Table 3 Summary of recommendations for future cancer detection dog studies

Limited number of training samples from unique donors promotes multiple sample learning rather than odour generalisation.

Pooling of samples from different donors to create new odour profiles.

Analytical work required to validate whether pooling of biological samples is effective in creating varied odour headspace.

Training dogs on several types of related disease odours from the outset (e.g. several forms of cancer).

Some evidence for the effectiveness of this technique based on training on two types of cancer odour[10].

Introducing dogs to disease and control odours concurrently from the outset of training rather than introducing the disease odour in isolation first.

Some studies have successfully trained dogs by presenting cancer samples in isolation in the early stages of training. However, training on disease and control samples from the outset may reduce the risk of reliance on multiple sample learning – further validation trials required.

Methods for rewarding dogs during double-blind trials may confound earlier training.

Independent referees should provide immediate feedback to trainer on correct/incorrect responses using remote system (e.g. sample allocation code over the telephone).

Successfully employed in the present study.

Training dogs to expect a reward only for a proportion of correct indications on positive samples.

Widely used technique in field of psychology and animal training.

Search-based discrimination tasks may not be optimal for encouraging sensitive disease odour indication behaviour.

The utility of alternative forms of discrimination task, such as habituation-dishabituation paradigms, should be explored.

Studies of novel paradigms for the measurement of olfactory discrimination in dogs and others species conducted[23].