Using Technology to Improve Matching Drug Details in Hospice Care

By Matt Phillion

For Wise Hospice Options, prescribers had been spending an average of 15 seconds matching each flagged drug and 20 seconds per drug entering missing sig, or label, details. Recently, they incorporated clinical-grade AI from DrFirst into automating this process, enabling those same providers to spend two to three seconds per drug, with fewer medications flagged for manual review by pill icons.

The improvement in numbers was better than anticipated: originally, Wise expected AI to codify about 92% of drugs, 80% of sigs, and 95% of allergies. After implementation, they found that AI was able to codify 99% of drugs, 85% of sigs, and 96% of allergies.

The AI implementation allowed Wise to standardize data from different systems into a workflow that allows clinicians to see complete information before ordering prescriptions, helping avoid delays and reduce errors. Clinicians continue to review medication information for accuracy and adjust based on discussions with the patient but require fewer clicks and keystrokes so they can make faster, more informed decisions.

“With meaningful use and interoperability standards in the 2010s, we started to see a lot of interfaces—we interface with 15 different EHRs right now,” says Brett Faubion, director of operations and finance with Wise Hospice Options. “Some have come and gone, some have been acquired, some have multiple products we interface with, and everyone seems to have a different data structure and different specifications for medications.”

EHRs deal with a lot of different data and engage with different areas of expertise, Faubion explains.

“We primarily deal with medications, so we can focus on that. EHRs don’t really have the capacity or resources to get as detailed as we’d like on medications,” he says.

Managing the various EHRs, data structures, and then editing them to fit into one model is difficult, Faubion explains.

“Some EHRs send over drug strength and formula with the drug name, some split it, some combine different fields. The last 10 years, we have tried to get champions at the EHR organizations to focus on and build out expanded details we get for hospice patients,” he says. “We’ve seen varying degrees of success. How can we improve control with the data we’re getting? AI became a very clear opportunity. Being able to leverage AI to basically interpret and codify medications, and allergies in particular, as well as sigs has been a huge benefit.”

What his organization has seen is a move from having every e-prescription auto-populated and then needing to be adjusted or tweaked manually to more than 80% being complete with all fields filled out.

“The clients we’ve onboarded have seen a night and day difference,” says Faubion. There will always be some issues we find with real life data or scenarios that the AI hasn’t been trained on before that we need to address, and our partner, DrFirst, has been great at fixing and monitoring for that.”

It isn’t a 100% reliance on technology, he notes, with the original prescription entered into the EHR for reference. What the AI does is save time manually editing and fixing the prescription, cutting back on hours of clicking and typing so the clinician can review.

The human element for verification is important, Faubion notes.

“Since every system is different, some systems, for example, might not provide codified allergies. That’s a big red flag, when you’re not getting codified allergies and then can’t check if the patient is allergic to the med,” he says. “Or if penicillin is spelled incorrectly and the system doesn’t recognize it, the automated check doesn’t exist to prevent patient harm. So being able to process misspellings, typos on complicated words, allergies and medications, and codifying those to make sure the check is accurate is a really big benefit to organizations.”

Why the impact on hospice is key

The first thing to consider, Faubion explains, is that when it comes to hospice patients, they are going to be on a lot of medications.

“You’re talking about 10-14, for most, somewhere in there, and some are hospital covered, and some are not,” he says. “The hospital may not be prescribing all 14. Let’s say you have seven of those on different drug databases, and they don’t match perfectly. You’re then having to edit the sig to make sure they match the EHR. It becomes a really cumbersome process.”

While taking 15-20 seconds to reconcile a medication may not sound like a lot of time, multiply it by seven medications per patient, 15 patients under your care, and it can lead to a lot of frustration and a lot of time spent, Faubion explains.

“It’s not a good provider experience,” he says. “A lot of standards that are used in other industries aren’t widely used in healthcare, and that’s why we have 15 different data structures. Providing a funnel that provides a similar user experience for all providers is hugely helpful.”

Another issue is alert fatigue. Already an issue in other parts of healthcare, working with hospice patients on so many medications is going to trigger a lot of alerts.

“And if that allergy is not codified, you may not see that alert. We’re trying to remove another barrier so providers can rely on that data,” says Faubion.

There’s always a fear that introducing AI into a process will cost human jobs, but Faubion says this is a case where the AI is doing work humans can’t.

“I don’t think this is a job we’d ever hire people to do,” he says. “Just looking at interface messages all day. There’s also some legal liability if someone mis-transcribes something. But we also want multiple checks in place. You don’t want to rely on a person or on AI too much. So, it’s not replacing prescribers, it’s not telling them what to prescribe.”

It’s converting what’s in one system to match another, and enhancing the provider experience rather than replacing them, allowing providers to focus on dosing and other important issues.

“In this specific instance I get the fear of replacement, but it’s not reducing anyone’s job or their opportunity for work. It’s helping those who are using the system,” he says.

Reaction to the technology has been interesting to see so far, Faubion says.

“Our first few clients have been e-prescribing with us for a while and were aware of data insufficiencies and the need to go in and reconcile meds, fixing the sigs, having to edit a lot of these medications before prescribing,” he says. “They’ve noted that it’s a better system than what they had before. There’s still plenty of situations where the AI is not going to be able to do what it needs to, such as tapering dosages or complex sigs, or times when compounding medication will require manual entry because they’re too complex to standardize. But we’ve seen that it cuts out a lot of issues they’ve traditionally experienced.”

The numbers being higher than expected is a pleasant surprise, Faubion says.

“We used a lot of historical data and thought our numbers would be accurate, but to implement it and see those numbers even higher was fantastic,” he says.

Next steps

Wise has been taking a metered approach to rollout to ensure that the AI can manage what is asked of it.

“We don’t want to roll this out to everyone at once and find out five EHRs handle things differently in a way that the AI isn’t trained for,” Faubion explains. “One of the benefits of a metered implementation is we can build out a plan to support it better and offer a more personalized approach.”

There are a lot of opportunities with this kind of clinical-grade AI, Faubion says.

“There’s ways it can fill the gaps in structures between systems that could expand beyond hospice,” he says.

Those opportunities continue to be ways to enhance the work prescribers do, he explains.

“By no means is this taking the job of the prescriber away. If you can provide the same level of care with a cheaper medication that is as beneficial, for example, by including pricing data, for example, there are a lot of opportunities in the future,” Faubion says. “It helps balance high quality with low cost. We want to provide a high level of security and safety for providers. That’s the biggest thing.”

Matt Phillion is a freelance writer covering healthcare, cybersecurity, and more. He can be reached at matthew.phillion@gmail.com.