ResMed Uses Amazon SageMaker to Personalize Sleep Therapy for Millions of Patients

ResMed provides continuous positive airway pressure devices and masks for people with sleep apnea, chronic obstructive pulmonary disease, and other sleep disorders. This cloud-connected equipment collects data on patients’ sleep patterns and shares it with patients through ResMed’s myAir patient engagement application. Then, myAir’s Smart Coaching feature uses AI/ML to launch customized recommendations to each patient to improve their outcomes.

In 2021, ResMed didn’t have the automated, unified AI/ML self-service solution to securely run inferences through the very large volumes of patient sleep data required to meet its 2025 goal. The first version of IHS was built alongside Manifold, an AWS Partner, with which ResMed had a strong track record of joint innovation. Although successful as a proof of concept, the container-based framework was developed by data scientists who each used different tools, which forced them to take responsibility for that infrastructure in perpetuity. “Leaving it to an individual developer to build their own toolbox isn’t scalable, nor will it lead to the rigorous quality we want in an end product,” says Badri Raghavan, ResMed’s vice president for AI and ML.

ResMed chose Amazon SageMaker to build a centralized, standardized AI/ML solution because it scaled globally and connected well with solutions the company was already using for data storage. In 2018, ResMed had built a data lake on AWS that was compliant with regional data regulations. Amazon SageMaker connects seamlessly with this data lake through AWS Glue, a serverless data integration service that makes it simple to discover, prepare, and combine data for analytics, ML, and application development.

Alongside Manifold, ResMed began building a second version of IHS, its next-generation ML solution, in early 2022. For guidance, the team took part in AWS Data Lab, which offers accelerated, joint engineering engagements between customers and AWS technical resources to create tangible deliverables that accelerate data, analytics, AI/ML, and application modernization initiatives. “The AWS Data Lab was great,” says Philomena Lamoureaux, senior manager of ML and AI at ResMed. “We had the time blocked out for our developers to focus only on the development and the education for this proof of concept.” After the AWS Data Lab, Amazon SageMaker adoption at ResMed more than doubled in 3 months. The prototype solution rolled out in April 2022, just 2 months after ResMed worked alongside the AWS Data Lab team, and the foundational AI/ML capabilities of the IHS solution on Amazon SageMaker were deployed within 6 months.

ResMed’s AI/ML solution uses Amazon SageMaker Processing to run preprocessing, postprocessing, and model evaluation workloads on fully managed infrastructure. ResMed takes advantage of many Amazon SageMaker features to train models and pipelines and to choose deployment types, including near-real-time and batch inferences. (See Figure 1 for more details on ResMed’s solutions architecture.) These ML models deliver near-real-time predictions to the myAir application that then tailors and delivers content to myAir users. Each ML model creates up to 2 million predictions per day. In addition to in-app notifications, myAir also sends personalized email campaigns to customers using Amazon Pinpoint, a flexible and scalable outbound and inbound marketing communications service.

“Previously, all myAir users would receive similar messages from the app,” says Urvashi Tyagi, chief technology officer at ResMed. “IHS has facilitated personalized interactions with patients through myAir based on which ResMed device they use, their waking hours, and additional contextual data.” Now, over 18.5 million patients enjoy tailored content and a personalized experience. “Our team can make sure patients get the benefit of all the data we have,” says Prakhar Shukla, director of data engineering at ResMed.

ResMed data scientists now have more time and flexibility. “The deployment, serving, and monitoring are streamlined and automated as much as possible so that data scientists can create a model without being tied to the infrastructure they build,” says Lamoureaux. “They can move on and have the space to be creative.” Using Amazon SageMaker, ResMed data scientists accelerate time to market by deploying ML models in days or weeks compared with months previously and by cutting time for AI/ML pipeline processing by several hours.

ResMed used Amazon SageMaker to rapidly build the AI/ML IHS solution that supports personalizing sleep therapy for over 18.5 million patients worldwide. “Prior to adopting Amazon SageMaker, all myAir users would receive the same messages from the app at the same time, regardless of their condition,” says Raghavan. “Amazon SageMaker has helped facilitate more personalized therapy for ResMed users. We took advantage of Amazon SageMaker features to train model pipelines and to choose deployment types, including near-real-time and batch inferences to deliver tailored content to myAir users.” In addition, says Raghavan, “Amazon SageMaker has helped us to achieve our key goal of embedding ML capabilities across our global organization by deploying ML models in days or weeks compared with months.”