Weekend Reads | A San Juan Island Program Proves Dogs Can Detect Parkinson’s Disease

This weekend’s read is the recently released report of a program to create a new test for Parkinson’s disease — a very unusual one. Dogs were trained to detect a unique scent produced by people with the disease. The program also has a local connection: It took place in Friday Harbor, San Juan Island.

Parkinson’s disease (PD) is a progressive neurological disorder in which the level of dopamine, an important hormone and neurotransmitter used by our brains to coordinate movement, gradually declines over time. It can cause tremors (a signature symptom), loss of balance, cramped handwriting, sleep disorders, and altered speech patterns. An estimated 10 million people have Parkinson’s disease worldwide, and approximately 90,000 new cases are diagnosed in the United States each year. 

The symptoms start appearing when the dopamine level in the brain drops about 60% from normal levels. It’s believed the gradual dopamine reduction begins about five to 10 years before that, though. There is no definitive screening test for Parkinson’s; doctors look at the combination of symptoms, test results, and hereditary and environmental factors — and 10% to 20% of cases are believed to be misdiagnosed. Once diagnosed, the disease is treated with drugs that slow (but don’t reverse) the reduction in dopamine and the progression of the disease. Because of this, it’s a priority for medical researchers to find ways to diagnose before symptoms first appear; they believe by doing so and starting treatment earlier, they can slow and delay the onset of symptoms and dramatically improve the quality of life for PD patients.

In 2015, a woman named Joy Milne did something remarkable. Milne has hyperosmia: an extremely sensitive sense of smell. She was able to distinguish with high accuracy between T-shirts worn by people with PD and those without the disease, purely by smelling the shirts. That established that there is a unique scent produced by people with Parkinson’s. And it raised a further question: Could dogs be trained to recognize that scent? 

It’s well-known that dogs have extremely sensitive senses of smell; they can detect chemical compounds at the parts-per-trillion level. That’s why they are employed by hunters to track prey, by police to track fleeing suspects, and by customs officials to detect contraband in luggage. So in 2016, a group of researchers and community members on San Juan Island created Parkinson’s alert dogs (PADs) to try to train dogs to detect the same scent that Milne was able to recognize. The project ran for seven years and wrapped up in March 2023.

Given the complex and costly nature of breeding and raising dogs solely for the purpose of training them for this kind of scent detection, they opted instead to use the household dogs of volunteer community members, in a humanely run program that only used positive reinforcement (no punishment). The 23 dogs enlisted in the program were of a variety of breeds.

Replicating the scenario with Joy Milne, the program collected T-shirts that had been worn overnight by donors, some who had been diagnosed with PD and some who had not and were believed not to have the disease. Over the course of time and using standard animal husbandry practices, the dogs were trained to identify the PD-positive T-shirts and signal to the trained when they had found one.

A lot of work went into designing the training protocol, since they couldn’t tell the dogs exactly what they were looking for — they could simply reward them when they gave the right answer. They also had to be careful to ensure the dogs weren’t detecting a scent produced by the effects of the drugs used to treat PD, so they needed to have T-shirt samples from both PD patients who were being treated as well as some who were not. And the training needed to be “blind”: The dog trainers couldn’t know which samples were which so that they couldn’t tip off the dogs (accidentally or intentionally). 

They also needed to ensure the dogs were not simply memorizing which samples were rewarded and which were not; they did so by having a “test” round after training using additional samples that the dogs had never trained on (and thus could not have memorized).

There were other complications with the training process. Animal training like this can “fade” over time if it isn’t constantly and consistently rewarded and reinforced. That proved to be a challenge with household dogs, since the dogs could potentially encounter the PD-related scent simply by interacting with people throughout their day, and their owners would not be able to reward “correct” behavior. For that reason, the program managers recommend against using household dogs for similar programs in the future.

Nevertheless, the program was a huge success. As a group, the collection of trained dogs had a 89% “sensitivity” rate (their rate of correctly identifying positive samples), and an 86% “specificity” rate (their rate of correctly identifying negative samples). Not all dogs did equally well (they had a variety of health issues, and age was a minor factor), but the top tier collectively had a 94% sensitivity rate, and the best dog reached 97.6% sensitivity and 97.1% specificity. The breed of dog wasn’t a factor; the best one was a poodle, but for breeds that were represented by more than one dog, there wasn’t a clear correlation or pattern.

Just as the researchers couldn’t tell the dogs what they were supposed to look for, the dogs couldn’t tell the researchers what they actually detected — which means that we know there is a unique scent that can be reliably detected, but we still don’t know what it is. A separate research study found about 500 separate chemical compounds that are unique to the skin secretions of PD patients, and the dogs could be detecting any of them, or some combination. Interestingly, this may be a case where artificial intelligence (AI) technology can assist in the future: This kind of training, in which the technology learns a complex pattern that humans can’t detect, is one kind of application AI excels at. The challenge will be to get enough data (samples from PD patients and chemical analyses of those samples) to use to train the AI.

It’s entirely possible, perhaps probable, that in time we will be able to use dogs to detect the signature scents of at least a handful of other diseases. There has been some success already with certain types of cancers, and there is certainly no lack of diseases that would benefit from early detection and treatment.

Plus, it’s inspiring that a breakthrough project like this can happen in a small island town in our neck of the woods. The PAD program has now been “rehomed” in a research institute in France that is continuing the work.

From small to tall: breed-varied household pet dogs can be trained to detect Parkinson’s Disease