Sub-Total: €0.00
Assessment of Remote Vital Sign Monitoring and Alarms in a Real-World Healthcare at Home Dataset
Nicole Zahradka 1,*,†, Sophie Geoghan 1,†, Hope Watson 1, Eli Goldberg 1, Adam Wolfberg 1, Matt Wilkes 2
Editor: Luca Mesin
- Author information
- Article notes
- Copyright and License information
PMCID: PMC9854741 PMID: 36671610
Abstract
The importance of vital sign monitoring to detect deterioration increases during healthcare at home. Continuous monitoring with wearables increases assessment frequency but may create information overload for clinicians. The goal of this work was to demonstrate the impact of vital sign observation frequency and alarm settings on alarms in a real-world dataset. Vital signs were collected from 76 patients admitted to healthcare at home programs using the Current Health (CH) platform; its wearable continuously measured respiratory rate (RR), pulse rate (PR), and oxygen saturation (SpO2). Total alarms, alarm rate, patient rate, and detection time were calculated for three alarm rulesets to detect changes in SpO2, PR, and RR under four vital sign observation frequencies and four window sizes for the alarm algorithms’ median filter. Total alarms ranged from 65 to 3113. The alarm rate and early detection increased with the observation frequency for all alarm rulesets. Median filter windows reduced alarms triggered by normal fluctuations in vital signs without compromising the granularity of time between assessments. Frequent assessments enabled with continuous monitoring support early intervention but need to pair with settings that balance sensitivity, specificity, clinical risk, and provider capacity to respond when a patient is home to minimize clinician burden.
Keywords: remote monitoring, alarm, vital sign, hospital at home, wearable
1. Introduction
Healthcare and Hospital at Home (HaH) programs have become popular during the COVID-19 pandemic, as hospitals exceeded their inpatient capacity, and the risk associated with in-person care increased [1]. Improvements in technology, such as medical-grade wearables and HIPAA-compliant communication platforms, alongside a clinical imperative to change practice, have made it possible to deliver acute care in remote settings [2]. Monitoring a patient’s overall status, including vital signs, is a standard of care in hospital settings to detect deterioration, facilitate intervention, and avoid adverse events. In a remote setting, a patient is less easily reached, so false alarms are potentially more disruptive and expensive. There is a premium on context and specificity. As we move care out of the brick-and-mortar hospital setting, and into patients’ homes, we must ask the question, is there a direct translation of vital signs and vital sign alarm settings between in-hospital and remote monitoring? Or do we need to develop a new paradigm of alerting that is more suited to this new environment?
One of the first indicators of clinical deterioration is a change in physiological status [3]. In the hospital, a patient’s overall status is assessed with a combination of objective and subjective information collected through vital sign measurements, reported symptoms, and clinical observation [4]. At-risk patients are identified with physiological track and trigger systems (PTTS) that use algorithms to assess vital signs. These algorithms range in sophistication from fixed individual vital sign thresholds, [5,6] to adaptive thresholds, [7] to summarizing multiple vital sign measurements and observations into one metric, such as early warning scores (EWS) [8]. PTTS allow clinicians to standardize assessments and responses to acute illness. The National Early Warning Score (NEWS), for example, is consistently used throughout the National Health Service in the United Kingdom (UK), and the latest iteration (NEWS2) incorporates respiratory rate, oxygen saturation, systolic blood pressure, pulse rate, level of consciousness or new confusion, and temperature. An aggregated score of 5–6 is a medium clinical risk and a key threshold for urgent response, while an aggregated score of 7 is a high clinical risk and escalated to an urgent or emergency response [9].
Assessments of physiological status are driven by the frequency of vital sign measurements, which are typically collected every 8 to 24 h by the clinical staff in a general unit [10]. Large gaps in time between vital sign measurements allow for clinical deteriorations to go undetected and may result in adverse clinical outcomes [3]. The low frequency of assessments may be compounded by incomplete sets of vital signs resulting from clinician selection of vital signs measured [11]. A previous study reported that as little as 21% of the 229 vital sign-related interactions between nurse and patient involved a full set of vital sign measurements [11]. Intermittent vital sign measurements generate sporadic information that is not consistently assessed, recorded, interpreted, or actioned [4,12,13]. Nurses agreed that continuous vital sign monitoring would enhance patient safety in the general ward [14,15]. When vital signs are measured by a nurse, an artifact may be introduced to the reading due to patient engagement; during manual observations, patients typically wake, sit upright, and remain still, resulting in vital sign measurements that may not be representative of their physiological status during activities of daily living.
The likelihood of early identification of a clinically significant change in patient status increases as vital signs are measured more frequently [16]. Hospital units where patients are likely to be medically unstable, such as intensive care, high dependency, and post-anesthesia care units, use continuous monitoring systems that often include invasive metrics, such as arterial or central venous blood pressure, with alarm settings that are highly sensitive to acute changes. The number of alarms reported for a patient ranged from 6.5–45.5 per hour on an ICU [17,18] and these units typically have a 1:1 or 1:2 nurse-to-patient ratio. A high nurse-to-patient ratio is required to be able to respond to alarms and intervene appropriately when alarms have high sensitivity. Patients admitted to these units spend most of their time stationary, as mobility is typically limited, either by pathology or equipment. Continuous monitoring is therefore comparably straightforward, as limited motion means the quality of continuous vital sign measurements is better, while tethered equipment is not as much of a hindrance as in other hospital units, where patients are encouraged to move more frequently as part of recovery [19].
Advances in wearable technology have created an opportunity for continuous monitoring to exist outside of high acuity [20] and hospital settings. This brings benefits but can also result in information overload [21]. New equipment, alarm settings, and the interpretation of increased vital sign measurements may end up being perceived as more of a burden than a benefit without proper training. Clinicians may not always recognize deterioration, but successful incorporation of continuous vital sign monitoring into decision making requires an understanding of its strengths and limitations—it is a new paradigm of monitoring, not simply an increase in the rate of intermittent observations [14]. Pairing alarms and EWSs with continuous monitoring may help clinicians recognize deterioration, but the tradeoff is an increased likelihood of false alerting and potential alarm fatigue, especially if alarm settings are not selected with the new context in mind. Actionable alarms are already a low percent (20–36%) of the total number of alarms triggered in adult ward settings [22]. The percent of unactionable alarms is likely to be higher when alarm settings traditionally used in hospitals are applied to vital sign measurements that are collected in a less controlled environment and do not account for factors such as physiological variability, activity, adherence to wearing the device, and measurement accuracy. Vital signs collected through wearables are susceptible to motion artifacts, which decrease the signal-to-noise ratio and can impact accuracy [23].
When implementing vital sign monitoring in HaH programs, clinicians are responsible for decisions that have tradeoffs between the risk of delayed/missed deterioration and alarm fatigue. A high frequency of vital sign measurements without appropriate alarm settings leads to a lot of data without actionable information (or too much actionable information), while a low frequency of vital sign measurements, regardless of the alarm settings, leads to late or missed deterioration. This balance becomes even more important during remote monitoring because care teams rely on patients to wear their devices correctly, and the subjective information collected through clinical observation is not as readily attainable as when a patient is in a hospital room. There is a limited amount of evidence on vital sign collection frequency and alarm recommendations for use in a HaH program. The purpose of this work is to evaluate the effects of vital sign observation rates and alarm settings on alarms using a method to simulate alarms in a real-world dataset collected remotely in a clinical setting. Alarm metrics under different simulated vital sign observation rates and alarm setting conditions demonstrate what would be observed and highlight the impact of the decisions a clinician makes when setting these parameters.
2. Materials and Methods
2.1. Current Health (CH) Platform
Current Health is a system that supports the remote delivery of care to patients in programs such as healthcare at home. The FDA 510(k)-cleared platform includes an upper-arm wearable that continuously monitors pulse rate (PR), oxygen saturation (SpO2), and respiratory rate (RR). Additional parameters, such as “Motion Level,” “Perfusion Quality,” and “Wearable-On-Arm,” are derived from the sensor signals in the CH wearable to provide context to the healthcare provider. Pairing vital signs (PR, SpO2, and RR) with movement and patient adherence to wearing the CH wearable offers some compensation for the clinical observation that is unavailable in a remote setting. These parameters are also used to improve the quality of vital sign observations by excluding those collected during unstable conditions, such as high levels of patient movement or incorrect wear. When using the Current Health Generation 2 (Gen2) wearable sensor, the CH platform outputs observations for PR, SpO2, and RR at rates of 30, 30, and 15 observations per minute, respectively. This generates 43,200 PR and SpO2 observations, and 21,600 RR observations every 24 h per patient. The CH platform uses a rolling median with an aggregation window (AW), the window of time the median was calculated over, and a minimum number of observations within AW to reduce variability in continuously collected vital sign observations [24]. The minimum number of observations was set to 20% of the expected number of observations for AW, determined by the observation rate (Table A1). The CH platform also integrates with peripheral devices to collect blood pressure, axillary temperature, lung function measures, weight, and patient-reported outcomes delivery via tablet; however, these data were excluded for dataset completeness, as not all HaH programs used peripheral devices.
2.2. HaH Program Dataset
Data from six HaH programs using the CH platform were screened for eligible patients. Inclusion criteria were HaH admission > 24 h, use of Gen2 wearable, and Gen2 wear time > 24 h. Exclusion criteria were multiple CH platform admissions, and test patients identified by “test” in first/last name, invalid health service ID, or invalid age (<20 or >130 years).
Seventy-six HaH patients with a variety of conditions who were admitted to the CH platform between April 21 and May 15 and discharged before 31 May 2021, were available to be included in the dataset. Patient demographics were limited to data available on the CH platform. The patients were 60 ± 16 years old (n = 42), 14 male and 11 female. The reported ethnicities were “Caucasian”: 17; “African American”: 4; “Southeast Asian”: 1; and “Other”: 3. Gender and ethnicity were not reported in 51 patients. Each patient’s dataset was composed of CH platform timestamps, such as CH platform admission timestamp (Ta) and CH platform discharge timestamp (Td); PR, SpO2, and RR values; and PR, SpO2, and RR observation timestamps, such as initial vital sign observation timestamp (Tvs_i) and final vital sign observation timestamp (Tvs_f).
2.3. Vital Sign Observations
The vital sign observation dataset (VSOD) was smoothed with a rolling median (AW = 5 min, Table A1), akin to the platform’s deployment in clinical practice. VSSD was used to create datasets with observations every 15 min (VS15), 1 h (VS1) [25], 4 h (VS4) [26,27], and 12 h (VS12) [9] to simulate vital sign observation frequencies that are common in hospital settings across the acuity spectrum (from the operating room and intensive care to general wards and ambulatory clinics). Observation time was factored into being as representative of in-person measures as possible. To simulate in-person measures, observations from VSSD were downsampled to every 12 h, 4 h, 1 h, and 15 min starting at 6am (Figure S1). The nearest observation was used when data were missing from the minute mark up to 30 min before or after the minute mark. For example, a vital sign observation from 6:05 am would be used for the 6am downsample time in the absence of any observations between 5:55 and 6:04 am.
2.4. Vital Sign Alarms
The physiological track and trigger component of the Current Health platform is designed to support algorithm customization. The trigger-based vital sign alarms are driven by a ruleset that contains vital sign rules; rules include vital sign threshold(s), logic statements, and an aggregation window. The ruleset can be tailored to meet the needs of the use case, taking into account the patient population, planned interventions, expected clinical course, physical distance and response times, and staffing capacity, and it is established prior to CH platform deployment. Rulesets may then be modified based on subsequent experience; vital sign thresholds may be modified at the patient level. The rulesets used in this evaluation were created to identify changes in SpO2 (hypoxia), changes in PR (tachycardia, bradycardia), and changes in RR (tachypnea, bradypnea). These changes are indicators of deterioration in a broad range of patients.
Vital sign alarms are affected by vital sign observation frequency, and alarm rules, thresholds, and aggregation windows. To compare the alarm output of different alarm parameters with each other and with in-person clinical monitoring (vital sign observation conditions), we replicated the alarm system so that previously collected patient data could be passed through retroactively (Figure S2). Three vital sign alarm rulesets were evaluated: a subset of NEWS2 rules (A1) [9], individual vital sign rules (A2) [5,6], and one primarily designed with combination rules (A3). Table 1 outlines the rules, including vital sign thresholds and combination rules, for each of the three rulesets. Each vital sign alarm ruleset was tested on VS15, VS1, VS4, and VS12 with an aggregation window set to 0. As described earlier, an aggregation window indicated how many data points to use when smoothing the vital sign dataset (VSOD) (Figure S1). Four aggregation window conditions were tested for each vital sign alarm ruleset: 5 min (VSSD), 15 min (AW15), 1 h (AW1), and 4 h (AW4). A timestamp log was generated for each test condition when the vital sign dataset was run through an alarm simulator (Python Software Foundation. Python Language Reference, version 3.8. Available at http://www.python.org (accessed on 10 November 2021) to indicate when an alarm ruleset condition was met. Timestamps were generated for each rule and then grouped for the vital sign alarm ruleset. Where the rulesets included logic statements with multiple vital signs, an overlap window of 30 s was used, so the vital signs needed to have breached their thresholds within 30 s of one another to trigger the alarm.
Read more here.
Article by, PubMed Central.
