Thursday August 18 2011
The dogs may have detected cancer medication rather than cancer itself
“Dogs can be trained to identify the scent of lung cancer long before symptoms develop,” the Daily Mail has today reported. The newspaper said that “sniffer dogs can be relied upon to find the unique smell of the disease in seven out of 10 sufferers”.
The claim is based on a study that had trained four dogs to detect breath samples from people with lung cancer among those taken from healthy people and people with the lung condition COPD. The researchers determined that when three out of the four dogs agreed on which samples indicated lung cancer then this consensus could correctly detect a cancer sample 72% of the time. The dogs could also correctly rule out cancer in healthy samples 94% of the time.
However, the researchers point out that the dogs may have been detecting the medication used by the cancer patients rather than substances indicating the disease itself. This casts doubt on how well the technique might detect undiagnosed cancer. The accuracy of the test is unlikely to be the same in an unselected group from the general population. Therefore further testing will be needed.
As it stands, it is not possible to say whether dogs will be useful to sniff out early lung cancer in a sample outside of a research setting, such as a random selection from the general population or from high-risk groups. Although a novel idea, researchers must see whether cancer-specific compounds are actually released when a tumour is present, and assess the practicalities of using the technique outside of a research setting.
Where did the story come from?
The study was carried out by researchers from the Ambulante Pneumologie in Stuttgart, Germany, and Schillerhoede Hospital, Gerlingen, Germany. The study was funded using the authors’ own money. One of the credited authors declared a possible conflict of interest due to them owning the training kennel used in the research. The study was published in the peer-reviewed medical European Respiratory Journal.
The Daily Mail and BBC News did not highlight the potential confounders identified by the researchers, such as the fact that the dogs may have been detecting drugs used to treat cancer rather than the presence of cancer itself.
What kind of research was this?
This research tested the ability of trained sniffer dogs to distinguish between breath samples from volunteers with confirmed lung cancer, from healthy volunteers and from people with the lung condition chronic obstructive pulmonary disease (COPD).
The researchers were interested in testing the sensitivity and specificity of this canine test. The sensitivity is the proportion of samples from people with lung cancer that the dogs correctly identified as having the condition. The specificity is the proportion of samples from people without lung cancer that the dogs correctly identified as not having the condition, i.e. correctly ruling out the presence of lung cancer.
The researcher said attention is periodically drawn to the concept that dogs may be able to detect the presence of cancer due to their highly sensitive sense of smell. There is current speculation that tumours may release unidentified volatile chemicals that dogs but not humans can smell. While ‘electronic nose’ sensor devices have been developed to try and distinguish patterns of volatile chemicals (smells), these require people to refrain from eating or smoking before the test. The researchers say that, as yet, no volatile chemicals specific to lung cancer have been identified.
What did the research involve?
Between December 2009 and April 2010 the researchers collected breath samples from people from a hospital and medical practice in Germany. The samples were collected from people with lung cancer (60 people), people with COPD (50 people) and healthy people (110 people). No restrictions were made regarding smoking behaviour or food ingestion prior to sampling. All participants provided their medical history so that the risk of lung cancer, other cancers and COPD could be assessed. They excluded people who had suspected or confirmed cancers other than lung cancer, as well as those who had previously had surgery of the chest or airways.
The participants breathed into a glass tube that contained a fleece material to absorb the smells. In order to minimise their degradation, the samples were kept at room temperature in the dark until testing.
Four family dogs (two German shepherd dogs, one Australian shepherd dog and one labrador retriever - two male, two female) were trained by a professional dog trainer by using rewards to indicate which samples were from patients with lung cancer. The dog indicated the lung cancer sample by lying in front of the sample tube with its nose touching the tube. Each testtube containing a human breath sample was used only once to prevent the dogs’ actions from being influenced by memories of each person’s unique odour signatures. The dog training phase used breath samples from 60 healthy volunteers and 35 patients with lung cancer. Samples from people with COPD were not used in training.
Following training, three types of test were performed:
- How well the dogs could identify a lung cancer sample placed alongside four healthy control samples.
- How well the dogs could identify a lung cancer sample placed alongside four COPD samples.
- How well the dogs could identify a lung cancer sample placed alongside four mixed samples from healthy controls and COPD patients.
Among the samples from people with lung cancer, 36% were from people with early stages of the disease. The majority of samples were from people who had a type of lung cancer called ‘adenomatous non-small cell lung cancer’, although the samples were from a mixture of lung cancer types.
What were the basic results?
In the first test, where lung cancer samples were hidden among healthy samples, the researchers used 10 lung cancer samples and 40 healthy samples over 10 rounds of testing. In the second test, researchers tested 10 lung cancer samples and 40 COPD samples. In the third test, researchers used five lung cancer samples, 10 healthy samples and 10 COPD samples across five rounds of the test.
The four dogs had a “hit rate” (identifying the lung cancer sample) of between 68 and 84% across the three types of test. The researchers calculated the sensitivity and specificity using a “corporate decision approach”, i.e. where an agreement was made when at least three dogs gave the same result.
Across all of the tests the sensitivity was 0.72 (confidence interval [CI] 0.51 to 0.88), meaning that the dogs could correctly identify the presence of cancer in a lung cancer patient 72% of the time. The specificity was 0.94 (CI 0.87 to 0.98), meaning that the dogs could correctly rule out a cancer in a non-lung cancer sample 94% of the time.
The researchers found that lung cancer detection was independent from COPD and the presence of tobacco smoke and food odours. However, further analysis identified nine drugs as potential confounders. Three of these drugs have been administered to patients with lung cancer and may have biased the study.
How did the researchers interpret the results?
The researchers said that “it must be assumed that a robust and specific volatile organic compound [or pattern] is present in the breath of patients with lung cancer”. They say that in order to create suitable scent-based screening tool, additional research efforts are required to overcome current technical limitations of electronic sensor technologies.
This small study adds to previous small studies which have shown that dogs can be trained to distinguish between breath samples from cancer patients and healthy controls. The researchers point out that their work was not intended to suggest that dogs would be used for cancer diagnostics but rather to encourage further research into developing ‘electronic nose’ sensor devices and identifying volatile chemicals that may be associated with the presence of tumours.
This study has strengths in so much as it looked for confounders that could have been associated with the dog’s ability to detect the cancer sample. However, the researchers highlighted that they found that nine drugs were potential confounders. Three of these were used for lung cancer, casting doubts on whether the dogs were detecting tumour-specific volatile compounds or just detecting the drugs being used to treat the cancer.
The trial was also relatively small, so the accuracy of these sorts of scent-based tests (with dogs or with electronic noses) will need to be tested in a large, unselected community sample before it can be said that it would be useful for screening.
Overall, further research would be needed to see whether dogs would identify samples of the cancer drugs as “cancer samples” and whether dogs were able to detect cancer on the breath of people who had not yet started treatment. If the technique were to show that it could detect cancer rather than cancer medication, researchers would then have to assess a number of key considerations, such as what stage of cancer it could reliably detect, how the technique could be practically used and whether or not it actually improves upon current diagnosis techniques. In short, the idea is certainly novel and interesting, but its use would still need to deliver in practical and clinical terms before it could be adapted into technologies to be used in clinical practice.