July 27, 2021 | Net Health

3 Minute Read

4 Early Lessons for AI Adoption in Healthcare Analytics

Artificial intelligence (AI), machine learning, and predictive analytics used to be the dream of the future for healthcare. But thanks to the exponential growth of technology and the drive of key innovators, it’s no longer just a dream—it’s become the reality of now.

As healthcare leaders, staying on the cutting edge of these technologies is a collision of excitement and calculated scrambling, as we look to see what and how these innovations should be implemented in our organizations.

In a recent webinar titled Modern Healthcare Innovation Leaders, several medical professionals on the cutting edge of AI innovation shared their thoughts about the changes. Today, we want to look at four key lessons from that webinar that are crucial for healthcare leaders to build a roadmap forward for AI implementation.

1. Have An Intimate Understanding of the Problem First

Often, AI is approached from a solutions-first angle. It’s the “I created this technology, let’s find a way to use it” approach. The problem with prescribing AI before defining the context it’s going to be utilized in is that it often misses the mark with results.

More specifically, if we don’t take the time to define the struggle and friction points from a perspective of intimate understanding first, then AI technology may not achieve the target goals and results.

An example shared in the webinar talks about the use of AI with obstetrical sepsis. The team working on the project did a phenomenal job in building data models to quickly identify patients at risk for the ailment. The problem, though, was that wasn’t the problem. Clinical teams already knew who was at risk, and the problem affecting mortality rates wasn’t a detection problem. It was a technology win but a project failure that could have been mitigated by better defining the problem first.

Actionable Step: Determine the problem you’re trying to solve first, get an intimate and deep understanding of that problem all the way down to the user level, and then start searching for technology-driven solutions.

2. Prepare For Data Bias

The lifeblood of any successful AI integration is the data driving the engine. If that data is bad, the best algorithm in the world is likely hopeless. One way in which data may be bad that’s not always clear from the get-go is data bias. And the reason the bias may be challenging to detect is that it may be subtle. It may be tied to a certain institution’s data, a certain protocol tied to an institution, or even to a manufacturer.

When this happens, the algorithm may succeed in the vacuum of testing, but when deployed to the real world—the accuracy of results falters.

Actionable Step: Dedicate time to the study of the actual data sets. The more locations and sources you can get, the better.

3. Develop a Plan of Action

If the most accurate algorithm imaginable is validated and tested by the best clinicians in the country, but it’s not implemented effectively in clinical workflows, what’s the value of that prediction? Zero. When we get valuable information, it only remains valuable if we have a plan of action in place to capitalize on the information. When we implement AI that drives business decisions or the clinical-decision making process, we have to be prepared to execute.

Actionable Step: Have a care model in place to execute on the actionable data coming out of your new implementations.

4. Be Prepared to ‘Kill Your Darlings’

In his book, Refounder, Net Health’s Co-founder Patrick Colletti references a famous William Faulkner quote that says, “In writing, you must kill all your darlings.” Colletti uses the quote to address the dangers of becoming overly attached to a particular idea. Instead, Colletti encourages readers to be open to using existing ideas as potential steppingstones for bigger and better ideas.

The translation to the adaptation of AI is perfect. While it’s okay to get excited about what new technologies and algorithms can bring to the table, be prepared to take a sober look at your realities and decide the best path forward without the constraints of overattachment.

Actionable Step: Ensure your team understands this mindset prior to beginning the process. The earlier and better you can ingrain this in your team, the lower the risk of overattachment.

How Can I Learn More?

If you’re interested in continuing the discussion, we’d encourage you to listen to the free complete Modern Healthcare Innovation Leaders webinar now. The panel discusses these topics more in-depth, as well as actionable steps that can be taken to move your organization forward with AI and healthcare analytics.

Modern Healthcare Innovation Leaders

How Top Health Systems Plan and Execute Innovation

 
 
 
Share this post

Subscribe and See More