A chatbot to complement a doctor’s visit? It’s not as radical as it sounds. Hospitals and health systems have plenty of incentive to further adoption of AI conversational agents; according to a survey published by research firm Statista, 27 percent of people would prefer to answer sensitive health questions posed by virtual assistants than by people. Another study — this one by Juniper Research — found that chatbots could lead to cost savings of over $8 billion by 2020.
One of the juggernauts looking to get in on the ground floor is none other than Microsoft. The Seattle company’s been experimenting with health care chatbots since 2017 — that’s the year Health Bot, a software-as-a-service offering within Microsoft’s Azure, made its debut. The idea was to provide a platform that would enable care providers to deploy text, voice, and touch-response experiences capable of replying to questions with natural language, and that would integrate relatively seamlessly with Skype, Cortana, Facebook, Slack, SMS, and others.
Health Bot had humble beginnings — as a hackathon project — but in roughly a year, Microsoft commercialized and launched it in private preview for select partners, including UPMC, Aurora Health Care, and Premera Blue Cross. Now, almost three years after its creation, it’s launching broadly in the Azure Marketplace — a digital marketplace for apps in Microsoft’s Azure cloud computing ecosystem — as the Healthcare Bot service.
Microsoft’s Healthcare Bot service is, at its core, a Dialogflow-like bot design suite tailor-made for health care scenarios (Microsoft Healthcare Israel head Hadas Bitran describes it as a “toolbox.”) To that end, it allows developers to build conversational flows with symptom-checkers that tap built-in triage protocol databases and surface information about previous conditions, health plan benefits and eligibility, and costs. Bots built with the Healthcare Bot service can handle appointment scheduling or look up doctors within driving distance. And the bots benefit from Azure’s NLP tech — chiefly language models that can handle interruptions, topic changes, human error, and complex medical questions.
“[Customers] don’t have to start from scratch,” said Botran, who previously worked on Microsoft’s Cortana and spearheaded development of the Healthcare Bot service. “[The Healthcare Bot service] has … content knowledge, such as a symptom-checker and information about conditions, medications, and procedures. It has language models trained to understand health care terminology. It understands if you are complaining or if you are asking about what doctor you should see or if you are thinking about side effects of a medication.”
With the Healthcare Bot service, providers, payers, insurers, and pharmaceutical companies can orchestrate tasks from a centralized, extensible portal and management API. It’s where they’re able to author new bots or modify and debug old ones, and integrate the bots with cognitive services like Microsoft’s Language Understanding Intelligent Service.
“Virtual assistants will never replace medical professionals,” Bitran said, adding that bots built with the Microsoft Healthcare Bot service never make a diagnosis or offer treatment. “That is not what they are for. Rather, virtual assistants help ease the burden from the health care system, helping medical professionals optimize their time.”
No matter how the chatbots in question are configured, Microsoft says they encrypt all data at rest and fully comply with the Health Insurance Portability and Accountability Act (HIPAA), the U.S. laws that provide data privacy and security provisions for safeguarding medical information. They also conform to the EU’s GDPR and ISO 27001 and ISO 270018, the industry codes of practice that concern the handling of personal data protection in cloud environments. Additionally, they have built-in privacy controls that allow users to ask Healthcare Bot service chatbots what sort of data they’ve collected and to tell them to forget that data.
In the coming weeks, Microsoft says partners will be launching bots built for scenarios like preparing for a doctor’s appointment or tests, answering pregnancy questions and tracking pre-natal appointments, and navigating the details of children’s treatments. And Microsoft says Healthcare Bot chatbots have already had a measurable real-world impact.
Multibillion-dollar diagnostic services provider Quest Diagnostics piloted a bot that helped people who visited the website during call center hours find testing locations, schedule appointments, and get answers to non-medical questions, such as whether to fast before a blood draw and when to expect results. (It autonomously handed exchanges off to human reps in the event it became confused or wasn’t able to help with something.) In a survey Quest Diagnostics conducted post-deployment, 50 percent of respondents said they vastly preferred to engage with such a chatbot than a website search box or FAQ section.
“[People] are really learning how to drive their health care experience, and they have a lot of questions,” said Jason O’Meara, senior director of architecture for Quest Diagnostics in Cary, North Carolina. “To find answers to their questions, many people don’t want to browse websites anymore if they can get to their answer more directly using a bot.”
The Healthcare Bot service is part of Microsoft’s Healthcare Next initiative, which enables its health care partners to create intelligent tools with the goal of “capturing new opportunities to apply AI to health care.”
The first strategic research partnership was with the University of Pittsburgh Medical Center. The company has also seen the launch of HealthVault Insights, an AI-based project that allows partners to generate insights about patient health and encourage patient engagement, along with Azure-powered genome analysis pipeline Microsoft Genomics and AI-driven radiotherapy planning tool Project InnerEye.