BACKGROUND: Lung cancer (LC) is one of the leading lethal cancers worldwide, with an estimated 18.4% of all cancer deaths being attributed to the disease. Despite developments in cancer diagnosis and treatment over the previous thirty years, LC has seen little to no improvement in the overall five year survival rate after initial diagnosis.
METHODS: In this paper, we extended a recent study which profiled the metabolites in sputum from patients with lung cancer and age-matched volunteers smoking controls using flow infusion electrospray ion mass spectrometry. We selected key metabolites for distinguishing between different classes of lung cancer, and employed artificial neural networks and leave-one-out cross-validation to evaluate the predictive power of the identified biomarkers.
RESULTS: The neural network model showed excellent performance in classification between lung cancer and control groups with the area under the receiver operating characteristic curve of 0.99. The sensitivity and specificity of for detecting cancer from controls were 96% and 94% respectively. Furthermore, we have identified six putative metabolites that were able to discriminate between sputum samples derived from patients suffering small cell lung cancer (SCLC) and non-small cell lung cancer. These metabolites achieved excellent cross validation performance with a sensitivity of 80% and specificity of 100% for predicting SCLC.
CONCLUSIONS: These results indicate that sputum metabolic profiling may have potential for screening of lung cancer and lung cancer recurrence, and may greatly improve effectiveness of clinical intervention.
|Number of pages||6|
|Journal||Biochimica et Biophysica Acta (BBA) - General Subjects|
|Issue number||11 (Part B)|
|Early online date||14 Jul 2016|
|Publication status||Published - 01 Oct 2016|
- lung cancer
- small vs non-small cell lung cancer
- artificial neural networks