Skip to Main Content
Don't have an account? Create Account

Predicting High-Need Patients Using Readily Available Data

With high utilization by the nation’s most complex patients and operating margins at or near the break-even point, essential hospitals must find innovative ways to deliver quality care with the scarcest of resources.

Researchers at the New York University (NYU) School of Medicine and NYC Health + Hospitals (NYC H+H) have found a way to do just that, using resources already at their fingertips. In a new study, they outline a lower-cost alternative to traditional methods for identifying high-risk patients that uses readily available data.

Identification is key: Just 5 percent of the population generates more than half of health care costs. These high-need patients often suffer from multiple chronic conditions and have unmet social needs, such as safe housing, access to transportation, or the ability to access healthy food. Research shows that proactively assisting high-need patients to meet their social and medical needs can lead to fewer costly emergency department visits and hospital stays.

But identifying potential high-risk patients for targeted interventions is a difficult task that typically requires access to large amounts of claims data and advanced analytics. Essential hospitals often lack electronic health records with such advanced capabilities, and the patients they treat include many uninsured people who lack accurate claims data.

NYU and NYC H+H researchers found that by leveraging administrative and clinical data the hospital already collects, their model could identify patients likely to be frequently hospitalized in the upcoming year. It works by assigning patients risk scores based on a set of variables that measure:

  • demographics;
  • past hospital visits;
  • chronic conditions;
  • missed clinic visits; and
  • how often their insurance or zip code changed.

Nonclinical elements of their model are designed to act as proxies for patients’ unmet social needs.

“We found that past hospitalizations and emergency room visits, older age, and certain clinical diagnoses were especially valuable for identifying high-risk patients at a large safety-net health care system,” said lead author Jeremy Ziring, a medical student at NYU and a former data analyst at NYC H+H.

The authors report that their method is more affordable, as well as quicker and easier to implement, than those based on claims data and proprietary algorithms. Also, their method does not rely on payer claims data, which allows physicians to apply risk stratification to all patients, potentially increasing provider buy-in.

“We hope that our model enables other providers serving vulnerable populations to better identify their highest needs patients and more effectively engage them in appropriate care management programs,” Ziring said.

Share

About the Author

Brian Roberson is a senior research analyst at America's Essential Hospitals.