A search of the Medicines and Healthcare Products Regulatory Agency website revealed no manufacturer Field Safety Notices or Medical Device Alerts for this device. No reports of adverse events were identified from a search of the US Food and Drug Administration (FDA) database: Manufacturer and User Device Facility Experience (MAUDE).
Fifteen relevant studies were identified, which took the form of journal publications, technical reports, conference abstracts and posters. Five of these were excluded because they were either out of scope for this briefing, were editorial or comment articles, or no results were presented. Four of the studies were excluded because they repeated data already found in the selected studies presented below. In these cases, peer‑reviewed journal publications, more recent publications and studies with larger cohort sizes (in ongoing studies with updates) were selected for inclusion. As a result, 6 studies are summarised in the briefing, of which 2 were available as abstracts and 1 was a company internal report. All included studies are tabulated in the appendix.
The study by Ben‑Ari (2010; table 1) compared the EarlySense system in both sleep laboratory and ICU settings in Israel with standard monitoring systems for each setting.
In the sleep laboratory, the standard comparative measuring device for respiratory rate (RR) was the Embla N7000 with Somnologica Studio Software System (Embla Systems Iceland). The standard measuring system for heart rate (HR) was the Embla Sleep Lab System. Patients included 16 adults (8 men, 8 women) and 41 children (32 boys, 9 girls). For HR, the EarlySense accuracy was 91.5% in children (5% absolute relative error [aRE] ±3%) and 94.4% in adults (3% aRE±3%) when compared with the HR data points collected by the standard measuring systems. For RR, accuracy was 91.8% in children (4% aRE±2%) and 93.1% in adults (4% aRE±1%) compared with standard RR measurement systems.
In the ICU setting, the standard method for HR was ECG monitoring (Datex/Ohmeda, GE Medical). Measurement of respiration for ventilated patients was done by end‑tidal CO2 (ET CO2) module, or was measured manually by trained research assistants. The study enrolled 42 adult patients (25 men, 17 women). The EarlySense system had 94% accuracy (3% aRE±0.3%) compared with the HR data points collected by the standard monitoring system, and an 82% accuracy (7% aRE±6%) for RR against ET CO2. EarlySense had 75% accuracy (8% aRE±8%) compared with RR manually monitored by research assistants.
The conference abstract by Sorkine (2008) describes an accuracy evaluation of the EarlySense system against standard ICU monitoring systems. Thirty‑eight critically ill patients (23 men and 15 women, aged 16−87 years) were recruited and simultaneously monitored with the EarlySense and the ICU's standard of care of ECG for HR and electrical plethysmography or ET CO2 for RR. Trained technicians also manually measured RR. The EarlySense system had a HR accuracy of 92.1% (aRE 3.6%). RR accuracy compared with ET CO2 and manual RR counts was 79.4% (aRE 9.6%) and 80.1% (aRE 12.6%) respectively.
The conference abstract by Frendl et al. (2013) describes an observational study to assess respiratory patterns, which may be used to predict respiratory failure. In the study, 37 adult critically ill surgical patients were continuously monitored while intubated or mechanically ventilated. Patients were retrospectively studied in 2 groups: those who were successfully extubated and those with failed extubation. EarlySense alerts were considered to be true positives if they were followed by a major clinical event within 24 hours. After extubation, abnormal respiratory patterns identified patients likely to need additional ventilatory support with 90% specificity, 50% sensitivity (positive predictive value [PPV]: 60%, negative predictive value [NPV]: 86%). Abnormal respiratory patterns prior to extubation showed 91% specificity, 44% sensitivity, a PPV of 78% and a NPV of 71%, where abnormal respiratory patterns are predictive of a failed extubation. The authors concluded that the EarlySense system can indicate which ventilated patients are not yet ready for extubation.
The Zimlichman (2012) study aimed to define cut‑off points for alarms using the EarlySense system, and assessed the predictive value of the alerts to detect clinical deterioration in a general (internal) medicine ward setting. The study enrolled 149 patients and data from patients with at least 30 hours of monitoring were used for the analysis (n=113). Major clinical events (defined as ICU transfer, intubation or cardiac arrest) were recorded, and the patients were constantly monitored with the EarlySense system. RR and HR alerts were based on the vital signs recorded and analysed retrospectively, and were set on a threshold basis. Trend analysis used grouped HR and RR readings for 6‑hour periods throughout the day with data collected every 3 minutes. The study authors compared the median of the readings for each period with the corresponding period of the previous day. Retrospective analysis showed that the optimal cut‑offs for the threshold alerts were HR below 40 or above 115 beats/min, and RR below 8 or above 40 breaths/min. Only 6‑hour time windows with at least 420 valid RR or HR results were included in the analysis. Staff did not make clinical decisions based on the EarlySense readings. EarlySense alerts were considered to be true positives if they were followed by a major clinical event within 24 hours. For the trend alerts, when comparing between time periods, retrospective analysis showed that a rise of 20 or more beats/min and 5 or more breaths/min corresponded with a maximal sensitivity and specificity. Nine out of 113 patients had a major clinical event. For HR threshold alerts, the sensitivity was 82% and the specificity was 67%. For RR, threshold alerts had a sensitivity of 64% and the specificity was 81%. For trend alerts, HR sensitivity was 78% and specificity was 90%. For RR trend alerts, sensitivity was 100% and specificity was 64%. The authors concluded that the EarlySense system is able to continuously measure RR and HR with low alert frequency, and provide timely prediction of patient deterioration.
The Brown et al. (2014) study compared 2 similar medical‑surgical units in the USA, 1 which used the EarlySense system as a monitoring system on all beds and the other which served as a control using standard nurse‑led monitoring. Outcomes from both wards were assessed for 9 months before and after the EarlySense system was deployed in 1 of the wards. Outcomes included ICU transfers, length of stay (total and ICU only), code blue events (a US hospital code used to indicate a patient needing immediate resuscitation), and Acute Physiology and Chronic Health Evaluation (APACHE II; Knaus et al. 1985) score. The APACHE II score is calculated from a patient's age, blood oxygen saturation, temperature, mean arterial pressure, arterial pH, HR, RR, serum sodium and potassium, creatinine and haematocrit levels, white cell count and Glasgow Coma Scale score. There were no statistically significant differences in ICU transfers or APACHE II scores between the 2 wards after the EarlySense system was introduced. A three‑armed comparison showed a significant reduction in days in ICU in the intervention unit post‑implementation (63.5 days/1000 patients compared with 120.1 and 85.36 days/1000 patients, for the implementation and control unit before implementation respectively; p=0.04). Length of stay in ICU for the control unit actually increased from 32.69 days per 1000 patients before implementation to 85.36 days per 1000 patients after implementation (p=0.01). The EarlySense‑equipped intervention unit had 6.3 code blue events per 1000 patients before implementation, improving to 0.9 events per 1000 patients after implementation (p<0.01). The control unit had 3.9 events per 1000 patients before implementation compared to 2.1 events per 1000 patients after implementation (p=0.36).
A company internal report (Zimlichman et al. 2009) describes an unpublished study in which the EarlySense system was used to detect clinical deterioration in medical‑surgical wards, and the system's level of false alarms. A total of 204 adult patients (99 men and 105 women) were continuously monitored in 3 hospitals (1 in the USA, 2 in Israel) for over 14,000 cumulative hours. Worsening of clinical conditions which led to interventions occurred in 29 patients (14%), with a total of 35 events of deterioration. Signs of worsening were detected by monitoring heart or respiratory rates (alerts and trends) in 31 of these events (88.6%). Of the 35 total events, 11 were defined as major. The EarlySense system detected signs of worsening in 100% of these major events.
Eight ongoing, completed or in‑development trials on the EarlySense system were identified in the preparation of this briefing.
NCT00361426 in patients with chronic obstructive pulmonary disease. Status: unknown.
NCT00382746 in patients with congestive heart failure. Status: completed.
NCT00361608 in patients with type 1 diabetes. Status: completed.
NCT01978340 in patients with obesity, sleep apnoea, obstructive central apnoea, sleep disorders, poor quality sleep. Status: not yet recruiting.
NCT02036996 in patients with overweight, sleep disorders. Status: not yet recruiting.
NCT02318004 in patient with myocardial infarction, cardiac surgery. Status: not yet recruiting.
NCT00640718 in patients post‑operatively, patients needing patient‑controlled analgesia, patients with respiratory conditions, patients at risk of falls. Status: unknown.
NCT01774708 in sleep (prediction of bed exit and fall prevention). Status: active but not recruiting.
During 2013–14, basic and advanced cardiovascular support accounted for 904,093 support days in England, whereas people needing basic and advanced critical respiratory support totalled 766,632 support days (HSCIC 2015). Some support days for cardiovascular and respiratory care may have been incurred concurrently, because people often need multiple system support. For non‑specific critical care, NHS reference costs 2013–14 state a range between £696 (XC07Z – adult critical care, 0 organs supported) and £1947 (XC01Z – adult critical care, 6 or more organs supported), based on generalised 'best case' and 'worst case' ICU referral costs. Adopting an effective early warning system could reduce the number of ICU transfers, cardiac arrests and other life‑threatening emergencies through detecting clinical deterioration.
Adopting an automated monitoring system could change the way that non‑critical care wards are organised, in that clinicians may need to do fewer scheduled rounds in order to monitor patients. This would be balanced by the time spent assessing alarm triggers and performing additional tests in response to alerts.
Setting up the EarlySense may need additional infrastructure such as networked computer terminals and wireless networking.
The currently available evidence for the clinical effectiveness of the EarlySense system was limited in both quantity and quality, and comprised 3 peer‑reviewed journal articles, 3 conference poster abstracts and 1 technical report. These studies were situated either in the USA or Israel, and no large UK‑based comparative studies were identified. The results may not be generalisable to the UK setting, because the hospitals in Israel or the USA may not be organised in the same way. They also may not use the NEWS system, and may therefore have less robust current patient monitoring in place.
The Ben‑Ari et al. (2010) study employed the EarlySense system in a sleep laboratory setting, which is not the intended setting for the use of the device. The patients were not randomised, which may lead to bias. The EarlySense system was compared with standard monitoring systems in both the sleep laboratory and ICU setting, which is an appropriate comparison for HR and RR accuracy measurements. Patient numbers were quite low, with 67 sleep laboratory participants and 42 ICU patients enrolled. The ICU patients were highly variable in their age (16−86 years), which could lead to a high level of variance in the results. This is a basic accuracy study and so does not show the effect of the device on patient outcomes such as ICU transfers or critical events.
Sorkine et al. (2008) is a conference poster abstract describing an accuracy evaluation of the EarlySense system compared with standard ICU monitoring systems. As an abstract, it lacks detail in both methodology and results reporting. The sample size was small, with 38 patients recruited and no control group or randomisation. The EarlySense system was compared with standard monitoring systems in the ICU setting, which is an appropriate comparison for HR and RR accuracy measurements. This was a basic accuracy study and so did not explore the effect of the device on patient outcomes such as ICU transfers or critical events.
Frendl et al. (2013) is a conference poster abstract describing an observational study to assess respiratory patterns, which may indicate an upcoming respiratory failure event. As a conference abstract, it lacks detail in both methodology and results reporting. The sample size was small, with only 37 patients recruited. There was no randomisation, but patients were split into 2 groups: those who were successfully extubated and those who were not. Only 16 unsuccessful extubations were available for analysis. This study does assess the EarlySense system's ability to predict respiratory failure, which is a possible application of this device in the NHS.
The Brown et al. (2014) paper describes a comparative study with the EarlySense system set up in 1 ward and a 'similar' control ward without the system. Various patient outcomes in both wards were also compared in the 9 months before the EarlySense was introduced. The sample size was large, with 2314 patients monitored in the EarlySense ward and 5329 across the 3 control arms. The authors state that the patients were referred alternately to the EarlySense or non‑EarlySense ward, but this is not true randomisation and the study was not blinded. The authors also note that the alarm thresholds for the EarlySense system could be altered by nurses if the patient was frequently triggering alarms. This could lead to bias. There is also an issue with the baseline similarities of the 2 selected wards. The baseline length of stay in ICU for the control unit was 32.69 days per 1000 patients, and the ward that subsequently introduced the EarlySense system had a baseline of 120.11 days in ICU per 1000 patients (p=0.06). Although not calculated as statistically significant, this is a large difference in ICU length of stay between the 2 wards at baseline, and may have caused bias.
The Zimlichman et al. (2009) publication is an unpublished report that has not, therefore, been peer‑reviewed. The sample size was reasonable, with 204 patients monitored, but only 29 patients had events that needed intervention. The results were based on a requirement for clinicians to set the system's high and low HR and RR thresholds. Although this allows tailoring of the device to suit an individual's specific HR and RR, it also has the potential to introduce a high degree of variability (because 1 clinician's concept of what constitutes a high or low RR and HR may differ from that of another). There was no randomisation because all enrolled patients were monitored with the EarlySense system.
The Zimlichman et al. (2012) study aimed to define cut‑off points for alarms using the EarlySense system and assess the usefulness of the alerts to predict clinical deterioration in a general (internal) medicine ward setting. Of the 149 patients enrolled, data from only 113 were analysed because 30 hours of monitoring was needed for analysis. The study recruited patients at an increased risk for respiratory failure, so the results from this study may not be applicable to a general medical ward. There was no randomisation because all enrolled patients were monitored with the EarlySense system. This study comprised a prospective data collection period, followed by retrospective data analysis. Therefore, trend analysis was carried out post‑hoc and was modelled to provide the optimal sensitivity and specificity. The authors showed that the modelled cut‑off values could predict clinical events, but the analysis was done on the same patients from which the thresholds were modelled. In order to prove the validity of these findings, these modelled thresholds should have been prospectively tested on a new patient cohort.