Screening of tuberculous pleural effusion by discriminant analysis

Pneumology Department of a 635-bed acute-care teaching hospital in Valencia, Spain.

To assess the value of discriminant analysis as a method of optimizing the discriminant power of routine radiographic features and a panel of laboratory parameters including biochemical analyses of pleural fluid for differentiation between tuberculous and non-tuberculous pleural effusion.

A series of 47 variables were retrospectively obtained from the medical records of 189 patients with exudative pleural effusion (tuberculous pleurisy, n = 78; non-tuberculous pleurisy, n = 111). A backward elimination method was applied until the best discriminant function was found.

The most powerful predictor of tuberculous pleural effusion was a function that consisted of four variables, as follows: age (years); tuberculin skin test (mm of induration at 48 hours); white blood cell count (cells/mm3); and bloodstained exudate (coded as yes = 1, no = 2). This function showed a sensitivity of 90%, specificity of 87%, positive predictive value of 83%, negative predictive value of 92%, and was able to give an 88% rate of correct classification.

The calculated discriminant function based on the patient's age, peripheral leukocyte count, tuberculin skin test and blood in the exudate is a simple, rapid and inexpensive method for screening tuberculous etiology in patients with pleural effusion.