Various cutoff levels of TNF-a were tested for the identification of CPPE, and 80 pg/mL yielded the best results (Table 2). This corresponded to a sensitivity, specificity, and accuracy of 78%, 89%, and 84%, respectively.
We next compared the relative diagnostic accuracies of pleural fluid pH, glucose, LDH, and TNF-a to categorize a PPE as complicated (Table 3). Decision thresholds for pH, glucose, and LDH were those widely accepted by consensus statements and expert opinions.’ We noted that pleural fluid TNF-a had the highest diagnostic accuracy as measured by the area under the receiver operating characteristic curve (AUC), 0.87, although overlapping confidence intervals (CIs) indicated that no single test could be clearly identified as being superior to others. Notably, pleural TNF-a offered a sensitivity that doubled that of pleural pH for predicting a CPPE. Only one patient (3%) with pleural pH < 7.20 and two patients (6%) with pleural glucose levels < 60 mg/dL exhibited concentrations of TNF-a < 80 pg/mL.
Logistic regression analysis from the biochemical pleural fluid variables showed that TNF-a at the prespecified threshold value was the most useful diagnostic indicator for CPPE (adjusted likelihood ratio [LR] +, 7.6; 95% CI, 3.1 to 10.8; adjusted LR—, 0.06; 95% CI, 0.01 to 0.27). The second best test was LDH (adjusted LR+, 3.6; 95% CI, 1.0 to 8.2; adjusted LR—, 0.23; 95% CI, 0.05 to 1.04).
Table 2—Operating Characteristics of Pleural TNF-а at Different Cutoff Values for Diagnosing Nonpurulent CPPEs

Table 3—Measures of Diagnostic Accuracy for Tests That Identify a Nonpurulent CPPE

Captcha Challenge