Home Coding Predicting death and other bad things

Predicting death and other bad things

Vital signs remain vital to hospital medicine

December 2017
Getty Images, Teekid

Published in the December 2017 issue of Today’s Hospitalist

OLD-TIMERS have been lamenting the death of the physical exam for decades. Confidence in stethoscopes continues to deteriorate in the era of iPhone EKGs and bedside echos. And let’s face it: Penlights and reflex hammers feel close to quackery when you can get a stat MRI or neurodiagnostics. This may be cynical to say, but you’d probably skip the physical exam on a lot of patients if it wasn’t required for billing.

From this perspective, vital signs are akin to phrenology. Medical use of thermometers began in the 1850s, and the data have questionable sensitivity and dismal specificity. Pulse and blood pressure are perhaps more useful, but they don’t hold a candle to cardiac biomarkers and telemetry. And respiratory rate? I’m always dubious of charted numbers when patients forever confound assessment by hiding under their covers and talking.

While individual symptoms and signs have limited utility, combinations can be extremely powerful.

But what if you don’t have the full resources of a hospital available to you? Or what if you’re concerned about cost and don’t want to scan everyone with vague neurologic symptoms, dyspnea or belly pain? Maybe we can learn something from correctional medicine.

Go to jail
Several months ago, I became the medical director of a county jail in central Minnesota. It’s a novel care environment with a familiar problem: excessive total cost of care. A lot of that spending is due to ambulance runs back and forth to the ED when inmates have potentially concerning symptoms.

Hospitals used to love that kind of volume, but not anymore. Most payer contracts have an upside carrot and downside stick. If you reduce total cost of care for your attributed lives, you get a chunk of the savings. But if costs remain static or continue to climb? Expect a financial hit. (Shared savings and penalties are also contingent on quality metrics, which prevents the financial shenanigans that riddled early HMOs.)

When the county contracted with the local health care system, the challenge was to deliver the right care at the right time in the right place. Easier said than done.

Clinical decision rules
I’ve been teaching evidence-based medicine at the University of Minnesota since 2004. I have a particular interest in evidence-based physical diagnosis; that is, the relative value of history and physical exam data for establishing and excluding disease. While individual symptoms and signs have limited utility, combinations can be extremely powerful. They’re often packaged into clinical decision rules that can lead to much more rational resource utilization.

For the jail, we rolled out clinical decision rules for specific complaints: chest pain, head injury, neck injury and so on. But I worried about more ambiguous circumstances. What if an inmate felt sick in the middle of the night when only corrections officers were around? How could we exclude a catastrophe without activating EMS?

Early warning scores
It turns out there is a rich literature on something called early warning scores. The one that we landed on, the National Early Warning Score (NEWS), was critically reviewed in the April 2013 issue of the journal Resuscitation. Professor Gary Smith and his colleagues wanted to see how NEWS stacked up against 33 other systems.

NEWS is decidedly low-tech, consisting of just seven variables, most of which can be obtained with automated devices.

In the case of respiratory rate, we’re using a capnometer in the jail, which also gives us oxygen saturation (SpO2) and end-tidal CO2. The latter isn’t part of NEWS but is a nice bonus. Only the level of consciousness contains a subjective element, but reliable assessment is easily learned.

Dr. Smith and his collaborators found that NEWS has a diagnostic accuracy of 89% and that it outperforms other systems. Moreover, patients with low NEWS scores have an exceedingly low risk of death (read: high negative predictive value). According to the model, one patient in roughly 2,300 will die in the next 24 hours, when the score is zero. Not impossible, but the odds overwhelmingly favor the house.

Note that Dr. Smith’s data were obtained from oldish hospitalized patients, who may have a higher probability of bad things occurring than other patients. NEWS likely overestimates the risk of death in jails, which is fine by me—but the numbers should be perfect for you. NEWS also contains data cuts for cardiac arrest and unplanned ICU admissions, although it doesn’t perform quite as well (72% and 86% accuracy, respectively). But I’d submit that’s still much better than clinical gestalt.

Clinical correlation recommended
Obviously, NEWS represents the beginning, not the end, of the patient assessment. You might have good reasons to alter care based on other clinical features. But if the story is bland and NEWS is low, you can probably stop worrying and move on to another patient.

David A. Frenz, MD, is a private practice physician and health care consultant. He was previously vice president and executive medical director for North Memorial Health in Robbinsdale, Minn. You can learn more about him and his work at www.davidfrenz.com or LinkedIn.

Leave a Comment

1 Comment on "Predicting death and other bad things"

Notify of
avatar
Sort by:   newest | oldest | most voted
S. Gregorio A. Cortes-Soto, M.D. (Linked In)
Guest
S. Gregorio A. Cortes-Soto, M.D. (Linked In)

Very good tool! Back to the basics.