Analytics in Action News

Enhancing Clinical Decision Support Tools with AI, Machine Learning

AI and machine learning could optimize clinical decision support, but organizations have several key things to consider before they can leverage these technologies.

Enhancing clinical decision support tools with AI, machine learning

Source: Getty Images

By Jessica Kent

- Clinical decision support systems are essential for organizations aiming to advance healthcare delivery, helping physicians flag potential problems and make more informed choices.

Although these tools can lead to more efficient, comprehensive patient care, clinical decision support systems have also come with significant challenges. Issues like clinician burnout and unintuitive interfaces have pushed leaders to apply advanced technologies to clinical decision support capabilities.

“Much of clinical decision support for clinical care has been somewhat interruptive,” Katherine Andriole, PhD, director of research strategy and operations at the MGH & BWH Center for Clinical Data Science, said during a recent episode of Healthcare Strategies, an Xtelligent Healthcare Media podcast.

“Machine learning and artificial intelligence can make these systems more seamless, and base outputs on much more information. With these technologies, we can examine more things in the EHR – things that maybe the physician isn’t even aware of, unless they specifically go hunting for that information.”

These innovative analytics technologies are already helping organizations streamline tedious processes and spot patient deterioration, Andriole noted.

Healthcare Strategies · The Evolving Role of AI and Machine Learning in Clinical Decision Support

“Many of the clinical decision support tools that are based in machine learning or AI that are already being used are for back office kinds of tasks that people may not even recognize are happening: Schedule optimization, predicting patient no-shows, and revenue cycle management,” she said.

“There are other things that are happening in clinical care, such as noting an urgent finding like a stroke. These are the types of things that we’re going to be seeing in the future.”

However, the industry still has some foundational problems to correct before these algorithms can become a routine part of clinical care. Many organizations lack the resources to train and develop these tools.

“We’re now in a phase that we call supervised learning, which means that we are training models by showing it cases where we know the answer. For example, we know that this patient has evidence of stroke, and we’ve circled it in the image. Then we test the model on cases that it hasn’t seen before,” Andriole said.

“The problem with that whole method is that we have to have the label or the answer on all the cases that we are using to train the models. That’s a costly process, because it requires expertise. You have to have a physician or an expert label those cases.”

Data access is a major roadblock to algorithm development as well.

“Getting access to enough data to train a model well is also a difficult task. The systems that we have the data in are made for clinical systems. They’re not necessarily made for taking data out and exploring that data, so there are some integration issues the industry has to overcome,” Andriole said.

Clearing some of the biggest hurdles to using AI and machine learning in clinical decision support will require the support and involvement of clinicians themselves, Andriole said.

“We need to educate not only our current physicians but also our trainee physicians and residents who have a great interest in learning more about these models. One of my fears is that physicians will blindly accept the outputs of these systems going forward, and that’s not what you want. You want them to be the last opinion on what happens to the patient,” she stated.

“Organizations will also need to have a clinical champion involved in this process, working with the vendor or the researcher who is developing the model. Clinicians can offer possible reasons why a model isn’t working in a particular case, and can help modify the algorithm based on their clinical expertise.”

As these technologies continue to evolve, clinical decision support systems will become more intuitive, leading to improved patient outcomes.

“Human beings detect things pretty well. It will be interesting to see whether some of these models can start to detect some of the things that the human eye cannot,” Andriole concluded.

Listen to the full podcast to hear more details about powering clinical decision support tools with machine learning and AI. And don’t forget to subscribe on iTunesSpotify, or Google Podcasts.