A study looking at whether dogs can help diagnose prostate cancer

Cancer type:

Prostate cancer

Status:

Results

Phase:

Other

This study looked at whether you can train dogs to smell prostate cancer in urine samples.

It was open for people to join between 2015 and 2019. The team published the initial results in 2020. They plan to publish more results in the future.

More about this trial

Dogs have a very good sense of smell. Cancer cells release substances called volatiles that have a distinctive smell. Previous research had shown that dogs might be able to detect cancer in urine samples.

In this study, researchers wanted to see if they could train dogs to smell prostate cancer. To do this, they collected urine samples from 2 groups of people:

people who had prostate cancer
people who did not have cancer (healthy volunteers)

The study involved training 6 dogs to detect the difference in how urine smells in someone who has cancer compared to someone who doesn’t.

Summary of results

In 2020, the study team published results about how they trained the dogs to detect cancer in urine samples. There is a summary about this below.

They plan to publish results about how well using detection dogs worked at a later date. We will include a summary of this once it’s available.

Training medical detection dogs

Stage 1
The team collected 236 urine samples:

21 from people with prostate cancer who were being seen at Milton Keynes University Hospital (MKUH)
215 from people who didn’t have cancer and attended volunteer events run by Medical Detection Dogs

They trained 6 dogs, using a reward system, to identify the difference between the samples. They did this by repeating the process many times over about 10 weeks.

The dogs were able to correctly identify cancer in more than 9 out of 10 samples (94%). And correctly identify where there was no cancer in just under 9 out of 10 samples (88%).

Stage 2
The team collected more samples from:

13 people with prostate cancer who were being seen at MKUH
79 people who didn’t have cancer who worked at MKUH, or were friends or relatives of people being seen at MKUH

The team continued to train the dogs with rewards, using these samples. Again they repeated the process with many samples over about 10 weeks.

By the end of this, the dogs were able to correctly identify cancer in slightly fewer samples, just under 9 out of 10 (88%). But they were only able to correctly identify where there was no cancer in less than 7 out of 10 samples (67%).

The research team wanted to find out why the accuracy had dropped.

Stage 3
The team looked at every step of the process to find out why the dogs had become less accurate at identifying prostate cancer in Stage 2 of the study.

Healthcare professionals (HCPs) tested each urine sample before it was used in the study. This was to check for signs of infection or other bladder or kidney problems. They use a small urinalysis stick called a dipstick to do this.

For the samples collected at the hospital, the HCPs put the dipstick in the sample pot to test the urine. But for those collected at volunteer events, the HCPs put a small amount of urine on the dipstick outside of the sample pot.

The team discovered that in Stage 1, the dogs had accidentally learnt to smell the difference between the samples which had come into contact with the dipstick, and those which hadn’t. And not the difference between samples that had come from people who had cancer, and those from people who didn’t. So the results of Stage 2 were less accurate.

Stage 4
The team felt that 2 of the dogs taking part weren’t able to tell the difference between the samples accurately enough. So they didn’t continue in the study.

They then tried to re-train the 4 dogs still in the study, making sure all the urine samples had been through the same process and tested in the same way.

To do this they used:

143 existing samples
217 new samples

They used a variety of different types of new samples:

22 new urine samples from people who had cancer
117 new urine samples from people who didn’t have cancer

They also had samples that were not urine to use as controls and help with training. These included water, and other things used in the process such as paper towels and gloves.

They were able to re-train 3 of the dogs to identify cancer more accurately. But they were unable to do this with one of the dogs.

Conclusion
The team concluded that it’s important to make sure the only difference in the samples is whether the person has cancer or not.

They also concluded that it’s important to continue to assess the accuracy of medical detection dogs during training. And that it is sometimes possible to re-train the dogs.

More detailed information
There is more information about this research in the reference below.

Please note, the information we link to here is not in plain English. It has been written for healthcare professionals and researchers.

A Lesson in Standardization – Subtle Aspects of the Processing of Samples Can Greatly Affect Dogs' Learning
C M Guest and others
Frontiers in Veterinary Science, 2020. Volume 7, article 525.

Where this information comes from
We have based this summary on the information in the article above. This has been reviewed by independent specialists (peer reviewed Open a glossary item ) and published in a medical journal. We have not analysed the data ourselves. As far as we are aware, the link we list above is active and the article is free and available to view.