At the AAMI 2014 conference, I had the opportunity to attend the breakfast symposium where Jim presented, Transforming Care in Non-ICU Settings through Disruptive Continuous Monitoring Technology. The following discussion centers on patient monitoring data analytics, pioneered by Sotera Wireless.
What is the value of data analytics applied to medical device alarms?
For many years caregivers have had to struggle under the weight of a large number of false and non-actionable alarms. The resulting cognitive overload often results in alarm fatigue. Sotera has determined that a very effective way to reduce non-actionable alarms is to optimize alarm default settings.
Before you can improve something, you must be able to measure it. Medical device manufacturers have always generated log files of patient data, alarms and other system data processed or generated by their patient monitoring systems. But this data was only used in product development, troubleshooting, and incident investigations. What is needed is to give clinicians access to this data – and tools to analyze the data – to reduce non-actionable alarms.
Presently, hospitals are forced to use a trial and error approach to alarm management, which means multiple iterations, which is not ideal. First, it takes a fair amount of labor. Secondly, it takes a long time because it’s an iterative approach. So that, in the absence of high fidelity analytics, customers experiment not knowing the consequences of their experiments, which means it potentially can move too far in one direction versus the other.
For example, if they make their alarms too sensitive, they open themselves up to more nuisance alarms. If they make their alarms less sensitive, then the specificity of detecting a patient who is truly deteriorating can be a patient safety concern. High fidelity alarm analytics fills that gap.
What do you mean by the term “high fidelity medical device data?”
High fidelity device data really means capturing all of the digitized information that the device is collecting at the origin. In the case of physiologic monitoring, that means the raw waveform. It also means all of the reduced data derived from the raw waveforms such as individual vital signs, like heart rate, respiration rate, pulse rate, SpO2, blood pressure, temperature and any alarms that occured. The reason high fidelity data is so important is that it allows retrospective simulations on that data and therefore avoids the iterative trial and error approach towards alarm management.
You mentioned simulating the results of alarm adjustments, how does that contrast with a conventional trial and error approach, and what’s the impact of your approach in clinical practice?
Well, there’s a significant difference between using high fidelity data analytics that do simulation versus taking an iterative approach, not that iterative approach is entirely bad, it just takes a long time, requires significant labor and time investment by the hospital.
If you have high fidelity data captured, this data represents the environment of use, and it represents the patient population of interest. Then you can take that high fidelity data and run “what-if scenarios” at different alarm configurations and see the difference in the number and types of alarms that are generated based on different alarm limit settings. This method avoids the iterative approach and the enormous time that it takes to do it.
How is the simulation actually done?
Sotera’s high fidelity analytics is an evidence-based approach to optimizing alarm settings. We upload de-identified high fidelity patient data into a secure private cloud. As of this date (late July, 2014) we have about 25,000 hours of data from the general care area across multiple care units, across multiple hospitals. By the end of 2015 we expect to exceed 100,000 hours of multi-parameter vital signs data.
We take the aggregate data and run large simulation scenarios in order to optimize the what-ifs, for the purpose of reducing false and nuisance alarms. Each new customer’s data is individually analyzed and compared to the ever-growing aggregate data. This comparison allows the customer to compare their results to the aggregate data. We have found this tool to be very effective in helping our customers rationalize their setting and to set expectations of alarm experience with ViSi Mobile before a broader adoption.
Considering less than 5% of alarms are clinically actionable, this tool allows the hospital to significantly reduce non-actionable events. We know within the aggregate data set there are no reported adverse events. But, we are not stopping here. Sotera is engaged in an IRB approved study to report on the types of actionable events that are identified by alarm signals. We hope to publish our findings next year.
Time stand-offs, where notification of a transient alarm is withheld for a predetermined period of time, has recently emerged as a key tool in reducing non-actionable alarms. How do time stand-offs work and what role do they play in reducing non-actionable alarms?
A time hold-off, or time delay, requires that whichever physiological parameter is violated stays in the alarm state for a persistent predetermined amount of time before an alarm is activated.
The human physiology is a wonderful system that often has temporary swings in physiology to compensate for a short-term condition. For example, the first time a patient ambulates after surgery places stress on their cardiovascular system. In response, we may see a transient change in heart rate and blood pressure. These changes may cause a true but non-actionable alarm. Likewise, patients recovering from anesthesia may experience short episodes of oxygen desaturation. These events are important to capture and display, but not necessarily cause an alarm condition because these alarms do not require an immediate intervention to avoid a harmful event. Time hold-offs provide a filter – the time delay – to help differentiate between those very short episodic changes and true harmful physiologic changes. Non-clinically actionable changes are filtered out of the alarm equation.
After a hospital completes an analysis of their high fidelity medical device data, what kinds of issues have emerged that have challenged these hospitals?
Before answering your question it is important to contrast ICU patients from non-ICU patients where ViSi Mobile is applied. In the ICU, the patient’s physiology is often being manipulated by drugs or external devices such as ventilators. In this environment of care clinicians are very concerned about very small deviations in physiology, and therefore, alarms are set to very sensitive levels. In the case of the general care area we have a very different alarm management challenge. The non-ICU patient is in a recovery period of their hospital stay. They are receiving medications that help them recover. They ambulate as part of the recovery process. We see from our high fidelity data that they occasionally have transient episodes of physiologic stress. What we are finding is that we can address the non-actionable alarm and alarm fatigue issue through this high fidelity data analytics.
What has surfaced in our early deployments of ViSi Mobile are people and process issues within the general care area. Our biggest challenge is partnering with our clinical customers in improving their clinical thinking skills in interpreting data – data that has a different context from higher acuity monitoring environments, and data that is new to the lower acuity general care areas.
For example, if a patient’s heart rate climbs above 160 beats per minute and we get an alarm, what does the nurse do at the bedside to correct that? It could be the patient is experiencing anxiety, or they forgot to disclose a medication they were taking at home prior to admission. Or, is this change an indication of the beginning of deterioration? So our focus really is in the area that we have termed transformation of care at the bedside where we are investing in the training of nurses to respond to alarms in a meaningful way, especially the actionable events.
So along with new data about their patients comes an increased need to be able to respond appropriately to that data?
Yes. So let me give you a couple of examples. What we’re finding is that ViSi Mobile is a disruptive technology to the non-ICU patient care area. The general care nurse is not accustomed to receiving real time physiologic information. So they’re discovering for the first time that their patients are experiencing the early stages of a harmful condition more often than they realize.
Sotera has discovered that we must first overcome the natural human element of denial. How could our patients have this many alarms or this many physiologic conditions that require our nurses’ response at the bedside? We have to overcome that barrier through training and investment in their day-to-day operation. And that often comes to working directly at the policy level within the nursing community. Let me give you an example of that.
It is very typical for nurses in the general care floor not to have within their scope of practice the ability to change alarm limits on a patient without a physician order. If you’ve ever worked in the general care floor, you’ll know that the nurses are very reluctant to call physicians for these kinds of permissions. So what typically happens is you get a few patients that are alarming all the time, and the nurses are reluctant to get a physician to write an order to change alarm limits.
As a result, we frequent engage our clinical customers in discussing policy issues that allow an extension of current scope of practice to allow clinical interventions (including changing alarm limits) within a limit defined by senior clinical leadership. In essence, we are empowering each nurse to intervene sooner to a deteriorating patient condition.
What’s the relative value of a device manufacturer’s own alarm analytics solution, like Sotera’s, and a patient-centric alarm analytic solution that accounts for all the devices attached to patients from a third party like a messaging middleware vendor?
Well, clearly from a workflow standpoint, the environment of care is more than just physiologic alarms. There are out of bed alarms, nurse call alarms, stat results from laboratories, and so forth. The true solution to the overall nuisance alarm problem really involves a new technology ecosystem that includes not only the individual devices and their alarm management at the source of the alarm, but also the integration of that information with other contextual information about the patient.
So, does a hospital need both kinds of analytics tools? Or is one better than the other?
In my opinion, it’s not an either or proposition, both compliment one another. Solving alarm fatigue requires strengthening each link in the system chain, starting from the choice of sensors and continuing all the way to how the nurse receives alarm information.
I think the device manufacturers are obligated to do whatever they can to strengthen their algorithms, to help customers analyze their device data to identify truly actionable events. Then the messaging middleware system has to take that data, combine it with other contextual data like demographics, admitting diagnosis, drug medications, comorbidities and consolidate all this information to create a higher level of decision support, such that nurses are only getting information that they have to act on, in a timely way to avoid harm.
Alarm settings are a key part of the clinical practice of alarms and a major contributor to minimizing non-actionable alarms. Once the hospital has gotten a handle on that, what other factors in effective alarm management must be addressed?
The answer to that question comes down to people, process and technology. So, through our alarm analytics and simulations we’re solving the technology component where only actionable information moves into the messaging or notification system. The next challenge is how do we use that information to cause sustainable behavioral and/or process changes within the institution. Our experience has shown us that the bigger elephant in the room is the investment in the critical thinking skills of the nurses at bedside.
It has been all too often in my career as a Clinical Engineer that hospitals will purchase a system with the expectation that it’s the technology that is the solution to nuisance alarm. That’s not entirely true. Technology plays a very important role in solving alarm fatigue, but if the hospital doesn’t invest in the training programs, the policy changes, the cultural changes, the process changes, at the bedside then any new technology in my opinion will be very short lived.
More often than not, hospitals buy these types of solutions only to abandon them later, because they’re not getting the improvement and outcomes in patients by whatever metric they decide. It is because they haven’t adequately invested in the process change, the policy change required to realize the full potential of the technology. In 2010 Dartmouth Hitchcock Hospital published a remarkable reduction in ICU transfers due to a multidisciplinary approach to early detection and interventions. They invested in people, process, and technology. Since implementation they have reported no unanticipated cardiopulmonary arrests, clearly an improvement in outcomes. Yet, no other institution has achieved similar results. Why? I can only conclude it was due to a deliberate improvement in the culture of care at Dartmouth that was enabled by a new technology.
We recently submitted an article that talks about a capability maturity model for organizations to address alarm management from a foundation to a sustainable level. It has been my experience that if you don’t go through those process changes and make those investments, then the hospital will struggle with realizing a sustainable solution.
Pictured above is the ViSi Mobile monitor.