Specially trained dogs can identify most patients with lung cancer by smelling their breath, researchers said.
Sniffing 100 breath samples from patients with biopsy-confirmed lung cancer, the dogs failed to flag only 29, reported Thorsten Walles, MD, of Schillerhoehe Hospital in Gerlingen, Germany, and colleagues online in the European Respiratory Journal.
Among 400 other samples from individuals without lung cancer, the canine sniff test gave false positives for just 28, the researchers found.
- Explain that specially trained dogs could identify most patients with lung cancer by smelling their breath.
- Point out that the findings may be more important for confirming that human exhalations contain markers for lung cancer, which eventually may be detectable by more conventional means.
However, Walles and colleagues suggested that the findings were most important for confirming that human exhalations contain markers for lung cancer, which eventually may be detectable by more conventional means.
“This is a big step forward in the diagnosis of lung cancer, but we still need to precisely identify the compounds observed in the exhaled breath of patients,” said Walles in a press release.
Several earlier studies have found that dogs, with their keen sense of smell, can identify patients with various forms of cancer, including tumors of the breast, colon, and lung merely by sniffing. The proposition originated in 1989 with a case report of a man whose melanoma was diagnosed because his dog kept sniffing the lesion.
The dogs used in the current study were young family pets — two German shepherds, one Labrador retriever, and one Australian shepherd. Using test tubes containing exhalations from 35 lung cancer patients and 60 healthy controls, a professional dog trainer taught the animals to lie down in front of tubes with samples from the patients.
During both the training and the subsequent testing phase, each sample was given to the dogs only once so that they would not simply learn to recognize individual participants’ characteristic odors.
The testing involved 50 healthy people, 25 patients with histologically confirmed lung cancer, and 50 patients with chronic obstructive pulmonary disorder (COPD). These were different individuals than those providing samples for the training phase.
In the cancer patient group, one patient had stage I disease, two each at IIa and IIb tumors, four were staged IIIa, five had IIIb disease, and 11 were at stage IV.
Blinded observers watched the dogs as they each sniffed at five tubes per session. If a dog appeared to hesitate in front of a tube, it was automatically recorded as an incorrect result. The investigators also took a variety of steps to keep from subtly influencing the dogs.
The testing was conducted in three phases. First, the dogs were presented with samples from healthy controls and lung cancer patients. Next, samples from lung cancer and COPD patients were presented. Finally, the dogs evaluated samples from all three groups.
Somewhat surprisingly, the dogs had the most difficulty in the first test, discriminating the lung cancer patient samples from those of healthy people. The total number of correct results from the four dogs was 22 versus 18 false results.
Accuracy was much better in the second phase, with the dogs correctly identifying 32 samples from the COPD and lung cancer patients against eight false results.
Results were better still in the third test, with 19 correct results and only one incorrect.
Overall, the sensitivity for detecting lung cancer was 71% overall (95% CI not reported). There was little difference in accuracy according to disease stage, the researchers indicated: All samples from the sole stage I patient had positive results in the sniff test. Accuracy rates for other disease stages were:
- 75% for stage IIa
- 75% for stage IIb
- 94% for IIIa
- 75% for stage IIIb
- 63% for stage IV
The dogs varied somewhat in their ability to sniff out cancer accurately, with one dog scoring 68% of samples correctly while another had 84% correct results (kappa statistic 0.436).
Specificity overall was 93% (95% CI not reported).
Walles and colleagues also calculated “corporate dog decision” accuracy by counting as accurate only those results on which three of the four dogs gave the same result. Sensitivity of these group decisions was 72% (95% CI 51% to 88%), with specificity of 94% (95% CI 87% to 98%).
The researchers determined as well that the dogs were as accurate in classifying smokers versus nonsmokers, indicating that the animals were not simply responding to tobacco-related breath components.
Because their study ruled out a role in the results for smoking and certain other potential confounders, the current study is an advance over previous research on dog sniff tests for cancer.
It “confirms the existence of a stable marker (or scent pattern) that is strongly associated with lung cancer and independent of COPD,” Walles and colleagues wrote, “reliably discriminated from tobacco smoke, food odors and (potential) drug metabolites.”
Whether dogs will ultimately be better than machines for breath analysis remains to be determined, they indicated.
“Electronic nose technologies” are not yet practical because of their complicated sampling procedures and vulnerability interference, the researchers commented.
Dogs, on the other hand, are “virtually on the verge of respectability” for disease detection. Yet without better understanding of what they are responding to, it will be impossible to develop a reliable screening test for lung cancer based on their abilities, Walles and colleagues suggested.
“Unfortunately, dogs cannot communicate the biochemistry of the scent of cancer,” they lamented.
By John Gever, Senior Editor, MedPage Today
Published: August 18, 2011