How is the healthcare industry using chatbots?

by Vagelis H. 07/14/2020

There is a lot of buzz around AI chatbots in the last few years, in most industries, but especially in healthcare. So, how are chatbots benefiting healthcare today? I’ll try to group healthcare chatbots in categories, based on their use cases:

1. Self-diagnosis chatbots

This is probably the most publicized category. Key players include UK-based Babylon Health, Your.Md, Ada Health, and Buoy Health. These chatbots use a knowledge base, and then ask the patient questions to try to make the right diagnosis. The knowledge base is typically created by analyzing a set of medical papers or cases.

Specifically, information extraction algorithms (such as MetaMap) are used to extract medical concepts (disorders, symptoms, drugs and so on), which are then linked to discover patterns, as shown in the image below , where nodes may be disorders, symptoms, drugs, and so on.

Deep learning is used to learn such patterns, and estimate the most probable diagnosis given the provided information and similar past cases (training data)

The appeal of self-diagnosis chatbots is that they can be used at any time, and have the potential to decrease healthcare cost, as they may substitute a physician for common medical cases.

The business model of many of these chatbots is to offer the self-diagnosis for free and then charge the patient for access to related services such as a doctor appointment or a remote consultation.

These chatbots are often hosted on a mobile app, or sometimes on Messenger, if they don’t share the data with a covered entity (so HIPAA is not an issue).

2. Healthcare navigation chatbots

The goal of these chatbots is to help users navigate the complex healthcare landscape; for example, find a doctor in their network, compare prices, or learn about their benefits.

These chatbots are typically available only to members of a healthcare system, so they are not freely accessible on the web. For example, Premera Scout by Blue Cross helps patients understand their benefits. Mayo Clinic is also very active in this area.

The technology of these bots can range from simple multiple-choice diagrams to deep learning models that are trained on existing customer service chats.

These bots are typically hosted as part of a patient portal or other secure environment, which limits their popularity, as many patients don’t know how to login to their patient portal. Jiseki Health tries to avoid this by relying on SMS chatbots to connect patients to services. SMS is also used to send reminders for appointments or prescription notifications.

3. Health coaching chatbots

These chatbots try to engage patients to achieve a more healthy lifestyle. For example, they may remind patients to exercise and eat healthy every day. Examples of such chatbots include Doppel and 1-million-strong-to-prevent-diabetes.

There are also nutrition bots like forksybot or bots that focus on mental health, such as Woebot. These bots often leverage existing clinical scales, such as PHQ9, which is a popular 9-question scale to detect depression. We can also include in this category chatbots that remind patients to take their medication or follow the doctor’s advice, such as Florence.

These bots generally have pre-defined flow and hence use less frequently advanced technologies like deep learning.

An interesting aspect of these bots is that they are usually hosted on Facebook Messenger, which allows them to easily send reminders and follow-up with the user.

4. Healthcare marketing chatbots

These bots typically collect information on leads (potential customers/patients), give them some information on a practice’s services or make an appointment. Such bots are more popular in out-of-pocket specialties, such as dentists, chiropractors, plastic surgeons or physical therapy, which are the specialties that invest more in marketing.

A key intuition here is that as patients search for services on Google, if a practice’s website has a chatbot that can engage the patient 24x7, it will make it more likely to attract this patient, especially in the case of millenials, where instant gratification is key.

In terms of technology, these bots use simple information extraction to extract fields like address or phone number from the user messages, and may use more complex deep learning Natural Language Processing (NLP) techniques for question answering (e.g., FAQs).

Such chatbots can be found on the pages of individual practices, and sometimes on their Facebook page.