Statistical models for evaluation of pleural effusions

Statistical models designed to aid in the identification of neoplastic effusions were formulated. A computer program utilizing n-chotomous, multivariate probit analysis was used for estimation of model parameters. These parameters were estimated for an illustrative sample of 88 patients with effusions of unknown etiology. The outstanding feature of the statistical approach is its potential as a diagnostic and research tool. Regarding its diagnostic potential, if carcino-embryonic antigen (CEA) levels of plasma (PLAS-CEA) and pleural fluid (PLFL-CEA) were used separately, PLAS-CEA values greater than 4.833 ng/ml and PLFL-CEA values greater than 4.193 ng/ml discriminated malignant from benign effusions. Elevated CEA levels in conjunction with cell block examination noticeably improved diagnostic efficacy. The thresholds are consistent with CEA levels found by other investigators. Furthermore, the method employed is specific in predicting the probability that a given effusion is malignant. Regarding its research potential, the efficacy of different sets of tests in predicting malignancy can be simultaneously and comparatively evaluated.