Joachim Schneider1. 1. Institut und Poliklinik für Arbeits- und Sozialmedizin der Justus-Liebig Universität, Giessen, Germany. Joachim.Schneider@arbmed.med.uni-giessen.de
Abstract
BACKGROUND: This study evaluates the diagnostic power of a fuzzy classifier and a tumor marker panel for the detection of lung cancers in comparison to pneumoconiosis patients at high-risk of developing lung cancer. METHODS: CEA, CYFRA 21-1, NSE, SCC and CRP were measured in newly diagnosed lung cancer patients of different histological types and stages in comparison to patients suffering from silicosis. First a fuzzy classifier was generated with a set of 216 primary lung cancer patients (158 non-small cell lung cancer (NSCLC) and 58 small cell lung cancer (SCLC) patients) and 127 patients suffering from pneumoconiosis (51 silicosis and 76 asbestosis). Subsequently, the classifier was validated on a second cohort of 38 NSCLC patients, 55 silicosis patients, 32 patients with chronic obstructive airway diseases (COLD) and 28 healthy control subjects. RESULTS: At 95%-specificity, NSCLC patients were detected in 50% at stage I (n= 30), in 64% at stage II (n=22), in 82% at stage III (n=56), in 88% at stage IV (n=50) and SCLC patients with limited disease status (n=21) in 71% and with extensive disease status (n=37) in 89% by use of the fuzzy classifier. Detection rates of single markers were below those. For the best single marker in NSCLC, CYFRA 21-1, sensitivities were 23.3% at stage I, 45.4% at stage II, 71.4% at stage III, 84% at stage IV (n=50), respectively. For the best single marker in SCLC, NSE, sensitivities were 61.9% at stage of limited disease and 81.1% at stage of extensive disease. In the validation set, the fuzzy classifier showed correct negative classification in 49 of the 55 cancer-free silicosis patients (specificity: 89%), in all COLD patients (specificity: 100%) and in all but one healthy subject (specificity: 96%). This confirmed an overall specificity of 93.9%. The sensitivity for lung cancer detection in high risk populations was 73.6%. All large cell carcinomas could be detected. The positive predictive value was 80%. The negative predictive value reached 91.5%. CONCLUSION: With the fuzzy classifier and a marker panel a reliable diagnostic tool for the detection of lung cancers in a high risk population is available.
BACKGROUND: This study evaluates the diagnostic power of a fuzzy classifier and a tumor marker panel for the detection of lung cancers in comparison to pneumoconiosispatients at high-risk of developing lung cancer. METHODS: CEA, CYFRA 21-1, NSE, SCC and CRP were measured in newly diagnosed lung cancerpatients of different histological types and stages in comparison to patients suffering from silicosis. First a fuzzy classifier was generated with a set of 216 primary lung cancerpatients (158 non-small cell lung cancer (NSCLC) and 58 small cell lung cancer (SCLC) patients) and 127 patients suffering from pneumoconiosis (51 silicosis and 76 asbestosis). Subsequently, the classifier was validated on a second cohort of 38 NSCLCpatients, 55 silicosispatients, 32 patients with chronic obstructive airway diseases (COLD) and 28 healthy control subjects. RESULTS: At 95%-specificity, NSCLCpatients were detected in 50% at stage I (n= 30), in 64% at stage II (n=22), in 82% at stage III (n=56), in 88% at stage IV (n=50) and SCLCpatients with limited disease status (n=21) in 71% and with extensive disease status (n=37) in 89% by use of the fuzzy classifier. Detection rates of single markers were below those. For the best single marker in NSCLC, CYFRA 21-1, sensitivities were 23.3% at stage I, 45.4% at stage II, 71.4% at stage III, 84% at stage IV (n=50), respectively. For the best single marker in SCLC, NSE, sensitivities were 61.9% at stage of limited disease and 81.1% at stage of extensive disease. In the validation set, the fuzzy classifier showed correct negative classification in 49 of the 55 cancer-free silicosispatients (specificity: 89%), in all COLDpatients (specificity: 100%) and in all but one healthy subject (specificity: 96%). This confirmed an overall specificity of 93.9%. The sensitivity for lung cancer detection in high risk populations was 73.6%. All large cell carcinomas could be detected. The positive predictive value was 80%. The negative predictive value reached 91.5%. CONCLUSION: With the fuzzy classifier and a marker panel a reliable diagnostic tool for the detection of lung cancers in a high risk population is available.