Where: The University of Chicago Medical Center, an academic tertiary care facility with 500+ beds.
The issue: Communicating the stability of patients during handoffs.
As at many academic centers, resident work-hour restrictions led to an increase in handoffs at the University of Chicago Medical Center. Aware of recent research showing that increased handoffs can lead to worse patient outcomes and that some bad outcomes on the wards, especially cardiac arrests, are predictable, the Chicago hospitalists looked for a reliable way to transmit vital information about patient status.
“One of the things that I've been really interested in is developing a predictive algorithm that uses vital signs and such to predict who might [have a cardiac arrest] so we can get to them earlier and intervene,” said Dana P. Edelson, MD, assistant professor of hospital medicine. However, this idea ran into some obstacles. First, incorporating such an algorithm into an electronic medical record would require expensive programming, and second, it's still unclear what information should be included to accurately predict events.
“In the meantime, it occurred to me that a fast and cheap way to potentially do the same thing is to rely on clinical intuition. I have this sense when I walk into a patient's room whether they're sick or not sick,” said Dr. Edelson. But, she noted, “There isn't a great way to communicate that when I leave the hospital at the end of the day.”
How it works
The solution to that communication barrier, developed by Dr. Edelson and colleagues, is very simple. Their Patient Acuity Rating (PAR) score is a number from 1 to 7, chosen in response to the question, “How likely is this patient to suffer a cardiac arrest or require emergent transfer to the ICU in the next 24 hours?”
“Really what you're asking is ‘How sick is this patient?’ Or ‘How likely are they to go bad quickly?’” said Dr. Edelson. The physician who is handing off the patient selects a response from 1 (extremely unlikely) to 7 (extremely likely) and passes that number along to the cross-covering physician, along with the usual signout information.
The goal is that the covering physicians will use this additional information to respond to changes in the patient's condition. “For example, if the nurse calls and says, ‘This patient's heart rate is up to 111 and it was 80 when I started my shift,’ that's concerning for any patient, but it's more concerning for a patient that's higher risk,” said Dr. Edelson. “A high PAR score is always a reason to go to the bedside, but [a low score is] never a reason not to go.”
As a preliminary test of the PAR system, 40 clinicians (including interns, residents, attendings and midlevel providers) created more than 6,000 scores. Their scores were then compared to outcomes—whether the patient had actually had a cardiac arrest or ICU transfer—over the following 24 hours. The study, published in the October Journal of Hospital Medicine, showed that the score fairly reliably predicted risk. A PAR of 4 or more had 82% sensitivity and 68% specificity for predicting negative outcomes.
One interesting finding was that residents performed worse on the scoring test than other clinicians. The average area under the receiver operator characteristics curve (AUROC) was 0.82 for everyone involved in the study, but for residents, it was 0.69.
“My hunch (had been) that the residents (were) the best predictors,” said Dr. Edelson. “I figured the lack of experience from the interns would affect their ability to predict well, and that for attendings lack of time at the bedside would affect their ability to predict. It turns out you need either time at the bedside or experience, and the relative lack of both for the residents is what stands out as being inferior.”
The interns' superior performance on the PAR trial (0.79 AUROC) is actually a good result, Dr. Edelson said. In the initial trial, designed to test the accuracy of the scoring, the PAR scores were not given to cross-covering physicians, but a study of the entire system is now underway. “In the subsequent study, I'm just collecting [the scores] from interns. That's who fills out the signout at the end of the day, so it makes sense,” she said.
The follow-up trial is also testing a version of Dr. Edelson's original project—the vital-signs-based algorithm for predicting cardiac arrest and ICU transfer. “We can compare the two and potentially see whether there's an additive benefit of using both at the same time,” she said. “My guess is that the best combination will be a model that uses vital signs, laboratory values and clinician intuition.”
Although research on the effects of the PAR score is ongoing, hospitalists can apply the system in their own facilities right now—indeed, some already are, Dr. Edelson said. “It's simple, intuitive and cheap to implement,” she said.