Operationalizing Predictive Analytics to Guide Healthcare Business Decisions

Increasingly, providers and health plans are turning to predictive analytics as a tool for identifying actionable insights that can be used to control costs and improve outcomes.

Previously, healthcare companies’ use of analytics focused on analyzing past data to create general insights for future decision-making. Today, says Christer Johnson (pictured at right), principal at EY, healthcare companies are using past data to create predictive models that are embedded into operational systems and can suggest next best actions in real-time.

“You want to remove subjectivity from the selection of variables and early on, put into the model as many variables as you possibly can, and let the algorithms decide,” Johnson said in an interview about his presentation (titled “Health care is lagging in data science. Why it won’t stay that way for long”) at HFMA’s 2018 Annual Conference. 

“Once you have a model that can predict something, operationalize it.”

Potential to Control Costs

By highlighting factors that cause clinical variation, predictive analytics can equip providers with the information they need to lower costs.

Consider the case of a health system that looks at average episode spend for joint replacements across its hospitals and finds that the difference between lowest and highest spend comes out to nearly $100,000—a 400 percent variance. Drivers of this variance range from avoidable to inevitable, and include:

• Case mix index
• Age and experience of surgeon
• Type of implant
• Patient demographics
• Day and time of surgery
• Post-acute care physician

After identifying this cost variance and some of the potential drivers, the health system could use predictive analytics to further explore drivers that can be adjusted, such as supply cost, timing of surgery, and staff training. 

“Predictive analytics can constantly run and look at all the variances, using exploratory data-mining techniques to highlight the factors that are influencing that variance,” Johnson said. Applying this method to the available information, the health system may have an opportunity to obtain actionable data and close the gap between the highest- and lowest-cost facilities.

Improving Outcomes by Engaging Patients

Healthcare stakeholders can use time-sequenced customer journey analytics to reach out to patients more effectively and impact behaviors that can improve outcomes, such as through case management for chronic conditions. 

Instead of leaving an organization to analyze only claims or EHR data, time-sequenced customer journey analytics includes data on past interactions with customers—such as phone, web, and mobile data—that is sequenced by customer. Analyzing this type of data allows a healthcare company to understand the claims and interaction patterns that suggest the best time to reach out to a customer, driving engagement in a case management or other program.    

For instance, health plans have operationalized predictive analytics in efforts to stem the increase of opioid addiction through customer engagement. In its ability to sift through a range of variables, predictive analytics has demonstrated that the timing of outreach can impact its success.

By examining interaction data on how and when customers engage with its resources, Johnson said, one health plan was able to prevent oversaturation of messaging, increasing the rate of positive responses to intervention efforts. Specifically, the health plan noted that 10 percent of customers responded if they were called at random, while that rate rose to 80 percent if the timing was based on customer journey data.

“You can engage with your patients and customers at a higher rate if you figure out when to intervene or reach out to them at the right moment,” Johnson said.

By applying predictive analytics to claims, clinical, and other data, a health plan can better understand which members are at risk for opioid addiction. Operationalizing its customer journey data, the plan may then intervene—through calls or emails, for example—to offer addiction prevention coaching and potentially reduce the rate of addiction.

Making Predictive Analytics Operational

Building the data and analytical platforms to support the timely development and deployment of predictive analytical models in operational systems will be a way for healthcare stakeholders to establish a competitive advantage. 

In the 2018 EY “Future of Health” survey of 2,455 US consumers, 152 physicians and 195 executives, 50 percent of stakeholder respondents indicated plans to use analytics to inform performance initiatives, and 47 percent reported that they seek to capture patient experience metrics within the next year. Johnson noted that while the survey covered all types of analytics, clients have recently sought ways to operationalize predictive analytics.

To implement this process, Johnson said, providers and health plans should consider establishing a centralized data warehouse and hiring analytics experts to build predictive models. Even if a stakeholder lacks a centralized warehouse, Johnson encourages users to build models with available data focusing on the highest-value issues, such as controlling costs, reducing variation, and improving customer experience. 

Johnson emphasized the importance of deciding on desired resolutions before implementing predictive analytics. “You can ask the right question, run the analytics to answer that question, and still not get any value from the analytics,” Johnson said. “You have to figure out the right decisions for which you need more insights.”

Elizabeth Barker is a digital communications professional and freelance writer in Chicago.

Interviewed for this article: Christer Johnson, principal, EY Analytics.