Personalized Medication Risk Mitigation
Not long ago, there was extensive hype around AI (artificial intelligence) and the potential life-saving applications it could provide in healthcare. Fast forward to 2021 and we’ve learned that the early entrants fell short of upholding their promises to the market. Anyone working in healthcare now has a heightened skepticism of the actual capabilities and real impact of operational AI in the real world —and rightly so.
As healthcare continues its digital transformation, there are some “winners” we can point to who have successfully applied AI in a meaningful, valuable way by borrowing best practices from other industries and tweaking them slightly to ensure they deliver value within the healthcare ecosystem. MedAware is applying a similar approach by leveraging clinician-driven AI to create a personalized medication safety net. This is why we succeed when so many other “like” AI solutions have failed. Some of you reading this are new to MedAware so I thought I should take the time in this post to expand a bit on what we do and why, put simply, it actually works.
Before I get into the nitty-gritty, let’s look at a “winner” in another industry who’s really doing an excellent job consuming vast amounts of disparate data elements to identify personal behavior patterns and associated personal risk. One of the areas AI has excelled and earned trust is financial fraud detection. Think of your bank. If I live in New York and go on a shopping spree in Tallahassee, FL, my bank sends me a text message that says “hey, we just saw something fraudulent: was this you?” In the world of risk detection, we call this an “outlier behavior”.
Now let’s apply the above thinking to healthcare. AI can be applied as a similar mechanism to identify outlier behavior patterns with medication ordering. In doing so, dangerous medication-related risks can be prevented and avoided. In the battle against prescription opioid dependency, AI can be applied to identify if a patient is at risk of future opioid dependency, based on their personal, patient-specific medical history. An alternate scenario is identifying that a medication may be inappropriate for a patient based on their lab result trends, medication history, and a variety of other personal measures that contribute to their composite health record.
There are a few factors that contribute to MedAware’s successful AI implementations:
- Asking the right clinical questions: to ensure the AI engine knows the correct questions to answer, we include hands-on, in-depth clinical involvement to understand workflows and mechanisms of error, defining the right questions that an AI engine can answer.
- Powerful foundational infrastructure: to ensure the analysis portion of the process has the best chance of success, exhaustive preparation is essential—including normalizing, cleaning, and coding clinical data in real-time.
- Continuous clinical assessment: to ensure accurate results that contain high clinical value and low alert fatigue, ongoing and continuous assessment of our AI engine’s performance is completed by expert clinicians.
- Active client engagement: to ensure the AI engine becomes smarter over time and continues to deliver clinical value, we maintain a regular cadence of dialog with clinical champions across our customer partner sites.
- Extensive and robust validation: to ensure we maintain best-in-class algorithms and continue to progressively improve our intelligence, we routinely test the AI engine in both clinical studies and as test use cases within live deployments—across millions of lives, multiple geographies, and various technologies.
Applying Clinician-Driven AI to Reduce Medication Risk
Consider some of the areas that touch medication decisioning: EHRs (Electronic Health Records), e-prescribing software, clinical decision support, dispensing software, infusion pumps, population health tools…the list goes on and on. So where does clinician-driven AI fit in the ecosystem? The answer is: any and all of the above—it just depends on the problem you’re trying to solve for.
As identified in our academic studies and across our customer sites, MedAware has seen success applying clinician-driven AI to determine personalized risk of (1) a patient having an adverse drug event or acquiring opioid use disorder (OUD), and (2) a provider prescribing erroneously. The lack of provider confidence in current clinical decision support (CDS) solutions can be seen from the heightened numbers surrounding provider alert fatigue. Studies have shown over 92% of medication-related alerts are clinically insignificant and don’t cause a change in prescriber behavior. As a result, clinicians tend to ignore most alerts altogether—including the “good” ones. For comparison’s sake, MedAware customers typically report prescribers change their behavior 43% of the time they see a MedAware alert.
Conventional systems trigger alerts primarily for drug-drug interactions and dosing errors. This means they’re not covering many additional medication-related risks—such as right drug-wrong patient, lab-result dependent irregularities, and opioid dependency risk. Moreover, adverse drug events (ADEs) that evolve post-prescribing and post-dispensing are not monitored and mitigated by current systems.
By factoring in personalization and examining the patient-specific and provider-specific context of the situation, clinician-driven AI can reduce the overall alert burden and false alarm rate and help providers practice more confidently and efficiently. Picture the window this will provide into the care management of your most at-risk patients and those with chronic diseases.
As the complexity of the healthcare environment grows and the complexity of a patient’s medication regimen expands, clinician-driven AI can help providers manage large amounts of disparate data elements across multiple care environments so they can focus only on exactly what’s required to make an informed decision at that precise moment in time. By honing in on the “outliers” to their personal behavior patterns—just as with the financial fraud detection example—we can identify what’s “normal” versus “abnormal” and prevent risk of a dangerous adverse drug event, opioid use disorder, and reduce provider alert fatigue.
If you’re interested in learning more about this topic, I invite you to take a look at our newest eBook, “Clinician-Driven AI Reduces Medication-Related Risks”. We’ve compiled learnings from some of the latest research on medication safety and also included best practices from our customer sites. I hope you enjoy!