In 1950, mathematician Alan Turing—perhaps best known for developing the technology that cracked the German Navy’s seemingly uncrackable Enigma code during WWII—published a paper titled “Computer Machinery and Intelligence.” His paper asked the question, “Can machines think?”
Today in healthcare, the need for automation, better quality, and shorter turnaround times throughout the revenue cycle management life cycle is leading to more and more interest within health information departments in investment in artificial intelligence (AI)—technology that does, to a certain extent, think.
When my department was determining what type of technology we needed to create efficiency around our scanning workflows, we set out to determine the key performance indicators (KPIs) for the investment.
Establishing KPIs for AI
We narrowed down those KPIs by looking at the pain points we were faced with operationally. In speaking with other health information leaders, common themes were issues related to turnaround time due to volume and resources, training end users, and low productivity.
Our operational pain points were similar.
Like them, we relied on a centralized intake of all remaining paper medical record documentation that needed to be scanned into the electronic health record (EHR) as close to discharge as possible to ensure continuity of care. We were impacted by factors like paper scanning backlogs, inconsistent indexing, and a heavy reliance upon bar coded forms for scanning recognition.
From there, we were able to identify where we needed greater efficiency, from the time the paper records were picked up, to when they were scanned and then indexed into the EHR.
The KPIs we created for the scanning technology we would choose were:
- Scalability: The ability to minimize courier costs through adoption of point-of-care scanning
- Efficiency: Turnaround time for scanning and indexing of less than an hour
- Accuracy/Quality: Less than 1 percent failure of auto-indexing
- ROI: Reduction in full-time employees and gain of efficiency—with return on investment (ROI) in less than six months
On the final KPI of ROI: It’s often overlooked when transitioning to a new technology that with the efficiency and automation come opportunities for leaders to create new career avenues for staff performing duties that are clerical in nature. Many times, these positions can pivot to less clerical, more analytical responsibilities.
Our OCR Scanning Solution ROI
The optical character recognition (OCR) scanning solution we implemented successfully met our KPIs for scalability, efficiency, accuracy, quality, and our established return on investment.
Its point-of-care scanning model helped us reduce our centralized health information management scanning and indexing staff, as we provided scanning devices to most of the areas that created the paper documentation.
Documents were scanned in sooner, as the scanning occurred at the point of care (POC). Paper scanning backlogs became a non-factor, as we started to see scanning within 24 hours of discharge with a one hour turnaround time.
We no longer depended on the end users who scanned to be indexing experts, able to scan the same exact document the same way every time. The OCR technology removed the need to rely on bar coded forms since the scanning software was able to convert images of typed, handwritten, or printed text into machine-encoded text, which resulted in a consistent 99-100 percent accuracy rate month over month. We started with scanning volumes of over 150,000 images a month and, after the first 45 days from our go-live, we saw consistent turnaround times of 20 minutes or less.
We also decreased our courier costs, as the clinical staff took on POC scanning and integrated the scanning solution to all net new site implementations.
Human ROI and AI
In “Computer Machinery and Intelligence,” Turing proposed what he called an “imitation game,” now often called “the Turing test.” Turing proposed that thinking is hard to define, so a better question than “Can computers think?” is: Can a computer’s end of a conversation with a human be distinguished to a third party from the human’s?
To Turing’s point, AI technology has a certain level of intelligence, but that deeper level of analytical intelligence still comes from humans.
In our case, our workforce members transitioned into duties that accessed the data we needed to analyze once documents were scanned—looking for targeted opportunities to better serve our POC users, manage ROUGE forms, perform exception-based QA/error resolution, and proactively get new forms added to the software’s model build for form recognition.
Dana A. Victor (firstname.lastname@example.org) is a health information administrator at Northside Hospital in Lawrenceville, GA.
Learn more about the role of artificial intelligence in healthcare, register for the webinar “Artificial Intelligence in Healthcare,” part of our Evolving Healthcare series.