28 Original article
Analysis of a data-fusion system for continuous vital sign monitoring in an emergency department Sarah J. Wilsona,c, David Wongb, Richard M. Pullingera, Rob Waya, David A. Cliftonb and Lionel Tarassenkob Objective The aim of the study was to evaluate the ability of a data-fusion patient status index (PSI) to detect patient deterioration in the emergency department (ED) in comparison with track-and-trigger (T&T). Materials and methods A single-centre observational cohort study was conducted in a medium-sized teaching hospital ED. Vital sign data and any documented T&T scores (paper T&T) were collected from adults attending the resuscitation room, majors or observation ward. For each set of vital signs, we retrospectively calculated T&T (eT&T). PSI was calculated retrospectively from the continuous vital sign data using a statistical model of normality. Clinical notes were examined to identify ‘escalation’ events, and the numbers of these escalations identified by paper T&T, eT&T and PSI were retrospectively calculated.
Conclusion Electronic data capture offers opportunities for increased detection of deteriorating patients in a busy clinical environment compared with paper charts. Sample size in this study is insufficient to determine which electronic method (eT&T or PSI) offers superior detection of the need for escalation. European Journal of Emergency Medicine 23:28–32 Copyright © 2016 Wolters Kluwer Health, Inc. All rights reserved. European Journal of Emergency Medicine 2016, 23:28–32 Keywords: detection of patient deterioration, early warning score, emergency department, patient status index, track-and-trigger, vital sign monitoring, vital signs a
Emergency Department, Oxford University Hospitals NHS Trust, Oxford, Department of Engineering Science, Institute of Biomedical Engineering, University of Oxford, Oxford and cEmergency Department, Heatherwood and Wexham Park Hospitals NHS Foundation Trust, Slough, UK b
Results Data from 472 patient episodes were examined. A total of 20 patients had PSI data at the time of an escalation related to vital sign abnormalities that occurred during their ED stay (vs. on arrival). Only four patient events were detected at the time by paper T&T. In all, 17 were detected retrospectively by eT&T and 15 by PSI. PSI had a calculated false-alert rate of 1.13 alerts/bed-day.
Correspondence to Sarah J. Wilson, MBChB, FCEM, c/o Tracey Pearson, John Radcliffe Hospital, Headley Way, Headington, Oxford OX3 9DU, UK Tel: + 44 1865 221173; fax: + 44 1865 222094; e-mail: [email protected]
Introduction
death and a low false-alert rate [2–4]. PSIs have not previously been evaluated in an ED.
Recognizing the deteriorating patient is key to preventing harm by correcting physiological abnormalities. However, identifying this group of patients is challenging, especially in a busy clinical environment. We have previously described [1] limitations of the paper-based track-and-trigger (T&T) system used to calculate early warning scores within an emergency department (ED) and highlighted the opportunity to use automated analysis of the outputs from bedside monitors. We now present further analysis from the same study, whereby data fusion technology is used to retrospectively calculate a patient status index (PSI) from continuous vital sign data collected from bedside monitors. In other settings (surgical step-down unit, high dependency area), prospective use of this approach has demonstrated appropriate detection of patient deterioration, a 40% reduction in critically unstable patients, reduced rate of unexpected
Supplemental digital content is available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal's website (www.euro-emergencymed.com). 0969-9546 Copyright © 2016 Wolters Kluwer Health, Inc. All rights reserved.
Received 30 December 2013 Accepted 10 April 2014
Objective The aim of the study was to evaluate the ability of a PSI to detect patient deterioration in the ED in comparison with documented T&T and retrospectively calculated T&T.
Materials and methods A single-centre observational cohort study was conducted in the ED of a medium-sized teaching hospital from January 2009 to January 2010 [1]. The study was approved by the UK National Research Ethics Service (ref: 08/H1307/56). We collected data from adults entering the resuscitation room, ‘majors’ and observation ward of the ED. Recruitment was restricted to times when members of the research team were available, typically daytime hours. Participants were excluded if they could not understand English or did not consent. Those unable to consent because of their clinical condition were followed up or had assent provided by next of kin or nominated consultee. DOI: 10.1097/MEJ.0000000000000166
Copyright r 2016 Wolters Kluwer Health, Inc. All rights reserved.
Data fusion for ED vital sign monitoring Wilson et al. 29
Hours of electronically acquired vital sign data for the study population
Table 1
Parameters HR RR SpO2 BP PSI
Number of hours of data 1626 1609 1643 1761 1904
The numbers of hours of available PSI data exceed that of other parameters, because it can be calculated from a partial set of vital sign data. BP, blood pressure; HR, heart rate; PSI, patient status index; RR, respiratory rate; SpO2, oxygen saturation.
collected for each parameter is shown in Table 1. The electronic data collected are incomplete. Reasons for this include lead disconnection and time spent away from the allocated bed space, for example during radiological investigations. Some data were also lost because of unexpected failure of the study data-collection server. The amount of PSI data is higher than for any single vital sign because it can be calculated from a partial set of vital signs (minimum three parameters).
Patient status index true alerts
As part of standard practice, vital sign and T&T data (paper T&T) were manually recorded at intervals onto paper observation charts by clinical staff (see Appendix, Supplemental digital content 1, http://links.lww.com/ EJEM/A79). Heart rate, blood pressure, respiratory rate (RR), oxygen saturation (SpO2), temperature and Glasgow Coma Scale (GCS) and paper T&T scores were collected retrospectively from observation charts. For each set of manually recorded vital signs, T&T was retrospectively calculated (eT&T). Continuous vital sign data (RR, heart rate, blood pressure, SpO2) were acquired using bedside monitors [Philips (Boeblingen, Germany)] and saved to a server. PSI was calculated retrospectively from the continuous vital sign data. The statistical model of normality used to calculate PSI (range 0.0–5.0) is derived from 3500 h of vital sign data recorded from hospital inpatients, as described fully elsewhere [5]. An alert would be generated for any PSI of at least 3.0 that persists for 4 out of 5 min. Patient deterioration was captured by recording escalations of care. An escalation was defined as any documented increase in the level of care associated with abnormal vital signs [1]. Two clinicians independently reviewed every set of clinical notes to identify escalations. A third clinician reconciled any discrepancies. Each escalation was further analysed independently by two clinical researchers to distinguish new deteriorations from persistent abnormalities following a previous escalation. This analysis considers only escalations that occurred after arrival in the ED. We excluded escalations occurring on arrival as they represent patients who were immediately recognized as unwell, and neither PSI nor T&T would offer any advantage. As a result, the number of escalations included in the analysis is relatively small. The outcome measure therefore was detection of escalation events associated with abnormal vital signs after arrival in the ED.
Results We collected T&T data from 472 patients. For 400 of these patients, we also collected continuous vital sign data over a period of 2169 h. The amount of data
We have only considered one event per patient to avoid analysing interdependent events. A total of 35 patients had an escalation after arrival in the ED; of them, 20 had PSI data around the time of the escalation. The remaining 15 patients were not considered further for the following reasons: (1) Seven patients were deemed to have ongoing conditions from arrival rather than a new deterioration (had been recognized as unwell at the time of arrival to ED and unlikely to benefit from a subsequent alert for the same parameters). (2) Two patients had no PSI data at all (one unexpected failure of software, one failure of third-party hardware). (3) Six patients had no PSI data at a time relevant to the escalation (two patients had moved to areas where there were no study monitors, four were only connected to study monitors after escalation). (4) Of the 20 patients with PSI data around the time of an escalation, four were detected by paper T&T, 17 were detected by eT&T and 15 were detected by PSI. In all, 12 events were detected by both eT&T and PSI. (5) PSI did not detect an event that eT&T detected in five patients. Four of these were predictable. (6) Two were for isolated pyrexia (temperature is not included in the PSI algorithm) and a further two were due to incomplete data such that PSI was unavailable at the time of the escalation. The remaining event was an episode of hypotension. (7) eT&T did not detect an event that PSI detected in three patients. One was an episode of brief fast atrial fibrillation documented in the text of the patient notes. The other two events were also documented in the patient notes but were not recorded on the corresponding observations chart. Figure 1 represents a patient for whom PSI would have alerted before T&T. The plot shows the electronically acquired vital signs, paper T&T, retrospective eT&T and PSI. T&T is displayed at times when observations were recorded on the paper chart. The grey-shaded block represents periods of time during which PSI would
Copyright r 2016 Wolters Kluwer Health, Inc. All rights reserved.
30 European Journal of Emergency Medicine 2016, Vol 23 No 1
Fig. 1
RR/BP/HR
SpO2
Patient: ED00570 100 95 90 200 190 180 170 160 150 140 130 120 110 100 90 80 70 60 50 40 30 20 10 0 T&T eT&T
SpO2 HR RR BP
0
1
4
6
1
0
1
4
6
1
PSI
5
0 14:00
15:00
16:00
Time A combined plot of the electronically acquired vital signs, paper T&T, eT&T and PSI for a patient who had a recorded escalation (denoted by solid vertical line). The grey-shaded block represents the period for which PSI generates an alert. BP, blood pressure; HR, heart rate; PSI, patient status index; RR, respiratory rate; SpO2, oxygen saturation; T&T, track-and-trigger.
generate an alert. The solid vertical line indicates the point at which an escalation was recorded.
Patient status index false alerts
To determine the false-alert rate for PSI alerts, we assumed that patients who did not have any documented increase in care (a wider criterion than ‘escalation’) did not warrant an alert, and therefore any alerts for this group are classified as false alerts. This will overestimate the false-alert rate, because we cannot be sure that all clinical deteriorations were documented. A total of 217 patients had no documented increase in care and had a cumulative length of stay in the ED of 43.5 days. There were 49 false alerts from 39 patients. The false-alert rate is thus 1.13 alerts/bed-day. A review of these ‘false’ alerts revealed that nine were likely to have been clinically relevant, which would further reduce the false-alert rate. Figure 2 shows the plot for one such alert, where the patient had abdominal pain requiring opiate analgesia. Further interpretation of the patient data suggests that the oxygen desaturation and hypotension at the time of the potential PSI alert should have prompted a patient review.
Discussion Each system for detecting patient deterioration has limitations. Paper T&T requires human input to accurately record the observations and assign scores to them. In a busy ED, events occur that are not documented. The resulting observations are only useful if they are recorded promptly and the resulting T&T score is calculated correctly. An electronically assisted T&T may eliminate these problems by automating the score calculation, ensuring that the observations are time stamped and can be configured to provide decision support. Both paper T&T and eT&T are limited by reliance on an imperfect scoring system. It is widely reported that most T&T scoring systems are not data driven [6,7]. However, the T&T score suffers from a more fundamental problem in that it reacts in a stepwise manner to continuous changes in patient physiology. For example, in our institution, a RR of 9 is ‘normal’ but a RR of 8 would trigger action to be taken (Appendix 1, Supplemental digital content 1, http://links.lww.com/ EJEM/A79). Unlike T&T systems, PSI is calculated using continuous data. A continuous electronic system should be able to detect deteriorating vital signs earlier than a system that
Copyright r 2016 Wolters Kluwer Health, Inc. All rights reserved.
Data fusion for ED vital sign monitoring Wilson et al. 31
Fig. 2
Patient: ED00188 SpO2
100 95 90 120 110 100 90
HR/RR/BP
80 70 60 50 40 30
SpO2 HR
20
RR BP
10 0 T&T eT&T
−1
−1
−1
3
1
0
PSI
5
0
21:00
22:00
23:00
00:00
Time
A combined plot of the electronically acquired vital signs, paper T&T, eT&T and PSI for a patient who had no recorded escalations. A PSI alert was generated at 22:15, which was initially classified as a false alert. However, further interpretation of the patient notes showed that the oxygen desaturation and hypotension at the time of the alert correspond to a period during which the patient reported abdominal pain. BP, blood pressure; HR, heart rate; PSI, patient status index; RR, respiratory rate; SpO2, oxygen saturation; T&T, track-and-trigger.
relies on documented intermittent observations. In addition, the PSI score is not limited to integer values, which means it can alert when some or all of the parameters begin to deviate from normality rather than when a single parameter becomes obviously abnormal. In theory, the ability to integrate data from multiple vital signs allows the system to increase the percentage of appropriate alerts. This sets it apart from standard bedside monitors, which use single-channel alerts. A primary disadvantage of PSI is its inability to automatically incorporate GCS. We considered adding GCS into the PSI calculations as it is part of standard T&T, but, because GCS must be assessed in person, it would not add value to the system. We have not included temperature in this PSI algorithm, although it would be possible to add if a consistently accurate temperature reading could be acquired electronically. The biggest barriers to the use of any electronic monitoring system are practical considerations including power failure, server failure, movement of patients or monitors within the ED or removal of monitoring leads
from the patient. The importance of these issues should not be underestimated. We are currently undertaking a larger study in the ED majors area looking at the feasibility of actively using eT&T and PSI to detect deterioration in real time. In this study, we assessed the ability of the three systems to detect escalation events. The fact that T&T was only documented for 4/20 events reiterates previous findings [1] and strongly supports the need to change the way vital signs are recorded in the ED. Both eT&T and PSI increased the detection rate (17/20 and 15/20, respectively). However, our sample size is too small to make any statistical comparison between PSI and eT&T. The poor sensitivity of the PSI to hypotension was recognized during the analysis, and the PSI algorithm has since been amended accordingly. Two cases in which eT&T detected an escalation and PSI did not were during periods of partial data acquisition related to technical and practical problems described above, reflecting the reality of clinical practice. We do not know whether other undocumented events were detected by PSI. However,
Copyright r 2016 Wolters Kluwer Health, Inc. All rights reserved.
32 European Journal of Emergency Medicine 2016, Vol 23 No 1
review of PSI alerts that were not associated with an escalation suggests that ∼ 25% may have warranted clinical review, but none appeared life threatening. Limitations
This study is dependent upon retrospective analysis of documentation in ED case notes. Our reporting is inherently biased in favour of eT&T because our study depends on the case notes and T&T chart to identify escalations. The recorded times of manually acquired vital signs are usually rounded to the nearest 5 min, from a clock that may vary from the true time by several minutes. If documentation is delayed, then the timings may reflect a ‘best guess’ as to when events occurred. This has been a detailed analysis of a small number of patients, whom we believe to be representative of our population. Future studies will include larger numbers of patients and should provide further evidence regarding the utility of different systems for the detection of deterioration among ED patients.
research nurses: Sally Beer and Soubera Yousefi to data collection and data entry, Karen Warnes to data analysis and Joe Crosbie, Elizabeth Oastler, Barbara Crowe, Rachel Bull, Gavin Shields, Lynsey Gander, Ursula Short and Sarah Carter for their assistance during the study. In addition, the authors acknowledge the valuable contribution of Jacqueline Birks (Centre for Statistics in Medicine, University of Oxford) during the consultation over study design. The work described in this paper has been funded by the National Institute for Health Research Biomedical Research Centre, Oxford. The funder played no part in the design of the study or in the analysis of the results. Conflicts of interest
Professor Tarassenko has a minority shareholding (less than 5%) in OBS Medical, the company which supplied the data fusion technology. There are no other conflicts of interest to declare.
References
Conclusion
1
Electronic data capture appears to offer improved detection of patient deterioration in a busy clinical environment compared with commonly used paperbased alternatives. Sample size is insufficient to determine which electronic method (eT&T or PSI) offers superior detection of the need for escalation. We are currently undertaking a larger study to address the feasibility of using these systems within a busy ED. Good clinical acumen cannot be replaced with electronic devices, but, in working environments that are becoming increasingly busy and complex, any aid to early detection of patient deterioration should be seriously considered.
Acknowledgements The authors thank the staff of the John Radcliffe Hospital ED, in particular the contribution of the
2
3
4
5 6
7
Wilson SJ, Wong D, Clifton D, Fleming S, Way R, Pullinger R, Tarassenko L. Track and trigger in an emergency department: an observational evaluation study. Emerg Med J 2013; 30:186–191. Hravnak M, Edwards L, Clontz A, Valenta C, Devita MA, Pinsky MR. Defining the incidence of cardiorespiratory instability in patients in step-down units using an electronic integrated monitoring system. Arch Intern Med 2008; 168:1300–1308. Hravnak M, Devita MA, Clontz A, Edwards L, Valenta C, Pinsky MR. Cardiorespiratory instability before and after implementing an integrated monitoring system. Crit Care Med 2011; 39:65–72. Watkinson PJ, Barber VS, Price JD, Hann A, Tarassenko L, Young JD. A randomised controlled trial of the effect of continuous electronic physiological monitoring on the adverse event rate in high risk medical and surgical patients. Anaesthesia 2006; 61:1031–1039. Tarassenko L, Hann A, Young D. Integrated monitoring and analysis for early warning of patient deterioration. Br J Anaesth 2006; 97:64–68. Tarassenko L, Clifton DA, Pinsky MR, Hravnak MT, Woods JR, Watkinson PJ. Centile-based early warning scores derived from statistical distributions of vital signs. Resuscitation 2011; 82:1013–1018. Royal College of Physicians of London. National early warning score (NEWS): standardising the assessment of acute-illness severity in the NHS – report of a working party, 2012. Available at: http://www.rcplondon.ac.uk/resources/ national-early-warning-score-news [Accessed 25 May 2013].
Copyright r 2016 Wolters Kluwer Health, Inc. All rights reserved.
Analysis of a data-fusion system for continuous vital sign monitoring in an emergency department Sarah J. Wilsona,c, David Wongb, Richard M. Pullingera, Rob Waya, David A. Cliftonb and Lionel Tarassenkob Objective The aim of the study was to evaluate the ability of a data-fusion patient status index (PSI) to detect patient deterioration in the emergency department (ED) in comparison with track-and-trigger (T&T). Materials and methods A single-centre observational cohort study was conducted in a medium-sized teaching hospital ED. Vital sign data and any documented T&T scores (paper T&T) were collected from adults attending the resuscitation room, majors or observation ward. For each set of vital signs, we retrospectively calculated T&T (eT&T). PSI was calculated retrospectively from the continuous vital sign data using a statistical model of normality. Clinical notes were examined to identify ‘escalation’ events, and the numbers of these escalations identified by paper T&T, eT&T and PSI were retrospectively calculated.
Conclusion Electronic data capture offers opportunities for increased detection of deteriorating patients in a busy clinical environment compared with paper charts. Sample size in this study is insufficient to determine which electronic method (eT&T or PSI) offers superior detection of the need for escalation. European Journal of Emergency Medicine 23:28–32 Copyright © 2016 Wolters Kluwer Health, Inc. All rights reserved. European Journal of Emergency Medicine 2016, 23:28–32 Keywords: detection of patient deterioration, early warning score, emergency department, patient status index, track-and-trigger, vital sign monitoring, vital signs a
Emergency Department, Oxford University Hospitals NHS Trust, Oxford, Department of Engineering Science, Institute of Biomedical Engineering, University of Oxford, Oxford and cEmergency Department, Heatherwood and Wexham Park Hospitals NHS Foundation Trust, Slough, UK b
Results Data from 472 patient episodes were examined. A total of 20 patients had PSI data at the time of an escalation related to vital sign abnormalities that occurred during their ED stay (vs. on arrival). Only four patient events were detected at the time by paper T&T. In all, 17 were detected retrospectively by eT&T and 15 by PSI. PSI had a calculated false-alert rate of 1.13 alerts/bed-day.
Correspondence to Sarah J. Wilson, MBChB, FCEM, c/o Tracey Pearson, John Radcliffe Hospital, Headley Way, Headington, Oxford OX3 9DU, UK Tel: + 44 1865 221173; fax: + 44 1865 222094; e-mail: [email protected]
Introduction
death and a low false-alert rate [2–4]. PSIs have not previously been evaluated in an ED.
Recognizing the deteriorating patient is key to preventing harm by correcting physiological abnormalities. However, identifying this group of patients is challenging, especially in a busy clinical environment. We have previously described [1] limitations of the paper-based track-and-trigger (T&T) system used to calculate early warning scores within an emergency department (ED) and highlighted the opportunity to use automated analysis of the outputs from bedside monitors. We now present further analysis from the same study, whereby data fusion technology is used to retrospectively calculate a patient status index (PSI) from continuous vital sign data collected from bedside monitors. In other settings (surgical step-down unit, high dependency area), prospective use of this approach has demonstrated appropriate detection of patient deterioration, a 40% reduction in critically unstable patients, reduced rate of unexpected
Supplemental digital content is available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal's website (www.euro-emergencymed.com). 0969-9546 Copyright © 2016 Wolters Kluwer Health, Inc. All rights reserved.
Received 30 December 2013 Accepted 10 April 2014
Objective The aim of the study was to evaluate the ability of a PSI to detect patient deterioration in the ED in comparison with documented T&T and retrospectively calculated T&T.
Materials and methods A single-centre observational cohort study was conducted in the ED of a medium-sized teaching hospital from January 2009 to January 2010 [1]. The study was approved by the UK National Research Ethics Service (ref: 08/H1307/56). We collected data from adults entering the resuscitation room, ‘majors’ and observation ward of the ED. Recruitment was restricted to times when members of the research team were available, typically daytime hours. Participants were excluded if they could not understand English or did not consent. Those unable to consent because of their clinical condition were followed up or had assent provided by next of kin or nominated consultee. DOI: 10.1097/MEJ.0000000000000166
Copyright r 2016 Wolters Kluwer Health, Inc. All rights reserved.
Data fusion for ED vital sign monitoring Wilson et al. 29
Hours of electronically acquired vital sign data for the study population
Table 1
Parameters HR RR SpO2 BP PSI
Number of hours of data 1626 1609 1643 1761 1904
The numbers of hours of available PSI data exceed that of other parameters, because it can be calculated from a partial set of vital sign data. BP, blood pressure; HR, heart rate; PSI, patient status index; RR, respiratory rate; SpO2, oxygen saturation.
collected for each parameter is shown in Table 1. The electronic data collected are incomplete. Reasons for this include lead disconnection and time spent away from the allocated bed space, for example during radiological investigations. Some data were also lost because of unexpected failure of the study data-collection server. The amount of PSI data is higher than for any single vital sign because it can be calculated from a partial set of vital signs (minimum three parameters).
Patient status index true alerts
As part of standard practice, vital sign and T&T data (paper T&T) were manually recorded at intervals onto paper observation charts by clinical staff (see Appendix, Supplemental digital content 1, http://links.lww.com/ EJEM/A79). Heart rate, blood pressure, respiratory rate (RR), oxygen saturation (SpO2), temperature and Glasgow Coma Scale (GCS) and paper T&T scores were collected retrospectively from observation charts. For each set of manually recorded vital signs, T&T was retrospectively calculated (eT&T). Continuous vital sign data (RR, heart rate, blood pressure, SpO2) were acquired using bedside monitors [Philips (Boeblingen, Germany)] and saved to a server. PSI was calculated retrospectively from the continuous vital sign data. The statistical model of normality used to calculate PSI (range 0.0–5.0) is derived from 3500 h of vital sign data recorded from hospital inpatients, as described fully elsewhere [5]. An alert would be generated for any PSI of at least 3.0 that persists for 4 out of 5 min. Patient deterioration was captured by recording escalations of care. An escalation was defined as any documented increase in the level of care associated with abnormal vital signs [1]. Two clinicians independently reviewed every set of clinical notes to identify escalations. A third clinician reconciled any discrepancies. Each escalation was further analysed independently by two clinical researchers to distinguish new deteriorations from persistent abnormalities following a previous escalation. This analysis considers only escalations that occurred after arrival in the ED. We excluded escalations occurring on arrival as they represent patients who were immediately recognized as unwell, and neither PSI nor T&T would offer any advantage. As a result, the number of escalations included in the analysis is relatively small. The outcome measure therefore was detection of escalation events associated with abnormal vital signs after arrival in the ED.
Results We collected T&T data from 472 patients. For 400 of these patients, we also collected continuous vital sign data over a period of 2169 h. The amount of data
We have only considered one event per patient to avoid analysing interdependent events. A total of 35 patients had an escalation after arrival in the ED; of them, 20 had PSI data around the time of the escalation. The remaining 15 patients were not considered further for the following reasons: (1) Seven patients were deemed to have ongoing conditions from arrival rather than a new deterioration (had been recognized as unwell at the time of arrival to ED and unlikely to benefit from a subsequent alert for the same parameters). (2) Two patients had no PSI data at all (one unexpected failure of software, one failure of third-party hardware). (3) Six patients had no PSI data at a time relevant to the escalation (two patients had moved to areas where there were no study monitors, four were only connected to study monitors after escalation). (4) Of the 20 patients with PSI data around the time of an escalation, four were detected by paper T&T, 17 were detected by eT&T and 15 were detected by PSI. In all, 12 events were detected by both eT&T and PSI. (5) PSI did not detect an event that eT&T detected in five patients. Four of these were predictable. (6) Two were for isolated pyrexia (temperature is not included in the PSI algorithm) and a further two were due to incomplete data such that PSI was unavailable at the time of the escalation. The remaining event was an episode of hypotension. (7) eT&T did not detect an event that PSI detected in three patients. One was an episode of brief fast atrial fibrillation documented in the text of the patient notes. The other two events were also documented in the patient notes but were not recorded on the corresponding observations chart. Figure 1 represents a patient for whom PSI would have alerted before T&T. The plot shows the electronically acquired vital signs, paper T&T, retrospective eT&T and PSI. T&T is displayed at times when observations were recorded on the paper chart. The grey-shaded block represents periods of time during which PSI would
Copyright r 2016 Wolters Kluwer Health, Inc. All rights reserved.
30 European Journal of Emergency Medicine 2016, Vol 23 No 1
Fig. 1
RR/BP/HR
SpO2
Patient: ED00570 100 95 90 200 190 180 170 160 150 140 130 120 110 100 90 80 70 60 50 40 30 20 10 0 T&T eT&T
SpO2 HR RR BP
0
1
4
6
1
0
1
4
6
1
PSI
5
0 14:00
15:00
16:00
Time A combined plot of the electronically acquired vital signs, paper T&T, eT&T and PSI for a patient who had a recorded escalation (denoted by solid vertical line). The grey-shaded block represents the period for which PSI generates an alert. BP, blood pressure; HR, heart rate; PSI, patient status index; RR, respiratory rate; SpO2, oxygen saturation; T&T, track-and-trigger.
generate an alert. The solid vertical line indicates the point at which an escalation was recorded.
Patient status index false alerts
To determine the false-alert rate for PSI alerts, we assumed that patients who did not have any documented increase in care (a wider criterion than ‘escalation’) did not warrant an alert, and therefore any alerts for this group are classified as false alerts. This will overestimate the false-alert rate, because we cannot be sure that all clinical deteriorations were documented. A total of 217 patients had no documented increase in care and had a cumulative length of stay in the ED of 43.5 days. There were 49 false alerts from 39 patients. The false-alert rate is thus 1.13 alerts/bed-day. A review of these ‘false’ alerts revealed that nine were likely to have been clinically relevant, which would further reduce the false-alert rate. Figure 2 shows the plot for one such alert, where the patient had abdominal pain requiring opiate analgesia. Further interpretation of the patient data suggests that the oxygen desaturation and hypotension at the time of the potential PSI alert should have prompted a patient review.
Discussion Each system for detecting patient deterioration has limitations. Paper T&T requires human input to accurately record the observations and assign scores to them. In a busy ED, events occur that are not documented. The resulting observations are only useful if they are recorded promptly and the resulting T&T score is calculated correctly. An electronically assisted T&T may eliminate these problems by automating the score calculation, ensuring that the observations are time stamped and can be configured to provide decision support. Both paper T&T and eT&T are limited by reliance on an imperfect scoring system. It is widely reported that most T&T scoring systems are not data driven [6,7]. However, the T&T score suffers from a more fundamental problem in that it reacts in a stepwise manner to continuous changes in patient physiology. For example, in our institution, a RR of 9 is ‘normal’ but a RR of 8 would trigger action to be taken (Appendix 1, Supplemental digital content 1, http://links.lww.com/ EJEM/A79). Unlike T&T systems, PSI is calculated using continuous data. A continuous electronic system should be able to detect deteriorating vital signs earlier than a system that
Copyright r 2016 Wolters Kluwer Health, Inc. All rights reserved.
Data fusion for ED vital sign monitoring Wilson et al. 31
Fig. 2
Patient: ED00188 SpO2
100 95 90 120 110 100 90
HR/RR/BP
80 70 60 50 40 30
SpO2 HR
20
RR BP
10 0 T&T eT&T
−1
−1
−1
3
1
0
PSI
5
0
21:00
22:00
23:00
00:00
Time
A combined plot of the electronically acquired vital signs, paper T&T, eT&T and PSI for a patient who had no recorded escalations. A PSI alert was generated at 22:15, which was initially classified as a false alert. However, further interpretation of the patient notes showed that the oxygen desaturation and hypotension at the time of the alert correspond to a period during which the patient reported abdominal pain. BP, blood pressure; HR, heart rate; PSI, patient status index; RR, respiratory rate; SpO2, oxygen saturation; T&T, track-and-trigger.
relies on documented intermittent observations. In addition, the PSI score is not limited to integer values, which means it can alert when some or all of the parameters begin to deviate from normality rather than when a single parameter becomes obviously abnormal. In theory, the ability to integrate data from multiple vital signs allows the system to increase the percentage of appropriate alerts. This sets it apart from standard bedside monitors, which use single-channel alerts. A primary disadvantage of PSI is its inability to automatically incorporate GCS. We considered adding GCS into the PSI calculations as it is part of standard T&T, but, because GCS must be assessed in person, it would not add value to the system. We have not included temperature in this PSI algorithm, although it would be possible to add if a consistently accurate temperature reading could be acquired electronically. The biggest barriers to the use of any electronic monitoring system are practical considerations including power failure, server failure, movement of patients or monitors within the ED or removal of monitoring leads
from the patient. The importance of these issues should not be underestimated. We are currently undertaking a larger study in the ED majors area looking at the feasibility of actively using eT&T and PSI to detect deterioration in real time. In this study, we assessed the ability of the three systems to detect escalation events. The fact that T&T was only documented for 4/20 events reiterates previous findings [1] and strongly supports the need to change the way vital signs are recorded in the ED. Both eT&T and PSI increased the detection rate (17/20 and 15/20, respectively). However, our sample size is too small to make any statistical comparison between PSI and eT&T. The poor sensitivity of the PSI to hypotension was recognized during the analysis, and the PSI algorithm has since been amended accordingly. Two cases in which eT&T detected an escalation and PSI did not were during periods of partial data acquisition related to technical and practical problems described above, reflecting the reality of clinical practice. We do not know whether other undocumented events were detected by PSI. However,
Copyright r 2016 Wolters Kluwer Health, Inc. All rights reserved.
32 European Journal of Emergency Medicine 2016, Vol 23 No 1
review of PSI alerts that were not associated with an escalation suggests that ∼ 25% may have warranted clinical review, but none appeared life threatening. Limitations
This study is dependent upon retrospective analysis of documentation in ED case notes. Our reporting is inherently biased in favour of eT&T because our study depends on the case notes and T&T chart to identify escalations. The recorded times of manually acquired vital signs are usually rounded to the nearest 5 min, from a clock that may vary from the true time by several minutes. If documentation is delayed, then the timings may reflect a ‘best guess’ as to when events occurred. This has been a detailed analysis of a small number of patients, whom we believe to be representative of our population. Future studies will include larger numbers of patients and should provide further evidence regarding the utility of different systems for the detection of deterioration among ED patients.
research nurses: Sally Beer and Soubera Yousefi to data collection and data entry, Karen Warnes to data analysis and Joe Crosbie, Elizabeth Oastler, Barbara Crowe, Rachel Bull, Gavin Shields, Lynsey Gander, Ursula Short and Sarah Carter for their assistance during the study. In addition, the authors acknowledge the valuable contribution of Jacqueline Birks (Centre for Statistics in Medicine, University of Oxford) during the consultation over study design. The work described in this paper has been funded by the National Institute for Health Research Biomedical Research Centre, Oxford. The funder played no part in the design of the study or in the analysis of the results. Conflicts of interest
Professor Tarassenko has a minority shareholding (less than 5%) in OBS Medical, the company which supplied the data fusion technology. There are no other conflicts of interest to declare.
References
Conclusion
1
Electronic data capture appears to offer improved detection of patient deterioration in a busy clinical environment compared with commonly used paperbased alternatives. Sample size is insufficient to determine which electronic method (eT&T or PSI) offers superior detection of the need for escalation. We are currently undertaking a larger study to address the feasibility of using these systems within a busy ED. Good clinical acumen cannot be replaced with electronic devices, but, in working environments that are becoming increasingly busy and complex, any aid to early detection of patient deterioration should be seriously considered.
Acknowledgements The authors thank the staff of the John Radcliffe Hospital ED, in particular the contribution of the
2
3
4
5 6
7
Wilson SJ, Wong D, Clifton D, Fleming S, Way R, Pullinger R, Tarassenko L. Track and trigger in an emergency department: an observational evaluation study. Emerg Med J 2013; 30:186–191. Hravnak M, Edwards L, Clontz A, Valenta C, Devita MA, Pinsky MR. Defining the incidence of cardiorespiratory instability in patients in step-down units using an electronic integrated monitoring system. Arch Intern Med 2008; 168:1300–1308. Hravnak M, Devita MA, Clontz A, Edwards L, Valenta C, Pinsky MR. Cardiorespiratory instability before and after implementing an integrated monitoring system. Crit Care Med 2011; 39:65–72. Watkinson PJ, Barber VS, Price JD, Hann A, Tarassenko L, Young JD. A randomised controlled trial of the effect of continuous electronic physiological monitoring on the adverse event rate in high risk medical and surgical patients. Anaesthesia 2006; 61:1031–1039. Tarassenko L, Hann A, Young D. Integrated monitoring and analysis for early warning of patient deterioration. Br J Anaesth 2006; 97:64–68. Tarassenko L, Clifton DA, Pinsky MR, Hravnak MT, Woods JR, Watkinson PJ. Centile-based early warning scores derived from statistical distributions of vital signs. Resuscitation 2011; 82:1013–1018. Royal College of Physicians of London. National early warning score (NEWS): standardising the assessment of acute-illness severity in the NHS – report of a working party, 2012. Available at: http://www.rcplondon.ac.uk/resources/ national-early-warning-score-news [Accessed 25 May 2013].
Copyright r 2016 Wolters Kluwer Health, Inc. All rights reserved.
