Late last year, data professionals from dozens of industries, including healthcare, convened in Washington, D.C. for the annual DGIQ (Data Governance & Information Quality) conference.
The main topic of conversation? How to support their organizations’ AI initiatives with quality data.
With sessions named things like “Data: The Missing Ingredient in Your AI Strategy (and Why it’s a Recipe for Failure)”, presented by IBM, and “Turning Metadata into a Competitive Advantage in the Age of AI”, by TELUS and Collibra, the takeaway was clear: Before an organization can pursue AI, they must have solid data practices in place.
Before an organization can pursue AI, they must have solid data practices in place.
What does this mean for healthcare organizations?
The healthcare industry is buzzing about ways AI could improve operations (with “AI” these days being used as a catch-all label to describe everything from predictive analytics to machine learning to basic statistical analysis).
Leaders would like to see AI doing things like summarizing clinical notes, automatically adding missing billing codes from those notes, suggesting treatment interventions in real-time, and identifying which patients are at risk for specific diseases. Just for starters.
But, compared to industries like retail and banking, the healthcare industry is lagging a bit when it comes to activating these initiatives.
There are obvious reasons for this. For one, there’s the incredibly broad scope of data—the data characteristics that can be gleaned from a human body are virtually endless. Plus, in healthcare, there’s a vast difference in the resources available for AI. Healthcare leaders are busy caring for patients, with fewer profits to spare, and life-or-death issues to focus on first.
Intentional or not, this has, in many cases, led to a general lack of urgency when it comes to investing in data governance.
Fortunately, there are consulting partners and data solutions available to help healthcare organizations efficiently prepare their data for AI, and there are clear steps that can be taken to prepare for effective AI solutions.
The good news is there are only a few core elements required for a solid AI foundation, and many healthcare organizations are putting these into place. Those who haven’t yet, still have plenty of time to catch up.
Three foundational elements to game-changing AI solutions
1. A solid data warehouse
Before a healthcare system can truly apply the power of AI to their data, they need to bring all of their data into one centralized location, where it can all be accessed easily.
Why is this critical?
“It’s basically impossible to operationalize anything like predictive analytics, machine learning, or natural language processing if the data is too hard to get to,” says Kevin Campbell, CEO of DTA Healthcare Solutions, who was a speaker at the DGIQ conference. “You have to start with a solid foundation for data integration.”
Let’s say a hospital wants to implement predictive analytics to determine the likelihood a heart attack patient will be readmitted after discharge.
In order to identify drivers of heart attack readmissions, the model will need to be able to access all the records that exist on past heart attack patients, including lab tests, blood pressure indicators, medications being taken, and patient demographics.
Then, in order to help clinicians prevent future readmissions, the model will need ongoing access to all of that same information for each new heart attack patient.
Obviously, this requires accessing information across a range of disparate systems—the EHR, billing, and pharmacy systems, to name a few.
A data warehouse solves the challenges involved with accessing all of the relevant data.
A strong data warehouse is the first step to supporting a wide range of AI initiatives.
In other words, if you want to implement widespread AI applications within your healthcare organization, you’ll need a data warehouse in place.
2. A strong data governance process
“Once you can access the data,” says Campbell, “the question is how do you make sure it’s accurate? How do you make sure it stays accurate? How do you make sure it gets updated when it should? Who has ownership of the data? Is anyone watching it to make sure it remains accurate?”
The answer to all of these questions is the same: data governance.
Garbage in, garbage out
Let’s say you want to be able to predict when patient volumes will be higher and when they will be lower, so you can optimize your staffing levels and forecast what medical supplies you’ll need. If your AI is pulling from flawed historical data, the predictions will be wrong.
Possibly, wildly wrong.
Here’s how that might look: Maybe your ED has labeled patient admissions differently than your surgical department, so only the ED’s patients are being counted by AI, and thousands of patients aren’t being factored in at all.
Or maybe both departments are tracking the same people as new patients, so the number of admissions being counted by AI is double what is true.
If no measures are in place for ensuring accuracy, there are endless ways for inaccurate data to skew AI results.
On the other hand, Campbell says, “If you can leverage your AI models against a trusted data source, you don’t have to wonder, ‘Is this data even right? Am I just working with garbage here? You can trust what AI delivers.”
3. A good data catalog
The third important element, after making sure your data is centralized and accurate, is to make sure it’s findable.
“Data lineage and provenance is critical for AI,” says Campbell. “In other words, you have to be able to account for exactly what fed the models. Where did this data come from?”
To do that, you need a data catalog, which can point you to any bit of data you’re trying to find and can also tell you where that data came from, who owns it, and what it means.
For the data professionals at the DGIQ conference, the need for a data catalog is obvious. It’s as essential to AI as bacon to a BLT.
But in the healthcare industry, data catalogs have been slower to catch on. Now, as more healthcare organizations hope to move toward AI solutions, it’s more important than ever that they implement a good data catalog, like, for example, Compendium, a data catalog that’s designed specifically for healthcare.
According to the data experts at the DGIQ conference, the most important way to support your AI dreams is to make sure you have high-quality, highly accurate, easily accessible data behind it. Which means that the healthcare organizations that have made the investment in their data warehouse, data governance, and data catalog are well ahead of the curve when it comes to leveraging AI.
And those that haven’t, know exactly where to start.
Does your healthcare data team need an extra hand with data governance?
The data pros at DTA Healthcare Solutions can help you meet your AI goals, quickly and painlessly. Learn more.
