Over 7% of all the search queries Google gets in a day are health-related inquiries
Improvements in machine vision technology and chatbots have enabled computers to get better and better at doing initial diagnosis of symptoms. Sometimes, these AIs can even catch things a human physician might miss, especially when looking for patterns in a large amount of cases.
AI symptom checkers work best when medical professionals team up with devs to create a custom API solution of their specific needs.
This is just one of the many ways in which artificial intelligence is redefining the healthcare industry. According to Accenture, investments in AI are projected to save U.S. healthcare providers $150 billion a year by 2026. AI symptoms checkers will play a key role in this process, improving the diagnosis, health and satisfaction of patients in addition to creating large savings for providers.
How Does It Work?
The most powerful AI symptom checkers are powered by one thing — lots and lots of data. Large data sets are critical, a symptom checker that only has data on the most common diseases is very likely to misdiagnose atypical cases.
“Every human being is different. If 100 people have the same disease, they will have different symptoms, different experiences of symptoms, different ways of explaining or expressing them and doctors will understand them differently and document them differently.”
— Dr. Jama Nateqi CEO at Symptoma
While it’s an option to have a symptom checker app that works using multiple choice forms that the patient fills out, the best ones allow patients to enter and describe their symptoms in writing, which is stored as free text. Then, the AI applies natural language processing to the text.
Based on this, the app conducts medical symptoms analysis then recommends next steps to the patient, whether that’s scheduling an appointment with a doctor, going immediately to the emergency room, or just taking it easy for a few days.
AI Training To Identify Symptoms For Diagnosis
The most technically challenging aspect of developing an AI symptom checker is training the AI on the data set, especially if you are going to be using free text.
The first thing an organization needs to do is to figure out which data is the most important, as the amount of data that healthcare providers generate is nearly infinite —- according to a recent IDC study the industry produced almost 2000 exabytes of data in 2020, and this amount is projected to grow 48% year over year.
To run all the patient data you have through an ML algorithm would take an impossible amount of computing resources — even for those who leverage external cloud computing solutions. The data-sets you use must be well curated.
“Proxet has deep experience across the healthcare industry, advising organizations on what data to collect and training ML algorithms on the data collected. We take special care never to apply a one-size-fits approach, developing a custom software solution for each specific case.”
— Vlad Medvedovsky, CEO & Founder at Proxet.
To the extent possible, you’ll also want to use a standard Electronic Medical Record (EHR) format. The wider variety of formats in the data set the more likely it is that the algorithm can misinterpret the data. So, proper AI training on clean data is critical.
When you have analyzed the data, now you have a data set that the patient’s symptoms will be checked against.
It’s also critical that you make sure the data is properly anonymized and HIIPA compliant.
The Components Of Symptom Checker Apps/Software
All AI Symptom Checker apps/software must have the following to work well:
- A database
- Diagnostic Engine
- A triage API
This is the information repository that powers the whole system. Patient inputs (like demographics, symptoms, and lab tests) are checked against the existing dataset. It connects this information with pieces of content in the database and returns a list of likely conditions (preliminary diagnosis),and care recommendations. To the patient, it just looks like a symptom finder program.
The diagnostic engine is the “thinking” part of the database. It uses the power of AI to analyze the patient’s inputs against the dataset.
Almost any AI symptoms tool is going to consist of a complex series of systems connected to each other by an API (Application Programming Interface). Simply put, an API allows one part of the system to send a request for information to another part of the system.
Triage refers to the sorting of patients by the urgency of their case.
A triage API sends patients symptoms to the dataset, where it is processed by the AI. Then the AI sends back a result with recommended next steps.
Large organizations need to pay special attention to how they build their APIs, which means taking a wholistic approach from the beginning. If you build one API for your customer chatbot, and another for internal use by practitioners, and a third for use during patients visits, that’s less than ideal. Multiple APIs built by different teams for different purposes can result in knowledge soloing, as different systems have important information that they are not sharing with each other.
This is why it’s best to work with experts from the beginning who can design a flexible API from the ground up. This way, if you need to add more APIs later, all your systems will still be able to talk to each other. This is a process known as integration.
- Triage tools. In medicine, the term triage means prioritizing patients according to the urgency of their case. Triage tools enable this process.
- Self-diagnostic apps or health chatbots. Often taking the form of an online symptom checker, these help people discover the possible causes of their symptoms and help them find the right type of care.
Pre-Diagnostic Decision Support Tools
Pre-diagnostic decision support tools are designed to be used by medical practitioners in a clinical setting. These can display key information from a patient’s EHR as needed and make AI powered suggestions to support the doctor’s decision making process.
A good example of medical API commonly in use is the Infermedica platform. You can see how it works in the graphic below. Most medical APIs operate on a similar model.
Note the arrows between the three grey boxes on the top right, and the blue box that says API. This illustrates communication between the AI-powered database and different types of user interfaces.
You may only need one user interface, or you may need several. Consider the following example:
John has been feeling ill for a few days now. At first he thought it was just some kind of mild stomach bug, but now he feels even worse. He doesn’t want to google his symptoms, because his past experience of checking symptoms on Google has usually returned results like “It’s probably nothing but it might also kill you”. It’s too broad a spectrum of answers to be useful.
Luckily, John’s has recently downloaded an app from his healthcare provider “Health.ly”.
He pulls up the app, where he has previously entered some key information about himself, such as his age, height, weight, and any important medical history. He taps the button to talk to the chatbot.
The chatbot will either ask him some multiple choice questions, or allow him to enter a description of his symptoms as free text. Because a lot of consideration has been put into developing a “bedside manner” for the bot, John feels comfortable talking to the app.
He tells the bot he has been feeling pain in his left abdomen for a few days, that the intensity of the pain has recently increased, and that seems to be moving in a downward direction.
The chatbot, using the API, sends this information to the database where it is checked against the symptoms calculator algorithm. It determines that John likely has a case of appendicitis — a serious and potentially life-threatening condition. It advises him that he should go to the hospital immediately.
If the AI is not highly confident in a specific result, it can provide a list of possible conditions, and links to more information about each. One of the major benefits of these chat bots is that patients can get an initial assessment quickly, avoiding delayed diagnosis with symptoms.
While perhaps the most common use of AI symptom checkers is to assist in diagnosis and triage, there are other applications as well.
For example, a patient can input their information and medical history to get a personalized health report, even in the absence of pressing symptoms.
These reports can be for the patient’s own use. But they can also be used in other industries. For example, insurance providers can use these reports when determining the risk-level of their client. According to a recent Accenture report, 77% of patients are willing to provide this data if they believe that they will benefit from cheaper premiums.
Proxet has significant experience in developing custom AI solutions in the healthcare and insurance industries. If you work with us, our experts will ensure that your organization is able to exploit these emerging technologies to the fullest.
Accurate parsing enables Q&A quality — but is it possible? No matter the industry or sector, businesses regularly deal with the question of how to efficiently process large amounts of info-heavy documents. Organization leaders, including CTOs, CDOs, and CPOs, are often looking for solutions to this question.
Build a modern data stack by following best practices from data engineering experts. Learn about data maturity, data stack components, and how to build.