Virta Health’s AI-Powered Coach Helps Patients Fight Diabetes

(Sergey Nivens/Shutterstock)

Type 2 diabetes afflicts tens of millions of Americans, with the disease inflicting severe impediments or even death on many — particularly in the wake of Covid, which is exacerbated by the condition. San Francisco-based Virta Health, founded in 2014, works to reverse Type 2 diabetes by helping people with the disease make nutritional changes and then, as those changes have the desired effects, carefully deprescribing medications. Virta, which has now treated tens of thousands of patients, powers this personalized care using Google Cloud-enabled AI and ML systems.

“Virta’s ML engineers have developed AI systems that comb through de-identified data like messages, biomarkers, labs, and health outcomes in order to identify patients that would most benefit from proactive care,” explained Aashima Gupta, director of Global Healthcare Solutions for Google Cloud, in a blog post. “These AI systems are part of a pipeline that ultimately alerts clinicians about ‘patient engagement opportunities’ and provides contextual information to guide personalized outreach. The pipeline is being continuously iterated on by a cross-functional team of Virtans including applied AI, clinical operations, design, and product.”

Before migrating to Google Cloud, Virta was using a hosted HIPAA-compliant container-as-a-service provider built on top of AWS’ EC2 hardware, the company told Datanami. Now, Virta uses Google Kubernetes Engine (GKE) to provision its training workloads, including six types of GPU nodes. Nasir Bhanpuri, senior machine learning engineer for Virta, explained that these node pools scale based on the workload and that training data are read directly from BigQuery or loaded from files hosted by Google Cloud Storage (GCS). Gupta said that this switch was motivated by greater flexibility and control, more detailed permissioning and useful managed tools like BigQuery and Google Cloud’s Healthcare API.

Virta providers engage both reactively and proactively with patients. Reactive engagement is in response to patient requests; proactive engagement, meanwhile, involves Virta reaching out after noticing data problems, patient achievements, or data thresholds that would warrant a change in medication. “Virta can flag which patients are most likely to leave the treatment, and alert providers to intervene early and keep patients engaged,” Bhanpuri explained. “They also use ML for estimating a patient’s HbA1c (a measure of blood sugar) to determine current progress toward diabetes reversal.”

Bhanpuri elaborated that last year, Virta honed in on these proactive “health coach” alerts, aiming to increase engagement. This included reducing the number of unuseful alerts, adding new types of alerts based on identified gaps, and modifying the AI-enabled prioritization algorithm.

Specifically, health coach alert prioritization was tailored to give higher priority to more time-sensitive engagement opportunities; patients who had not received a coach message recently; and patients who began treatment more recently. Bhanpuri said that these shifts in the algorithm increased health coach-driven action by about 20%, leading to greater patient success as measured by metrics like medication reduction and glycemic control.

“We’re excited to be such a large part of helping Virta achieve its goal of reserving Type 2 diabetes in millions of patients,” Gupta said. “In 2022, the company anticipates triple-digit year-over-year customer and patient growth. Increasing patient volume will require Virta to lean even more heavily into ML and AI to keep patients safe, engaged, and successful on the diabetes reversal treatment, while continuing to streamline workflows for clinicians and health coaches.”

Related Items

Why AI Has Yet to Deliver on Promises in Healthcare

Putting Data to Work to Deliver the Best Care Everywhere

What Will AI’s Biggest Contribution to Healthcare Be?