Over the last decade, the use of AI and digital tools within the healthcare sector has grown exponentially, from improving a product for monitoring or treating conditions such as a heart monitor or an artificial pancreas, to creating a system that can help a doctor to do their job in the most efficient way, such as a symptom checker or a differential diagnosis tool. There are now a large range of symptom checkers on the market, and a growing interest in their use, not just at home by patients, but as a tool in doctors' surgeries and hospitals to help with care direction, initial presentation of symptoms, and patient engagement. With so many tools available, it is difficult to distinguish the different types of symptom checkers, what they can offer and how they can be used in a professional medical environment. We’ve written a white paper on patient engagement and symptom checkers, outlining the need for such tools, the different types available, and the best way to test and implement a symptom checker in an institution. In this blogpost, we will discuss one element of that white paper, touching on the different types of symptom checkers you can find on the market today, and their pros and cons.
The majority of tools use a rules-based system to power their symptom checker. These are built on decision trees, following a strict pre-programmed route to narrow down options and eventually provide the patient with a diagnosis and/or advice. They are easy to use and follow a basic simple structure but their simplicity does leave them open to some issues. They need constant monitoring and updating manually to improve their accuracy and scope, and the more symptoms or conditions added, the more complicated and unwieldy the decision trees get. This can mean the end user is left answering a large number of questions, many with no relevance, in order to reach the end goal of one diagnosis. That one diagnosis may be an option, but with many symptoms there are several possible diagnoses requiring further work from a medical professional to determine the final diagnosis. Generally, these systems are good for helping patients know what to do for single symptoms and for diagnosing very common illnesses for which the symptoms and diagnosis easily match up. They could help reduce the amount of people coming in with conditions that do not require a doctor, particularly if they have a triage tool attached to them. On the other hand, the results could be inaccurate and hiding a much more serious condition that does require medical attention.
Chatbot systems are becoming more prevalent, particularly as AI and chat windows are growing in popularity and ability in all industries across the internet. At the moment, however, all chatbot symptom checkers are a design front on top of a basic rules based system. The tree is below an AI chat system, which asks the questions in a friendly, and often quite intelligent, manner. This is better for the end user as they feel they are being heard, and it can be easier to understand the questions. It can, however, fall short if the message given by the chatbot does not correspond with the patient’s messages. For example, a patient could enter several symptoms at once in one message, and the chatbot system may only be able to pick up on one of those in order to fit it into the decision tree. This miscommunication with the computer is irritating and off-putting at best, and at worst it can lead to an incorrect answer being entered into the decision tree and patient being pushed early down an inappropriate pathway.
These systems use an interactive image of a body to help patients explain where the pain is or where they are experiencing symptoms. Again, this is more attractive to the end user and potentially easier to use initially, but they are using a rules based system underneath, and this can lead to confusion. Symptoms such as a pain in the ankle and a headache are easy to enter on an avatar, however, more ambiguous symptoms such as tiredness or dizziness, or mental health symptoms such as a constant sad mood, are harder to attach to an image of a body.
Rather than following a strict order of set queries and conditions like the decision trees in rules based systems, deep learning symptom checkers will take a much wider view. The term deep learning refers to technology that has been trained to have deep knowledge in a specific area. For the patient, they simply enter their symptoms as a list in free text and the system takes these symptoms and matches them up to the possible diseases using pattern recognition. The patient is then presented with a list of the likely diseases that could be the cause of this constellation of symptoms. Essentially, this is the first step a doctor will take when examining a patient - they will take all the clinical features, including age and location information, and using their own database of knowledge, come up with a differential diagnosis, a list of things it could be, which can then be narrowed down with further questions and actions. Systems like this have been found to be the most accurate overall. They do require a significant initial setup to ensure the database is properly trained, but once finished and tested, these systems require far less updating. Both Isabel tools, the professional Isabel DDx Generator, and the Isabel Symptom Checker which is based off the professional tool, use an enhanced deep learning system, and at the time of writing this blogpost and as far as we are aware, we are the only symptom checker to be using deep learning technology. Our DDx Generator has been rigorously tested and the system has been in action since 2001, with continual development and improvement, as well as being medically validated.
You can learn more about the Isabel Symptom Checker by downloading the full white paper below. We go into detail on how the system works, its integration capabilities, and why we believe symptom checkers like Isabel can help improve patient engagement and the diagnostic process. We also talk about the other symptom checkers available right now, and some studies that explain the need and demand for symptom checkers in healthcare institutions.
Jason is the CEO and Co-founder of Isabel. Prior to co-founding Isabel, Jason spent 12 years working in finance and investment banking across Europe. His daughter, Isabel, fell seriously ill following a misdiagnosis in 1999 and this experience inspired Jason to abandon his city career and create Isabel Healthcare Ltd.