MedCity Influencers, Artificial Intelligence

3 must-haves when considering a chatbot for healthcare

Although there is a wide range of healthcare chatbot applications, such as providing medical information and mental health assistance, typical chatbot solutions are often inadequate, leaving both their users and owners frustrated.

One of the biggest challenges in the healthcare industry is patient accessibility to healthcare due to various limitations, such as limited availability of physician office hours and limited education about care sites (as outlined by this article).  The accessibility challenge is further exacerbated by the Covid-19 pandemic, as care resources have become even more scarce. Government agencies and healthcare organizations have started using or exploring how technologies can help improve patient accessibility as well as ease the burdens of care providers. One of such technologies is chatbot.

Although there is a wide range of healthcare chatbot applications, such as providing medical information and mental health assistance, typical chatbot solutions are often inadequate, leaving both their users and owners frustrated. The intended care accessibility is lost with chatbots ignoring or misunderstanding user inquiries or not being able to complete assigned tasks (e.g., care triage). It also diminishes the original purpose of easing care provider burdens, if it requires painstaking amounts of time and resources to set up and maintain a chatbot. With perhaps hundreds of chatbot platforms in the market, herein lies the challenge — how to select a chatbot platform for healthcare?

With improving patient accessibility to care in mind, government agencies and healthcare organizations should look for chatbot platforms with three characteristics: quality of conversation engagement, ease and speed of chatbot set up, and continuous, non-interruptive chatbot improvements. 

  • Quality of conversation engagement

Unlike many other service industries, healthcare provides the services that make people feel better, physically and mentally, literally. When a chatbot represents a care facility to interact with its visitors and patients, it must be instilled with a sense of responsibility and empathy just as if the visitors and patients were interacting with a real person. In other words, not only must a chatbot carry on a conversation, but it should also engage with its users in a quality and fruitful conversation, improving care accessibility.  Below are several examples of demonstrating the characteristics of quality and fruitful interactions. The chatbot that can support such quality conversations are also often known as cognitive AI chatbots.

Actively listening to users

To engage with users in a productive conversation, an effective chatbot should actively listen to its users and respond to user requests responsibly and empathetically, instead of forcing users to follow a pre-fixed, rigid path to navigate care options. The following example shows that a user interrupts an urgent care triage chatbot by asking a question. In this case, the chatbot answers the user question promptly and then continues the conversation.

An interaction between an urgent care triage chatbot and a user. During the triage flow, the user interrupts the flow with a question. The chatbot is able to interpret the question and respond to it promptly and then continue with the flow. 

A patient is already in a very stressful situation, emotionally and physically,  when s/he inquires about care services. Just like an empathetic care provider who could help alleviate the stressful situation and deliver better care outcomes (see this article on the role of empathy in healthcare), a healthcare chatbot should also feel what a patient feels and respond empathetically.  The example below shows the interaction between a chatbot and a patient in the course of mental health assessment. The chatbot is able to actively listen to and respond to a user empathetically.

An interaction between a chatbot and a patient in the course of mental health assessment, during which the chatbot actively listens to the patient and responds to her empathetically. 

Reading between the lines

Studies show that patient personalities are correlated with medication adherence in both older adults and adolescents, which informs the need of individualized care management regimens.  When employing a chatbot as part of care management regimens (e.g., monitoring patient status and encouraging treatment adherence), such a chatbot should also be able to deeply understand each user and use the insights to personalize each engagement.

The example below shows that a personal wellness chatbot interacts with two different users and encourages both users to stay with the wellness program with different motivational statements based on the users’ personality, which is inferred by the chatbot during their respective conversations.

A healthcare chatbot is able to infer personality insights from a conversation and use the insights to personalize each engagement. On the left, the chatbot engages with a self-motivated, independent person while on the right, the chatbot engages with a caring, family oriented person. 

To support the deeply personalized interaction shown above, a healthcare chatbot must analyze user input beyond the surface meaning to automatically infer users’ unspoken needs and wants as well as personality. The chatbot can then use the inferred insights to personalize each engagement. Typically such inferences are made by combining big data analytics with modern psychometric theories. Field studies have also shown the usefulness of such inferences to predict team performance.

  • Ease and speed of chatbot setup

If a chatbot platform claims to offer the quality of conversation as described above, it is then important to ask how easily and quickly to set up and deploy such a chatbot in a production environment. This is because building quality chatbots from scratch requires AI expertise and a sophisticated IT team, not to mention large amounts of training data and intensive computing resources. Many organizations underestimate the importance of time to value associated with a chatbot project and the importance of using time to value to evaluate a chatbot platform. Otherwise, they could become very disappointed with the actual timeline needed or frustrated with the chatbot quality given the amount of time they spent on building it, assuming they ever get to a successful outcome.

As few healthcare organizations have the required AI/IT expertise or resources, it is important to select a chatbot platform that can support end-to-end, no-code design, development, and deployment of chatbots that can also deliver quality of conversation engagement, such as active listening capabilities.

Moreover, it is important to check if a no-code chatbot platform is enabled with reusable AI, which will drastically speed up the setup process since a chatbot can be quickly customized with pre-built AI modules (e.g., pre-built active listening engine) without the need of building everything from the ground up.

  • Continuous, non-interruptive chatbot improvements

No chatbot is perfect and it requires human supervision and continuous maintenance and improvements. Moreover, healthcare situations may change rapidly, such as the pandemic situations and corresponding care policies. On the other hand, every healthcare-related conversation is a critical conversation, no matter whether a patient is inquiring about treatment options or is in the middle of a health assessment. A healthcare organization simply cannot afford to interrupt any ongoing conversations in order to teach its chatbot new knowledge (e.g., informing patients about new care policies during the pandemic) to improve its capabilities.

Thus, it is also important to select a chatbot platform that can support continuous, non-interruptive improvements. For example, if a user asks a question that a healthcare chatbot is unable to answer, such a platform should inform the human supervisor of the chatbot about the situation in real time and enable him/her to improve the chatbot instantly without interrupting any ongoing conversations.

Summary: Due diligence questions

Overall, healthcare services are human-centered services. When healthcare organizations select a chatbot solution to augment its workforce and improve patient accessibility to healthcare, it is important to select a chatbot platform that can enable a human-centered engagement at scale with 3 must-haves: (1) delivering quality and productive engagements, (2) easy and quick to set up, and (3) support of continuous, non-interruptive chatbot improvements.

The following questions could be used to perform due diligence when considering or evaluating a chatbot platform for healthcare:

Assessing quality of conversations

  • I want to use a chatbot for patient engagement. How does it handle arbitrary user interruptions? How does it resume the chat flow from user interruptions?
  • How does it handle user free-text questions?
  • How does it handle complex questions that require a multi-turn conversation?
  • What can it understand about its users through the conversation?

Assessing ease and speed of setup

  • How fast can it be set up?
  • By whom?
  • What resources do I need to do so?

Assessing chatbot improvements

  • How can I know if my chatbot made a mistake or cannot answer certain things?
  • How can I improve my chatbot?
  • Do I need to take down my chatbot or interrupt the ongoing chats in order to improve it?

Photo: venimo, Getty Images


Avatar photo
Avatar photo

Michelle Zhou

Dr. Michelle Zhou is a co-founder and CEO of Juji, a cutting-edge artificial intelligence company that powers cognitive AI assistants in the form of chatbots. She is an expert in human-centered AI, an interdisciplinary field that intersects AI and human-computer interaction (HCI). Zhou has authored more than 100 scientific publications and 45 patent applications on subjects including conversational AI, personality analytics, and interactive visual analytics of big data. Prior to founding Juji, she spent 15 years at IBM Research and the Watson Group, where she led the research and development of human-centered AI technologies and solutions, including IBM Watson Personality Insights.

Zhou serves as editor in chief of ACM Transactions on Interactive Intelligent Systems and as associate editor of ACM Transactions on Intelligent Systems and Technology. She is an ACM distinguished member and was formerly the steering committee chair for the ACM International Conference Series on Intelligent User Interfaces. She received a Ph.D. in computer science from Columbia University.

This post appears through the MedCity Influencers program. Anyone can publish their perspective on business and innovation in healthcare on MedCity News through MedCity Influencers. Click here to find out how.

Shares0
Shares0