Removing human bias from medical evaluations with conversational AI

The Challenge

Dialogue’s mission is to improve humanity’s well-being by reducing barriers to quality care. As Canada’s leading virtual health care platform, Dialogue offers integrated healthcare services for employers to keep their employees happy, healthy, and performing at their highest potential. A full range of health professionals, from nurses to physicians and allied health practitioners, are available at the click of a button via mobile phone or computer to help employees optimize their work-life balance.

Dialogue has enjoyed rapid adoption in recent years, but with success comes the challenge of delivering great care at scale.

To address this challenge, Dialogue set out to automate the patient intake process by developing a conversational AI assistant capable of gathering information about patients’ medical conditions through a conversational evaluation before introducing them to a member of the care team.

The Solution

During the patient intake evaluation, Dialogue’s contextual assistant uses Rasa NLU to classify patients’ chief complaint and uses Rasa Core for generating conversational responses. The assistant supports three different languages—English, French, and German—depending on user preference and can support multiple users in a single conversation.

The ability to integrate with existing systems and modify conversational AI components was a key requirement of the project. The Rasa action server integrates external services that support the patient experience, like a patient profile service. Dialogue also extended the Rasa Form Policy to integrate an external inference engine for managing the patient evaluation dialogue.

Multimedia and innovative message types provide a richer patient experience within the Dialogue Virtual Clinic. Moving beyond simple text questions with suggested replies, patients can upload files and images through the interface, and the Dialogue Virtual Clinic even allows patients to select a specific part of the body where they experience pain from a visual representation.

Dialogue’s tech stack includes Mattermost for messaging and relies on Docker images containing Rasa components—Core, NLU, and action server—that are deployed using Kubernetes on AWS EKS, through a CircleCI deployment pipeline. Dialogue Virtual Clinic includes a variety of client-side applications that interface with the conversational agent, including, implementations written using React, React Native, Android Native, and Electron.

The Results

Dialogue leverages technology to help the care team deliver the safest and highest quality care to patients. The Virtual Clinic assistant automates, standardizes, and removes human bias from key steps in the patient experience so the care team can focus on what matters most: patients and their safety.

Patients can now complete an evaluation anytime, anywhere, in only a few minutes on average—compared to the time-consuming process patients are used to with the traditional medical system. This process helps the care team consult patients efficiently, so they can continue to provide quality care to a greater number of patients in a smarter, more effective way.

Conversational systems greatly support medical professionals to better respond to patients’ needs. Rasa platform allowed us to focus on creating fast conversational experiences and deliver them under a human-in-the-loop regime. This way we ensure that medical decisions are always taken by medical professionals using evidence-based methods that lower the bias. And the administrative and data collection tasks are done via an intuitive chatbot.

Rasa’s open-source model and a clean extensible architecture make it an ideal choice for product development as well as applied research projects.

Alexis Smirnov, Co-founder and CTO of Dialogue