It is no secret that AI is an often-overhyped buzzword in healthcare. Over the last two years, KLAS has been watching the space closely, doing extensive research and interviewing providers to try to understand how real AI is and in which ways it is impacting healthcare. We specifically wanted to know if and how healthcare systems and payers use AI to make a difference. As part of that research process, we also clearly defined what healthcare AI means. KLAS defines AI as software that provides machine learning (ML) or natural language processing (NLP).
So is AI just another shiny object? We tackle this question head-on in our Healthcare AI 2019 report, the first to dig into the reality of AI tools being used by the healthcare customers. As part of our research, we spoke with 57 health organizations who currently partner with AI vendors. Although most users are still in a pioneering phase, we have uncovered many that are driving better patient outcomes and experience.
Over a series of blogs, I will be sharing some key findings from the Healthcare AI report, including:
- Top AI use cases and outcomes
- Common misconceptions
- End user and vendor partner best practices
A Mile Wide and an Inch Deep
The question we were asked most often during our research was this: how are my peers using their AI tools, if at all? The chart below shows the areas where interviewed users are leveraging AI. In all, there are about 90 total use cases, with 37 distinct uses across 11 categories in the clinical, financial, and operational realms.
While it is exciting to see AI being adopted across a wide variety of use cases, it is still too early to say whether they can be scaled across broader customer bases. Providers are just dipping their toes into the possibilities, creating a market that is a mile wide and an inch deep, so to speak.
Clinical Use Cases and Outcomes
The most common use cases in AI are clinical, with tools focusing on readmission predication and prevention (a top priority for all healthcare organizations using AI or not), avoidable ED visits, and the prevention of hospital-acquired diseases, to name a few. We also found tools that help with the discovery of treatment best practices, chronic disease management, clinical research, clinical trial matching, patient engagement, and education.
But what outcomes are these tools driving? One VP of Data and Analytics told KLAS, “We automated some of our back-end processes using AI. We produce actionable lists that pull in all the patient information that the care managers need so that they do not have to log in to multiple systems or go to multiple sources. This saves them quite a few hours of work each year, and those hours can be redirected to follow up with patients. In fact, the hospitals save over 1,300 hours per year in staff time. That was the first, immediate impact. Originally, we had a list coming out of the EMR, and we also looked at the LACE scores. Using our AI tool, we identified more true positive cases, and we reduced our false positives compared to the list that was being generated by the EMR.”
Financial and Operational Use Cases
Although we have seen far fewer reported use cases in the financial and operational areas, we anticipate this area to experience tremendous growth as data become more available and normalized when compared to clinical areas. Judging by some of the commentary on early financial and operational AI tools, it looks like those tools will start to have a large impact on operational efficiency.
For now, a few top use cases in this area include propensity to pay, claims-fraud detection, and predictions for patient cost, bed availability, length of stay, no-shows, ED patient throughput, and nurse turnover.
A CMIO shared with us, “We started using AI to calculate the geometric length of stay, and we used that as a proxy to start calculating avoidable inpatient days on the first day a patient arrives in the hospital. That is amazingly useful for our case managers, and we are all very excited about it. Our case managers can look at a patient list and decide which patient they should focus on first based on the predictions from the system. They can see which patients are going to be discharged soon and how many avoidable inpatient days are predicted for each patient. With that information, they can know how to focus their efforts to provide the most value for the hospital.”
To see more specifics about AI vendors and validated outcomes, I recommend reading the full report. Stay tuned for more information in future blogs about misconceptions and best practices as customers and vendors work together to achieve outcomes associated with AI.
This piece was written by Lois Krotz, Director of Research Strategy, and Financial & Services Research with KLAS. To follow her on Twitter, click here.
Share Your Thoughts
You must be logged in to post a comment.