November 21, 2017

Canine Scent Detection in the Diagnosis of Lung Cancer:

Revisiting a Puzzling Phenomenon

Many scientists have believed for a while that there exists a molecule that is detectable that can diagnose lung cancer. Science has moved in the way of creating an artificial nose to detect said molecules, but has been ineffective at doing so thus far. To try a different approach to verify the existence of said molecule, a volatile organic compound or VOC for short, sniffer dogs were tested.

Breath samples of 220 patients (110 healthy, 60 with lung cancer and 50 with COPD) were gathered. Various limits and exclusions were set on patients to ensure the most accurate results. No restrictions were made on food, drinking or smoking. Definitions and standards were established for lung cancer and COPD. The participants were divided into three categories: A (healthy), B(lung cancer) and C (COPD). Their breaths were collected in cylindrical glass tubes with a polypropylene fleece. Those who had lung cancer had multiple levels of lung cancer and the data of the dogs accuracy per stage of cancer is broken up in the analysis of the results.

Four family dogs were designated as sniffer dogs : 2 German shepherds, 1 Australian Shepherd, and 1 Labrador Retriever. These dogs were trained on a reward-based approach and were taught to lie down with their muzzle to the tube to indicated a positive sample. Every test tube was used only once to avoid scent memory. Test tubes with COPD were not used in training.

Three tests were performed. In the first test, the dogs had to find the breath sample of the lung cancer patient among  four healthy test tubes. The second test was of a similar nature, but instead of four healthy controls, it was the breaths of four patients with COPD, to see if the dogs could distinguish between lung cancer and COPD. The final test was to see if they could determine the 1 sample with lung cancer among a mixture of healthy and COPD test tubes. Location of the test tubes were determined by rolling a dice and various measures were taken to ensure the dogs didn’t have an advantage given by a human. Indications had to be definite and hesitation free for it to count.

Overall sensitivity was 71% and specificity was 93%. State I cancer was predicted with 100% accuracy, stage IIa and IIb: 75%. Stage IIIa was 94%. 75% for stage IIIb and 63% for stage IV. Data was additionally analyzed with a corporate dog decision, meaning three dogs had to make the same decision. Using this second approach, the sensitivity was 72% and specificity was 94%.

The experiment proves the existence of a marker that can be associated with lung cancer, and distinguished from COPD. Because of the way the experiment was set up, things like smoking, drugs and food are able to be excluded. There exists subtle markers between COPD and lung cancer that must be carefully focused on to ensure proper diagnoses.  Using breath diagnosis will be a very noninvasive way to determine lung cancer. This work was done primarily to exclude potential confounders and bias that existed in other studies. Nine confounders were identified in the experiment. 

http://erj.ersjournals.com/content/erj/39/3/669.full.pdf

Ehmann, R., Boedeker, E., Friedrich, U., Sagert, J., Dippon, J., Friedel, G. and Walles, T. (2011). Canine scent detection in the diagnosis of lung cancer: revisiting a puzzling phenomenon. European Respiratory Journal, 39(3), pp.669-676.

Reviewed by: Anneliese Ceisel