Wise Hospice Options uses AI to reduce e-prescribe time from 20 seconds to 2



As a health IT company, Wise Hospice Options integrates with many different systems, particularly electronic health records. The company has integrated with more than 15 EHRs throughout the 21 years it has operated. The only constant it has found is that each system has its own standard and way of doing things.

In hospice care, Health Level 7 and Fast Healthcare Interoperability Resources specification standards are not tightly followed, and fields may be used in different ways by the different EHRs. When the company started implementing e-prescribing in its system in 2013, these differences became apparent very quickly.

Each system sends different levels of detail with identifiers from different drug compendia. The lack of universal standardization left the Wise Hospice system and IT team unable to completely fill the gaps that interoperability promised it would fill.

Improving accuracy, user experience of integrations

Over the years, Wise staff have tried many different solutions to improve the accuracy and user experience of the integrations. They have worked with various EHRs to improve the data they receive; however, the EHR companies have their own IT priorities and capacities, so Wise staff often were unable to get the full data needed for complete resolution of the data gaps.

This led to unrecognized drugs and instructions (sigs) that the e-prescribing system could not parse or handle. Even with NCPDP working toward a Structured and Codified Sig Standard, adoption by EHRs would be difficult as medications are not their sole focus. The NIH has RxNorm APIs available to assist with drug database standardization, but the adoption in the hospice space has been slow and is not regularly used.

Out of the 15 EHR systems Wise has integrated with, only one was willing and able to work with the vendor to send the detailed and specific data for a truly functional medication integration that supports near-seamless e-prescribing – unfortunately, this lone system was acquired and is now being sunset, said Brett Faubion, director of operations and finance at Wise Hospice Options.

“With the available standards not being adopted and the difficulty in getting a multitude of organizations together to work toward a medication message standard, we had to look at what options were in our control to improve the experience of our users and fulfill the promise of interoperability: reducing duplicate entry and the risks associated,” he explained.

“One method that other systems have taken is to simply accept and load whatever free-text entries are received from the EHR,” he continued. “While doing this streamlines the users’ e-prescribing flow, it comes with significant risks and reduced functionality. Free-text sig entries are not able to be checked for dosing in our e-prescribing tool since dosing and frequency are not in their own distinct fields.”

The only place risks might be mitigated

Allergy checks could also be compromised by abbreviations, misspellings or typos of the allergy (for example, Penicillin is often misspelled) and could directly lead to patient harm. Not every EHR offers these checks on the orders and medications entered into their system, making the e-prescribing system the only place these risks might be mitigated.

“We do not believe the simplified e-prescribing workflow justifies the increased risk to our users and the hospice patients and our clients resoundingly agreed,” Faubion said. “With reducing functionality not being an option, we needed to look for more innovative solutions.

“Our goals coming out of this situational review were to improve the user experience of e-prescribing in our system, improve the effectiveness of medication and allergy interfaces without requiring EHR-side updates, and to streamline the channels and number of contacts involved in the system,” he continued.

Too much energy was being wasted on 10 different fronts, and Wise staff needed to compile that energy into a single source as a solution, either as internal development or using a single-source external tool, he added. This led staff to look at AI technologies for the various types and structures of data Wise receives for medications, sigs and allergies. The company needed a tool that could codify medications and allergies, and parse free-text sigs into codified fields.

Integrated e-prescribing for hospices

Wise Hospice Options decided to work with health IT vendor DrFirst to offer integrated e-prescribing to hospices. Wise has worked together to improve the user experience and data flow, and minimize many issues over the years, but the data is only as good as what it gets from the data source, the EHR. As Wise pivoted from trying to collaborate with more than 10 different organizations to one, it discussed the issue with DrFirst staff.

“This is when their clinical-grade AI tool first came up,” Faubion noted. “This tool was built to codify medications from medication names, strengths and forms, codify allergies from free-text entries, and parse sigs into individual fields used by the DrFirst e-prescribing system. This seemed like a perfect fit for our needs with a trusted partner that demanded less development time from our IT team.

“After seeing a demonstration of the AI and the data backing its results, we discussed prior use cases, reviewed test data and finalized the data flow,” he continued. “This would be the first application of the clinical-grade AI tool in a real-time setting, processing the data as we receive from the EHRs. To accommodate the differences in each EHR’s data formatting, the AI model would need to be segmented and trained separately for each integration.”

While the coding infrastructure to be implemented would be standardized and toggled on or off for each account, handling each integration uniquely allowed for greater accuracy and better results. This decision did increase the work for both the Wise and DrFirst teams, but they determined it was worth the improved performance for clients.

A major improvement for users

“The goal of simplifying the collaboration needed for improvement was met, and while an AI tool will not provide 100% conversions, an 80% conversion rate for sigs and even higher for medications and allergies was a major improvement for our users,” Faubion reported. “Even if the EHRs enhanced their medication interfaces, there would still be a portion of medications that would be complex and not transfer well, such as compounds or complex sigs with multi-part instructions.

“While using an AI tool does not fully eliminate duplicate entry or the risk of mistranslation, it significantly reduced the amount of manual entry or correction needed,” he continued. “E-prescriptions would still need to be reviewed for accuracy, but the majority would no longer need to be adjusted or edited. This improvement occurs completely behind the scenes and does not require any manual activation or intervention by a user.”

Wise Hospice Options worked with the DrFirst team to adapt their AI tool to the majority of EHRs Wise has integrated with and is continuing to expand the supported systems. Expanding service to an integration involves analyzing the data received, processing a significant load of batch data, and reviewing the results for potential pitfalls, improvements and unique patterns for a client.

“For example, some EHRs split the strength of a medication from the name while others lump them together,” Faubion explained. “Some of the differences occur at the client level, such as one client selecting multiple routes for a medication. Treating each structure differently allows the AI to be even more accurate for each client and system. The downside is that initially, we have had to limit our onboarding of this tool to an EHR basis.

“We open the offering to each applicable client as we support a specific EHR with this tool,” he continued. “The additional accuracy is absolutely worth the slower rollout time. This also allows us to take a more individualized approach and identify the gaps the AI might have for a client’s specific data or ordering practices that may not be currently supported by the AI tool. We can then work directly with the client to find a resolution and ensure all parties are satisfied with the data flow.”

No additional user entry or intervention

The way Wise built the process and data flow allows for the AI to be used without any additional user entry or intervention. It is a tool in the background that enhances the data received from the EHR to make it more compatible with the e-prescribing system.

“This changes the workflow from a ‘fix then prescribe,’ with various clicks and potentially multiple screens to manually match a medication, to ‘review then prescribe,'” Faubion said. “Users still need to pay attention to what is being prescribed and to what alerts may occur, such as an allergy interaction or overdosing.”

In Wise Hospice Options’ initial pre-live test batches of medication and allergy data, it saw the AI codify 92% of medications, fully parse 80% of sigs and codify 95% of allergies. The company did not expect 100% for any of these categories as there are instances that are complex or custom and need specific attention. For medications, there are compounds that will not easily translate or items that might be entered as a medication that do not have a codified identifier, like oxygen.

“Complex sigs exist that the e-prescribing system is not designed to handle by default, such as non-standard frequencies,” he explained. “Allergies can’t always be codified and may not be relevant to prescribing medications, such as ‘dust’ or ‘seasonal allergies.’ Given these nuances, we were excited about the potential performance.

“Once we implemented the DrFirst AI with several accounts, we saw even better performance metrics than expected,” he continued. “The AI tool has codified 99% of medications, 85% of sigs and 96% of allergies. The almost complete prevention of unmatched or invalid medications has been a great result and major improvement to the user experience.”

Workflow times add up

Before, any medication that was received with old or outdated NDCs, drug identifiers from a compendium Wise does not use, or OTC medications that may not be in the drug database yet, would require a user to find and manually select the correct medication in the e-prescribing system to ensure the correct medication was prescribed. This workflow would take roughly 15 seconds for an experienced user and several clicks to work through, which adds up quickly when each patient has an average of seven medications.

“Add the time needed to rewrite the sigs for each medication compared to simply reviewing the sig, and we reduced the time to e-prescribe from 20 seconds per e-prescription to two to three seconds,” Faubion reported. “We have been very happy with these results and the improvement this AI tool has made to our users’ experience, workflow and data accuracy.

“While we have seen very positive results, there were still some bumps and improvements that we’ve worked through since implementation,” he added. “Early during implementation, our clients reported a handful of AI errors and we quickly implemented an escalation system with our users and the DrFirst team to ensure these errors were corrected quickly and did not become reoccurring.”

Wise also has seen data variance at the hospice level as some organizations use manual entry or create their own drug entries compared with others that stick to a prepopulated list. The DrFirst team also has worked on improving the processing of the Reasons section of an e-prescription, as this was a field that was rarely used in prior use cases and its utilization varies significantly even among Wise’s client pool.

“This has led to many improvements, including handling multiple reasons being listed,” Faubion noted. “As we work together to improve data processing, we are excited to see the results continue to improve over time and as more data is fed through the AI.”

Follow Bill’s HIT coverage on LinkedIn: Bill Siwicki
Email him: bsiwicki@himss.org
Healthcare IT News is a HIMSS Media publication.

WATCH NOW: A Chief AI Officer must be chiefly accountable



Source link

Leave a Comment