The NYT reports:
In the small world of people who train dogs to sniff cancer, a little-known Northern California clinic has made a big claim: that it has trained five dogs - three Labradors and two Portuguese water dogs - to detect lung cancer in the breath of cancer sufferers with 99 percent accuracy. Skip to next paragraph Peter DaSilva for The New York Times
The study was based on well-established concepts. It has been known since the 80's that tumors exude tiny amounts of alkanes and benzene derivatives not found in healthy tissue.
Other researchers have shown that dogs, whose noses can pick up odors in the low parts-per-billion range, can be trained to detect skin cancers or react differently to dried urine from healthy people and those with bladder cancer, but never with such remarkable consistency.
The near-perfection in the clinic's study, as Dr. Donald Berry, the chairman of biostatistics at M. D. Anderson Cancer Center in Houston, put it, "is off the charts: there are no laboratory tests as good as this, not Pap tests, not diabetes tests, nothing."
As a result, he and other cancer experts say they are skeptical, but intrigued.
I don't have an opinion on the validity of this particular claim, but a more general point is why the emphasis merely on training dogs? Why not also breed the most trainable dogs to create a cancer-sniffing breed? These dogs could be useful in third world settings where conventional tests are too expensive, or they could be used for quick and dirty screenings of large numbers of people. The dog world is transfixed by the assumption that a breed has to be homogenous in appearance, but that should be secondary to selectively breeding a higher degree of accuracy.