0
Communications to the Editor |

Classic, Abbreviated, and Modified Light’s Criteria : The End of the Story? FREE TO VIEW

José Manuel Porcel, MD; Manuel Vives, MD
Author and Funding Information

Affiliations: University Hospital Arnau de Vilanova Lleida, Spain Clínica Recoletas Albacete, Spain,  *Medical University of South Carolina Charleston, SC

Correspondence to: José Manuel Porcel, MD, Professor and Chairman, Departament de Medicina, Universitat de Lleida, Av. Alcalde Rovira Roure 80, Hospital Universitari Arnau de Vilanova, 25198 Lleida, Spain; e-mail: jporcelp@medynet.com



Chest. 1999;116(6):1833-1836. doi:10.1378/chest.116.6.1833-b
Text Size: A A A
Published online

To the Editor:

Heffner and colleagues (April 1997)1 performed a meta-analysis involving 1,448 patients in order to determine the appropriate decision thresholds and diagnostic accuracies for pleural fluid (PF) tests that discriminate between exudative and transudative pleural effusions. In this study, cutoff points derived from the receiver operating characteristics (ROC) analysis differed from previously reported values in PF for lactate dehydrogenase (LDH-PF) (> 0.45 of the upper limits of normal instead of more than two thirds of the upper limits of normal), protein (P-PF) (> 2.9 g/dL instead of> 3 g/dL), and cholesterol (CHOL-PF) (> 45 mg/dL instead of higher values). Their meta-analysis concluded that Light’s criteria provide good discriminative properties, although the PF cutoff point should be set at > 45% of the laboratory’s upper limit of normal for serum LDH (“modified” Light’s criteria). Likewise, LDH-PF values could be removed from Light’s criteria (“abbreviated” Light’s criteria) without affecting diagnostic accuracy. Finally, if physicians choose to avoid obtaining a serum sample, paired (LDH-PF or CHOL-PF) or triplet (LDH-PF or CHOL-PF or P-PF) test combinations making use of the new cutoff values are suitable alternatives to Light’s criteria.

Although we have previously observed that changing the classic Light’s criteria with different cutoff points offers no advantages for discriminating between transudative and exudative pleural effusions,2 we decided to determine whether a similar conclusion could be obtained following the recommendations of Heffner and colleagues. For this purpose, we retrospectively collected data from 455 consecutive adult patients undergoing thoracentesis at the University Hospital Arnau de Vilanova during the last 4 years. Ninety-one patients were excluded from the study due to nondefinitive diagnoses or to the existence of more than one potential etiology for the PF accumulation. The remaining 364 pleural effusions consisted of 72 transudates (19.8%) and 292 exudates (80.2%). There were 210 men (57.7%) and 154 women (42.3%) in the study, with a mean age of 57.5 years (range, 15 to 97 years). The causes of the transudates were congestive heart failure (48 patients), liver cirrhosis (14 patients), and nephrotic syndrome (10 patients), whereas exudates were secondary to neoplastic conditions (104 patients), pneumonia (94 patients), tuberculosis (54 patients), and other etiologies (40 patients). The diagnoses of the patients were defined by predetermined criteria.2The results of applying the meta-analysis ROC cutoff points and the previously reported cutoff points are shown in Table 1.3 We chose 50 mg/dL as the cutoff for CHOL-PF based on our previously reported data.4 The number of test results available for each of the PF parameters was as follows: P-PF, 357; LDH-PF, 359; CHOL-PF, 171; PF/serum P ratios, 347; and PF/serum LDH ratios, 342.

Several conclusions can be drawn from this investigation. First, the abbreviated form of Light’s criteria has an accuracy similar to the classic Light’s criteria, but the former offers no advantages over the latter since LDH-PF is required to calculate the LDH ratio anyway. Second, the application of the alternative cutoff values suggested by Heffner and colleagues produced a relatively large decrease in specificity for an insignificant gain in sensitivity in our study. Thus, Heffner’s findings probably cannot be generalized to a particular practice setting, and so we think that each laboratory must determine its own best cutoff points. Finally, we reinforce the concept that classic Light’s criteria remain the benchmark test combination for differentiating exudates and transudates. To paraphrase an editorial in CHEST 3 years ago,5 it is time to move on in the clinical investigation of the pleural space and to stop focusing our efforts on finding ever more perfect criteria for separating transudates and exudates.

Table Graphic Jump Location
Table 1. Diagnostic Accuracy for Test Combinations that Identify Exudative Pleural Effusions*

“Modified” refers to the adoption of the cutoff values suggested by Heffner et al.1

* 

CI = confidence interval; PV = predictive value; LR+ = likelihood ratio positive; LR− = likelihood ratio negative; OR = odds ratio.

Heffner, JE, Brown, LK, Barbieri, CA (1997) Diagnostic value of tests that discriminate between exudative and transudative pleural effusions.Chest111,970-980. [CrossRef]
 
Vives, M, Porcel, JM, Vicente de Vera, MC, et al A study of Light’s criteria and possible modifications for distinguishing exudative from transudative pleural effusions.Chest1996;109,1503-1507. [CrossRef]
 
Heffner, JE Evaluating diagnostic tests in the pleural space. Differentiating transudates from exudates as a model.Clin Chest Med1998;19,277-293. [CrossRef]
 
Gázquez, I, Porcel, JM, Vives, M, et al Comparative analysis of Light’s criteria and other biochemical parameters for distinguishing transudates from exudates.Respir Med1998;92,762-765. [CrossRef]
 
Bartter, T, Santarelli, RJ, Pratter, MR Transudate vs exudate: genug!Chest1996;109,1419-1421. [CrossRef]
 
To the Editor:

We appreciate the interest shown by Drs. Porcel and Vives in our meta-analysis of tests that discriminate between transudates and exudates (April 1997).1We also found interesting their retrospective data, presented in their letter and their previous report,2 that compare different testing strategies in an effort to improve the accuracy of pleural fluid categorization. Several comments in their letter merit further discussion.

We performed our meta-analysis to address the analytic errors commonly committed by studies that attempted to “best” Light’s criteria with new pleural fluid tests. These studies almost always incorrectly analyze their data with hypothesis-testing statistics (comparison tests for proportions) rather than with analytic techniques recommended for diagnostic test research (eg, receiver operating characteristics [ROCs] analysis).35 Additionally, these studies report underpowered, small data sets of < 500 patients and frequently recommend new tests on the basis of their higher specificity compared with Light’s criteria. This latter recommendation fails to recognize that multiple tests combined in “or” rules (eg, Light’s criteria) always have a higher sensitivity but lower specificity compared with noncombination single tests when each of the test components of the combination and the“ new” single test have similar discriminative properties.6

The use of hypothesis-testing statistics runs the risk of incorrectly concluding that one test or test combination is “better” than another when no differences exist, or vice versa. Porcel and Vives run afoul of this error in their original report.2 They noted no differences among various test combinations in this underpowered study of a small sample of 195 patients analyzed with hypothesis-testing statistics (Z distribution test for comparing proportions). Interestingly, they contend in their letter, without employing any statistical analyses, that the “lactate dehydrogenase (LDH)-[pleural fluid] PF or [cholesterol] CHOL-PF” rule has a higher specificity than the “modified LDH-PF or CHOL-PF” rule. Examination of the overlapping 95% confidence intervals for the specificity results of these two combinations (Table 1 in their article,2), however, indicates that no statistically significant differences exist. Had they provided appropriate statistical analyses, no differences would have been noted among most, if not all, of the studied test combinations.

We should emphasize that the lack of differences among reported tests that discriminate between exudates and transudates was exactly the point of our meta-analysis. The ROC analysis that we performed demonstrated no differences among any of the test combinations examined, excluding those that used pleural fluid bilirubin. We concluded that Light’s criteria have excellent discriminative properties but that other available tests and test combinations also have those properties. Investigators, therefore, might slow their rush toward trying to find even better tests for this indication. We did not suggest that clinicians should use any of the new test combinations based on any differences in their diagnostic accuracies; indeed, we emphasized that no differences existed. We did state, however, that alternative testing strategies with similar diagnostic accuracies might provide benefits in some clinical settings unrelated to their diagnostic performance. For instance, the combination rule of “P-PF or LDH-PF or CHOL-PF” avoids the need for blood test results because the calculation of the ratios of pleural fluid to serum is unnecessary.

We did propose in our meta-analysis that clinicians could use an abbreviated form of Light’s criteria (ie, eliminate either the LDH/PF or the LDH-PF/LDH serum ratio). This recommendation was based on the principles of diagnostic testing, which argue against the use of two tests in “or” rule combinations if the tests are highly correlated.6 Because the “LDH-PF” and“ LDH-PF/LDH-serum” in Light’s criteria are “mathematically linked” (and highly correlated in our meta-analysis) since each includes “LDH-PF”, clinicians can abbreviate Light’s criteria by excluding either “LDH-PF” or “LDH-PF/LDH-serum” without sacrificing diagnostic accuracy. Indeed, Porcel and Vives prove this point in Table 1 in their article by showing nearly identical point estimates and overlapping 95% confidence intervals for the diagnostic performances of the standard and abbreviated forms of Light’s criteria.

The use of the abbreviated form of Light’s criteria is more than a purist’s attempt to avoid cluttering diagnostic rules with unneeded elements; clinicians can obtain good diagnostic accuracy using two components of Light’s criteria (ie, “LDH-PF or[ protein] P-PF/P-serum”) by excluding LDH-PF/LDH-serum when a serum LDH level is unavailable. Often clinicians need such options for the care of their patients.

We are interested in the efforts of Porcel and Vives to validate or disprove the cutoff points recommended in our meta-analysis of 1,448 patients by an examination of their much smaller data set (n = 301 to 360 patients) without the assistance of statistical analyses. Based on their conclusion that differences existed among test combinations, they state that the cutoff points suggested in our study “cannot be generalized to a particular practice setting, and we think that each laboratory must determine its own best cutoff points.” There are several problems with this recommendation.

First, the confidence intervals in Table 1 of their article2 indicate that no differences existed among test combinations that were the same except for the use of different cutoff points. Second, the small sample size of their report (and the experimental design discussed below) probably accounts for the appearance of differences in point estimates, as suggested by the wide confidence intervals, which are especially notable for their odds ratios. Third, the application of diagnostic tests for all disorders might be enhanced if every institution performed rigorous test evaluations on their local population, but this goal is not often feasible. Fourth, large sets of patient-level data analyzed in a meta-analysis based on high-quality primary studies from multiple institutions are more likely to provide generalizable recommendations than are small data sets, such as those reported by Porcel and Vives, except when the characteristics of the patient populations differ markedly among examined populations. As emphasized in our meta-analysis, however, we found multiple design flaws in the primary studies that limit our recommendations.6The studies by Porcel and Vives, unfortunately, share these flaws (see below). Fifth, the “best” cutoff points are often figments of our imagination. Cutoff points are selected on the basis of what we want a test to do and on characteristics of the tested population. If we do not want to miss cases (screening approach), we adjust cutoff points to maximize sensitivity and to sacrifice specificity, and vice versa. Our meta-analysis used a Bayesian technique for determining cutoff points that considered prevalence and misclassification costs.7 The technique used by Porcel and Vives selected cutoff points that were 1 SD from the mean of the sample that they were using to validate our recommendations. Deriving a new cutoff point from a small validation sample and comparing its performance to a general recommendation from a larger, independent sample biases the sample toward the new cutoff point (which is circular logic). Their approach does not consider misclassification cost, which is an important element in selecting cutoff points.

Our meta-analysis also attempted to point out methodological weaknesses of studies that assess the performance of pleural fluid testing strategies. The most important flaw in these investigations is their retrospective design, which allows clinicians who determine disease classification to have knowledge of test results that are under evaluation (in this instance, Light’s criteria). Such designs use circular logic and bias results (review bias). The Porcel and Vives studies were retrospective without efforts to prevent or control for this source of bias. An additional weakness is the exclusion of patients who have indeterminate disease classifications, which is a major source of bias.6 We note that 91 of the 455 evaluated patients reported in their letter were excluded from evaluation without an effort to estimate the impact of these exclusions on the study results.

We thank Drs. Porcel and Vives for initiating this dialogue. We think such discussions are important to improve the investigation of diagnostic tests, which has been shown to be in need of improvement in pulmonary research.8 We should conclude that little is gained by parsing cutoff points, sensitivities, or specificities in the specific instance of studying tests that evaluate exudative and transudative effusions: our meta-analysis indicates that clinicians already have many acceptable approaches for making these determinations. We do not think, however, that the contention of Drs. Porcel and Vives that we have had enough research in this area is important. We argue that we need better research that conforms to diagnostic test evaluation standards. Pertinent to the evaluation of exudates and transudates, however, we still teach the Light’s criteria approach for pleural disease research, which has several advantages over other strategies, to medical students. It works. It is venerable. And it recognizes the contributions of a giant in the field of pleural disease research.

References
Heffner, JE, Brown, LK, Barbieri, C Diagnostic value of tests that discriminate between exudative and transudative pleural effusions.Chest1997;111,970-979. [CrossRef]
 
Vives, M, Porcel, JM, de Vera, MV, et al A study of Light’s criteria and possible modifications from transudative pleural effusions.Chest1996;109,1503-1507. [CrossRef]
 
Guyatt, GH, Tugwell, PX, Feeny, DH, et al A framework for clinical evaluation of diagnostic technologies.Can Med Assoc J1986;134,587-594
 
Nierenberg, AA, Feinstein, AR How to evaluate a diagnostic marker test: lessons from the rise and fall of dexamethasone suppression test.JAMA1988;259,1699-1702. [CrossRef]
 
Centor, RM, Schwartz, JS An evaluation of methods for estimating the area under the receiver operating characteristic (ROC) curve.Med Decis Making1985;5,149-156. [CrossRef]
 
Heffner, JE Evaluating diagnostic tests in the pleural space: differentiating transudates from exudates as a model.Clin Chest Med1998;19,277-293. [CrossRef]
 
Zweig, MH, Campbell, G Receiver-operating characteristic (ROC) plots: a fundamental evaluation tool in clinical medicine.Clin Chem1993;39,561-577
 
Heffner, JE, Feinstein, D, Barbieri, C Methodologic standards for diagnostic test research in pulmonary medicine.Chest1998;114,877-885. [CrossRef]
 

Figures

Tables

Table Graphic Jump Location
Table 1. Diagnostic Accuracy for Test Combinations that Identify Exudative Pleural Effusions*

“Modified” refers to the adoption of the cutoff values suggested by Heffner et al.1

* 

CI = confidence interval; PV = predictive value; LR+ = likelihood ratio positive; LR− = likelihood ratio negative; OR = odds ratio.

References

Heffner, JE, Brown, LK, Barbieri, CA (1997) Diagnostic value of tests that discriminate between exudative and transudative pleural effusions.Chest111,970-980. [CrossRef]
 
Vives, M, Porcel, JM, Vicente de Vera, MC, et al A study of Light’s criteria and possible modifications for distinguishing exudative from transudative pleural effusions.Chest1996;109,1503-1507. [CrossRef]
 
Heffner, JE Evaluating diagnostic tests in the pleural space. Differentiating transudates from exudates as a model.Clin Chest Med1998;19,277-293. [CrossRef]
 
Gázquez, I, Porcel, JM, Vives, M, et al Comparative analysis of Light’s criteria and other biochemical parameters for distinguishing transudates from exudates.Respir Med1998;92,762-765. [CrossRef]
 
Bartter, T, Santarelli, RJ, Pratter, MR Transudate vs exudate: genug!Chest1996;109,1419-1421. [CrossRef]
 
Heffner, JE, Brown, LK, Barbieri, C Diagnostic value of tests that discriminate between exudative and transudative pleural effusions.Chest1997;111,970-979. [CrossRef]
 
Vives, M, Porcel, JM, de Vera, MV, et al A study of Light’s criteria and possible modifications from transudative pleural effusions.Chest1996;109,1503-1507. [CrossRef]
 
Guyatt, GH, Tugwell, PX, Feeny, DH, et al A framework for clinical evaluation of diagnostic technologies.Can Med Assoc J1986;134,587-594
 
Nierenberg, AA, Feinstein, AR How to evaluate a diagnostic marker test: lessons from the rise and fall of dexamethasone suppression test.JAMA1988;259,1699-1702. [CrossRef]
 
Centor, RM, Schwartz, JS An evaluation of methods for estimating the area under the receiver operating characteristic (ROC) curve.Med Decis Making1985;5,149-156. [CrossRef]
 
Heffner, JE Evaluating diagnostic tests in the pleural space: differentiating transudates from exudates as a model.Clin Chest Med1998;19,277-293. [CrossRef]
 
Zweig, MH, Campbell, G Receiver-operating characteristic (ROC) plots: a fundamental evaluation tool in clinical medicine.Clin Chem1993;39,561-577
 
Heffner, JE, Feinstein, D, Barbieri, C Methodologic standards for diagnostic test research in pulmonary medicine.Chest1998;114,877-885. [CrossRef]
 
NOTE:
Citing articles are presented as examples only. In non-demo SCM6 implementation, integration with CrossRef’s "Cited By" API will populate this tab (http://www.crossref.org/citedby.html).

Some tools below are only available to our subscribers or users with an online account.

Related Content

Customize your page view by dragging & repositioning the boxes below.

  • CHEST Journal
    Print ISSN: 0012-3692
    Online ISSN: 1931-3543