ISTANBUL, Turkey—A blood test that measures specific blood proteins can accurately distinguish lung cancer from other smoking-related lung diseases, researchers from France said at the 31st Congress of the European Society for Medical Oncology (ESMO) (Late Breaking Abstract 8). “The findings could help overcome a major problem with cancer diagnosis, and perhaps improve treatment outcomes for lung cancer patients,” said William Jacot, MD, of Hôpital Arnaud de Villeneuve, Montpellier, France.
Dr. Jacot and colleagues used SELDITOF MS (surface-enhanced laser desorption ionization-time of flight mass spectrometry) to evaluate serum proteomic profiles in patients with lung cancer and tobacco-induced chronic pulmonary disease (CPD). SELDI separates, detects, and identifies peptide and protein peaks from different biological samples.
The researchers collected serum samples from 170 patients, including 147 with pathologically confirmed lung cancer and 23 with chronic pulmonary disease, from the Thoracic Oncology Unit of Montpellier University Hospital. When each sample had been measured using SELDI, the output was analyzed with software designed to detect protein peaks. The discriminative power of differentially expressed proteins was assessed using a classification algorithm and a regressiontree algorithm.
More than 200 protein peaks, ranging from 2 to 80 kDa, were generated for each serum sample, 31 of which significantly differed between the two study groups (P < .001). Of these, 19 were significantly higher in patients with lung cancer, and 12 were increased in the chronic pulmonary disease group. The diagnostic value of each of these 31 peaks was high, Dr. Jacot said, with an AUCROC ranging from 0.69 to 0.84. Using these data, the researchers were able to develop a serum proteomic signature for lung cancer, compared with chronic pulmonary disease.
“The results were interesting,” Dr. Jacot said. “Using the peaks significantly different between the two groups, we built a regression-tree algorithm using two protein peaks [at 3204 and 7166 Da] as splitters. Using this algorithm, we were able to correctly classify 88.2% of the sera in the learning set as coming either from lung cancer patients or those with chronic pulmonary disease.”
The validity of this classification tree algorithm was then challenged in the test set phase: 73.9% of controls and 86.5% of lung cancer samples were correctly identified.