Multiple markers for lung cancer diagnosis: validation of models for localized lung cancer

Sera from 71 patients with localized lung cancer, from 70 normal controls, and from 73 patients with benign lung diseases were analyzed for 10 substances to detect lung cancer: ferritin, lipid-bound sialic acid, total sialic acid, beta 2-microglobulin, lipotropin, the alpha and beta subunits of human chorionic gonadotropin, calcitonin (two assays), parathyroid hormone, and carcinoembryonic antigen (CEA). Individual markers were studied, and optimal combinations of markers were sought for discriminating patients with localized lung cancer from normal controls and from patients with benign lung disease. Both logistic regression and recursive partitioning methods for discrimination were tried. The best rules involved only CEA and ferritin for discriminating patients with lung cancer from normal controls, and CEA and age for discriminating patients with lung cancer from those with benign lung diseases. The performance of these rules was validated on an independent serum panel containing sera from 56 patients with localized lung cancer, 75 normal controls, and 75 patients with benign lung diseases. Three rules designed to achieve 95% specificity against normal controls attained 14%-36% sensitivity for localized lung cancer in the validation panels, whereas three rules designed to achieve 95% specificity against benign lung diseases attained 30%-39% sensitivity. Some aspects of potential clinical applications are discussed.