×  

by Tim Rowan, Editor

 

Raise your hand if this is a familiar scene in your Home Health agency:

 

On a chaotic Friday afternoon, the referring physician's or hospital's EMR sends your intake department a patient's referral documents, which may include a "history & physical," discharge notes, medical record, and other documents. Your overworked, overwhelmed intake nurse looks with dismay at the up to 300 pages, knowing most of it is irrelevant but that critically important information is buried somewhere inside.

She suppresses a scream.

Somehow, she needs to find the 40 or 50 words she needs to get a glimpse of an elderly person's primary and secondary diagnoses, ten or twelve relevant co-morbidities, and most recent procedures or treatments.

The intake nurse takes a deep breath and tries to decide: Do I spend the next 90 minutes reading the whole thing? Or do I skim through it, hoping I catch those 40 or 50 words without missing any critical information? "Darn those hospital EMRs," she whispers under her breath. "They store everyone's entire life story, including page after page of data irrelevant to us, and they make it easy to send the whole thing to us rather than just what we need."

Sound familiar? It gets worse. Upon closer examination, the intake nurse's job is even more challenging. Referral documents arrive in various formats. There is no standardization. Intake departments experience variation in number of pages, often in "free-text" narratives rather than structured data, and every unique EMR delivers it in its own format. Slow, human reading of these documents is a guarantee of error.

 

New AI Tool to the Rescue

Before long, that intake nurse, along with in-house coders and OASIS nurses, will be freed from the burden of overwhelming referral documentation; freed from both the inaccuracy of rushed scanning and the inefficiency of time-consuming, careful reading. In fact, for some professional coders, those Tolstoy-length narratives that physician and hospital EMRs produce are already a thing of the past, as of just a few months ago.

Home Health coders working for outsource services firm Select Data can now read a one- or two-page summary of each patient's referral documents instead of having to plod through all 100, 200, or 300 pages. In spite of the efficiency gained, coding accuracy continues to increase among this group of expert coders because the application from their revolutionary artificial intelligence system, called "SmartCareAI™," is no mere summary.

SmartCareAI™ reads those lengthy referral documents in seconds. It can process text documents, PDFs, and images of documents. Identification of diagnosis and co-morbidities are highlighted in the report, regardless of how many different terms or abbreviations a doctor might use to name the same condition.¹

Even more remarkable, when the referring physician mentions a condition or complaint not present, SmartCareAI™ is able to read the surrounding text and know what the physician meant when a sentence includes phrases such as "did not" or "does not." It highlights those clinical terms in a different color for easy identification.

"This is not just NLP,"² Select Data CEO Ed Buckley emphatically told us. "This is artificial intelligence. The more data we feed it, the smarter it becomes. NLP recognizes words, AI models learn to understand meaning. That is why our coders rarely miss an important sentence, even when it is buried deep into a document, and why they are less likely to mistake conditions and complaints not present, should a physician mention them, with diagnoses and conditions the patient is experiencing."

 

CLINICAL DOCUMENTATION IMPROVEMENT

Select Data's team of data scientists did not stop there but built the new tool on the shoulders of years of coding methodology that has always exceeded Home Health's minimum requirements. Buckley explained, "Our coders adhere to the tenants of Clinical Documentation Improvement (CDI) standards found in acute care settings. When you outsource your coding to Select Data, we go to extra lengths to ensure that your home health documentation is in-line with documentation from the patient's physician. This rigor readies you for the higher-bar standards coming from PDGM, HHVBP, and RCD."

Again with barely disguised passion, Buckley asserted that the new SmartCareAI™ Artificial Intelligence tools, combined with his company's longtime methods and expert team, help protect clients from audits, boost reimbursement and, most importantly, outcomes.

Augmented Home Health

Buckley calls the net result of the AI analysis the "Augmented Home Health Workforce." He explains it with an excitement that borders on passion. "Value-Based Purchasing will raise the quality bar higher than PDGM did. The balance between accurate documentation and productivity, coupled with the nurse shortage, threaten to cripple Home Health in more ways than just lowering reimbursement for most agencies. With VBP, the necessity will be stronger than ever for full understanding of referral documents, accurate coding and code sequencing, and accurate OASIS assessments with complete, bullet-proof documentation."

In the product demonstration we saw, the AI system did read hundreds of referral pages in seconds. The summary report it presented was intuitive and easily understood with minimal instruction. Most impressive, however, are the outcomes.

"SmartCareAI™has transformed our coders' experience," he said, "so just imagine if all coding services used it. I believe it would transform Home Health." He likened the AI summaries to "intelligent Cliff Notes," remembering that a lot of college students passed a lot of courses without ever reading Shakespeare or War & Peace.

Select Data coders and reviewers using SmartCareAI™ maintain a high level of accuracy. On average, coders and reviewers using SmartCareAI™ identify 16 percent more comorbidities than when using a manual review process. On average, they identify 12 OASIS inconsistencies per record and wind up recommending 10.7 codes per record. Home Health agencies served by Select Data coding services have improved their star ratings up to 15 percent and seen a up to 35 percent reduction in hospital readmissions.

 

---

¹ During our demo, the physician writing the sample document mentioned diabetes, diabetes mellitus, type II diabetes, DM, and DM2. The AI system recognized every variation as naming the same diagnosis.

 

² Natural language processing (NLP) refers to the branch of computer science — and more specifically, the branch of artificial intelligence or AI — concerned with giving computers the ability to understand text and spoken words in much the same way human beings can.

 

©2022 by Rowan Consulting Associates, Inc., Colorado Springs, CO. All rights reserved. This article originally appeared in Home Care Technology: The Rowan Report. homecaretechreport.com One copy may be printed for personal use; further reproduction by permission only. editor@homecaretechreport.com