Want to keep learning?

This content is taken from the Taipei Medical University's online course, AI and Big Data in Global Health Improvement. Join the course to learn more.

Skip to 0 minutes and 15 seconds We were interested in Australia as our hospitals are starting to implement these systems to understand how do they make decisions about which alert to include in their system. So we recently did a survey of 26 hospitals, and we asked them what alerts have you introduced, so nearly all of them had implemented allergy alerts, drug-drug interaction alerts, and a lot of them had implemented dose range checking. And we asked them why did you choose to put these alerts in,

Skip to 0 minutes and 46 seconds and most of them tended to say: “Well, we think there’s good evidence that these alerts are effective,” and then we asked them, “Would you have any evidence that those alerts have made a difference in your hospital?” And they said, “Well, no we haven’t because we haven’t evaluated it.” So we started to say well what is the evidence that these types of alerts are effective. So here we did a systematic review, and we found, for example, that there were six studies, and find out that it looked like drug condition alerts and five of those had shown some positive effect on prescriber behavior or outcomes.

Skip to 1 minute and 23 seconds There were six studies that have looked at drug-drug interaction alerts, and two of those six studies had shown some improvement, and we looked at corollary order alerts, and one of six studies showed some benefit, but what was very interesting is that there have been no studies that demonstrate what happens when you put all these alerts in that you keep on adding alert upon alert. Does the system overall become more effective? Or not? So we’re interested in saying can we take a more evidence-based approach to decision support selection, and we wanted to focus on one type, drug-drug interaction alerts.

Skip to 2 minutes and 6 seconds For any of you who have these systems in place in your hospitals, you’ll know that you can often choose to have maybe about a hundred drug-drug interaction alerts, or up to 15,000. One of our hospitals has 15,000 drug-drug interaction alerts. So we were saying what is the size of the problem of drug-drug interaction in hospitals? so we did a systematic review. And this was published this year. These are potentially quite frequent, so around a third of patients experience a potential drug-drug interaction alert, and ICU patients experience a potential rate, but there are very few studies to demonstrate whether there is any harm that patients gain from these types of errors.

Skip to 2 minutes and 56 seconds In the few studies that have looked at, they found 2% of patients were harmed as a result of a drug-drug interaction alert, and most of that harm was quite minor. So for this problem, one of our hospitals is throwing 15,000 alerts at the doctors. So we can start to ask the question, is the solution warranted by the size of the problem? Now we’re moving forward to an era of AI approaches, and you’re going to hear a lot about this from the other speakers today about all the potential things that we can do to provide even more sophisticated decision support for our healthcare providers.

Skip to 3 minutes and 44 seconds Some of you will heard of things such as that the Deep Patient Project in the US, where they’re using over 700,000 patient records to start being able to predict the probability of diseases, to be able to make recommendations about what is the most appropriate treatment for patients. So these this is the type of decision support that we’re going to see in the future. Alright, I think the question is that as we move in this direction, the challenges that we face with trying to understand how do we have the right mechanisms and model, how do we incorporate this decision support into clinical workflows remains? It’s the same challenge that we have now.

Skip to 4 minutes and 27 seconds And we must pay attention to it because unless we do that, we cannot transfer this enormous potential from AI into real clinical actions and improvements in outcomes.

Studies Review on Hospitals' Medical Decision Support

This video, Dr. Johanna Westbrook explains a result from a survey that hospitals’ decision on what alerts should be implemented. Nearly all of the hospitals had implemented allergy alerts, drug-drug interaction alerts, and a lot of them had implemented dose range checking. The hospitals believe that these alerts seem efficient.

Then, the educator presented many interesting questions on hospitals’ medical decisions. There were six studies that have looked at drug-drug interaction alerts, and two of those six studies had shown some improvement. However, in another study, the researcher found 2% of patients were harmed as a result of a drug-drug interaction alert, though most of that harm was quite minor. These results show that development is still ongoing.

Deep Patient Project in the US is another example. They used over 700,000 patient records to start being able to predict the probability of diseases, to be able to make recommendations for the most appropriate treatments for patients. This is the type of decision support that OHDSI intended to see in the future.

In the final part, Dr. Westbrook pointed out the challenges of AI implementing in the current system. How do hospitals have the right mechanisms and models? How do they incorporate decision support into clinical workflows? It’s the same challenge that we have now.

From the example that Dr. Westbrook presented, how do you view the potential of AI implemented in the medical decisions now?

Share this video:

This video is from the free online course:

AI and Big Data in Global Health Improvement

Taipei Medical University

Get a taste of this course

Find out what this course is like by previewing some of the course steps before you join: