Press "Enter" to skip to content

Predictive analytics may ease the burden of flu outbreaks

Each year the public health impact of the influenza season can be as unpredictable as the virus itself.

For hospitals, such uncertainty can present capacity challenges that lead to shortages in supplies, equipment, beds and staffing, as well as higher patient utilization costs.

The problem can be especially daunting in major urban areas, where the population density can cause flu to spread quickly and have severe health consequences for those in lower-income, medically vulnerable neighborhoods.

Detroit Medical Center this year launched a project seeking to reduce some of the unpredictability of the flu season.

Through the health system’s electronic health record, the initiative identifies patients who have consistently visited the emergency department for flu-like symptoms and uses predictive analytics based on their clinical, personal and demographic information to estimate their likelihood of having been vaccinated.

The Centers for Disease Control and Prevention’s flu surveillance information gives DMC weekly updates as to when flu activity may be high, which the system can use to schedule a targeted flu vaccination campaign in areas with high concentrations of at-risk patients.

Using predictive analytics lets DMC identify patients who are more than 75% likely to develop flu over the next two to three months, said Dr. Leonardo Lozada, chief medical officer at DMC’s Detroit Receiving, Harper University and Hutzel Women’s hospitals.

“We can go after them by sending them messages or even starting campaigns for vaccination early on to a certain population,” Lozada said. The vaccination campaign consists of offering flu shots in concentrated areas, such as community centers, churches or schools. Other means of outreach include calling or texting patients or in some cases visiting their homes to encourage them to get vaccinated.

Lozada is hopeful preventive measures to limit the flu’s impact will lead to improved quality and patient satisfaction and reduced healthcare costs related to treating flu.

“We are in a world where we have to take consideration of our resources,” Lozada said. “How do we maximize our resources—the best way to do it is by correlating and predicting the effect of our treatments on our patients.”

Only in recent years has the healthcare industry begun embracing predictive analytic tools to improve disease management, diagnosis and patient care.

DMC is one of a few providers to expand the use of predictive analytics for conditions that have a high impact on healthcare, such as seasonal influenza. Most providers view such tools as a way to mitigate the costly burden caused by chronic health conditions such as heart disease, cancer and diabetes.

Hospitals could benefit from knowing why some communities have higher rates of vaccinations than others, said Dr. William Schaffner, professor of preventive medicine and health policy at Vanderbilt University Medical Center. “If you can define those differences, you can then go the next step and start doing research to figure out what the barriers are,” he said.

Although some progress has been made toward increasing flu vaccination rates overall, rates have stalled in the last few years among groups at highest risk for flu complications, Schaffner said.

Roughly two-thirds of adults age 65 and older get vaccinated, while half of pregnant women and two-thirds of young children get a flu shot each year.

Schaffner said computer records indicating which employees have been vaccinated have played an important role in raising flu vaccination rates among healthcare professionals, which climbed by 15 percentage points from the 2010-11 season to the 2017-18 season.

The financial stakes are high. Each year flu costs the nation an estimated average of $3.2 billion in direct medical costs, according to a 2018 study published in the journal Vaccine, and an average of $8 billion more a year in lost productivity.

“That’s where our big bang for the buck truly is,” Lozada said. “The idea of planning for something is always less costly than reacting to something.”

Between 12,000 and 59,000 people die during an average flu season. But a sudden, unpredictable change in the virus, or an outbreak within a large unvaccinated population can quickly turn an average season into one like the 2017-18 season, which resulted in nearly 80,000 deaths and more than 950,000 hospitalizations.

The health impact of this year’s flu season thus far appears to be less severe than last season. On Jan. 11, the CDC reported approximately 7 million flu cases had occurred between Oct. 1 and Jan. 5, resulting in an estimated 84,000 hospitalizations, which the agency stated fell within the parameters of an average flu season.

Vaccination rate ceiling

Most hospitals rely on the CDC to track the virus’ progression. That information combined with the known demographics of those usually at severe risk of flu complications—young children, pregnant mothers and the elderly—has traditionally helped shape vaccination campaigns targeting those groups.

Yet flu vaccination rates have never risen above 70% among any demographic despite such efforts. Lozada and others believe predictive analytics may help achieve more effective vaccination coverage. They contend such tools may allow stakeholders to get a better understanding of the factors commonly associated with patients who are less likely to get a flu shot, which will help providers and payers tailor their outreach approaches.

With vaccinations remaining controversial for some, any way to raise the rates would be welcome. “Influenza vaccine is one of the most available preventive services there is,” said Dr. Joshua Sclar, chief medical officer for BioIQ, a population health management platform company. “That eliminates a significant number of barriers for most people, and yet uptake is still pretty low.”

Sclar said payers have a growing interest in using social determinant risk factors such as poverty, unemployment, low education, housing instability and food insecurity in identifying vulnerable populations; payers see such data as a way of helping predict the likelihood of hospitalizations if those patients get the flu.

“It’s a resource-allocation game here,” Sclar said. “You have finite resources to try and get the population vaccinated, and you want to dedicate those resources as appropriate to the highest-risk populations you can identify.”

For Dr. Amesh Adalja, senior scholar at the Johns Hopkins University Center for Health Security, the biggest potential for such advanced technologies in flu surveillance comes from the ability to provide real-time, community-level data about flu activity, which could be essential in an emergency such as a flu pandemic to understand the burden of the illness.

Such tools could also can raise clinicians’ awareness about when and where to test for and treat the virus and help hospitals better allocate resources during flu season, which Adalja said is usually done haphazardly.

“They don’t staff for flu season the way they do for when a marathon or the Super Bowl is in town,” Adalja said. “They don’t think of it that way, and I think you could do that much better if you had data that was real-time and accessible so that people can see not what happened with flu two weeks ago but what’s happening today.”

Urban poor at risk

But Adalja believes there’s still a need to focus on improving clinical practices that encourage more patients to get flu vaccinations among the groups already well-known to be at higher risk for flu complications.

“You have many vulnerable populations that are left unvaccinated even before you even think about doing all this very high, big data targeting,” said Adalja, who pointed out vaccination rates among pregnant women remained low—around 49% in 2018—despite their increased flu risk being well-known to clinicians.

And Vanderbilt’s Schaffner said the evidence is clear that communities that are both poor and have higher household density are associated with a higher risk of more severe influenza.

He said while DMC and a few others have been able to use predictive analytics and socio-economic data to more precisely identify populations more vulnerable to the flu’s effects, the question remains whether such information can be applied toward developing newer, more effective interventions.

“How can public health marshal their resources to develop new programs—these will have to be innovative—that can go in and proactively provide preventive health services to those areas that now can be defined,” Schaffner said.

“These things won’t be easy, and they won’t be cheap.”