How AI/NLP transformed healthcare faxes into actionable insights – for faster treatment
In 2020, as America’s hospitals became overrun by patients presenting with COVID-19 symptoms, providers and labs began sending notifications of positive cases to their local health departments (LHDs). The notices provided these government agencies with valuable information on where to mobilize resources to slow potential outbreaks.
But in addition to the sheer number of COVID-positive notices they were receiving, LHDs faced another challenge. Many providers were transmitting their notifications by fax. Hardcopy records piled up on LHDs’ fax machines. Public health officials needed to manually sift through and read each fax document to triage the most serious cases.
Dr. Umair Shah, executive director of Houston’s Harris County Public Health Department, put it this way: “Picture the image of hundreds of faxes coming through, and the machine just shooting out paper.”
This is a familiar story in healthcare: fax pages everywhere, containing unstructured data that needs to be reviewed and manually re-entered into another system before the care team can take their next steps to help the patient. But fortunately for several health departments dealing with COVID faxes in 2020, there was a positive twist.
Adding artificial intelligence to faxes.
Researchers at the Stanford University School of Medicine created an artificial intelligence (AI) program to help public health officials respond to potential COVID outbreaks more quickly and reliably. How? By enabling technology to identify and flag the acute notifications embedded among the thousands of fax pages scattered all over their office floors.
Here’s how it worked. The researchers created a series of AI-powered algorithms to review the standard COVID-positive forms (called confidential morbidity reports, or CMRs) that providers were faxing to their local health departments.
First, the program converted each incoming fax to digital format, as a PDF. From there, the AI algorithms were able to perform several key steps automatically, including:
- Read the fax; detect handwriting and translate it into words and digits.
- Detect and segregate all fax pages pertaining to COVID reporting.
- Review the details of these faxes for key elements indicating the level of acuteness.
- Prioritize faxes containing references to emergencies and urgent issues.
- Send a priority notification to health officials for the most serious cases.
Stanford’s test shows AI can improve healthcare faxing efficiency and outcomes.
The test was a success. In 2021, Stanford Medicine published a report summarizing how their AI algorithms helped several local health departments more quickly identify and act on the most urgent COVID faxes. Here are a couple of the report’s key findings:
In one test pilot, the AI program was able to successfully identify and prioritize 49 of the 59 high-urgency faxes that it reviewed.
The program didn’t catch every high-priority fax, but it did identify 83% of them. In doing so, the Stanford’s AI code helped to save public health officials significant time and effort sifting through those fax reports manually.
One health department estimated the AI tool helped them respond to acute cases 20 hours sooner than if they reviewed each fax manually.
In their pilot program, the Stanford researchers activated the fax AI program just before a long holiday weekend. When those public health officials returned to their offices, they found more than 400 COVID-related CMRs on their fax machine.
But the officials noted that the AI program had already identified highly serious cases among those faxes. They also concluded that if they had been using their legacy process—”The machine just shooting out paper,” as Dr. Shah put it—the team would have been delayed in responding to those urgent cases by roughly 20 hours.
But the Stanford artificial-intelligence algorithm was limited by design.
Stanford Medicine’s real-world test run of a fax AI tool provides an illuminating example of what artificial intelligence can do to make healthcare faxing more efficient and effective. But it’s important to keep in mind that this was a limited solution—In fact, Stanford’s researchers themselves noted that there were several limitations inherent in the test case. Here’s one major example noted in their report:
“Our system benefits from the fortunate fact that, for the forms used, the information used to determine the acuity of a case is contained in structured fields (checkboxes). Forms that captured this same information in free-text fields instead of structured fields would pose a much more challenging problem. The ability to extract free-text information on these forms could enable further efficiency, especially related to saving time for data entry, and is a potential avenue for future work.”
Stanford’s COVID test case demonstrates that moving unstructured documents to structured documents can add efficiencies—and even lead to better outcomes—in one specific workflow. But clearly, artificial intelligence doesn’t need to be limited to scanning checkboxes in a specific type of fax report. Learn more about building your own efficiencies.