Combining clinical and analytical parameters improves prediction of malignant pleural effusion

The usefulness of a panel of tumour markers and clinical-radiological criteria for diagnosing malignant pleural effusion (MPE) is not clearly stated. Our purpose was to assess the performance of those parameters in the diagnosis of MPE.

Consecutive patients with exudative PE were enrolled and divided into two groups: MPE and non-MPE. Logistic regression analysis was used to estimate the probability of MPE. Four prognostic models were considered: (1) clinical-radiological variables; (2) analytical variables; (3) combination of clinical and analytical variables; and (4) simpler model removing some analytical variables. Calibration and discrimination (receiver operating characteristics curves and AUC) were performed.

A total of 491 pleural exudates were included: tuberculous (n = 72), malignant (n = 211), parapneumonic (n = 115), empyemas (n = 32), or miscellaneous (n = 61). The AUC obtained with Model 1 (absence of chest pain and fever and radiological images compatible with malignancy), Model 2 (CEA, NSE, CYFRA 21-1, and TPS), Model 3 (sum of the variables of models 1 and 2), and Model 4 (the variables of model 1 plus CEA) were 0.918, 0.832, 0.952 (all with a P < 0.05), and 0.939 (P < 0.01 compared to models 1 and 2), respectively. The correct classification rate for Models 1, 2, 3, and 4, was 87.2, 79.5, 88.4, and 87.6 %, respectively.

All models analysed had a good diagnostic yield for MPE, being greater in those that combined radiological and analytical criteria. Although Model 3 obtained a higher yield, the simplest model (Model 4) is very attractive due to its simplicity of use in daily practice.