Predictive analytics has changed the landscape of healthcare. It has revolutionized the way providers approach hospital readmissions, patient outcomes, late payments, and missed appointments. It has enabled healthcare organizations to estimate the likelihood of future outcomes and answer many other questions based on patterns in historical data. And all of this can be achieved using just a single predictive analytics model.
In a recent HealthITAnalytics.com webcast, Sriram Parthasarathy, Logi’s Chief Product Owner of Predictive Analytics, outlined all the ways healthcare application teams are using predictive analytics to improve the quality of care, revenue cycle management, and resource management.
“For people who have built applications, you will know that you normally have a number of developers looking at data logic and creating all the rules to create the application,” Parthasarathy says. While in previous years developers had to analyze data logic and create rules if they wanted to generate new applications, machine learning has changed all that.
“Now no one is coding that logic,” explains Parthasarathy. “You point historical data to the algorithm, and the algorithm will automatically look at your data and create those duos. You keep feeding in new data, and it’ll constantly be updating the rules to be in sync with the changing patient behavior.”
Healthcare providers primarily use predictive analytics to determine what actions they can take to avoid adverse health outcomes. Knowing the positive outcomes tied to certain behaviors creates pathways to prevention.
A common use case involves predicting the number of patients that may be readmitted to the hospital within 30 days of discharge. Data analytics platforms can identify those patients who are at risk of readmission and recommend ways to avoid this outcome.
“It’s not only predicting what is going to happen but enabling the healthcare providers—giving them enough tools to act upon particular insights,” Parthasarathy says.
This use case is an example of a classification predictive model, which answers questions by analyzing data based on historical answers. Classification models are typically binary, meaning they categorize data into two distinct groups. In the example above, we have one group identifying patients who may be readmitted and one identifying those who likely will not be readmitted.
There are 10 use cases for a classification model, each falling into one of three categories, including:
- Predicting something that is going to improve patient outcomes, such as screening for a particular condition or chronic disease
- Employing predictions to facilitate more proactive clinical decision-making
- Identifying and reducing fraud (transactions, claims, invoices)
When initiating a predictive process in healthcare, Parthasarathy recommends choosing a problem with a significant return on investment.
“Once you’ve picked the right problem, it’s important to pick the right dataset,” he advises. “Start with what you have, and create the predictive model with the key part.”
When you have insights in hand, it’s equally important to distribute them to all stakeholders. These may include the CFO, line manager, healthcare providers, customer support—everyone who touches the patient at some point in time.
After training a predictive model, Parthasarathy suggests healthcare leaders prioritize running their businesses based on the predictive insights they’ve gained.
“For example, for those who have high risk for a chronic condition—maybe patients with an 80 percent or more probability—prioritize having them come in for a screening next week,” he says. “Or, if a hospital can only screen 20 patients for this chronic condition in a week, wouldn’t you rather spend those valuable resources on patients who have a high risk of actually getting this chronic condition?”
Categorizing patients in this way can help healthcare organizations make well-informed decisions and ultimately spend their time and resources more effectively and efficiently.
If you’re looking to bring predictive analytics into your business, Parthasarathy says, “The key recommendation is to pick a use case that has gathered acceptability within your company and has a good return on investment. Then build a very simple proof of concept and roll that out to your customer base to get feedback.”