How does artificial intelligence work in the Medgate App?
Why does Medgate use artificial intelligence?
First of all: we’re not talking about C-3PO from Star Wars or self-driving cars! Artificial intelligence – AI for short – is a term that is used in many ways. Terms such as “machine learning” and “natural language processing” (NLP) are much more appropriate in our context.
But let’s start at the beginning: our starting point is a patient who has a medical concern and is looking for help or advice. Medgate doctors are a popular point of contact because they are available around the clock, take time to listen to the patient and can often offer assistance without the patient needing to visit a practice.
However, not every issue can be treated with telemedicine. There are conditions that require physical examination or treatment. Nobody contacts Medgate if it’s already clear that a wound needs to be sutured. But for some complaints it’s not clear what needs to be done.
So our question was, how can we identify patients as soon as possible who we cannot help with telemedicine and for whom we would recommend an in-person consultation? It would save time for the patient and be more efficient for the health system if such patients could be identified as soon as possible.
What kind of data form the basis for Medgate’s artificial intelligence?
Medgate doctors have cared for several million patients since our founding in 2000. In each case, the patient’s symptoms were recorded, a decision was made and then recorded in the medical files. If a patient contacts us with a concern, it is very likely that we have had some similar cases in the past. If that is the case, we can see whether the doctors who dealt with these past patients provided them with telemedicine support or recommended an in-person consultation. We can also see whether (and with what urgency) the doctors would have recommended a consultation with a family doctor, a specialist or a hospital.
The Medgate App’s artificial intelligence takes on this task. The App asks the patient questions about his or her concern. Artificial intelligence helps to determine which questions should be asked in order to find as many similar cases as fast as possible, and to exclude dissimilar cases from the comparison. The App then recommends the point of contact: should the patient have a telemedicine consultation? Or is it very likely that the patient cannot be treated with telemedicine, and should contact his family doctor instead? Or perhaps it’s an emergency, and the patient should call Medgate immediately to be on the safe side instead of making an appointment?
Many anonymized medical case files have been processed and the separated data compiled in the form of a “knowledge graph”, so that the Medgate App’s artificial intelligence can triage such cases. The App’s AI relies on this knowledge when ‘interviewing’ new patients.
So it is not really artificial intelligence that is used, but human intelligence and medical expertise. The machine doesn’t invent anything; it just organizes decisions made by doctors in the past. We can think of this as asking all the doctors who have looked after Medgate’s patients over the years to help diagnose the new patient! This technology also makes it possible to incorporate findings from publications and guidelines.
Are you planning to expand the application of AI?
We’re just at the beginning. We are currently working on extracting further characteristics and information from the case data and including it in the model. This makes the system more meaningful and efficient.
There are many paths for expansion – we are taking one step at a time.
What is the current status of AI in Swiss medicine? And what about patient acceptance?
AI is increasingly used in the field of image-assisted diagnostics, and many devices and software tools already contain AI-based algorithms.
In principle, many patients have nothing against AI, as long as it doesn’t make the decisions. Patients inform themselves about AI, and are impressed that artificial intelligence can help to recognize symptoms and assess the urgency of their concerns. But ultimately, they want to be treated and cared for by a real person. Doctors should definitely make use of the best tools available in their work – and these tools may be supported by AI.
I share this view, as a digital health technologist and as a patient!
To what extent can AI reduce healthcare costs?
We’re looking for ways to improve results while reducing costs. That sounds contradictory at first, because most people think that high quality means high costs. I’ll give you two examples.
With image-based screening programs, it has been shown that machine image recognition is much more precise, faster and cheaper than human image recognition.
With drug treatment of hypertension or diabetes, it is important that patients receive the most suitable drugs at the correct dosage. Frequent checks enable adjustments that are much faster and more precise. It would be sensible to combine measurements that are collected online with an algorithm that monitors progress. Then the algorithm could notify the doctor if the patient’s values exceed a certain limit. This can be achieved with simple stored rules, but AI would make it possible to coordinate the patient’s therapy even better.
Let’s look ahead. How do you think that AI might change medicine during the next ten years?
There will be two parallel developments: AI will need to be established and developed further as increasing quantities of data become available. The potential lies in combining the two developments.
An increasing number of doctors will make more and more AI-based decisions.
Medicines are automatically prescribed for certain diseases, even if the prescription needs to be approved by a real person.
In radiology and pathology, a significant proportion of the findings are created automatically. This makes diagnosis faster, cheaper and more precise for certain tests.
In health economics, billing approaches can be more differentiated, fairer, and at the same time simpler from a human perspective. Lump-sum billing will no longer be used for certain interventions in which specific aspects of the treatment need to be considered and adjusted afterwards with an extremely complicated billing process. AI could derive useful information from the costs of similar previous cases, and this could overcome rigid and costly accounting rules.
In the future, a patient will meet a device using artificial intelligence at least once during treatment – and take this for granted.