Have you worried that your headaches are the result of a brain tumour, or that your child’s leg pain is caused by cancer? You’re not alone. You may well be a cyberchondriac: “a person who compulsively searches the Internet for information on real or imagined symptoms of illness.”  If this sound familiar, you are in good company.

If you search “child leg pain”, google will auto-complete your search with “leukemia” – not because it is the most likely cause of your child’s leg pain, but because people who have searched “child leg pain” in the past were most likely to have clicked on links correlating this phrase with leukemia (probably because they wanted to understand the worst-case scenario). That’s how machine learning works – it pushes up the article that was most popular among other readers.

It makes sense to push up an article that most previous users clicked on – this is one of the best proxies for relevance to new users. However, the engineers behind search engines realise this isn’t necessarily beneficial for google users:

• It’s scary – the average reader may assume cancer is the most common cause of child leg pain, or brain tumours are a common reason for headaches. Cyberchondriacs get even more paranoid.
• It can encourage harmful behaviour – imagine if you search “best way to kill myself” and the top hits documented in detail the most painless way to die. Will the information push you over the edge in your decision?

Engineers behind search engines have to make a choice on what information to present to users – what people want (the traditional way) versus what they may need. 

The Making of “Dr Google” 

It was my pleasure to have Evgeniy Gabrilovich, Senior Staff Research Scientist working on health-related searches at Google, shed light on how Google thinks about it’s responsibilities to users. Evgeniy is addressing a sizeable group of Google’s customers. 5% of all google searches are health-related, 20% of which are people who type in a symptom hoping to find a cause.

Evgeniy’s team works on The Health Knowledge Graph, which aims to give users the best facts when they enter their symptoms.  The Health Knowledge Graph does not replace traditional web search, it complements it. Try it out: Type in “chest pain”, “depressed” or “child leg pain” and you will get a side bar on the right which covers the ranked list of likely conditions, how common or critical the condition is, incidence by age group, etc. The center section still presents traditional web-search results.

 

When you type in a symptom you’re experiencing “child leg pain“, Evegeniy’s team aims to give you the most accurate diagnosis while minimising cyberchondria “Growing pains”.

Google realised that they didn’t have the expertise to do this on their own. It’s a huge technical challenge because of the large number of conditions and symptoms, and the overlaps between them. Furthermore, people use colloquial language to describe their symptoms, which the machine needs to decipher. Finally, user intent is often unclear. For example, if someone types in “weight loss” – are they trying to lose weight? Are they describing a side effect of medication?

Together with doctors from Harvard Medical School and the Mayo Clinic, they used machine learning to establish correlations between symptoms, conditions and treatments such that when you type in your symptom, you will get information that closely mirrors what a doctor might tell you (although it doesn’t go so far as to diagnose you… yet). Just to make sure, every result is evaluated by real doctors, who are asked “would you be comfortable with google showing these results”? 

What does this mean for the medical profession? 

Fifteen years ago, very few would have trusted medical advice that wasn’t from a doctor. Ten years ago, people started turning to the search engines for advice it wasn’t ready to give. Now, search engines are training themselves to give professional medical advice. They will only get better.

What’s next? I recently met a start-up, Mendel Health, which automates matching cancer patients to clinical trials through personal medical history and genetic analysis. Founder Karim Galil was previously a medical doctor. He was motivated by the fact that a single doctor’s brain cannot capture all information about diseases, possible treatments and clinical trials. He had patients die because he, as their doctor, was not aware of a clinical trial that could have saved their life.

Let’s take Karim’s idea a step further – suppose all my genetic, medical information and daily physical conditions (heart rate, glucose levels…) are constantly updated in a database that is linked to all potential interventions, treatments and medications.

• While I am healthy, I can be alerted to risk factors and preventative actions (for example – you have a 50% chance of becoming diabetic in the next year. If you do X, Y and Z, the probability drops to 20%).
• When I am ill, I can understand all my treatment options and the probability of success.

When a machine can diagnose me and recommend potential treatments, what will be the role of my doctor?

• Much of what a primary care doctor does – assessing my condition, referring me to other specialists or recommending basic medications – can be encoded in software and search engines. Will they simply be a ‘stamp of approval’ – a safety blanket of sorts – before I take my next steps to get treatment?
• Perhaps new roles for doctors will open up – for example, in training and verifying Dr Google as more and more people rely on it.
• Complex surgical procedures will likely still require human attention. However, with robotic technologies like Verb Surgical, which enable top surgical expertise to be propagated across many doctors, will the average level of surgical skill required by each doctor be lower than before?

Why does this matter?

I honestly can’t envision a world with no doctors. Health is so close to our hearts that it requires a personal and emotional touch. However, it is important to understand how technology will change the role of the doctor:

This this will have large impacts on how countries train doctors (e.g. how long? what skills?),  allocate resources (e.g. primary care vs specialists), and design incentives in their healthcare system (e.g. if patients have access to so much information, will there be a trend towards over-consumption of medical services? Do co-payments have to change?). 

I am certainly not an expert in the field of medicine or medical technology, but would like to continue exploring this topic – especially from the perspective of what countries need to know, and how they should respond. Ping me if you are a doctor / work in healthcare and medical technology – I would love to hear your thoughts.