At Vanderbilt University Medical Center (VUMC) there are about 486 distinct operational units involved in inpatient care and evaluation.
Pick any patient: Which of Vanderbilt’s many operational units are apt to collaborate in the care of that patient?
In a new study appearing in the Journal of the American Medical Informatics Association, Vanderbilt researchers used health care billing codes, electronic health record (EHR) user logs and unsupervised learning techniques to get a bird’s eye view of Vanderbilt’s collaborative care teams and their constituent operational units.
Starting with four continuous months of hospital data, the investigators used billing codes to quickly place 17,947 de-identified inpatients under 1,413 diagnosis groupings. EHR logs furnished records of which employees of which Vanderbilt operational units viewed the patients’ records. These data were fed into various computer models and algorithms used for network analysis.
Finally, a survey of clinical and administrative experts was used to verify that the inferred clusters constituted care teams.
The study, led by You Chen, Ph.D., assistant professor of Biomedical Informatics, found 34 Vanderbilt collaborative care teams in all, comprising 317 operational units. (Approximately 35 percent of Vanderbilt inpatient operational units, 169 units in all, didn’t cluster with a care team.)
The study would appear to provide material for sketching parts of a patient-centered care delivery system. U.S. health care payers, with the federal government in the lead, have begun pushing new payment models that transfer financial risk to health care providers, and this is expected to result in a care delivery system that’s more organized around patients, instead of around lines of reimbursable services, as is currently the case.
“It’s very hard to model the patient-centered health care system. This is just a start. We want to show health care organizations how their EHR data can provide guidelines for establishing some type of patient-centered system,” Chen said.
In the study, some of the care teams inferred by network analysis were given familiar labels like Urology or Nephrology, while others were given less familiar labels such as Infection Related, Oncology I, Oncology II and Women and Babies.
Chen and colleagues surveyed 23 clinical and administrative experts, testing whether respondents could distinguish the inferred clusters from randomly generated clusters. Of the 34 inferred clusters, 27 were recognized by respondents as non-random. The most sprawling of the verified care teams, with 39 operational units, was labeled Interventional Cardiology and Vascular Institute, followed by Neuroscience with 25 operational units. Of the 27 verified care teams, 16 had five or fewer operational units.
Chen hopes eventually to compare care team structures and patient outcomes at different institutions.
“We want to look for any significant associations that may exist between the cost of care and team structures, while also taking patient care quality into account. This should be helpful for shaping more efficient team structures.”
He said that for the time being the problem of connecting these sorts of data from multiple institutions remains to be solved.
Joining Chen in the study were Nancy Lorenzi, Ph.D., Warren Sandberg, M.D., Ph.D., Kelly Wolgast, DNP, and Bradley Malin, Ph.D.
The study was supported by the National Institutes of Health (grants LM011933 and LM010685).
