Predictions lead to better prevention

BOSTON—On any given day, 1 in 25 hospital patients has a hospital-acquired infection and more than 750,000 of those infections are preventable, said Todd Schlesinger, vice president of business development for Jvion, speaking at the Big Data Healthcare Analytics Forum. With a 1 percent Medicare penalty for these infections, “it doesn’t take long for hundreds of dollars of these penalties per patient to turn into millions of dollars in losses for an organization.”

Predictive analytics can foresee these events and apply deep machine learning technologies before clinical signs are present but “we found that satisfying this equation is harder than it may seem.”

That’s because the solution has to fit into the real-world of the hospital, be easily deployed, fit into the workflow and require little time and effort from internal hospital resources, Schlesinger said. Data have to be easily extracted and the solution must have a clear return on investment. “Knowing when the model is wrong is as important as knowing when the model is right. That seems obvious but it’s extremely critical in healthcare. When technology incorrectly identifies a risk, it undermines everything.”

Schlesinger discussed his firm’s success with Baptist Health in Montgomery, Ala. Jvion helped the organization garner insight from data to mitigate patient risk of infection. The clinician is at the core of a successful effort, he said. They engaged the clinicians at Baptist by making sure the solution fit into their workflow and then demonstrated the tool’s accuracy.

After one year, Baptist Health had a 35 percent drop in discharges and 31 percent drop is length of stay for Stage III and IV pressure ulcer patients. They avoided more than $1.84 million in unreimbursable costs across four conditions in the first year.

The hospital also worked on a model to predict heart attacks. One percent of its patients have a heart attack within a year of discharge. The model allowed them to take the entire population and narrow it down to a targeted group of 1.5 percent of those patients at the great risk. They have seen 96 percent accuracy at current target thresholds. “We’re very proud of that,” he said.

While he acknowledged that the best prevention models are based on 10 years of risk assessment that include lifestyle modifications, Baptist’s current solution allows the hospital to apply what they know about preventing heart attacks that occur in the next year where lifestyle modifications might not be effective.  

Of Baptist Health’s predictive analytics effort, Schlesinger said, “This is a very strong example of a forward-thinking organization seeing real results and positioning themselves to be successful under the new quality and value based systems.”

With all the pressures on the healthcare industry, he said there is a “completely different landscape. The best way to promote health is to prevent illness. Treatment without prevention is unsustainable and you can’t prevent what you can’t predict.”