Tech

How doctors and artificial intelligence can work together

24 February 2022 in Tech

It is unlikely that grand predictions of AI replacing doctors will come true in our lifetimes - at Healthily, we certainly hope they don’t. But there is a strong case for their collaboration, particularly if outcomes include improved patient care, system efficiency and reduced human error.

In a new and insightful article for Forbes Technology Council, Rena Christina Tabata outlines where AI and medical professionals can dovetail for the benefit of both patients and healthcare systems.

As Tabata points out, Medical AI is already showing that it can detect traditionally difficult to identify or diagnose conditions.

At Healthily, we have designed our Smart Symptom Checker to focus on symptoms and influencing factors rather than probable conditions. This means we can look beyond seemingly innocuous symptoms and detect symptom combinations or influencing factors that might signify early signs of cancer. Though results might surprise users, being less probable doesn’t make them impossible.

Looking for AI health solutions for your business? Get in touch to speak to an expert.

Smart Symptom Checkers

As smart symptom checkers have almost infinitely extensible datasets, we will evolve to include external factors such as social determinants and lifestyle models. For instance, smoking is already an influencing factor in the Healthily algorithm.

This may lead to early presentation and diagnosis, significant increases in life expectancy, and a reduction in healthcare spending in the long run.

For Tabata, deciding when a model is ready to be released to the big wild world is complex.

We would agree. After six years of development, Healthily launched the first AI Smart Symptom Checker in 2015, and while there would have been benefits to waiting longer, at some point, you have to allow actual users to play to learn how to keep improving your product.

As a result, we’ve been able to identify how users describe symptoms (rather than, say, pointing to a pain point) and interpret conclusions from their final report. We can establish what further questions we need to ask to understand conditions and how to improve our user experience fully - and our technology and intelligence grows.

At Healthily, we take the development of our AI as seriously as a pharmaceutical company takes the development of a drug. This is difficult for digital investors who are used to a culture of moving fast and breaking things. But in medical AI, we can’t afford to break things, as it could lead to loss of life.

Will people ever trust AI chatbots?

As Tabata points out, trust in non-human digital assessment is the main barrier to their acceptance - no matter how good it is.

The gold standard of face-to-face medical consultation is embedded in developed societies and coveted in emerging economies. Many emerging economies have swiftly jumped to mobile money, but mobile health is a trickier selling proposition.

Another pertinent question Tabata asks is what margin of error would we accept? It probably should be at a similar rate to that of humans, but the issue is we don’t actually know what the human error rate is in primary care assessments. Soon Healthily will publish a literature review paper looking at studies of medical fallibility - and in truth, the research base in primary care is pretty sketchy.

Looking for AI health solutions for your business? Get in touch to speak to an expert.

What are we testing against?

Healthily can now match GPs on vignette tests (mini-medical stories). Still, we often get marked down in academic studies because our symptom checker has asked more questions than the GP, elicited more information, and concluded it differs from a pre-ordained test answer.

Imperial College London’s Self-Care Research Unit currently has a paper in peer review that looks at this issue and concludes that the current process for testing AI symptom checkers is flawed for this and other reasons.

That said, we need a benchmark for safety. At Healthily, we have arbitrarily decided that we need to be 98% safe to attain any form of trust with our users. We mean 98% of the time, our triage advice will be appropriate or over-cautious. We are currently at 97% according to internal testing, so we still have work to do.

How do Healthily users find the AI experience?

Currently, our users tell us they find the symptom checker and final report useful and helpful - mainly because it eases worries about symptoms and helps decide what next steps they should take. The final report is also a helpful discussion tool if they need to see a pharmacist or doctor. So it seems trust can be established if the Healthily Smart Symptom Checker is clear that is it a triage for the best next steps, not a replacement for healthcare professional support.

Healthily AI solutions for your business

Could award-winning digital health technology, AI symptom checking, assessments and triage from Healthily help your customers, sales or process efficiency? You could help your users understand faster whether their next steps should be self-care, doctor care or urgent attention, or use our AI triage and reports to pre-screen patients for an online consultation.

Visit our business page to find out more, or get in touch to speak to an expert.

Important: Our website provides useful information but is not a substitute for medical advice. You should always seek the advice of your doctor when making decisions about your health.

The self-care app
Check your health, manage 100s of conditions, track your progress and much more. Download today.
Top