G Model
ARTICLE IN PRESS
RESUS 6432 1–7
1
Resuscitation xxx (2015) xxx–xxx
Contents lists available at ScienceDirect
Resuscitation journal homepage: www.elsevier.com/locate/resuscitation
2
Clinical Paper
3
Chest compression rate feedback based on transthoracic impedance夽
4
Q1
5 6
Digna M. González-Otero a , Sofía Ruiz de Gauna a,∗ , Jesus Ruiz a , Mohamud Daya b , Lars Wik c , James K. Russell d , Jo Kramer-Johansen e , Trygve Eftestøl f , Erik Alonso a , Unai Ayala a a
Department of Communications Engineering, University of the Basque Country, UPV/EHU, 48013 Bilbao, Spain Oregon Health & Science University, 97239-3098 Portland, OR, USA Norwegian National Advisory Unit on Prehospital Emergency Medicine, Oslo University Hospital, Oslo, Norway 9 10 Q2 d Russell Biomedical Research Consulting, USA e Norwegian Centre for Prehospital Emergency Care (NAKOS), Oslo University Hospital and University of Oslo, 0424 Oslo, Norway 11 f Department of Electrical Engineering and Computer Science, Faculty of Science and Technology, University of Stavanger, 4036 Stavanger, Norway 12 7 8
b c
13
14 27
a r t i c l e
i n f o
a b s t r a c t
15 16 17 18 19
Article history: Received 11 February 2015 Received in revised form 24 April 2015 Accepted 26 May 2015
20
26
Keywords: Transthoracic impedance CPR quality Chest compression Basic life support Automated external defibrillator
28
1. Introduction
21 22 23 24 25
29Q3 30 31 32 33 34 35 36
Background: Quality of cardiopulmonary resuscitation (CPR) is an important determinant of survival from cardiac arrest. The use of feedback devices is encouraged by current resuscitation guidelines as it helps rescuers to improve quality of CPR performance. Aim: To determine the feasibility of a generic algorithm for feedback related to chest compression (CC) rate using the transthoracic impedance (TTI) signal recorded through the defibrillation pads. Methods: We analysed 180 episodes collected equally from three different emergency services, each one using a unique defibrillator model. The new algorithm computed the CC-rate every 2 s by analysing the TTI signal in the frequency domain. The obtained CC-rate values were compared with the gold standard, computed using the compression force or the ECG and TTI signals when the force was not recorded. The accuracy of the CC-rate, the proportion of alarms of inadequate CC-rate, chest compression fraction (CCF) and the mean CC-rate per episode were calculated. Results: Intervals with CCs were detected with a mean sensitivity and a mean positive predictive value per episode of 96.3% and 97.0%, respectively. Estimated CC-rate had an error below 10% in 95.8% of the time. Mean percentage of accurate alarms per episode was 98.2%. No statistical differences were found between the gold standard and the estimated values for any of the computed metrics. Conclusion: We developed an accurate algorithm to calculate and provide feedback on CC-rate using the TTI signal. This could be integrated into automated external defibrillators and help improve the quality of CPR in basic-life-support settings. © 2015 Published by Elsevier Ireland Ltd.
High quality chest compressions (CCs) are recommended during cardiopulmonary resuscitation (CPR). Target goals for CC are rate between 100 and 120 compressions per minute (cpm), depth of at least 50 mm, full chest recoil between compressions, and minimization of all interruptions.1 Several studies have shown that rescuers often provide poor quality CCs.2–5 Monitoring and providing feedback on rescuers’ performance improve CC quality.6 Advanced systems based on accelerometers
夽 A Spanish translated version of the summary of this article appears as Appendix in the final online version at http://dx.doi.org/10.1016/j.resuscitation.2015.05.027. ∗ Corresponding author. E-mail address: sofi[email protected] (S. Ruiz de Gauna).
or on magnetic induction can measure the chest displacement and rate and guide the rescuer to perform adequate CCs.7,8 The use of CPR-aid devices contributes to improve the adherence to recommendations for CC metrics.9–11 They are relatively expensive and not generally used or available as part of most automated external defibrillators (AEDs). The transthoracic impedance (TTI) recorded through the defibrillation pads fluctuates for each compression around the patient’s baseline impedance. This signal has been proposed as a possible option for calculating the CC-rate and the chest compression fraction (CCF).12 However, TTI cannot be used to reliably estimate the CC-depth.13 Recently, automatic detectors of individual CCs in the TTI signal have been developed.14,15 Each TTI fluctuation was automatically classified as CC or not, depending on its amplitude and duration. The ability of the algorithm to provide feedback on CC-rate was
http://dx.doi.org/10.1016/j.resuscitation.2015.05.027 0300-9572/© 2015 Published by Elsevier Ireland Ltd.
Please cite this article in press as: González-Otero DM, et al. Chest compression rate feedback based on transthoracic impedance. Resuscitation (2015), http://dx.doi.org/10.1016/j.resuscitation.2015.05.027
37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52
G Model RESUS 6432 1–7
ARTICLE IN PRESS D.M. González-Otero et al. / Resuscitation xxx (2015) xxx–xxx
2
63
determined. However, these algorithms were optimized and tested with subsets of a single database. The TTI signals used in each study had common properties regarding the front-end for the signal acquisition, the type of pads, and the rescuers’ CPR performance, all of which may influence the characteristics of the TTI fluctuations. 16 The generalizability of these algorithms to other devices and settings is unclear. Our aim was to determine the accuracy of a generic method to calculate and provide feedback on CC-rate for any system recording the TTI signal. We tested the method on three independent out of hospital cardiac arrest (OHCA) databases.
64
2. Materials and methods
65
2.1. Data collection
53 54 55 56 57 58 59 60 61 62
66 67 68 69 70 71 72 73
74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93
94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111
Data were extracted from three OHCA studies compiled in different Emergency Medical Service (EMS) systems and time periods during which different releases of international resuscitation guidelines (years 2000,17 2005,18 and 201019 ) were in effect. Each database was derived from distinct monitor-defibrillators with different signal acquisition characteristics. Data were provided anonymous and use of the data was approved by the local institutional review board or Ethics committees. • Sister database2,20 : 364 episodes acquired in three locations (Akershus in Norway, Stockholm in Sweden, and London in UK) between 2002 and 2004, according to 2000 guidelines. Modified Heartstart 4000 defibrillators (Philips Medical Systems, Andover, USA) were used to record the episodes. The TTI was stored with a sampling frequency of 500 Hz and a bandwidth of 0–80 Hz. • TVF&R database: 623 episodes acquired by the Tualatin Valley Fire & Rescue, a first response ALS fire agency serving nine incorporated cities in Oregon, USA, between 2006 and 2009, according to 2005 guidelines. Heartstart MRx monitordefibrillators (Philips Medical Systems, Andover, USA) were used to record the episodes. Nominal specifications for the acquisition of the TTI signal when the devices were produced were a sampling frequency of 250 Hz and a bandwidth of 0–20 Hz. • Oslo EMS database: 370 episodes acquired in Oslo (Norway) between 2012 and 2013, according to 2010 guidelines. Lifepak 12/15 defibrillators in semiautomatic mode (Physio-Control, Redmond, WA, USA) were used to record the episodes. The TTI was stored with a sampling frequency of 61 Hz and a bandwidth of 0.3–30 Hz. All episodes contained the ECG and the TTI signals acquired through the defibrillation pads. The SISTER and TVF&R databases included a force signal that was acquired using an auxiliary chest pad containing an accelerometer and a pressure sensor.20,15 We extracted the episodes that had the required signals correctly recorded and with at least 1000 CCs: 156/364 in Sister, 136/623 in TVF&R and 267/370 in Oslo EMS. We randomly selected 60 episodes per database for annotation and review, a sample expected to be sufficient to make clinically relevant distinctions based on our prior experience.14,15 CC instants were annotated in the 180 episodes used in the study. For the Sister and TVF&R episodes we applied an automatic peak detector on the force signal with a fixed threshold of 2 kg. Two reviewers (JR and SRG) manually corrected the marks when necessary. For the Oslo EMS episodes we had the CC instants provided by the Code-Stat data review software (Physio-Control, Redmond, WA, USA). This software uses the TTI and the ECG signals for automated detection of the CCs. Authors DMG, JR and SRG
independently reviewed these signals to correct the annotations if necessary. We discarded the signal in time intervals where the TTI was disconnected: 313 intervals in Sister (4.1% of the time), 5 in TVF&R (0.1%) and 137 in Oslo EMS (4.5%) and intervals in which CC instants could not be reliably annotated: 14 intervals in Sister (0.3%), 23 in TVF&R (0.7%) and 26 in Oslo EMS (0.3%). The top panels in Fig. 1 show 8 s of one episode from the TVF&R database (A) and another from the Oslo EMS database (B). The annotated compressions are represented by dashed lines in the compression force (CF) and in the ECG, respectively. 2.2. Calculation of the CC-rate We designed the method under the following specifications: ability to report CC-rates between 60 and 210 cpm, minimum TTI sampling frequency of 20 Hz, and configurable feedback interval. The method is based on direct calculation of CC-rate in consecutive 2-s time intervals (analysis windows). For each window, the gold standard for the rate was computed as the inverse of the mean time interval between the annotated CCs. In the top panels of Fig. 1 there are three compressions in the selected analysis window (indicated by the rectangle). In example (A) the mean time interval between CCs is 0.58 s, so the rate, computed as the inverse, is 1.72 compressions per second, i.e., 103.4 cpm. The TTI signal (panel 2) is first filtered to suppress fluctuations out of the band of interest (1–3.5 Hz, 60–210 cpm), yielding the processed TTI (pTTI) signal (panel 3). To decide whether each analysis window contains CCs, the amplitude of a sinusoidal signal with the same energy as the pTTI is computed. The decision is then made comparing this amplitude with a dynamic threshold. Windows containing CCs are analysed in the frequency domain to identify the dominant frequency (highest peak) in the band of interest. Depending on the TTI waveform, the dominant frequency may directly correspond to the mean CC-frequency (CC-rate in Hz) in the analysis window, as shown in the bottom panel of Fig. 1A. In other cases, the TTI waveform presents a strong second harmonic, so in order to identify the mean CC-frequency the method looks for another peak of the spectrum around half the dominant frequency. If such a peak is found, its frequency is considered the mean CC-frequency. This is illustrated in Fig. 1B. The online appendix provides a detailed description of the method. 2.3. Performance evaluation Episodes were divided into non-overlapping 2-s analysis windows. Then, the performance of the method was assessed in terms of sensitivity (Se), i.e., the percentage of windows that did have compressions for which a CC-rate was reported; positive predictive value (PPV), defined as the percentage of windows that truly contained CCs from the total of windows for which CC-rate was reported; and an accuracy factor (AF) that indicates the percentage of analysis windows for which CC-rate was provided with an error below 10% of the gold-standard rate, calculated as described in Section 2.2. For the analysis windows for which CC-rate was provided, the errors were plotted as a function of the gold-standard rate, and the 95 percentile of unsigned errors was computed. The power of the method to provide alarms for non-adequate CC-rates was evaluated. We adopted a tolerance of 10 cpm for the lower limit to avoid exposing the rescuer to excessive ‘low-rate’ alarms, as suggested by Kramer-Johansen et al.21 Thus, the limits used for analysis were 90–120 cpm. The CCF was computed for each episode as the percentage of analysis windows containing CCs. Intervals with spontaneous circulation were identified using the ECG, the force, and the TTI signal, and were excluded for the CCF calculation.
Please cite this article in press as: González-Otero DM, et al. Chest compression rate feedback based on transthoracic impedance. Resuscitation (2015), http://dx.doi.org/10.1016/j.resuscitation.2015.05.027
112 113 114 115 116 117 118 119 120 121 122
123
124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151
152
153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172
G Model
ARTICLE IN PRESS
RESUS 6432 1–7
D.M. González-Otero et al. / Resuscitation xxx (2015) xxx–xxx
3
Fig. 1. Graphical example of the method using two intervals from the TVF&R (A) and the Oslo EMS (B) databases. Gold-standard and estimated rates (GS and CC-rate, respectively) are shown in the bottom panel expressed in cpm.
Table 1 Summary of characteristics of the episodes of each database: duration (s), percentage of the episode in which the patient was intubated, chest compression fraction (CCF), percentage of the episode in which the patient had spontaneous circulation (ROSC period), mean CC-rate, range of the CC-rate and mean CC-depth. The last two rows show the CCF and the mean CC-rate estimated by the method. All values are expressed as mean (standard deviation). An asterisk (* ) next to the name of the metric indicates statistically significant differences between databases. NA stands for Not Available. Sister (n = 60) Metrics (from gold standard) 1708.9 (542.2) Duration (s) 78.5 (26.5) Intubation period (%) 66.8 (14.2) CCF* (%) 5.6 (14.3) ROSC period (%) * Mean CC-rate (cpm) 120.3 (13.9) 60.0–200.0 Range CC-rate (cpm) Mean CC-depth* (mm) 35.7 (7.6) Metrics (estimated) 67.2 (13.7) CCF* (%) 119.7 (13.1) Mean CC-rate* (cpm)
173
TVF&R (n = 60)
Oslo EMS (n = 60)
1551.5 (536.0) 85.8 (20.5) 81.7 (7.4) 7.9 (13.7) 110.0 (11.3) 60.0–178.1 38.5 (6.4)
1916.3 (896.0) 78.3 (27.7) 87.9 (5.9) 7.6 (14.9) 111.6 (8.9) 62.2–197.1 NA (NA)
80.5 (8.7) 110.8 (10.3)
84.2 (5.9) 111.7 (8.6)
2.4. Statistical analysis
178
Between-database comparisons were made using ANOVA for variables that passed the Lilliefors normality test, and using the Kruskal–Wallis test otherwise. The estimated and the goldstandard metrics were compared using the Mann–Whitney U-test. P-values
ARTICLE IN PRESS
RESUS 6432 1–7
1
Resuscitation xxx (2015) xxx–xxx
Contents lists available at ScienceDirect
Resuscitation journal homepage: www.elsevier.com/locate/resuscitation
2
Clinical Paper
3
Chest compression rate feedback based on transthoracic impedance夽
4
Q1
5 6
Digna M. González-Otero a , Sofía Ruiz de Gauna a,∗ , Jesus Ruiz a , Mohamud Daya b , Lars Wik c , James K. Russell d , Jo Kramer-Johansen e , Trygve Eftestøl f , Erik Alonso a , Unai Ayala a a
Department of Communications Engineering, University of the Basque Country, UPV/EHU, 48013 Bilbao, Spain Oregon Health & Science University, 97239-3098 Portland, OR, USA Norwegian National Advisory Unit on Prehospital Emergency Medicine, Oslo University Hospital, Oslo, Norway 9 10 Q2 d Russell Biomedical Research Consulting, USA e Norwegian Centre for Prehospital Emergency Care (NAKOS), Oslo University Hospital and University of Oslo, 0424 Oslo, Norway 11 f Department of Electrical Engineering and Computer Science, Faculty of Science and Technology, University of Stavanger, 4036 Stavanger, Norway 12 7 8
b c
13
14 27
a r t i c l e
i n f o
a b s t r a c t
15 16 17 18 19
Article history: Received 11 February 2015 Received in revised form 24 April 2015 Accepted 26 May 2015
20
26
Keywords: Transthoracic impedance CPR quality Chest compression Basic life support Automated external defibrillator
28
1. Introduction
21 22 23 24 25
29Q3 30 31 32 33 34 35 36
Background: Quality of cardiopulmonary resuscitation (CPR) is an important determinant of survival from cardiac arrest. The use of feedback devices is encouraged by current resuscitation guidelines as it helps rescuers to improve quality of CPR performance. Aim: To determine the feasibility of a generic algorithm for feedback related to chest compression (CC) rate using the transthoracic impedance (TTI) signal recorded through the defibrillation pads. Methods: We analysed 180 episodes collected equally from three different emergency services, each one using a unique defibrillator model. The new algorithm computed the CC-rate every 2 s by analysing the TTI signal in the frequency domain. The obtained CC-rate values were compared with the gold standard, computed using the compression force or the ECG and TTI signals when the force was not recorded. The accuracy of the CC-rate, the proportion of alarms of inadequate CC-rate, chest compression fraction (CCF) and the mean CC-rate per episode were calculated. Results: Intervals with CCs were detected with a mean sensitivity and a mean positive predictive value per episode of 96.3% and 97.0%, respectively. Estimated CC-rate had an error below 10% in 95.8% of the time. Mean percentage of accurate alarms per episode was 98.2%. No statistical differences were found between the gold standard and the estimated values for any of the computed metrics. Conclusion: We developed an accurate algorithm to calculate and provide feedback on CC-rate using the TTI signal. This could be integrated into automated external defibrillators and help improve the quality of CPR in basic-life-support settings. © 2015 Published by Elsevier Ireland Ltd.
High quality chest compressions (CCs) are recommended during cardiopulmonary resuscitation (CPR). Target goals for CC are rate between 100 and 120 compressions per minute (cpm), depth of at least 50 mm, full chest recoil between compressions, and minimization of all interruptions.1 Several studies have shown that rescuers often provide poor quality CCs.2–5 Monitoring and providing feedback on rescuers’ performance improve CC quality.6 Advanced systems based on accelerometers
夽 A Spanish translated version of the summary of this article appears as Appendix in the final online version at http://dx.doi.org/10.1016/j.resuscitation.2015.05.027. ∗ Corresponding author. E-mail address: sofi[email protected] (S. Ruiz de Gauna).
or on magnetic induction can measure the chest displacement and rate and guide the rescuer to perform adequate CCs.7,8 The use of CPR-aid devices contributes to improve the adherence to recommendations for CC metrics.9–11 They are relatively expensive and not generally used or available as part of most automated external defibrillators (AEDs). The transthoracic impedance (TTI) recorded through the defibrillation pads fluctuates for each compression around the patient’s baseline impedance. This signal has been proposed as a possible option for calculating the CC-rate and the chest compression fraction (CCF).12 However, TTI cannot be used to reliably estimate the CC-depth.13 Recently, automatic detectors of individual CCs in the TTI signal have been developed.14,15 Each TTI fluctuation was automatically classified as CC or not, depending on its amplitude and duration. The ability of the algorithm to provide feedback on CC-rate was
http://dx.doi.org/10.1016/j.resuscitation.2015.05.027 0300-9572/© 2015 Published by Elsevier Ireland Ltd.
Please cite this article in press as: González-Otero DM, et al. Chest compression rate feedback based on transthoracic impedance. Resuscitation (2015), http://dx.doi.org/10.1016/j.resuscitation.2015.05.027
37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52
G Model RESUS 6432 1–7
ARTICLE IN PRESS D.M. González-Otero et al. / Resuscitation xxx (2015) xxx–xxx
2
63
determined. However, these algorithms were optimized and tested with subsets of a single database. The TTI signals used in each study had common properties regarding the front-end for the signal acquisition, the type of pads, and the rescuers’ CPR performance, all of which may influence the characteristics of the TTI fluctuations. 16 The generalizability of these algorithms to other devices and settings is unclear. Our aim was to determine the accuracy of a generic method to calculate and provide feedback on CC-rate for any system recording the TTI signal. We tested the method on three independent out of hospital cardiac arrest (OHCA) databases.
64
2. Materials and methods
65
2.1. Data collection
53 54 55 56 57 58 59 60 61 62
66 67 68 69 70 71 72 73
74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93
94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111
Data were extracted from three OHCA studies compiled in different Emergency Medical Service (EMS) systems and time periods during which different releases of international resuscitation guidelines (years 2000,17 2005,18 and 201019 ) were in effect. Each database was derived from distinct monitor-defibrillators with different signal acquisition characteristics. Data were provided anonymous and use of the data was approved by the local institutional review board or Ethics committees. • Sister database2,20 : 364 episodes acquired in three locations (Akershus in Norway, Stockholm in Sweden, and London in UK) between 2002 and 2004, according to 2000 guidelines. Modified Heartstart 4000 defibrillators (Philips Medical Systems, Andover, USA) were used to record the episodes. The TTI was stored with a sampling frequency of 500 Hz and a bandwidth of 0–80 Hz. • TVF&R database: 623 episodes acquired by the Tualatin Valley Fire & Rescue, a first response ALS fire agency serving nine incorporated cities in Oregon, USA, between 2006 and 2009, according to 2005 guidelines. Heartstart MRx monitordefibrillators (Philips Medical Systems, Andover, USA) were used to record the episodes. Nominal specifications for the acquisition of the TTI signal when the devices were produced were a sampling frequency of 250 Hz and a bandwidth of 0–20 Hz. • Oslo EMS database: 370 episodes acquired in Oslo (Norway) between 2012 and 2013, according to 2010 guidelines. Lifepak 12/15 defibrillators in semiautomatic mode (Physio-Control, Redmond, WA, USA) were used to record the episodes. The TTI was stored with a sampling frequency of 61 Hz and a bandwidth of 0.3–30 Hz. All episodes contained the ECG and the TTI signals acquired through the defibrillation pads. The SISTER and TVF&R databases included a force signal that was acquired using an auxiliary chest pad containing an accelerometer and a pressure sensor.20,15 We extracted the episodes that had the required signals correctly recorded and with at least 1000 CCs: 156/364 in Sister, 136/623 in TVF&R and 267/370 in Oslo EMS. We randomly selected 60 episodes per database for annotation and review, a sample expected to be sufficient to make clinically relevant distinctions based on our prior experience.14,15 CC instants were annotated in the 180 episodes used in the study. For the Sister and TVF&R episodes we applied an automatic peak detector on the force signal with a fixed threshold of 2 kg. Two reviewers (JR and SRG) manually corrected the marks when necessary. For the Oslo EMS episodes we had the CC instants provided by the Code-Stat data review software (Physio-Control, Redmond, WA, USA). This software uses the TTI and the ECG signals for automated detection of the CCs. Authors DMG, JR and SRG
independently reviewed these signals to correct the annotations if necessary. We discarded the signal in time intervals where the TTI was disconnected: 313 intervals in Sister (4.1% of the time), 5 in TVF&R (0.1%) and 137 in Oslo EMS (4.5%) and intervals in which CC instants could not be reliably annotated: 14 intervals in Sister (0.3%), 23 in TVF&R (0.7%) and 26 in Oslo EMS (0.3%). The top panels in Fig. 1 show 8 s of one episode from the TVF&R database (A) and another from the Oslo EMS database (B). The annotated compressions are represented by dashed lines in the compression force (CF) and in the ECG, respectively. 2.2. Calculation of the CC-rate We designed the method under the following specifications: ability to report CC-rates between 60 and 210 cpm, minimum TTI sampling frequency of 20 Hz, and configurable feedback interval. The method is based on direct calculation of CC-rate in consecutive 2-s time intervals (analysis windows). For each window, the gold standard for the rate was computed as the inverse of the mean time interval between the annotated CCs. In the top panels of Fig. 1 there are three compressions in the selected analysis window (indicated by the rectangle). In example (A) the mean time interval between CCs is 0.58 s, so the rate, computed as the inverse, is 1.72 compressions per second, i.e., 103.4 cpm. The TTI signal (panel 2) is first filtered to suppress fluctuations out of the band of interest (1–3.5 Hz, 60–210 cpm), yielding the processed TTI (pTTI) signal (panel 3). To decide whether each analysis window contains CCs, the amplitude of a sinusoidal signal with the same energy as the pTTI is computed. The decision is then made comparing this amplitude with a dynamic threshold. Windows containing CCs are analysed in the frequency domain to identify the dominant frequency (highest peak) in the band of interest. Depending on the TTI waveform, the dominant frequency may directly correspond to the mean CC-frequency (CC-rate in Hz) in the analysis window, as shown in the bottom panel of Fig. 1A. In other cases, the TTI waveform presents a strong second harmonic, so in order to identify the mean CC-frequency the method looks for another peak of the spectrum around half the dominant frequency. If such a peak is found, its frequency is considered the mean CC-frequency. This is illustrated in Fig. 1B. The online appendix provides a detailed description of the method. 2.3. Performance evaluation Episodes were divided into non-overlapping 2-s analysis windows. Then, the performance of the method was assessed in terms of sensitivity (Se), i.e., the percentage of windows that did have compressions for which a CC-rate was reported; positive predictive value (PPV), defined as the percentage of windows that truly contained CCs from the total of windows for which CC-rate was reported; and an accuracy factor (AF) that indicates the percentage of analysis windows for which CC-rate was provided with an error below 10% of the gold-standard rate, calculated as described in Section 2.2. For the analysis windows for which CC-rate was provided, the errors were plotted as a function of the gold-standard rate, and the 95 percentile of unsigned errors was computed. The power of the method to provide alarms for non-adequate CC-rates was evaluated. We adopted a tolerance of 10 cpm for the lower limit to avoid exposing the rescuer to excessive ‘low-rate’ alarms, as suggested by Kramer-Johansen et al.21 Thus, the limits used for analysis were 90–120 cpm. The CCF was computed for each episode as the percentage of analysis windows containing CCs. Intervals with spontaneous circulation were identified using the ECG, the force, and the TTI signal, and were excluded for the CCF calculation.
Please cite this article in press as: González-Otero DM, et al. Chest compression rate feedback based on transthoracic impedance. Resuscitation (2015), http://dx.doi.org/10.1016/j.resuscitation.2015.05.027
112 113 114 115 116 117 118 119 120 121 122
123
124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151
152
153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172
G Model
ARTICLE IN PRESS
RESUS 6432 1–7
D.M. González-Otero et al. / Resuscitation xxx (2015) xxx–xxx
3
Fig. 1. Graphical example of the method using two intervals from the TVF&R (A) and the Oslo EMS (B) databases. Gold-standard and estimated rates (GS and CC-rate, respectively) are shown in the bottom panel expressed in cpm.
Table 1 Summary of characteristics of the episodes of each database: duration (s), percentage of the episode in which the patient was intubated, chest compression fraction (CCF), percentage of the episode in which the patient had spontaneous circulation (ROSC period), mean CC-rate, range of the CC-rate and mean CC-depth. The last two rows show the CCF and the mean CC-rate estimated by the method. All values are expressed as mean (standard deviation). An asterisk (* ) next to the name of the metric indicates statistically significant differences between databases. NA stands for Not Available. Sister (n = 60) Metrics (from gold standard) 1708.9 (542.2) Duration (s) 78.5 (26.5) Intubation period (%) 66.8 (14.2) CCF* (%) 5.6 (14.3) ROSC period (%) * Mean CC-rate (cpm) 120.3 (13.9) 60.0–200.0 Range CC-rate (cpm) Mean CC-depth* (mm) 35.7 (7.6) Metrics (estimated) 67.2 (13.7) CCF* (%) 119.7 (13.1) Mean CC-rate* (cpm)
173
TVF&R (n = 60)
Oslo EMS (n = 60)
1551.5 (536.0) 85.8 (20.5) 81.7 (7.4) 7.9 (13.7) 110.0 (11.3) 60.0–178.1 38.5 (6.4)
1916.3 (896.0) 78.3 (27.7) 87.9 (5.9) 7.6 (14.9) 111.6 (8.9) 62.2–197.1 NA (NA)
80.5 (8.7) 110.8 (10.3)
84.2 (5.9) 111.7 (8.6)
2.4. Statistical analysis
178
Between-database comparisons were made using ANOVA for variables that passed the Lilliefors normality test, and using the Kruskal–Wallis test otherwise. The estimated and the goldstandard metrics were compared using the Mann–Whitney U-test. P-values
