Providence uses Azure OpenAI Service to decrease clinician burnout and expedite patient care

A flood of electronic communications

Electronic health messaging has changed the landscape for patients and caregivers. Taking advantage of the flexibility and familiarity of communications available 24/7 using mobile apps and/or email, patients no longer need to wait for office hours to check in with their healthcare providers. 

Electronic messaging has become the most common way for patients to stay in touch with their caregivers. But this popularity is a mixed blessing. While convenient for both patient and provider, electronic messages require providers to spend time triaging and classifying the influx. Overflowing inboxes can bog down caregivers, slowing down their availability to deliver care directly to patients. 

Providence is a healthcare organization employing 117,000 caregivers who serve in 51 hospitals and 1,085 clinics. It provides a comprehensive range of health and social services across Alaska, California, Montana, New Mexico, Oregon, Texas, and Washington State. Providence is committed to providing high-quality, compassionate healthcare for everyone—regardless of coverage or ability to pay. 

Medical assistants and other caregivers at Providence had found themselves overwhelmed by the volume of electronic messages they were receiving daily. The number of patient messages had increased threefold since the start of the COVID-19 pandemic. In December of 2022, electronic communications surpassed phone calls as the primary means of communication between patients and providers. Simultaneously, due to inflation, an increased reliance on agency nurses and medical assistants, and a shortage of qualified medical assistants, the cost of managing these messages rose. Some U.S. medical centers began charging patients for electronic messages as a way of offsetting this expense. Providence’s mission for equity drove it to evaluate alternative, innovative solutions.

Providence committed to exploring all possibilities for managing these important messages from patients and modernizing its approach. In the past, caregivers considered messages in their order of arrival. As many of the incoming messages look alike, it was essential to separate those that did not require a doctor (such as questions about the opening hours of the clinic) from those that urgently needed a doctor’s attention. “Our goal was to take care of all messages promptly and to enable Providence caregivers to work ‘top of license,’ delivering the patient care they’re uniquely qualified to provide,” explains Ford Parsons, MD, Chief Operational Analytics at Providence. 

Taming the influx

It was important to find a solution that meshed with the way Providence caregivers work, rather than creating a new set of tools that would require separate access and training. As third-party solutions did not fulfill that requirement, a team of Providence clinicians, informaticists, and AI specialists developed a new product, ProvARIA, based on Azure OpenAI Service. The team was able to take advantage of industry-leading AI technology, and a cloud-scale data and app platform of Azure to deliver adaptive, responsive, and personalized experiences by building and modernizing intelligent applications. 

ProvARIA categorizes the inbox messages by content and urgency so the right type of caregiver can handle that message. For example, the solution prioritizes responding to an email from a patient experiencing severe or concerning symptoms ahead of a query about clinic hours. It uses an advanced natural language processing (NLP) engine to sort the messages and a customized user interface (UI) within the electronic health record (EHR) system to streamline the triage of the categorized messages. The NLP engine is based on OpenAI’s GPT architecture and runs in Providence’s Microsoft Azure cloud. Azure OpenAI Service is accessible and unified with the workflow.

The Providence team recognized that the use of large language models (LLM) such as OpenAI’s present both opportunities and challenges in the healthcare setting. ProvARIA uses LLM as a document classifier, which lends itself to a rules-based verification process and minimizes the risks present in other applications of LLMs (i.e., document generation, document summarization). They believe this approach—AI with the guardrails of rules—represents a responsible use of AI in healthcare.

Context-specific workflow recommendations manage each message more efficiently, as well as provide documentation for best practices for message triage right in the workflow. For example, ProvARIA categorizes the message, enabling it to be prioritized and/or delegated to specialized caregiver groups, such as delegating all ‘refill request’ messages to a pharmacy technologist. The new AI-enabled process replaces the previous practice of printing out document triage flowcharts and posting them in caregiver workspaces.

Parsons says, “A key differentiating feature of ProvARIA is its deep integration of the AI-generated content into the clinical workflow.” The solution configures messages to display context-specific ‘quick action’ buttons, shortcuts to common workflows. For example, a one-click button to refer a patient to physical therapy is available for all ‘back pain’ messages. Category-specific knowledge base articles are presented directly below the message. This level of integration requires a diverse set of skills—data science, engineering, and clinical informatics. Azure OpenAI Service simplifies and modernizes these administrative workflows, freeing up clinicians’ time for patient-facing work.

Making a difference to patients and caregivers

Providence piloted ProvARIA with four separate clinics representing 27 different doctors and nurse practitioners. All their electronic communications funneled into a centralized inbox in a single office, where a group of medical assistants worked together to address the messages. This group of medical assistants processed about 10,000 messages a month, resulting in a 35 percent improvement in turnaround time. With such dramatic results, other groups within Providence enthusiastically joined the pilot. The new system introduced efficiencies that allowed the clinics to manage messages with fewer medical assistants, a boon when the limited number of qualified medical assistants is in high demand. Providence medical assistants, with support from ProvARIA, are now processing 5,000 messages per day, covering 145 Providence clinics and 650 providers.  

“The most profound outcome is one we cannot measure and hadn’t anticipated: our caregivers have gained peace of mind from the new messaging solution,” says Parsons. “There's a certain amount of fear that you’ve only gotten to number 300 of the 1,000 emails in your inbox. What if message 350 is someone having a heart attack and asking for immediate help—and you didn’t get to that patient because you were too busy responding to queries about the clinic hours? The new system means that all our caregivers are helping patients who need it most. Our process is no longer getting in the way of good patient care.”

“Best of all, patients, especially those with concerning symptoms, are experiencing faster turnaround times. That means more same-day consultations and/or appointments for urgent issues,” says Parsons. “ProvARIA empowers caregivers to respond more quickly and accurately and frees up time for them to do what they do best—provide personalized, face-to-face care to patients in need.” 

Find out more about Providence on Twitter, Facebook, and LinkedIn.