8
9
concentrations comparing the sensitivities and specificities of the new tests with the three-test combination of Light’s criteria.
Unfortunately, these studies and expert opinion recommend a single cutoff point to dichotomize pleural effusions into transudative or exudative categories. This approach loses much of the diagnostic information contained in laboratory tests that have continuous integer values.11
12
13
Moreover, recommendations for the use of test combinations in "or" rules increase the sensitivity but decrease the specificity of the pleural fluid analysis for identifying exudates.12
This increased sensitivity is intended to avoid false-negative results when evaluating patients with exudative effusions, which have greater diagnostic import than transudative effusions. The lower specificity, however, has resulted in the coinage of the confusing term "pseudoexudates" for effusions in patients who appear clinically unlikely to have an pleural fluid exudate but fulfill diagnostic criteria for an exudative effusion on pleural fluid analysis.14
An alternative strategy for improving the discriminative properties of diagnostic tests is to generate multilevel likelihood ratios using cutoff points and apply these ratios to convert pretest probability to posttest probability of an exudative effusion.13
With this strategy, laboratory results in the borderline exudative range with the use of single cutoff points generate low likelihood ratios that would not misclassify patients if the pretest suspicion for an exudative effusion were low.
We previously reported the results of a meta-analysis15
using discrete, patient-level data previously reported in primary studies to compare the operating characteristics of different pleural fluid tests for identifying exudative effusions. We have reanalyzed these data to generate likelihood ratios for pleural fluid exudates across the range of results for commonly used pleural fluid tests. These analyses allow a Bayesian approach to diagnosis that improves the accuracy of pleural fluid categorization and avoids subcategories of effusions, such as pseudoexudates.
|
Materials and Methods
|
|---|
The literature search and collection of data from primary investigators were previously reported.15
Briefly, a MEDLINE search and a search of the investigators’ files retrieved articles published since 1976 that reported the operating characteristics of tests that discriminated between exudative and transudative effusions. Primary investigators were contacted and asked to provide their primary data in spreadsheet format. Data were accepted into the analysis if two examiners determined that: (1) data points were available from the primary investigators or could be extracted from dot plots in the original reports, (2) discrete data points could be linked to individual patients, and (3) an acceptable reference standard was used to categorize the patients’ effusions as exudates or transudates.
Data were pooled into statistical software (JMP; SAS Institute; Cary, NC) that grouped patients into exudative and transudative categories subgrouped by discrete strata of test result ranges. Cutoff points for the test result strata were selected to provide multiple likelihood ratios within their most clinically useful range from 0 to 10. Test result strata that did not contain test values for patients with transudative effusions were reported as "noncalculable" because the resultant likelihood ratio calculations would require division by zero. These values were estimated from a prediction equation derived from linear regression with natural log transformation of the calculable likelihood ratios.
Likelihood ratios were calculated by standard methods.13
The study received an exemption from institutional review board approval.
|
Results
|
|---|
As previously reported,15
11 studies3
4
5
6
7
8
9
10
14
16
17
were identified in the literature, of which 4 studies containing 408 patients10
14
16
17
were excluded because original data were no longer available or because investigators did not enroll consecutive patients. The remaining seven studies3
4
5
6
7
8
9
provided data on 1,448 patients and eight pleural fluid tests, which included pleural fluid protein (1,187 patients), ratio of pleural fluid to serum protein (1,393 patients), pleural fluid LDH expressed as a fraction of the upper limits of normal for the serum assay (LDH-PF) [1,438 patients], LDH-R (1,388 patients), pleural fluid cholesterol (1,248 patients), the ratio of pleural fluid to serum cholesterol (1,123 patients), and the gradient of serum to pleural fluid albumin (386 patients). The ratio of pleural fluid to serum bilirubin (303 patients) was also available in the data set, but these data were not examined in this study because of their previously reported poor discriminative properties.15
The patients’ underlying diagnoses have been previously reported15
and are shown in Table 1
. Seventy-four percent of the effusions were exudates, and 26% were transudates. Likelihood ratios calculated from the primary data are shown in Tables 2 3
4
5
, along with the numbers of patients in each stratum of test results.
|
Discussion
|
|---|
The medical literature commonly describes the operating characteristics of a diagnostic test by dichotomizing test results into normal and abnormal values and calculating the sensitivity and specificity of the test.18
Unfortunately, knowledge of the sensitivity and specificity of a test offers little clinical utility when evaluating individual patients because these indexes do not describe the probability of disease if the test result is positive.19
Dichotomization of test results treats borderline abnormal and extreme test results the same and ignores important clinical information.11
13
15
Moreover, test dichotomization does not assist in assessing to what degree a test result alters a clinician’s estimation of the pretest probability of disease.
Likelihood ratios more accurately convey how a specific test result identifies the target disorder by quantifying the likelihood of a given test result in patients with a condition compared with the likelihood of the same result in patients without the condition.13
18
Likelihood ratios also have advantages over the indexes of positive and negative predictive values because these latter values depend on the prevalence of the target condition in the population being tested. Our study reports the multilevel likelihood ratios for the commonly performed tests that discriminate between transudative and exudative pleural effusions.
To our knowledge, the data set from which the likelihood ratios were calculated is the largest available, being collected from the primary investigators who provided discrete patient-level test results. Large sample sizes enhance the accuracy of likelihood ratio estimates.13
18
All of the commonly used tests evaluated in this study provide likelihood ratios that extend beyond the most clinically useful range of 0 to 10. This finding supports the conclusion that each of these tests provides comparable diagnostic information when used in a single test strategy.15
Clinical practice and expert opinion have established Light’s criteria as the "gold standard" for discriminating between exudative and transudative pleural effusions.20
21
This three-test combination has been stated to have a lower specificity as compared with the use of more recently recommended single tests, such as pleural fluid cholesterol.22
23
The lower specificity of Light’s criteria,2
however, is an expected consequence of combining multiple tests in a parallel manner using an "or" rule in an effort to enhance sensitivity.12
Moreover, two of the components of Light’s criteria—pleural fluid LDH and LDH-R—demonstrate multicollinearity (Pearson coefficient of correlation = 0.84) in the data of the present study,15
and would not be expected to perform well when combined in a diagnostic test model.24
This multicollinearity results from the inclusion of LDH-PF as one of the terms in the calculation of LDH-R (LDH-PF/serum LDH). Likelihood ratios allow sequential application of tests avoiding the loss of specificity when tests are combined in a dichotomous manner with "or" rules.
The likelihood ratios from the present study climbed to > 1, indicating an increased probability of an exudate, at the previously reported cutoff points for each of the examined pleural fluid tests from this data set,15
with the exception of LDH-PF. In the present study, the likelihood ratio rose to > 1 at an LDH-PF range of 0.61 to 0.70, which differs from the cutoff point of 0.45 we previously reported.15
Our previous report15
selected cutoff points that maximized sensitivity while maintaining the highest diagnostic accuracy for the tests. LDH-PF represents a special case, however, because of its multicollinearity with LDH-R.15
Because of this multicollinearity, the cutoff point for LDH-PF that produced the highest sensitivity and overall diagnostic accuracy when combined with the other two tests in Light’s criteria was 0.45, even though a cutoff point of 0.67 performed better when LDH-PF was considered by itself.
On the basis of our findings, how should patients with pleural effusions be evaluated? Clinicians first estimate the probability of an exudative effusion on the basis of their knowledge of a patient’s clinical presentation.11
This pretest probability is converted to pretest odds by the following equation: probability/(1 - probability). The pretest odds is then multiplied by the likelihood ratio, which generates the posttest odds. The posttest odds is then converted to the posttest probability by the following equation: posttest odds/(posttest odds + 1). Likelihood ratio normograms are available to quickly convert pretest to posttest probabilities at the bedside.11
If the posttest probability is diagnostically indeterminate with a midrange value, it can be treated as a new "pretest" probability, converted to pretest odds, and multiplied by a likelihood ratio from the results of a second pleural fluid test. This calculation generates a new posttest odds, which can be converted to a posttest probability. A clinical example using serial likelihood ratios is shown in Table 6
.
A limitation of the study derived from our efforts to partition the sample into multiple levels so as to provide clinically useful likelihood ratios. This resulted in a small number of patients in several of the test result levels. Absence of patients with transudative effusions within a test result level results in a noncalculable likelihood ratio because of the need to divide by 0. These values, however, were estimated by a nonlinear regression analysis of the other likelihood ratios for that test. The small number of values within some extreme test result levels caused several calculated likelihood ratios to fall out of the expected integer series. As shown in Table 5
, for instance, albumin gradient values between 0.19 to 0.20 had a higher likelihood ratio than albumin gradient values between 0.17 to 0.18. Any inaccuracies in calculated likelihood ratios at these extreme ranges of test result values, however, would not affect posttest odds to a clinically important degree because of the low values of the likelihood ratios.
The study results are also limited by weaknesses in the experimental designs of the primary studies that were reported in our original meta-analysis.15
The results of the primary studies, however, are entrenched into clinical practice with the standard use of their proposed tests to discriminate between exudative and transudative effusions. The likelihood ratios proposed in this study present a more useful approach for using existing data until more rigorously designed studies are reported in the future.
In conclusion, aggregate data from our previous meta-analysis allow the calculation of multilevel likelihood ratios for pleural fluid tests used to discriminate between exudative and transudative pleural effusions. These likelihood ratios can be used in a Bayesian manner to estimate the posttest probability of an exudative effusion in a strategy that improves diagnostic accuracy and quantifies diagnostic confidence.
|
Acknowledgements
|
|---|
The authors thank the primary authors for providing their primary, patient-level data.
|
Footnotes
|
|---|
Abbreviations: LDH = lactate dehydrogenase; LDH-PF = pleural fluid lactate dehydrogenase expressed as a fraction of the upper limits of normal for the serum assay; LDH-R = ratio of pleural fluid to serum lactate dehydrogenase
Work performed at the Medical University of South Carolina.
Received for publication August 13, 2001.
Accepted for publication December 5, 2001.
|
References
|
|---|
-
Sahn, SA (1988) State of the art: the pleura. Am Rev Respir Dis 138,184-234[ISI][Medline]
-
Light, RW, MacGregor, I, Luchsinger, PC, et al (1972) Pleural effusion: the diagnostic separation of transudates and exudates. Ann Intern Med 77,507-513
-
Costa, M, Quiroga, T, Cruz, E (1995) Measurement of pleural fluid cholesterol and lactate dehydrogenase: a simple and accurate set of indicators for separating exudates from transudates. Chest 108,1260-1263[Abstract/Free Full Text]
-
Hamm, H, Brohan, U, Bohmer, R, et al (1987) Cholesterol in pleural effusions: a diagnostic aid. Chest 92,296-302[Abstract/Free Full Text]
-
Romero, S, Candela, A, Martín, C, et al (1993) Evaluation of different criteria for the separation of pleural transudates from exudates. Chest 104,399-404[Abstract/Free Full Text]
-
Suay, VG, Moragón, EM, Viedma, EC, et al (1995) Pleural cholesterol in differentiating transudates and exudates: a prospective study of 232 cases. Respiration 62,57-63[Medline]
-
Valdés, L, Pose, A, Suàrez, J, et al (1991) Cholesterol: a useful parameter for distinguishing between pleural exudates and transudates. Chest 99,1097-1102[Abstract/Free Full Text]
-
Burgess, L, Maritz, FJ, Taljaard, JJF (1995) Comparative analysis of the biochemical parameters used to distinguish between pleural transudates and exudates. Chest 107,1604-1609[Abstract/Free Full Text]
-
Roth, BJ, O’Meara, TF, Cragun, WH (1990) The serum-effusion albumin gradient in the evaluation of pleural effusions. Chest 98,546-549[Abstract/Free Full Text]
-
Meisel, S, Shamiss, A, Thaler, M, et al (1990) Pleural fluid to serum bilirubin concentration ratio for the separation of transudates and exudates. Chest 98,141-144[Abstract/Free Full Text]
-
Sackett, DL, Straus, SE, Richardson, WS, et al (2000) Evidence-based medicine: how to practice and teach EBM. Churchill Livingstone Edinburgh, UK.
-
Heffner, JE (1998) Evaluating diagnostic tests in the pleural space: differentiating transudates from exudates as a model. Clin Chest Med 19,277-293[CrossRef][ISI][Medline]
-
Jaeschke, R, Guyatt, G, Sackett, D (1994) Users’ guides to the medical literature: III. How to use an article about a diagnostic test. JAMA 271,703-707[CrossRef][ISI][Medline]
-
Chakko, SC, Caldwell, SH, Sforza, PP (1989) Treatment of congestive heart failure: its effect on pleural fluid chemistry. Chest 95,798-802[Abstract/Free Full Text]
-
Heffner, JE, Brown, LK, Barbieri, C (1997) Diagnostic value of tests that discriminate between exudative and transudative pleural effusions. Chest 111,970-979[Abstract/Free Full Text]
-
Peterman, TA, Speicher, CE (1984) Evaluating pleural effusions: a two-stage laboratory approach. JAMA 252,1051-1053[CrossRef][ISI][Medline]
-
Scheurich, JW, Keuer, SP, Graham, DY (1989) Pleural effusion: comparison of clinical judgment and Light’s criteria in determining cause. South J Med 82,1487-1491
-
Sox, HC, Jr, Blatt, MA, Higgins, MC, et al (1988) Medical decision making. Butterworths Boston, MA.
-
Riegelman, RK, Jr, Hirsch, RP (1996) Studying a study and testing a test: how to read the health science literature. Little, Brown Boston, MA.
-
Bartter, T, Santarelli, RJ, Pratter, MR (1996) Transudate vs exudate: genug! Chest 109,1419-1421[Free Full Text]
-
Antony, VB, Holm, KA (1995) Testing the waters: differentiating transudates from exudates. Chest 108,1191-1192[Free Full Text]
-
Romero, S, Martinez, A, Hernandez, L, et al (2000) Light’s criteria revisited: consistency and comparison with new proposed alternative criteria for separating pleural transudates from exudates. Respiration 67,18-23[CrossRef][ISI][Medline]
-
Light, RW (1995) Pleural disease. Williams and Wilkins Baltimore, MD.
-
Katz, MH (1999) Multivariable analysis: a practical guide for clinicians. University Press Cambridge, UK.
-
Hulley, SB, Cummings, SR (1988) Designing clinical research. Williams and Wilkins Baltimore, MD.
TOP
Abstract
Introduction
