About one to three percent of all emergency department patients present as clearly urgent, cannot-wait cases requiring immediate life-saving intervention. They're Emergency Severity Index level 1: critically injured trauma patients, patients with clear signs of an acute MI, the ones unresponsive after an overdose. They are dying.
Another 40 to 70 percent will be triaged as ESI-2 or ESI-3 needing immediate or relatively immediate attention. But many of those patients will present with relatively vague or nonspecific complaints, like weakness, dizziness, or feeling unwell.
It's easy to dismiss cases like these as positional vertigo or an ear infection, but six percent of patients with vague complaints ended up dying within 30 days of presentation. The Basel Non-Specific Complaints (BANC) study, in fact, found 218 nontrauma ESI 2 and 3 patients presented with nonspecific complaints. Thirty days later, 128 (59%) were diagnosed with a serious condition, and 12 (6%) died. That mortality rate, the investigators noted, is comparable with that of patients with community-acquired pneumonia. (Acad Emerg Med 2010;17:284.) “Sensitive risk stratification tools are needed to identify patients with potentially adverse health outcomes,” they wrote.
Therese Djärv, PhD, an emergency medicine researcher at Sweden's Karolinska Institute, came to a similar conclusion. Her study of more than 20,000 patients seen by a major Stockholm ED found that those presenting with a “decreased general condition” — a nonspecific decline in health — had a fourfold risk of dying while hospitalized compared with patients who had other presenting complaints. (Eur J Emerg Med 2014 Jun 6. [Epub ahead of print].)
“Since triage systems most commonly are based on vital signs and elderly people with conditions might not present with affected vital signs, they end up with a low priority even if care personnel use triage systems perfectly,” she said. “This is a weakness of today's system, and we need these kinds of studies to highlight when we need to step out of the box and assess the patient further than today's triage systems do.”
A variety of tools have been proposed to make that stratification easier. The stress marker copeptin, for example, has been reported in several studies to be a promising nonspecific biomarker of disease severity in patients with vague complaints. The potential of this and other biomarkers is tantalizing, but they need more study. Dr. Djärv is in the planning phase of such a project.
But what if there were just a few questions that a triage nurse or emergency physician could ask to discern the really sick from the not-so-sick? Researchers at the Center for Adaptive Rationality at the Max Planck Institute for Human Development in Berlin, Germany, are now working on a short algorithm that they hope will accurately identify patients with vague complaints who are most in need of immediate attention.
Physicians at University Hospital Basel in Switzerland asked Planck psychologist Mirjam Jenny, PhD, to develop the tool, presenting her with 88 signs of serious illness from a pool of more than 1,200 emergency patients who had described vague problems. Dr. Jenny used a complicated program to sort those signs into 14 pieces of evidence that identified which patients needed quick treatment while not misclassifying the truly ill as not needing attention. These were a mixture of test results, demographic data, comorbidities, and other data about the patient, such as whether they live alone.
Dr. Jenny presented her preliminary findings at the Summer Institute on Bounded Rationality, hosted by the Planck Institute in June, explaining that a set of four questions identified by a much simpler computer algorithm was able to classify patients almost as reliably as the more complicated program.
It's called a “fast and frugal decision tree,” she explained. “They are mostly studied by psychologists — increasingly also by economists and bioinformaticians — and have been applied to a diverse set of problems, such as the detection of depression, the detection of suicide attackers, and [to identify fragile banks in] the financial sector.”
The preliminary four-question decision tree for patients with vague complaints asks:
- Does the patient look ill?
- Does the patient have a Charlson Comorbidity Index of five or higher?
- Is the patient at least 65 years old?
- Is the patient male?
A patient is identified as “highly morbid” if the answer is yes to the first question and any one of the other three questions.
Dr. Jenny stressed that the tool is preliminary. “For further analyses, we are using more computationally powerful tools and expect new results that might differ from these, so whether we will find the same 14 predictors with our more thorough analysis is not yet clear,” she said.
“Looking ill,” which is likely to remain a linchpin of the decision tree going forward, could be problematic, said David Newman-Toker, MD, an associate professor of neurology at the Johns Hopkins University School of Medicine. “Decisions will likely vary with the experience and risk tolerance of the observer. This makes it hard to draw generalizable inferences outside of the center where it was developed, and maybe even within. They would need to do inter-rater reliability statistics.”
Nonetheless, he said, a tool that could accurately and reliably segregate the “sick” from the “not sick” among ED patients who are neither obviously sick (such as bleeding out from a massive femur fracture) nor obviously well (complaining of a small laceration on the hand) would be of benefit in the ED.
“For this approach to be used, authors of this new tool will need to rigorously demonstrate that their approach represents an accurate, reliable improvement over existing triage scales,” he said.
And Dr. Newman-Toker remained skeptical of the idea that any sort of prevalence-based approach will generalize well. “Just as one example, here at Johns Hopkins, endocarditis is about 1,000 times more common among our emergency patients than at Sinai Hospital, 20 minutes away, because we draw from a heavily drug-addicted population. Those young patients might present with nonspecific complaints like fatigue or malaise. There are variations like these even within a single city, so I'm not sure prevalence-based approaches to stratification can be used in a general way that will contain costs without reducing quality.”
Dr. Jenny said fast and frugal decision trees can be tailored to specific hospital populations because they are so simple and flexible. “In short, they can be adapted to a specific hospital and its patient population to achieve increased stratification accuracy,” she said.
Access the linksin EMN by reading this on our website or in our free iPad app, both available atwww.EM-News.com. Comments?Write to us email@example.com.