Can early warning systems predict (and prevent) cardiac arrest?
The increasingly data-saturated modern health care milieu has been catnip to technologists and statisticians. If only we could manage and analyze the data better, as this appealing narrative has it, we could improve health outcomes in the hospital.
Predictive modeling algorithms represent the apotheosis of this paradigm, offering hope to detect patients' impending deterioration and cardiac arrest at an early preventable moment, better than ordinary care or clinical judgment.
The latest is eCART, a scoring system using a secret sauce of 28 variables including labs, vital signs and other data. eCART was developed at the University of Chicago and its creator founded Chicago based tech firm Quant HC in hopes of marketing it throughout the world.
The Quant HC team has published troves of data analysis, recently purporting to show that among 3,789 patients transferred to the ICU, if eCART had been in place, it would have identified the need for ICU transfer sooner and with better outcomes. The longer that those sluggish humans waited to transfer the crashing patient after eCART would have sounded the alarm, the worse the patient did.
The authors have published impressive performance for the eCART algorithm when using tranches of retrospective data, i.e., patients who already went to the ICU, had cardiac arrest, or died. Setting their specificity of 90%, their model could have predicted half or more of impending cardiac arrest, more than a day before the events occurred.
What they don't say is how often eCART would sound a false alarm among the increasingly ill and continually unstable baseline patient population in U.S. hospitals.
They seem to think: probably a lot. As their team told the Chicago Tribune in an interview:
We have to balance our desire to want to save every single life with the need to not over-notify clinicians.
The way we’ve been thinking about it — how many nurses are there versus the number of patients? What would be a reasonable number of alarms for them to respond to?
Hmm. What do you say we start with "head exploding" and dial it back to just under "accelerated burnout," and we'll go from there?
It goes without saying, all preventable cardiac arrests should be prevented. But before allowing your hospitals to deploy eCART or any new warning system, physician and nursing leaders should insist on seeing the data on the false alarm rates in a real-world, real-time cohort.
There are patients with reversible illnesses who could return home to weeks, months, or years of a life they would want to live, but experience preventable cardiac arrest and death in the hospital. Every one of these events is a terrible tragedy that all health care teams must strive together to prevent.
But they are a fraction of the hundreds of thousands of patients lingering in and out of hospitals near the end of life, whose regular perturbations of vital signs and labs could easily overwhelm the discriminatory power of eCART or any other early warning system, and with it, the sanity of the health care teams responding to the alarms.
Deployment on medical ward patients would likely result in a large increase in use of intensive care resources and delivery of "palliative critical care" -- possibly well worth it for the lives saved, but a discussion worth having before, rather than after your city's competing hospital administrators start their "me-too" rush to adopt.
The premise of these algorithms is seductive, and their use is probably inevitable. Today though, eCART and other early warning systems mainly serve to perpetuate the American myth that all death is preventable with the right technology.
Association between intensive care unit transfer delay and hospital mortality: A multicenter investigation. J Hosp Med. 2016 Nov;11(11):757-762.
Multicenter development and validation of a risk stratification tool for ward patients. Am J Respir Crit Care Med. 2014 Sep 15;190(6):649-55.
Quant HC tries to predict cardiac arrest with an algorithm. Chicago Tribune.