AI, Healthcare, and C-3PO

In this special guest feature, David Sellars, CEng, Principal, Product Innovations at DrFirst, discusses how artificial intelligence has been lauded as the coming revolution in healthcare. David has over 20 years of healthcare experience with a deep emphasis in Big Data, AI, healthcare interoperability, programming, and systems optimization. He has served as Principal, Product Innovations, for DrFirst since 2010, leading R&D for the Product Team. His experience in healthcare ranges across all five of the top EHR’s, including implementing clinical modules and large multi-hospital migrations.

Artificial intelligence has been lauded as the coming revolution in healthcare, and a quick web search reveals that top minds in academia, tech, and industry are actively welcoming and discussing its potential. Yet when you move from searching the web to talking to clinicians, you may get a different impression. If you ask a clinician in your local hospital or pharmacy if they use AI in their regular workflows, you are likely to find that AI is curiously absent.

While most people think of “caregiving” as a hands-on endeavor, those hands are often on a keyboard doing data entry and data synchronization for records that healthcare providers are sharing with each other. Of course, relieving people from tedious, repetitive, time-consuming tasks is one area where AI can shine, primarily by addressing key concerns such as professional burnout and providing more time for direct patient care, all while saving millions of dollars a day for the U.S. healthcare system.

Let’s do the math:

• Family medicine physicians spend 157 minutes per day on clerical and administrative tasks such as documentation, order entry, billing, and coding¹  
• At $100 per hour on average², this means doctors are paid approximately $262/day to enter data into the electronic health record (EHR)³ instead of talking to patients about their symptoms and treatment plans
• With roughly 66,000 family medicine physicians practicing in primary care offices in the U.S.², that adds up to over 172,000 hours and $16.8 million per day that these highly trained professionals are spending in front of a keyboard instead of face-to-face with a patient

But it gets worse. The numbers above don’t account for nurses, pharmacists, physician assistants, and other clinicians, who also burn countless hours on data entry, leading to frustration and burnout.

Not surprisingly, most clinicians detest this aspect of their jobs.³ But how does this affect the patient? Say you develop a kidney stone. At each stop in your healthcare journey, someone asks you to recite what medications you take. The first time may be at your initial doctor’s appointment for a diagnosis. Then once more if you see a specialist. And yet again if you then go to the hospital for treatment. Each time, a clinician re-records data you have already given to the previous provider. While this information is critical to help avoid drug interactions and reduce hospital readmissions, it can be frustrating for patients to repeat the same information over and over while clinicians re-enter it. And it means healthcare providers are spending time entering data that already exists elsewhere, which takes time away from “hands-on” caregiving and introduces the risk of keyboard errors that can lead to medication errors.

Capturing data in the EHR used to be the most daunting industry challenge. However, over the past 20 years, what used to be a digital information drought has swelled to a flood of hand-entered, electronic data. With so much uncodified data in today’s EHRs, the problem has become that clinicians can’t make use of it all.

It gets even worse when this data is shared between providers caring for the same patient. Clinics, hospitals, and pharmacies can often share data, but because their respective systems don’t use the same terminology, this data is unusable when it gets to the receiving organization. For computers, it’s like speaking Portuguese to someone who only speaks Hindi (of course, AI can help with this, too). For the receiving clinician, it’s more work to translate and transcribe outside data than it is to re-gather and manually enter it. 

For example, if you are prescribed an antibiotic for kidney stones, the medication instruction could be entered as “1QD” by the doctor, from the Latin for “one a day.” This order gets sent electronically to the pharmacy, where the pharmacist has the job of interpreting the clinical intent of “1QD” and then re-entering it into the pharmacy system using their own short codes, “tk 1 t po qd.” The pharmacy’s short codes ultimately result in a patient-friendly instruction, such as “take 1 tablet by mouth daily” printed on the bottle. Although each person is speaking the same language, the abbreviations and codes used have programmatical meaning only to the computer systems that use them daily, resulting in a complex communication problem.

And it’s made more complex by the mind-boggling number of permutations for a single, simple prescription instruction, such as “take 1 tablet by mouth once daily,” which has 832 permutations commonly used in the wild, as noted by researchers at the University of Michigan. Their study revealed that pharmacy staff needs to manually adjust more than 80% of prescription instructions, which impinges on time with patients and can introduce keystroke errors.

Organizations such as HL7 International and the National Council for Prescription Drug Programs (NCPDP) have dedicated years to establishing standards (such as FHIR, Fast Healthcare Interoperability Resources) for exchanging data. However, because these standards continuously evolve, the healthcare industry, as a whole, is never fully aligned on a single standard. This is further complicated because these standards aren’t directly used in the average clinical workflow. For example, a healthcare provider wouldn’t use the standard national drug code to search for a medication to prescribe, just as a shopper wouldn’t typically use a UPC code to search for a kitchen gadget to buy.

Companies making headlines for AI are often promising more from the technology than it can provide, leading to missed opportunities for healthcare to benefit from more practical and focused uses of AI that can make small but meaningful contributions to improving healthcare workflow. As a result, highly educated clinicians are still doing work that others could do, such as transcription and data entry.

In the U.S., because peoples’ healthcare data is distributed across different hospitals and physician practices, which often use different EHRs with different preferred terminology, there is no single source of truth for all of a patient’s information. For example, each time you have a new healthcare visit, your records may be requested and assembled by your clinician. But the data sent likely uses different terminology than the receiving EHR. So, to use the information in a meaningful way, beyond simply storing it, someone has to read it and re-input it manually into the EHR, using the local terminology required for it to perform safety checks and other actions.

What if AI could interpret these medical records and codify the data into the receiving doctor’s EHR terminology, regardless of the two or three external systems sending the information?

Here is where we turn to C-3PO, the Star Wars translator droid that is “fluent in over six million forms of communication.” One significant future use of AI is a healthcare IT version of C-3PO that can receive medical jargon from many sources and provide a meaningful output to each unique receiver in real time. 

Some innovative pharmacies are using AI to do this work today. Galva Pharmacy, serving a small town in Illinois, was the first pharmacy to leverage AI directly in the pharmacist workflow. Early results show their overall data entry time has decreased by more than 45% when AI technology is used versus not. This is an example of what’s to come for clinicians in every aspect of healthcare. Using AI for ordinary clinical tasks can reduce human-induced errors, decrease the mundane work, and allow highly trained medical experts to spend more of their time doing skilled medical work. 

1. ^{https://www.annfammed.org/content/15/5/419.full}
2. ^{https://www.bls.gov/oes/current/oes291215.htm}
3. ^{Stanford Medicine Harris Poll https://med.stanford.edu/content/dam/sm/ehr/documents/EHR-Poll-Presentation.pdf}

Sign up for the free insideBIGDATA newsletter.

Join us on Twitter: @InsideBigData1 – https://twitter.com/InsideBigData1