Clinical

Using EHR data to predict post-acute care placement


 

Editor’s Note: The Society of Hospital Medicine’s (SHM’s) Physician in Training Committee launched a scholarship program in 2015 for medical students to help transform health care and revolutionize patient care. The program has been expanded for the 2017-2018 year, offering two options for students to receive funding and engage in scholarly work during their first, second, and third years of medical school. As a part of the longitudinal (18-month) program, recipients are required to write about their experience on a monthly basis.

When patients are admitted to the hospital, the focus for the first 24 hours is on the work-up: What do the data point values tell you about how sick this patient is, and what will they need to get better? While the goal for this information is to develop the appropriate treatment and management for the patient’s acute problem, it could be leveraged to help with other parts of the patient’s hospital stay as well. In particular, it could help avoid unnecessarily long stays in the hospital caused by patients’ waiting for a bed at a lower level of care.

Ms. Monisha Bhatia

Post-acute care placement is a major issue in discharge planning because it involves extensive coordination of resources not just from within the hospital but from other institutions as well, such as skilled nursing facilities and long-term acute care hospitals. About one in four Medicare patient hospitalizations result in a post-acute care placement. Discharge planning is a time-consuming process that can result in an unnecessarily increased length of stay, which can pose risks to the patient and tie up resources in the hospital. Discharge planning does not necessarily have to start late in the hospital stay. What if it could start within a day of admission?

My research mentor, Eduard Vasilevskis, MD, created a rough scoring system for predicting post-acute care placement using admission data, just based on his clinical gestalt. Even at this preliminary stage, the model has already functioned well without much refinement; however a validated, statistically robust model could potentially transform the way that we initiate the discharge planning process. Jesse Ehrenfeld, MD has helped us develop it further by giving us access to a curated database of deidentified EHR data, which contains all of the variables we would like to assess.

The strengths of this potential model are manifold. First, it relies on data collected early in the patient’s hospital course. Second, it relies on routinely collected information (both at our home institution and elsewhere, making it potentially generalizable). And third, it relies on objective patient data rather than requiring providers use their impressions of the patients’ functional status to guess whether they will require discharge planning services. Although such prediction models have been generated before, this model would be among the first to incorporate information routinely collected by nursing staff, such as the Braden Scale, instead of relying on additional instruments or surveys. In addition to predicting placement destination, the model may also be predictive of in-hospital mortality.

With this information, we hope to give hospital teams an additional tool to help mobilize resources toward patients who need the most attention – not just while they’re in the hospital, but also on their way out.

Monisha Bhatia, a native of Nashville, Tenn., is a fourth-year medical student at Vanderbilt University in Nashville. She is hoping to pursue either a residency in internal medicine or a combined internal medicine/emergency medicine program. Prior to medical school, she completed a JD/MPH program at Boston University, and she hopes to use her legal training in working with regulatory authorities to improve access to health care for all Americans.

Next Article:

   Comments ()