The Time Is Now for Precision Patient Monitoring

June 2019Andrew Schwartz

Xiao Hu and Michele Pelter of the UCSF School of Nursing believe the era of precision patient monitoring is about to begin.

The two lead a multidisciplinary research collaborative – the Hu Research Lab and the ECG Monitoring Research Lab (Pelter Lab), clinical nurses and advanced practice nurses at UCSF Helen Diller Medical Center at Parnassus Heights and the School’s Center for Physiologic Research (CPR) – that has become a national leader in efforts to create bedside monitors that both eliminate the unnecessary signals responsible for alarm fatigue and predict life-threatening events before they occur. (Made possible by a generous gift from David Mortara, the CPR, according to its director, Fabio Badilini, aims to address alarm fatigue by establishing an annotated database from bedside physiologic data collected at UCSF intensive care units [ICUs] and to work with the Food and Drug Administration [FDA] to establish performance standards within its Medical Device Development Tools program.)

As Hu recently explained in a perspective paper in Digital Medicine, advances in machine learning, the increasingly sophisticated ability to read and translate signals from the body and the widespread adoption of electronic health records (EHRs) have created an unprecedented opportunity to make the vision of precision patient monitoring a reality. His team’s work on a “SuperAlarm” is proof of concept, having demonstrated the ability to achieve 90 percent sensitivity in predicting when an ICU patient will be in need of resuscitation.

Now, says Hu, “Rapidly testing and implementing these types of monitors must be guided by three principles, a four-part algorithm and close partnership with frontline clinicians, particularly nurses.”

Three Principles

In his Digital Medicine paper, Hu identified the three characteristics that must be present for the next generation of patient monitors:

  1. Precise: The monitors must reliably identify true clinical concerns and generate actionable alarms that alert clinicians in time to enable further diagnostic workups.
  2. Predictive: “It will no longer be enough to identify patient conditions while they are happening, which is often too late,” says Hu, citing today’s ventricular fibrillation alarms as an example.
  3. Interpretable: Clinicians must be able to trace the clinical and physiologic data responsible for the alarming condition and have it in a format that supports further diagnostic workup.

These principles, says Hu, are the touchstone for a four-part algorithm that researchers should use in devising the next generation of monitors.

Four-Part Algorithm

The first element in the algorithm assumes that today’s monitors are here for the foreseeable future and so must suppress false alarms more effectively than they do today. This is possible, say Pelter, because we’ve gained a much stronger understanding of which alarms serve a useful clinical purpose and which do not.

That understanding informed a project that Pelter – along with her former students Sarah Berger and Amy Larsen, clinical nurse specialists at UCSF Helen Diller Medical Center – recently led to determine how best to program new monitors in adult ICUs. They used research data from the Hu and Pelter labs to drive the evidence-based decisions for monitor setting.

Michele Pelter “For example, we’ve come to believe that even true alarms for some arrhythmias – such as accelerated ventricular rhythm (AVR) – can be a nuisance, because they may not be clinically actionable,” says Pelter. With that in mind, she and her team recently completed a study in which they systematically analyzed EHR data around the time of alarms in five adult ICUs and found that a third of audible electrocardiographic (ECG) alarms were for AVR, 95 percent of which were false. Going one step further, they examined the small number of true AVR alarms and found that none were clinically actionable or led to a code blue or death, which suggests this alarm could be turned off. The team published its findings in the American Journal of Critical Care; if verified in larger studies, these findings can play a significant role in reducing false and unnecessary alarms.

The second algorithm element involves designing new alarms within today’s monitors that incorporate a deeper analysis of physiologic data streams, including ECG and other hemodynamic signals. And, says Hu, “Because we are envisioning that our precise patient-monitoring algorithm will be hosted in a connected environment, we don’t need to limit ourselves to physiologic data. The new alarms can incorporate data from additional sources, in particular, the EHR.”

The third element of the algorithm combines the first two with another well-established algorithm – in principle, the same one used by Amazon to find items that customers tend to purchase together. This will enable the monitors to identify co-occurrence patterns that take place frequently before a clinical event of interest – and much less frequently among control patients. Once the monitor identifies such a “SuperAlarm pattern,” it can then identify ensuing data streams with matching patterns.

This, says Hu, is where the fourth element of the algorithm enters. Because a single pattern is not typically enough to trigger an actionable alarm, the monitors must recognize a sequence of patterns. At that point, says Hu, the monitor can issue a clinician alert that is precise, predictive, interpretable and actionable – assuming, that is, that all of the work up to this point has been informed by clinicians who understand what is clinically relevant and how the alarms figure into their workflow.

Making It Work in the Real World

The last piece is why Hu and Pelter believe it is essential to engage clinicians early on in the research process. In particular, they are fostering a series of partnerships between clinical nurses at UCSF Helen Diller Medical Center and faculty from the School to direct where the research needs to go – and to test and shape emerging advances.

“We have seen a series of promising innovations turn into disappointments when we don’t account for how clinicians use these monitors,” says Pelter.

Multidisciplinary patient-monitoring team (from left): Hu, Hannah Jang, Larsen, Jacob Abba (software engineer), Nate Tran (PhD degree student), Maximilian Vuong (lab assistant), Kelly Bushman One current partnership brings Hu and Pelter together with nurses in the neurological ICUs at UCSF Helen Diller Medical Center and the Institute for Nursing Excellence at UCSF Health to create something they hope will be a model for future projects. Backed by a Clinical Nurse Research grant – one of two new UCSF Health funding mechanisms aimed at fostering collaborative research among the School and UCSF Health – the team is conducting a study on intracranial pressure (ICP) monitoring.

ICP monitoring is a critical element in averting secondary injury after patients experience conditions such as a subarachnoid hemorrhage or traumatic brain injury. Nurses are primarily responsible for this time-intensive procedure, which involves closing an external ventricular drain (EVD) used to evacuate excess cerebrospinal fluid and blood, waiting for equilibrium, measuring the ICP and then reopening the EVD to ensure that excess fluid drains. A pilot study found that far too often, nurses record measurements before the proper waiting period transpires, so the team is using an interdisciplinary approach to design, develop and test a software intervention and clinical protocol to help ensure more accurate ICP readings.

While the project is still in its early stages, Hu and the bedside nurse leading the effort, Vaughn Mouton, have already identified how they will work together moving forward. “The nurse will review the current EVD and ICP measurement protocols in the context of the pilot study. He is also helping to design the prototype of the tool, and he will play the crucial role of soliciting feedback from clinical nurses at the bedside,” says Kelly Bushman, unit director of neurological ICUs at UCSF Helen Diller Medical Center.

Hannah Jang, associate chief nurse researcher and clinical inquiry manager at the Institute for Nursing Excellence, says the project exemplifies three important ingredients for making research successful in the clinical environment.

“The first is that we have built a culture where nurses are knowledgeable about how research can change clinical practice and care for the better,” she says. “The second ingredient is the support of our stakeholders, specifically nursing leadership and unit directors, and Kelly has been enormously supportive. And the third is funding, and that’s why we’ve instituted the grants we have this year, which aim to foster a deeper relationship with the School of Nursing, one where faculty can reach out to conduct research along their lines of interest and direct care nurses can reach out to the School for support on their projects.”

“The ICP project shows what’s possible with the right support and the right players,” says Bushman. “Ultimately, we are all working to improve patient care.”

Hu agrees and says the project is but one demonstration of why he is confident that precision patient monitoring has reached an inflection point.

“Our approach is a tested road map for rapidly transforming today’s patient monitors into precision patient monitors with reliable, real-time therapeutic decision support capacity,” he says. “SuperAlarm demonstrated proof of concept. Now we need to invest in reaching that next level, with our School and the medical center playing a leading role.”