Applying AI Based Outlier Detection to Healthcare – Interview with Dr. Gidi Stein from MedAware

Most people who receive healthcare understand that healthcare is as much art as it is science. We don’t expect our doctors to be perfect or know everything because the human body is just too complex and there are so many factors that influence health. What’s hard for patients to understand is when obvious human errors occur. This is especially true when technology or multiple layers of humans should have caught the obvious.

This is exactly why I was excited to interview Dr. Gidi Stein, CEO and Co-founder of MedAware. As stated on their website, their goal is to eliminate prescription errors. In the interview below, you’ll learn more about what MedAware and Dr. Stein are doing to achieve this goal.

Tell us a little about yourself and MedAware.

Early in my career, I worked in the Israeli high-tech industry and served as CTO and Chief Architect of several algorithm-rich startups. However, after many years working in technology, I decided to return to school and study medicine. In 2002, I graduated from Tel Aviv University Medical School with a specialization in internal medicine, treating patients and teaching students and residents in one of Israel’s largest hospitals.

After working as a physician for several years, I heard a heartbreaking story, which ultimately served as my motivation and inspiration to found MedAware. A physician was treating a 9-year-old boy who suffered from Asthma. To treat the symptoms, the physician entered the electronic prescribing environment and selected Singulair from the drop-down menu, a standard treatment for asthma. However, unfortunately, he accidentally clicked Sintrom, an anticoagulant (blood thinner). Tragically, neither the physician, pharmacist nor parent caught this error, which resulted in the boys’ untimely death. This avoidable, medication-related complication and death was caused by a typo.

Having worked as a physician for many years, I had a difficult time understanding that with all the medical intervention and technological support we rely on, our healthcare system was not intelligent enough to prevent errors like this. This was a symptom of a greater challenge; how can we identify and prevent medication related complications before they occur? Given my combined background in technology and medicine, I knew that there must be a solution to eliminate these types of needless errors. I founded MedAware to transform patient safety and save lives.

Describe the problem with prescription-based medication errors that exists today.  What’s the cause of most of these errors?

Every year in the U.S. alone, there are 1.5 million preventable medication errors, which result in patient injury or death. In fact, medication errors are the third leading cause of death in the US, and errors related to incorrect prescription are a major part of these. Today’s prescription-related complications fall into two main categories: medication errors that occur at the point of order entry (like the example of the 9yr old boy) and errors that result from evolving adverse drug events (ADEs). Point of order entry errors are a consequence of medication reconciliation challenges, typos, incorrect dosage input and other clinical inconsistencies.

Evolving ADEs are, in fact, the bulk of the errors that occur – almost 2/3 of errors are those that happen after a medication was correctly prescribed. These are often the most catastrophic errors, as they are completely unforeseen, and don’t necessarily result from physician error. Rather, they occur when a patient’s health status has changed, and a previously safe medication becomes unsafe.

MedAware uses AI to detect outlier prescriptions.  It seems that everything is being labeled AI, so how does this work and how effective is it at detecting medication errors?

AI is best used to analyze large scale data to identify patterns and outliers to those patterns. The common theme in industries, such as aviation, cyber security and credit card fraud, is that they are rich with millions of transactions, 99.99% of which are okay. But, a small fraction of them are hazardous, and these dangers most often occur in new and unexpected ways. In these industries, AI is used to crunch millions of transactions, identify patterns, and most importantly, identify outliers to those patterns as potential hazards with high accuracy.

Medication safety is similar to these industries. Here too, millions of medications are prescribed and dispensed every day, and in 99.99% of cases, the right medication is prescribed and dispensed to the right patient. But, on rare occasions, an unexpected error or oversight may put patients at risk. MedAware analyzes millions of clinical records to identify errors and oversights as statistical outliers to the normal behavioral patterns of providers treating similar patients. Our data shows that this methodology, identifies errors and ADEs with high accuracy and clinical relevance and that most of the errors found by our system would not have been caught by any other existing system.

Are most of the errors you find obvious errors that a human could have detected but just missed or are you finding surprising errors as well?  Can you share some stories of what you’ve found?

The errors that we find are obvious errors; any physician would agree that they are indeed erroneous. These include: prescribing chemotherapy to healthy individuals, not stopping anticoagulation to a bleeding patient, birth control pills to a 70-year-old male and prescribing Viagra to a 2-year-old baby. All of these are obvious errors, so why didn’t the prescribers pick these up? The answer is simple: they are human, and humans err, especially when they are less experienced and over worked. Our software is able to mirror back to the providers the crowdsourced behavioral patterns of their peers and identify outliers to these patterns as errors.

You recently announced a partnership with Allscripts and their dbMotion interoperability solution.  How does that work and what’s the impact of this partnership?

Today’s healthcare systems have created a reality where patient health information can be scattered across multiple health systems, infrastructures and EHRs. The dbMotion health information exchange platform aggregates and harmonizes that scattered patient data, delivering the information clinicians need in a usable and actionable format at the point of care, within the provider’s native and familiar workflow. With dbMotion, all of the patient’s records are in one place. MedAware sits on top of the bdMotion interoperability platform as a layer of safety, accurately looking at the thousands of clinical inputs in the system and warning with even greater accuracy. MedAware catches various medication errors that would have been missed due to a decentralized patient health record. In addition to identifying prescription-based medication errors, MedAware can also notify physicians of patients who are at risk of opioid addiction.

This partnership will allow any institution using Allscripts’ dbMotion to easily implement MedAware’s system in a streamlined manner, with each installation being quick and effortless.

Once MedAware identifies a prescription error, how do you communicate that information back to the provider? Do you integrate your solution with the EHR vendor?

Yes, MedAware is integrated with EHR platforms. This is necessary for error detection and communication of the warning to the provider. There are two intervention scenarios: 1) Synchronous – when errors are caught at the point of order entry, a popup alert appears within the EHR user interface, without disturbing the provider’s workflow, and the provider can choose to accept or reject the alert. 2) Asynchronous – the errors/ADE is caught following a change in the patient’s clinical record (i.e. new lab result or vital sign), long after the prescription was entered. These alerts are displayed as a physician’s task, within the physician’s workflow and the EHR’s user interface.

What’s next for MedAware?  Where are you planning to take this technology?

The next steps for us are:

  1. Scale our current technology to grow to 20 million lives analyzed by 2020
  2. Create additional patient safety centered solutions to providers, such as opioid dependency risk assessment, gaps in care and trend projection analysis.
  3. Share our life-saving insights directly to those who need it most – consumers.

About the author

John Lynn

John Lynn is the Founder of HealthcareScene.com, a network of leading Healthcare IT resources. The flagship blog, Healthcare IT Today, contains over 13,000 articles with over half of the articles written by John. These EMR and Healthcare IT related articles have been viewed over 20 million times.

John manages Healthcare IT Central, the leading career Health IT job board. He also organizes the first of its kind conference and community focused on healthcare marketing, Healthcare and IT Marketing Conference, and a healthcare IT conference, EXPO.health, focused on practical healthcare IT innovation. John is an advisor to multiple healthcare IT companies. John is highly involved in social media, and in addition to his blogs can be found on Twitter: @techguy.

   

Categories