Reliability and accuracy of the thoracic impedance signal for measuring cardiopulmonary resuscitation quality metrics

Erik Alonso, Jesús Ruiz, Elisabete Aramendi, Digna González-Otero, Sofía Ruiz de Gauna, Unai Ayala, James K Russell, Mohamud Daya. Reliability and accuracy of the thoracic impedance signal for measuring cardiopulmonary resuscitation quality metrics. Resuscitation, Volume 88, 2015, Pages 28-34, ISSN 0300-9572, https://doi.org/10.1016/j.resuscitation.2014.11.027. (https://www.sciencedirect.com/science/article/pii/S0300957214008818) Abstract Aim: To determine the accuracy and reliability of the thoracic impedance (TI) signal to assess cardiopulmonary resuscitation (CPR) quality metrics. Methods: A dataset of 63 out-of-hospital cardiac arrest episodes containing the compression depth (CD), capnography and TI signals was used. We developed a chest compression (CC) and ventilation detector based on the TI signal. TI shows fluctuations due to CCs and ventilations. A decision algorithm classified the local maxima as CCs or ventilations. Seven CPR quality metrics were computed: mean CC-rate, fraction of minutes with inadequate CC-rate, chest compression fraction, mean ventilation rate, fraction of minutes with hyperventilation, instantaneous CC-rate and instantaneous ventilation rate. The CD and capnography signals were accepted as the gold standard for CC and ventilation detection respectively. The accuracy of the detector was evaluated in terms of sensitivity and positive predictive value (PPV). Distributions for each metric computed from the TI and from the gold standard were calculated and tested for normality using one sample Kolmogorov-Smirnov test. For normal and not normal distributions, two sample t-test and Mann-Whitney U test respectively were applied to test for equal means and medians respectively. Bland-Altman plots were represented for each metric to analyze the level of agreement between values obtained from the TI and gold standard. Results: The CC/ventilation detector had a median sensitivity/PPV of 97.2%/97.7% for CCs and 92.2%/81.0% for ventilations respectively. Distributions for all the metrics showed equal means or medians, and agreements >95% between metrics and gold standard was achieved for most of the episodes in the test set, except for the instantaneous ventilation rate. Conclusion: With our data, the TI can be reliably used to measure all the CPR quality metrics proposed in this study, except for the instantaneous ventilation rate. Keywords: Automated external defibrillator; Cardiopulmonary resuscitation quality; Chest compression; Thoracic impedance; Ventilations. Reliability and accuracy of the thoracic impedance signal for measuring cardiopulmonary resuscitation quality metrics Erik Alonso∗,a, Jes´us Ruiza, Elisabete Aramendia, Digna Gonz´alez-Oteroa, Sof´ıa Ruiz de Gaunaa, Unai Ayalaa, James K. Russellb, Mohamud Dayac Affiliation and addresses: aCommunications Engineering Department. University of the Basque Country UPV/EHU. Alameda Urquijo S/N 48013 Bilbao, Spain bPhilips Healthcare. Bothell, WA 98021, United States cDepartment of Emergency Medicine. Oregon Health & Science University. 97239-3098 Portland, OR, United States Corresponding author: ∗ Erik Alonso email: erik [email protected] Tel. : +34946017384 Fax. : +34946014259 Word counts: Abstract: 258 words Paper : 3085 words Abstract Aim: To determine the accuracy and reliability of the thoracic impedance (TI) signal to assess cardiopulmonary resuscitation (CPR) quality metrics. Methods: A dataset of 63 out-of-hospital cardiac arrest episodes containing the compression depth (CD), capnography and TI signals was used. We developed a chest compression (CC) and ventilation detector based on the TI signal. TI shows fluctuations due to CCs and ventilations. A decision algorithm classified the local maxima as CCs or ventilations. Seven CPR quality metrics were computed: mean CC-rate, fraction of minutes with inadequate CC-rate, chest compression fraction, mean ventilation rate, fraction of minutes with hyperventilation, instantaneous CC-rate and instantaneous ventilation rate. The CD and capnography signals were accepted as the gold standard for CC and ventilation detection respectively. The accuracy of the detector was evaluated in terms of sensitivity and positive predictive value (PPV). The reliability of the TI to measure CPR quality metrics was analyzed by comparing statistically the metrics obtained from the TI with those obtained from the gold standard. The accuracy of each metric was analyzed individually for each episode. Results: The CC/ventilation detector had a median sensitivity/PPV of 97.2%/97.7% for CCs and 92.2%/81.0% for ventilations respectively. Distributions for all the metrics showed equal means or medians, and agreements >95% between metrics and gold standard was achieved for most of the episodes in the test set, except for the instantaneous ventilation rate. Conclusion: With our data, the TI can be reliably used to measure all the CPR quality metrics proposed in this study, except for the instantaneous ventilation rate. Keywords Cardiopulmonary Resuscitation Quality, Thoracic Impedance, Chest Compression, Ventilations, Automated External Defibrillator 1. INTRODUCTION1 Delivery of high quality cardiopulmonary resuscitation (CPR) is a key component of current2 resuscitation guidelines.1,2 Minimally interrupted chest compressions (CCs), with a rate of at least3 100 min−1and at least 5 cm depth, are recommended along with allowance for full chest recoil on4 completion of each compression. Guidelines also recommend oxygenation to be provided with a5 ventilation rate of 8-10 min−1in intubated patients. Several studies have revealed that low quality6 CCs, pauses in CCs, and hyperventilation are associated with poorer outcomes in both animals3–9 7 and humans.10–14 Nevertheless, suboptimal quality of CPR is common in both in-hospital and8 out-of-hospital cardiac arrest (OHCA).7,9,12–18 9 Procedures such as episode debriefing or the incorporation of feedback systems into defibrillators10 have been developed to improve CPR quality. A standardized review of resuscitation episodes11 permits reporting of rescuer performance in terms of CPR quality metrics related to CCs and12 ventilations. Efforts have been made to streamline these reports by defining the CPR quality13 metrics that should be reported.19 Systems for real time feedback on CPR have proven effective14 to improve metrics.20 The feedback on too slow CCs or hyperventilation can help guide rescuers15 towards metrics recommended by guidelines.16,21–24 16 In the context of advanced life support (ALS), the monitor/defibrillators may be equipped with17 acceleration/force sensors as well as capnography and pulse oximetry modules. These may be used18 to assist the rescuer on the CPR quality. By contrast, in the context of basic life support (BLS),19 some automated external defibrillators (AEDs) also offer acceleration/force sensors for feedback,20 but these are relatively expensive accessories for widespread use. Often the only patient interface21 is defibrillation pads, and only the electrocardiogram (ECG) and TI signals are available.22 Every CC causes a fluctuation visible in the TI signal. The TI signal can be used to identify23 CCs16,25,26 and ventilations.16,27–30 Algorithms for CC detection using the TI signal are integrated24 into commercial software for episode reviewing, and algorithms for ventilation detection have been25 recently proposed using the TI.27,28 Their global performance is provided in terms of sensitivity26 and positive predictive value (PPV).25 Nevertheless, detailed analysis of the utility of the TI signal27 for CC/ventilation metrics is needed.28 The purpose of this study was to analyze retrospectively OHCA episodes to determine the29 reliability and accuracy of using the TI to evaluate seven quality metrics of the CCs and the30 ventilations. An efficient detector for CCs and ventilations was developed, and the reliability and31 accuracy of each of the seven CPR metrics was evaluated globally and for each patient.32 2. MATERIALS AND METHODS33 2.1. Data materials34 The dataset used in this study was a subset of a large out-of-hospital cardiac arrest (OHCA)35 registry containing 623 episodes maintained by the Tualatin Valley Fire & Rescue agency (Tigard,36 Oregon, USA). The episodes, one per patient, were collected using the Philips HeartStart MRx37 monitor/defibrillator between 2006 and 2009. From the original database only 199 episodes had38 the compression depth signal (CD), the TI signal and the capnogram. Those episodes where39 the three signals were concurrent for at least 20 min were selected. A total of 63 episodes with40 mean (SD) duration 41 (11) min comprised the dataset. The CD signal (sampling rate 250 Hz)41 was computed from the force and the acceleration recorded through a CPR assist pad. The TI42 signal was recorded through the defibrillation pads by applying a sinusoidal excitation current43 (32 kHz, 3 mA peak-to-peak) with a resolution of 0.74 mΩ per least significant bit, a bandwidth44 of 0 −80 Hz and a sampling rate of 200 Hz. The capnogram was acquired using Microstream45 (sidestream acquisition) with a frequency rate of 40 Hz and a resolution of 0.004 mmHg per bit.46 For the CC-detection the CD was used as gold standard. The instants of CCs were automatically47 marked applying a fixed threshold of -15 mm and manually reviewed (panel a in Fig. 1). For48 ventilation detection, the capnogram was used as gold standard and ventilations were annotated49 independently by three experienced biomedical engineers relying on visual inspection (panel a in50 Fig. 1). Intervals where capnography was disconnected (0.90% of the time) or uninterpretable51 because the ventilation pattern was not clearly recognizable (11.65% of the time), were excluded52 from the analysis. A total of 2575 minutes were analyzed, which included 110286 CCs and 1758653 ventilations. Episodes were randomly allocated to training and test sets, 32 and 31 episodes54 respectively.55 2.2. Simultaneous CC/ventilation detector56 An algorithm that simultaneously detects CCs and ventilations was developed based on the57 TI signal. The TI signal, z[n], where nis the time sample number, was digitally preprocessed.58 Local maxima were identified, and waveform features characterizing the signal in their vicinity59 were extracted. A decision algorithm based on the waveform features determined whether or not60 each local maximum was CC or ventilation (Fig. 2).61 Signal preprocessing62 The z[n] signal was preprocessed as shown in Fig. 2. For CC detection, a low-pass filter with63 a cut-off frequency of fc1= 1.8 Hz was applied in order to remove high frequency noise and to64 retain fluctuations due to CCs and ventilations. The resulting signal is denoted as zc[n]. For65 ventilation detection, z[n] was low-pass filtered at a cut-off frequency of fc2= 0.6 Hz aiming to66 suppress fluctuations due to CCs as well as high frequency noise. In this case, the signal obtained67 is expressed as zv[n].68 Peak detection and feature extraction69 The zc[n] and zv[n] signals were analyzed independently to detect the instants of the local70 maxima. For each local maximum detected in zc[n] and zv[n], which might be a potential CC or71 ventilation respectively, the following waveform features were computed:72 •A1:The trough-to-peak amplitude of the fluctuation.73 •A2:The peak-to-trough amplitude of the fluctuation.74 •d1:The duration of the trough-to-peak rise.75 •d2:The duration of the peak-to-trough fall.76 The features and locations of each fluctuation in zc[n] and zv[n] were stored in the vectors77 vcand vvrespectively (Fig. 2). Panel b in Fig. 1 illustrates, from top to bottom respectively,78 examples of z[n], zc[n] and zv[n] signals, where the extracted features are depicted.79 Decision algorithm80 For CC detection, the decision algorithm decided whether a local maximum was considered as81 CC, based on the extracted features and location of each fluctuation, vc. The algorithm evaluated82 the duration of the fluctuation (d1+d2) against a static threshold and the mean amplitude, mean83 value of A1and A2, against a dynamic threshold given by:84 Thc=Wc· Nc X k=1 A1k+A2k 2(1) Thcrepresents the weighted average of the mean amplitude of the last 6 detected CCs, and85 Wcdenotes the weighting factor. The fluctuation was classified as CC if its duration and mean86 amplitude were above the thresholds established by the decision algorithm. A refractory period87 between consecutive CCs was considered to avoid possible false positives.88 For ventilation detection, the decision algorithm classified a local maximum as ventilation based89 on the extracted features and location of each fluctuation, vv. The decision algorithm evaluated the90 inflation time (d1) against a static threshold, and the inflation amplitude, A1, against a dynamic91 threshold defined as follows:92 Thv=Wv· Nv X k=1 min(A1k, A2k) (2) Thvdescribes the weighted average of the minimum amplitude of the last 17 detected93 ventilations, and Wvdenotes the weighting factor. The fluctuation was classified as ventilation94 if its inflation time and inflation amplitude were above the thresholds established by the decision95 algorithm. A refractory period between consecutive ventilations was considered in order to avoid96 possible false positives.97 The instants of CCs and ventilations detected by the decision algorithm were denoted as tcand98 tvrespectively (Fig. 2).99 2.3. CPR quality metrics100 Seven different CPR quality metrics were computed using tcand tvinstants:101 •Mean CC-rate: Mean frequency of CCs during CC-series. Consecutive CCs separated by102 less than 1.5 s formed a CC-series.19 A single value per episode was reported.103 •Fraction of minutes with inadequate CC-rate (FMIR): Proportion of minutes with104 CC-rate below 90 cpm or above 120 cpm.19 A tolerance of 10 cpm from the lower limit105 (100 cpm) recommended by the 2010 resuscitation guidelines1was established to consider106 a CC-rate as inadequate. A single value per episode was reported.107 •Chest compression fraction (CCF): Proportion of time without spontaneous circulation108 during which CCs were provided.31 A single value per episode was reported.109 •Mean ventilation rate: Mean value of the ventilations provided every minute of the110 episode.19 A single value per episode was reported.111 Acknowledgements257 This work received financial support from the Ministerio de Econom´ıa y Competitividad of258 Spain through the projects TEC2012-31928 and TEC2012-31144, from the University of the Basque259 Country (UPV/EHU) through the unit UFI11/16 and from the Programa de Formaci´on de Personal260 Investigador del Departamento de Educaci´on, Universidades e Investigaci´on del Gobierno Vasco261 through the grants BFI-2010-174, BFI-2010-235 and BFI-2011-166.262 References263 [1] Nolan JP, Soar J, Zideman DA, et al. European Resuscitation Council Guidelines for Resuscitation 2010 Section264 1. Executive summary. Resuscitation 2010;81(10):1219–1276.265 [2] Meaney PA, Bobrow BJ, Mancini ME, et al. Cardiopulmonary Resuscitation Quality: Improving Cardiac266 Resuscitation Outcomes Both Inside and Outside the Hospital A Consensus Statement From the American267 Heart Association. Circulation 2013;128(4):417–435.268 [3] Babbs CF, Voorhees WD, Fitzgerald KR, Holmes HR, Geddes LA. Relationship of blood pressure and flow269 during CPR to chest compression amplitude: Evidence for an effective compression threshold. Annals of270 Emergency Medicine 1983;12(9):527–532.271 [4] Fitzgerald KR, Babbs CF, Frissora HA, Davis RW, Silver DI. Cardiac output during cardiopulmonary272 resuscitation at various compression rates and durations. American Journal of Physiology - Heart and273 Circulatory Physiology 1981;241(3):H442–H448.274 [5] Sunde K, Wik L, Naess PA, Grund F, Nicolaysen G, Steen PA. Improved haemodynamics with increased275 compression-decompression rates during ACD-CPR in pigs. Resuscitation 1998;39(3):197–205.276 [6] Cheifetz IM, Craig DM, Quick G, et al. Increasing tidal volumes and pulmonary overdistention adversely affect277 pulmonary vascular mechanics and cardiac output in a pediatric swine model. Critical care medicine 1998;278 26(4):710–716.279 [7] Milander MM, Hiscok PS, Sanders AB, Kern KB, Berg RA, Ewy GA. Chest compression and ventilation rates280 during cardiopulmonary resuscitation: the effects of audible tone guidance. Academic emergency medicine:281 official journal of the Society for Academic Emergency Medicine 1995;2(8):708–713.282 [8] Yu T, Weil MH, Tang W, et al. Adverse Outcomes of Interrupted Precordial Compression During Automated283 Defibrillation. Circulation 2002;106(3):368–372.284 [9] Aufderheide TP, Sigurdsson G, Pirrallo RG, et al. Hyperventilation-induced hypotension during285 cardiopulmonary resuscitation. Circulation 2004;109(16):1960–1965.286 [10] Wik L, Steen PA, Bircher NG. Quality of bystander cardiopulmonary resuscitation influences outcome after287 prehospital cardiac arrest. Resuscitation 1994;28(3):195–203.288 [11] Gallagher EJ, Lombardi G, Gennis P. Effectiveness of bystander cardiopulmonary resuscitation and survival289 following out-of-hospital cardiac arrest. JAMA: the journal of the American Medical Association 1995;290 274(24):1922–1925.291 [12] Abella BS, Sandbo N, Vassilatos P, et al. Chest compression rates during cardiopulmonary resuscitation are292 suboptimal: a prospective study during in-hospital cardiac arrest. Circulation 2005;111(4):428–434.293 [13] Edelson DP, Abella BS, Kramer-Johansen J, et al. Effects of compression depth and pre-shock pauses predict294 defibrillation failure during cardiac arrest. Resuscitation 2006;71(2):137–145.295 [14] Wik L, Kramer-Johansen J, Myklebust H, et al. Quality of cardiopulmonary resuscitation during out-of-hospital296 cardiac arrest. JAMA: the journal of the American Medical Association 2005;293(3):299–304.297 [15] Abella BS, Alvarado JP, Myklebust H, et al. Quality of cardiopulmonary resuscitation during in-hospital cardiac298 arrest. JAMA: the journal of the American Medical Association 2005;293(3):305–310.299 [16] Kramer-Johansen J, Myklebust H, Wik L, et al. Quality of out-of-hospital cardiopulmonary resuscitation with300 real time automated feedback: a prospective interventional study. Resuscitation 2006;71(3):283–292.301 [17] Perkins GD, Boyle W, Bridgestock H, et al. Quality of CPR during advanced resuscitation training.302 Resuscitation 2008;77(1):69–74.303 [18] Kampmeier TG, Lukas RP, Steffler C, et al. Chest compression depth after change in CPR guidelines-Improved304 but not sufficient. Resuscitation 2014;85(4):503–508.305 [19] Kramer-Johansen J, Edelson DP, Losert H, K¨ohler K, Abella BS. Uniform reporting of measured quality of306 cardiopulmonary resuscitation (CPR). Resuscitation 2007;74(3):406–417.307 [20] Hostler D, Everson-Stewart S, Rea TD, et al. Effect of real-time feedback during cardiopulmonary resuscitation308 outside hospital: prospective, cluster-randomised trial. BMJ 2011;342.309 [21] Beckers SK, Skorning MH, Fries M, et al. CPREzy improves performance of external chest compressions in310 simulated cardiac arrest. Resuscitation 2007;72(1):100–107.311 [22] Yeung J, Meeks R, Edelson D, Gao F, Soar J, Perkins GD. The use of CPR feedback/prompt devices during312 training and CPR performance: A systematic review. Resuscitation 2009;80(7):743–751.313 [23] Skorning M, Beckers SK, Brokmann JC, et al. New visual feedback device improves performance of chest314 compressions by professionals in simulated cardiac arrest. Resuscitation 2010;81(1):53–58.315 [24] Kern KB, Stickney RE, Gallison L, Smith RE. Metronome improves compression and ventilation rates during316 CPR on a manikin in a randomized trial. Resuscitation 2010;81(2):206–210.317 [25] Stecher FS, Olsen JA, Stickney RE, Wik L. Transthoracic impedance used to evaluate performance of318 cardiopulmonary resuscitation during out of hospital cardiac arrest. Resuscitation 2008;79(3):432–437.319 [26] Aramendi E, Ayala U, Irusta U, Alonso E, Eftestøl T, Kramer-Johansen J. Suppression of the cardiopulmonary320 resuscitation artefacts using the instantaneous chest compression rate extracted from the thoracic impedance.321 Resuscitation 2012;83(6):692–698.322 [27] Risdal M, Aase SO, Stavland M, Eftestøl T. Impedance-based ventilation detection during cardiopulmonary323 resuscitation. IEEE transactions on bio-medical engineering 2007;54(12):2237–2245.324 [28] Edelson DP, Eilevstjønn J, Weidman EK, Retzer E, Hoek TLV, Abella BS. Capnography and chest-wall325 impedance algorithms for ventilation detection during cardiopulmonary resuscitation. Resuscitation 2010;326 81(3):317–322.327 [29] Losert H, Risdal M, Sterz F, et al. Thoracic impedance changes measured via defibrillator pads can monitor328 ventilation in critically ill patients and during cardiopulmonary resuscitation. Critical care medicine 2006;329 34(9):2399–2405.330 [30] Roberts K, Srinivasan V, Niles DE, et al. Does change in thoracic impedance measured via defibrillator electrode331 pads accurately detect ventilation breaths in children? Resuscitation 2010;81(11):1544–1549.332 [31] Iyanaga M, Gray R, Stephens SW, et al. Comparison of methods for the determination of cardiopulmonary333 resuscitation chest compression fraction. Resuscitation 2012;83(5):568–571.334 [32] Idris AH, Daya M, Owens P, et al. High incidence of chest compression oscillations associated with capnography335 during out-of-hospital cardiopulmonary resuscitation. Circulation 2010;122:A83.336 [33] Ayala U, Eftestøl T, Alonso E, et al. Automatic detection of chest compressions for the assessment of337 CPR-quality parameters. Resuscitation 2014, DOI: 10.1016/j.resuscitation.2014.04.007.338 Figure Legends339 Figure 1 Examples of the signals comprised in each episode: CD, capnogram340 and TI signals from top to bottom in panel a. Reviewers’ annotations341 of CCs and ventilations are depicted as red dotted lines in CD and342 capnogram signals respectively. Panel b shows (from top to bottom)343 in more detail the segment colored in blue in the TI signal of panel344 a, z[n], the same segment processed to enhance CCs, zc[n], and the345 same segment processed to enhance ventilations, zv[n]. The extracted346 features are also represented in both zc[n] and zv[n].347 Figure 2 Scheme used to compute the instants of both CCs and ventilations348 using exclusively the TI signal, z[n].349 Figure 3 Box plots showing the performance of the simultaneous350 CC/ventilation detector. The sensitivity and PPV values for351 CCs and ventilations for the test set are depicted in panels a and b352 respectively.353 Figure 4 Bland-Altman plots for CC-rate, FMIR, CCF, ventilation rate and354 FMH are represented in a, b, c, d and e panels respectively. The 95%355 LOA is depicted in black dashed lines.356 Figure 5 Results obtained from the instantaneous rate analysis. a and357 d panels show the Bland-Altman plots for instantaneous CC-rate358 and instantaneous ventilation rate respectively. The percentage359 of instantaneous rates with an error above 10% in CC-rate and360 percentage of instantaneous rates with an error greater then 2 min−1 361 in ventilation rate are represented per episode in panels b and e. The362 percentage of false rates for instantaneous CC-rate and instantaneous363 ventilation rate are shown in panels c and f respectively.364 Figure 6 (Supp.) Examples of uninterpretable intervals in the capnogram.365 From top to bottom intervals with: questionable inflation/deflation366 pattern, CPR artifact, unusual waveform, and CPR artifact in the367 10 −40 s interval.368 Figure 7 (Supp.) Two different time intervals corresponding to the same369 episode are shown in panels a and b. The CD, z[n], and zc[n] signals370 can be observed in both panels from top to bottom respectively.371 Red dotted lines depict the instants of the CCs marked in the372 gold standard. Red triangles represent the instants of the correctly373 detected CCs (true positives), whereas black triangles represent the374 instants of the incorrectly detected CCs (false positives). In zc[n] of375 panel a, the errors in the detection due to the second harmonic of the376 TI fluctuation are illustrated. However, there are no detection errors377 in panel b as there is no second harmonic in the TI.378 16 Figure 8 (Supp.) Two different time intervals corresponding to the same379 episode are shown in panels a and b. The capnography, z[n], and zv[n]380 signals can be observed in both panels from top to bottom respectively.381 Red dotted lines depict the instants of the ventilations marked in382 the gold standard. Red triangles represent the instants of the383 correctly detected ventilations (true positives), whereas black triangles384 represent the instants of the incorrectly detected ventilations (false385 positives). In zv[n] of panel a, there are no detection errors.However,386 there are two detection errors in panel b due to fluctuations in the TI387 not corresponding to ventilations as can be seen in the capnogram.388 CD (mm) −30 −15 0 Capnogram (mmHg) 0 10 20 Time (s) a) TI (Ω) 010 20 104 107 110 z[n] (Ω) 104 107 110 A2 A1 d1 d2 zc[n] (Ω) 103 106 109 Time (s) b) A2 A1 d1 d2 zv[n] (Ω) 17.11 19.11 21.11 105 106 107 Figure 1: Examples of the signals comprised in each episode: CD, capnogram and TI signals from top to bottom in panel a. Reviewers’ annotations of CCs and ventilations are depicted as red dotted lines in CD and capnogram signals respectively. Panel b shows (from top to bottom) in more detail the segment colored in blue in the TI signal of panel a, z[n], the same segment processed to enhance CCs, zc[n], and the same segment processed to enhance ventilations, zv[n]. The extracted features are also represented in both zc[n] and zv[n]. z[n] Low-Pass Filter fc1 Low-Pass Filter fc2 zc[n] zv[n] Peak Detection & Feature Extraction vc vv Decision Algorithm Decision Algorithm tc tv Figure 2: Scheme used to compute the instants of both CCs and ventilations using exclusively the TI signal, z[n]. a) SENS (%) PPV (%) 60 80 100 b) SENS (%) PPV (%) 60 80 100 Figure 3: Box plots showing the performance of the simultaneous CC/ventilation detector. The sensitivity and PPV values for CCs and ventilations for the test set are depicted in panels a and b respectively. Gold standard (cpm) Mean CC ratea) Error (cpm) 90 100 110 120 130 −20 −10 0 10 Gold standard (fraction) FMIRb) Error (fraction) 00.2 0.40.60.8 −0.1 0 0.1 0.2 Gold standard (fraction) CCFc) Error (fraction) 0.4 0.60.81 −0.05 0 0.05 0.1 Gold standard (min−1) Mean ventilation rated) Error (min−1) 4 8 12 16 −4 −2 0 2 Gold standard (fraction) FMHe) Error (fraction) 00.2 0.40.6 −0.04 0 0.04 0.08 Figure 4: Bland-Altman plots for CC-rate, FMIR, CCF, ventilation rate and FMH are represented in a, b, c, d and e panels respectively. The 95% LOA is depicted in black dashed lines.