Biomarker Breath Test Correctly Predicts Lung Cancer

ORLANDO — A 2-minute breath test to detect volatile organic compounds (VOCs) successfully predicted lung cancer in patients with stage I-IV disease, according to a study reported by Michael Phillips, MD, at the American Society of Clinical Oncology 41st Annual Meeting (abstract 9510). "The breath test may potentially become a useful adjunct to lung cancer detection," said Dr. Phillips, clinical professor of medicine, New York Medical College, Valhalla.

Prior studies using gas chromatography and mass spectroscopy have identified VOCs associated with lung cancer. Dr. Phillips had previously identified oxidative stress products (alkanes and methylated alkanes) as being sensitive and specific biomarkers for lung cancer.

Lung Cancer Development

Lung cancer development is thought to require a high-risk genotype and a high-risk exposure to a toxin, Dr. Phillips said. Combined, they induce enzyme production that results in a high-risk phenotype. These induced enzymes metabolize the toxins into carcinogens, resulting in lung cancer. Normal oxidative stress generates VOCs, which are excreted in the breath.

"Our hypothesis [see Figure] is that these compounds undergo rapid accelerated clearance by the induced enzymes, thereby resulting in detectable changes in the breath VOCs," he said.

Dr. Phillips and his colleagues sought to determine whether the breath VOCs could be clinically useful biomarkers for lung cancer. They enrolled 213 smokers older than age 60 in the control group and 238 patients with suspected primary lung cancer, untreated and in their first episode with the disease.

The control group received a spiral CT scan of the chest and a 2-minute breath test. The suspected lung cancer patients underwent biopsy and the same breath test. Breath samples were analyzed with gas chromatography and mass spectroscopy. Patients with no evidence of disease after CT scan or biopsy (the controls) and patients with documented primary lung cancer underwent a 2 to 1 split into a training set (to construct a predictive model based on the breath VOCs) and a prediction set (to test the model).

The team ended up with 212 controls and 195 patients with primary lung cancer, primarily non-small-cell lung cancer. Patient characteristics (sex, age, smoking history) were similar in the cancer group and the controls.

Fuzzy Logic

Dr. Phillips used "fuzzy logic" for a multivariate analysis of the data to identify biomarkers of lung cancer. "Despite its name, fuzzy logic is a rigorous and precise technique," he said.

In the training set, fuzzy functions generated typicality matrixes for the controls and the lung cancer patients. Those data were placed in the prediction set along with new breath samples. Using the typicality matrixes, the investigators determined two values, one for lung cancer and one for no disease.

The fuzzy logic identified 28 VOC markers of lung cancer. Ten were alkanes and methylalkanes (biomarkers of oxidative stress); 20 were alkane derivatives (ie, downstream metabolites of these products of oxidative stress). Seventeen of the compounds showed reduced abundance in lung cancer patients. The researchers entered the breath VOCs into the predictive model.

Study Results

"Breath VOCs predicted lung cancer in the prediction set with a sensitivity of 90.6%, a specificity of 82.7%, and an area under the curve (AUC) of 0.91," Dr. Phillips said. The team also split the datasets using a leave-one-out method. This method produced similar results, with sensitivity of 81.5%, specificity of 87.3%, and identical AUC (0.91). The predictive curves were comparable for all stages of cancer, Dr. Phillips said, "although we did see the best predictive value in early-stage cancer." The predictions were not affected by tobacco smoking.

Dr. Phillips noted that the breath test was also positive for lung cancer in other groups, including those with cancer metastatic to the lung, postoperative lung cancer, mesothelioma, and suspicious imaging with a negative biopsy. He explained that excision of the lung cancer does not affect extrapulmonary sites, which are the sources of most of the oxidative stress in the body.

Dr. Phillips calculated that if the breath test were administered to 100,000 smokers older than age 60, assuming a 2% prevalence of lung cancer, about 14% of the tests would be positive and 86% negative. The positive predictive value would be 11.6% and the negative predictive value 99.6%. The high negative predictive value is important, he noted, because the test identifies the majority of patient who do not have lung cancer.

"This means that if a breath test is positive, it would make sense to do chest imaging, because the positive predictive value will greatly improve and thereby improve the detection of lung cancer," Dr. Phillips said. "However, if the breath test is negative, it probably would not make sense to do imaging, because the negative predictive value is unlikely to improve much more than 99.6%." Thus, he concluded, use of the breath test could lower costs and decrease patients’ exposure to radiation by reducing the need for CT screening in smokers.