Innovations in industries like retail, communication, transportation, and entertainment, have forever changed customer expectations. Information and services can now be accessed immediately from nearly any device in any location. For healthcare, this means patients want new ways to conveniently and quickly access care… and taking their business elsewhere if those needs are not met. In fact, it has been reported that 30% of patients left an appointment because of long wait times, and 20% of patients permanently changed providers for not being serviced fast enough.
To meet these expectations, some healthcare organizations are now adopting a “digital front door” strategy to engage with patients at every touch point of their healthcare journey. Doing so allows organizations to create a personalized and cohesive digital relationship with patients. This is challenging because healthcare data is complex, compliance laws have to be followed, information must be accurate and secure, and costs kept down while meeting appropriate staffing needs. The great news is that AWS provides services that allow healthcare professionals to engage with their patients. Many of the same services that allowed for rapid innovations in other industries were built on AWS and are now being utilized to transform healthcare organizations.
Chatbots are one tool that should be considered in a digital front door strategy for solving some of these patient demands. They can act as virtual workers, provide immediate and accurate information, triage patient needs, schedule appointments, and have the ability to scale out and service any number of patients. In this post, we discuss how healthcare organizations can deploy a chatbot that scales to handle increased demand, automates data collection and presentation, meets compliance laws, and only costs a few dollars to get started.
The first task we must do, before actually creating a chatbot, is determine what use cases our chatbot should handle. This provides us a direction that aligns with business and clinical needs. Ask yourself about what problems need to be solved at the digital front door. What questions could a patient ask and how would a bot respond to those questions? Take the time to document these in a spreadsheet and group them based on subject matter. An example of this type of logic planning is shown in the table below.
Remember that you are building a digital representative of your organization. Keep answers concise, professional, and written as if they are from the same person. This is also a good time to research other applications that make up your organization’s front door strategy to ensure that workflows and data are consistent. Creating an automated bot is helpful so long as the answers it provides are correct.
With a plan in place, we can now create our bot. AWS makes it easy to get started through pre-built solutions. For this example, we use a sample solution documented in a previous AWS blog, the QnABot. Follow the instructions within the QnABot blog and launch the AWS CloudFormation stack. You will have a fully functional bot ready to customize. The solution is entirely serverless, so you only pay for resources you use and the bot will scale without having to worry about managing servers. All of these services are HIPAA eligible and can be configured with encryption at-rest, in-transit, and logging to meet your obligations under the AWS Business Associates Addendum. The figure below shows the high level architecture diagram of a QnABot. If you would like to dive deeper in to the architecture, you can refer back to the QnABot blog here.
For the customer facing part of the bot, Amazon Lex provides a web client so that the same bot can be embedded within any website using simple HTML. This allows healthcare organizations with multiple websites, even those targeting different patient populations, to have one centralized bot that is servicing all questions from visitors at their digital front door.
Chatbots are conversational interfaces, and they work on the concept of defining intents and utterances. Think of an intent as a function or response that your bot performs. Think of utterances as what people say in order to trigger an intent. As you can imagine, it’s difficult to predict all of the ways someone may ask a question. Fortunately, thanks to how conversational interfaces work, we can take our questions and generalize them into broader key words to serve more questions with the same answer. For example, think of how many ways our question “When are you open?” could be asked. While we can enter this question verbatim within our bot, we can also define alternative phrases for it, such as “hours” or “what time”, so we can funnel additional user questions to the correct answer. It would enter as this within our bot’s content management designer…
Patients can now ask our chatbot about operating hours in a variety of ways and the correct answer is returned.
We can customize our bot for additional capabilities by modifying its code. One of the core components of this particular bot is the Bot Fulfillment AWS Lambda Function. The purpose of this function is to interpret responses from the user. To gain insights into the needs of our patient population, we can modify the function to store all user responses in Amazon DynamoDB, a NoSQL database service. From Amazon DynamoDB, Amazon Kinesis Data Firehose can store the user interaction data in Amazon Simple Storage Service (Amazon S3) so it can then be displayed as a dashboard using Amazon QuickSight. We can even add functionality so that if a user asks “Can I speak to a real person?”, the function writes to a queue in Amazon Connect, a cloud contact center, so users can be connected to a professional. Below you will find the fulfillment function architecture.
We now have a fully functional chatbot containing our curated information. However, healthcare organizations have additional information within thousands of unstructured documents that may be beneficial for the public to search. This could include directory information, clinician biographies and specialties, or illness and symptom information. This is where Amazon Kendra, a machine learning powered enterprise search service can help. Please note that Amazon Kendra is not HIPAA eligible so be cognizant of the data you decide to target. Amazon Kendra can extract information from unstructured and structured documents in Amazon S3, relational databases, websites, and other popular sources. It then returns the document or source where it can surface the answer to the query. Amazon Kendra provides the ability for keyword searches such as “headache cause” or natural language questions such as “what are common causes of a headache?”. You can boost the relevance of document freshness so that more recent documents are given priority over stale documents. If there are conflicting or multiple answers to a query, Amazon Kendra ranks the documents it finds and returns the list of ranked documents. Users can even vote on results to improve Amazon Kendra’s querying ability further. By utilizing a search service such as Amazon Kendra, healthcare organizations can give patients additional knowledge at the digital front door. You can read about adding Amazon Kendra to the QnABot here.
AWS offers additional ways you can service patients with chatbots in a secure (using encryption at rest and encryption in transit) and compliant way. Here are some other possibilities and integration points you could utilize a chatbot in a healthcare setting:
- Webpages and portals – Chatbots on websites and patient portals are common these days. As patients continue to take greater responsibility for their care decisions and planning, chatbots can create a personalized conversion for a patient while empowering them with more information.
- SMS – Over 7 trillion SMS messages are sent every year and over 6.1 billion people around the world have an SMS enabled phone. By integrating a bot in a two-way SMS format, you can leverage this channel to engage with patients all over the world in a more personal way. Refer to this prior AWS blog on how you can build two-way sms chat bots with Amazon Pinpoint’s SMS channel. Please remember that while Amazon Pinpoint is HIPAA eligible, excluding voice message capabilities, we do not recommend sending healthcare information through SMS. Keep messages generic, and redirect users to systems they need to authenticate with (such as electronic health record applications) to read their personalized message.
- Amazon Alexa – A voice enabled interaction with an NLP based search on the backend can offer patients a more natural and conversational approach to finding information. Amazon Pollywith Amazon Lex can help you build a voice enabled bot that can serve as the digital front door for patients to access remotely.
- Amazon Connect – As demonstrated above, Amazon Connect can be integrated for a callback feature. If a bot were screening for symptoms, for example, it could connect the patient to the right caregiver in a relatively short time-frame without having to have a patient wait on hold.
- Mobile Applications – Amazon Lex offers mobile SDK’s so that your health bot can easily be integrated into your existing mobile applications.
- Amazon Comprehend Medical – An extension of your bot with Amazon Comprehend Medical (natural language processing service for medical vocabulary) can help with quickly and accurately gathering information such as medical condition, medication, dosage, and frequency. Medical ontologies such as ICD-10-CM and RxNorm can also be linked with the extracted information.
As healthcare continues to rapidly evolve, health systems must constantly look for innovative ways to provide better access to the right care at the right time. Applying digital technologies, such as rapidly deployable chat solutions, is one option health systems can use in order to provide access to care at a pace commiserate with patient expectations. In this post, we showed you how to quickly get started with an existing solution and customize it for a healthcare organization. We encourage you to take a deeper look at the services mentioned and to start experimenting with chatbots in order to see how they can fit within your own healthcare organization.