Intelligent Alarms and Putting Data into Context for Care at Home
By Adam Wolfberg, M.D., Chief Medical Officer, Current Health
Every clinician has had this experience: one of the panels on the vitals monitoring screen at the nurses’ station starts to blink, indicating that something is wrong. A patient’s oxygen saturation is in the 80s. You go to the door of the patient’s room and they are asleep, breathing easily, normal color in their face. Wrapped in sheets, the pulse oximeter is askew.
Part of clinical education and practice is placing data in context. We learn to do it almost without thinking during our training, and it becomes second nature. Blood pressure 164/84 in a known hypertensive patient and you trust the value; the same reading in a previously normotensive patient, and you check the cuff size, and confirm she was seated with her legs uncrossed. You repeat the measurement in the second patient because a single outlier measurement must be placed in context, reviewed, and repeated—before it can be believed. When the heart rate begins to climb in an ICU patient? Wait for 60 seconds to see if the increase represents heart rate variability or a trend. Is there a benign explanation, such as the patient suddenly becoming agitated?
Nothing replaces the context provided by a walk to the foot of the bed where the clinician can lay eyes on the patient, and if necessary, have a conversation or conduct an examination: Is his breathing labored? Can he speak normally? How is his color? Can he get out of bed? This basic ‘reality check,’ conveys a wealth of information to an experienced clinician.
What about the monitored patient at home?
When a patient is receiving care while at home, you can’t go to the foot of the bed. You can’t repeat the measurement. You have to trust that the patient is appropriately positioned, and that they used the correct device in the correct way. A conversation requires a phone call, text message, or a video visit. A physical examination requires sending a clinician to the home.
How clinicians adapt
At Current Health, we find that our clinical colleagues develop a new set of contextual sign posts to help them identify the sick patients, without overwhelming others with unnecessary outreach. Remote, continuous monitoring has a learning curve that necessitates a change in mindset for the clinician used to reviewing spot observations in hospital.
Clinicians learn to interpret the continuous vital sign data from Current Health in the context of the patient’s clinical condition and their life circumstances, keeping in mind how active they are, how frequently they leave the home, and even how strong network connectivity is in their neighborhood. The patient who goes to the grocery store each day generates a different level of concern when vital sign data stop streaming, compared to the patient who is unable to leave their home.
Configuring alarms for sensitivity and specificity
Alarms play an outsized role in the safety of monitoring patients in their homes, both because clinicians frequently are not watching vital signs in real-time, and because, when an alarm fires, there is no opportunity to contextualize the patient in person.
Patients in the home are usually more mobile than those in hospital, and have to undertake more of their activities of daily living themselves. These additional physical challenges, coupled with postural changes and natural circadian variation, lead to a broader range of observations than would be seen in a patient sitting in a hospital chair. Monitoring continuously, rather than intermittently, also reveals incidental findings, such as nocturnal sleep apnea. Alarm settings need to account for these sources of variability.
Managing Clinical Risk
Finally, alarms are also higher stakes when the patient is in the home, because alarm triage typically requires a telephone or video call, and overnight, that means waking the patient up.
Given the high stakes of alarms and the additional variability in physiology, dialing in sensitivity (making sure the alarm fires if the patient is deteriorating) while optimizing specificity (making sure the alarm doesn’t fire if the patient is fine) is particularly important.
In our experience working with health systems, we find that alarms are invariably optimized for sensitivity in the beginning, and then over time, as the disruption of false-positive alarms sets in, clinicians (and their patients) increasingly value specificity (and often accept the recommendations that the Current Health team made in the first place). Accepting this balance of sensitivity and specificity is at the heart of the learning curve in remote patient care.
Gathering Context for Alarms
Over time, we find that clinicians get to know the daily habits of their patients, and their vital sign trends during the day and overnight. Current Health technology provides context in our dashboards, alerting clinicians to periods of time when the patient is active or the quality of the data coming through is poor, allowing the clinician to take the available data in context. We also help reduce the barriers to gaining additional clinical context by providing in-app messaging and one-click video calls.
Our most experienced colleagues are adjusting the alarm settings on outlying patients, reviewing signal quality, tracking activity, and understanding trends. In doing so, they balance sensitivity, specificity, clinical risk and capacity for response in these adjustments. Over time, providers develop that same automatic approach to placing data in context as they did during their internships.
Want to learn more about designing intelligent alarms for care at home? Check out our on-demand webinar.
Responding to Real-Time Health Data