Andoni Elola, Elisabe e A amendi, Unai I us a, E ik Alonso, Yuanzheng Lu, Ma y P. Chang,
Pamela Owens, Ahamed H. Id is. Capnog aphy: A suppo ool o he de ec ion o e u n o
spon aneous ci cula ion in ou -o -hospi al ca diac a es . Resusci a ion, Volume 142, 2019,
Pages 153-161, ISSN 0300-9572, h ps://doi.o g/10.1016/j. esusci a ion.2019.03.048.
(h ps://www.sciencedi ec .com/science/a icle/pii/S0300957219301327)
Abs ac
Backg ound
Au oma ed de ec ion o e u n o spon aneous ci cula ion (ROSC) is s ill an unsol ed p oblem
du ing ca diac a es . Cu en guidelines ecommend he use o capnog aphy, bu mos
au oma ic me hods a e based on he analysis o he ECG and ho acic impedance (TI) signals.
This s udy analysed he added alue o E CO2 o disc imina ing pulsed (PR) and pulseless
(PEA) hy hms and i s po en ial o de ec ROSC.
Ma e ials and me hods
A o al o 426 ou -o -hospi al ca diac a es cases, 117 wi h ROSC and 309 wi hou ROSC, we e
analysed. Fi s , E CO2 alues we e compa ed o ROSC and no ROSC cases. Second, 5098
a e ac ee 3-s long segmen s we e au oma ically ex ac ed and labelled as PR (3639) o PEA
(1459) using he ins an o ROSC anno a ed by he clinician on scene as gold s anda d. Machine
lea ning classi ie s we e designed using ea u es ob ained om he ECG, TI and he E CO2
alue. Thi d, he cases we e e ospec i ely analysed using he classi ie o disc imina e cases
wi h and wi hou ROSC.
Resul s
E CO2 alues inc eased signi ican ly om 41 mmHg 3-min be o e ROSC o 57 mmHg 1-min
a e ROSC, and E CO2 was signi ican ly la ge o PR han o PEA, 46 mmHg/20 mmHg
(p < 0.05). Adding E CO2 o he machine lea ning models inc eased hei a ea unde he cu e
(AUC) by o e 2 pe cen age poin s. The combina ion o ECG, TI and E CO2 had an AUC o he
de ec ion o pulse o 0.92. Finally, he e ospec i e analysis showed a sensi i i y and
speci ici y o 96.6% and 94.5% o he de ec ion o ROSC and no-ROSC cases, espec i ely.
Conclusion
Adding E CO2 imp o es he pe o mance o au oma ic algo i hms o pulse de ec ion based on
ECG and TI. These algo i hms can be used o iden i y pulse on si e, and o e ospec i ely
iden i y cases wi h ROSC.
Keywo ds: Re u n o spon aneous ci cula ion (ROSC), ROSC de ec ion, Capnog aphy, End- idal
CO2 (E CO2), Elec oca diog am (ECG), Tho acic impedance
Capnog aphy: A suppo ool o he De ec ion o Re u n o Spon aneous
Ci cula ion in Ou -o -Hospi al Ca diac A es
Andoni Elola∗,a, Elisabe e A amendia, Unai I us aa, E ik Alonsoa, Yuanzheng Lub, Ma y P.
Changc, Pamela Owensc, Ahamed H. Id isc
A ilia ion and add esses:
aCommunica ions Enginee ing Depa men .
Uni e si y o he Basque Coun y UPV/EHU.
48013 Bilbao, Spain
bEme gency and Disas e Medicine Cen e .
The Se en h A ilia ed Hospi al, Sun Ya -sen Uni e si y.
Shenzhen, China
cDepa men o Eme gency Medicine.
Uni e si y o Texas Sou hWes e n Medical Cen e (UTSW).
Dallas, Uni ed S a es
Co esponding au ho :
∗
Andoni Elola
email: [email protected]
Tel. : +34946013956
Fax. : +34946014259
Wo d coun s: 3255
Abs ac wo d coun s: 281
Abs ac
Backg ound: Au oma ed de ec ion o e u n o spon aneous ci cula ion (ROSC) is s ill an
unsol ed p oblem du ing ca diac a es . Cu en guidelines ecommend he use o capnog aphy,
bu mos au oma ic me hods a e based on he analysis o he ECG and ho acic impedance (TI)
signals. This s udy analysed he added alue o E CO2 o disc imina ing pulsed (PR) and pulseless
(PEA) hy hms and i s po en ial o de ec ROSC.
Ma e ials and me hods: A o al o 426 ou -o -hospi al ca diac a es cases, 117 wi h ROSC and
309 wi hou ROSC, we e analysed. Fi s , E CO2 alues we e compa ed o ROSC and no ROSC
cases. Second, 5098 a e ac ee 3-second long segmen s we e au oma ically ex ac ed and labelled
as PR (3639) o PEA (1459) using he ins an o ROSC anno a ed by he clinician on scene as
gold s anda d. Machine lea ning classi ie s we e designed using ea u es ob ained om he ECG,
TI and he E CO2 alue. Thi d, he cases we e e ospec i ely analysed using he classi ie o
disc imina e cases wi h and wi hou ROSC.
Resul s: E CO2 alues inc eased signi ican ly om 41 mmHg 3-min be o e ROSC o 57 mmHg
1-min a e ROSC, and E CO2was signi ican ly la ge o PR han o PEA, 46 mmHg/20 mmHg
(p < 0.05). Adding E CO2 o he machine lea ning models inc eased hei a ea unde he cu e
(AUC) by o e 2 pe cen age poin s. The combina ion o ECG, TI and E CO2had an AUC o he
de ec ion o pulse o 0.92. Finally, he e ospec i e analysis showed a sensi i i y and speci ici y o
96.6% and 94.5% o he de ec ion o ROSC and no-ROSC cases, espec i ely.
Conclusion: Adding E CO2imp o es he pe o mance o au oma ic algo i hms o pulse
de ec ion based on ECG and TI. These algo i hms can be used o iden i y pulse on si e, and
o e ospec i ely iden i y cases wi h ROSC.
Keywo ds
Re u n o spon aneous ci cula ion (ROSC), ROSC de ec ion, capnog aphy, end- idal CO2(E CO2),
elec oca diog am (ECG), ho acic impedance
1. In oduc ion1
The main goal o esusci a i e e o s du ing ou -o -hospi al ca diac a es (OHCA) is o achie e2
e u n o spon aneous ci cula ion (ROSC). Those e o s include high quali y ca diopulmona y3
esusci a ion (CPR), du ing which ches comp essions should be minimally in e up ed o ac ions4
like hy hm analysis o pulse checks. Cu en pulse de ec ion me hods such as ca o id pulse check,5
o checking o signs o li e as ecommended by he cu en guidelines, a e bo h ime consuming and6
inaccu a e [1–5]. The e is he e o e a need o accu a e and au oma ed pulse de ec ion me hods [6]7
ha can be used by eme gency medical pe sonnel as a decision suppo ool o iden i y ROSC. Such8
me hods would con ibu e o imp o e he apy, educe and sho en pauses in ches comp essions,9
and inc ease su i al a es [7, 8].10
Cu en guidelines suppo he use o capnog aphy o ea ly de ec ion o ROSC [9]. Highe 11
alues o end idal CO2(E CO2), and sudden inc eases in E CO2ha e been linked o ROSC in12
OHCA [10–13]. Al hough some medical algo i hms exis o he de ec ion o ROSC using E CO2
13
alues [14], he only au oma ic me hod based on capnog aphy was ecen ly p oposed [15].14
Mos au oma ic me hods o he de ec ion o pulse in OHCA es on he analysis o he ECG and15
he ho acic impedance (TI). The TI signal shows low ampli ude luc ua ions o e e y e ec i e16
hea bea [16], so ea u es cha ac e izing he TI signal ha e been p oposed alone [17–19], o in17
combina ion wi h ECG ea u es [20–22] o he de ec ion o pulse. In his con ex , de ec ion o 18
pulse is amed as a classi ica ion p oblem wi h wo ypes o o ganized hy hms: pulse-gene a ing19
hy hms (PR) and pulseless elec ical ac i i y (PEA).20
The pu pose o his s udy was o e alua e he added alue o capnog aphy o he classi ica ion21
o PR/PEA du ing OHCA. Fi s , E CO2 alues we e au oma ically de ec ed in o de o compa e22
he alues be ween pa ien s wi h and wi hou ROSC, and o analyse how E CO2changed as23
he pa ien app oached ROSC. Then, he added alue o E CO2 o PR/PEA classi ica ion was24
e alua ed by de eloping machine lea ning PR/PEA classi ie s.25
2. Ma e ials26
Fo his s udy we analysed 1561 OHCA episodes e ospec i ely, ea ed by he Dallas27
Fo Wo h Cen e o Resusci a ion Resea ch (UTSW, Dallas) using he Philips Hea S a MRx28
de ice be ween 2012 and 2016. The de ice iles included he ECG and TI eco ded h ough29
he de ib illa ion pads wi h sampling equencies o 250 Hz/200 Hz espec i ely, and capnog aphy30
eco ded h ough sides eam acquisi ion wi h a sampling equency o 125 Hz. The elec onic iles31
we e linked o clinical anno a ions and ROSC was de ined as palpable pulse in any essel o any32
leng h o ime. The i s ROSC ins an anno a ed by he escue on scene was he gold s anda d;33
based on ha ins an PR and PEA anno a ions we e made au oma ically and pa ien s wi h ROSC34
and wi hou ROSC we e classi ied.35
The ollowing pa ien inclusion/exclusion c i e ia we e applied. Only episodes wi h TI, ECG36
and capnog aphy we e conside ed (n=835). Cases whe e ROSC was suspec ed bu no anno a ed37
by clinicians on si e we e excluded, which comp ised pa ien s anspo ed o hospi al (n=252), o 38
episodes wi h long pe iods (>2 min) wi hou comp essions p esen ing an o ganized hy hm wi h39
E CO2abo e 25 mmHg (n=26). Episodes wi h suspec ed in e mi en ROSC we e also excluded,40
hese we e episodes in which shocks o ches comp essions (>2 min) we e deli e ed a e he41
anno a ed onse o ROSC (n=76). Fo ou analysis o he ROSC cases, he capnog am had o42
be a ailable a leas 4 minu es be o e and 1 minu e a e he onse o ROSC. I no , he case43
was excluded (n=55). The inal da ase con ained 426 episodes, 117 wi h ROSC and 309 wi hou 44
ROSC.45
Figu e 1 shows a 3-minu e in e al om wo cases o he s udy da ase . In he ROSC case ( op46
panel) E CO2inc eases a ROSC onse , and a e ROSC he hea a e inc eases and he e is pulse47
ela ed ac i i y in he TI. In he no-ROSC case (bo om panel) E CO2is always below 20 mmHg,48
and al hough he hea a e changes du ing PEA he e is no pulse ela ed ac i i y in he TI.49
3. Me hods50
Th ee analyses we e conduc ed: E CO2le els in episodes wi h and wi hou ROSC, de elopmen 51
and e alua ion o a PR/PEA classi ie using ECG/TI segmen s and E CO2 alues, and a case s udy52
o he use o he classi ie o e ospec i ely iden i y cases as ROSC/no-ROSC.53
3.1. Analysis o E CO2le els54
Onse and o se o each en ila ion we e au oma ically delinea ed in he capnog am using55
a me hod in oduced in a p e ious s udy [23]. Fo each en ila ion, E CO2was au oma ically56
calcula ed as he maximum CO2 alue du ing he al eola pla eau (see Figu e 1).57
In ROSC cases, median E CO2le els we e compu ed e e y minu e (ME CO2) in a i e minu e58
in e al a ound ROSC (4-min be o e o 1-min a e ). Simila ly, o pa ien s wi hou ROSC, he59
ME CO2 alues we e compu ed o each one o he las i e minu es o he episode. The ME CO2
60
alue o he las minu e o he episode co esponds o he las minu e be o e he EOE, i.e. he61
ins an when he moni o /de ib illa o was disconnec ed.62
3.2. PR/PEA machine lea ning classi ie 63
Following he classical scheme p oposed in p e ious s udies [20–22], he de ec ion o ROSC64
implies he disc imina ion be ween PR and PEA once an o ganized hy hm is iden i ied by he65
shock ad ice algo i hm. I is he e o e a wo class classi ica ion p oblem, o which, i s , he da ase 66
o PR/PEA segmen s was de ined, and hen a classi ie was designed using ea u es ex ac ed om67
he ECG, TI and capnog aphy signals.68
3.2.1. PR/PEA segmen da ase 69
PR and PEA segmen s o 3.2-s du a ion we e ex ac ed du ing in e als wi h no ches 70
comp ession a e ac s. Pauses in ches comp essions we e au oma ically de ec ed using he71
comp ession dep h signal om he CPR assis pad when a ailable [24], o he TI o he wise [25].72
Segmen s wi h la ge ECG ampli ude oscilla ions (>3.5 mV) we e disca ded as noisy, and hen73
o ganized hy hms (PEA o PR) we e de ec ed du ing he pauses using an o line e sion o a74
hy hm analysis algo i hm o a comme cial au oma ed ex e nal de ib illa o (AED) [26]. In ROSC75
cases, all segmen s be o e ROSC onse we e labelled as PEA, and hose a e ROSC onse as PR76
(see Figu e 1, panel a). In no-ROSC cases, all segmen s we e labelled as PEA (see Figu e 1, panel77
b). A minimum sepa a ion be ween consecu i e segmen s o 20-s was en o ced o os e ECG and78
TI wa e o m di e si y in he segmen s.79
3.2.2. Machine lea ning PR/PEA classi ie 80
Nine PR/PEA classi ica ion ea u es we e compu ed om he mos ecen ly p oposed81
algo i hms, six ECG ea u es in oduced in [27], and h ee TI ea u es [17, 22, 28]. These ECG82
and TI ea u es a e desc ibed in de ail in Appendix A. The ME CO2, he median E CO2in he83
minu e be o e he analysis window (pause in ches comp essions wi h o ganized ECG hy hm) was84
also added. The ea u es we e combined in a Random Fo es (RF) classi ie , a machine lea ning85
algo i hm based on he agg ega e o e o se e al independen ly designed unco ela ed decision ees86
[29]. RF classi ie s ha e shown excellen pe o mance in many classi ica ion p oblems, including87
PR/PEA classi ica ion [27], and a e obus agains anno a ion e o s.88
All pa ien s we e weigh ed equally o ain he RF classi ie and 300 ees we e used. Fo 89
each segmen , he RF classi ie compu es he p obabili y o being PR (pp ), and segmen s we e90
classi ied as PR o pp >0.5 and as PEA o he wise. The classi ie was ained and es ed using91
a pa ien wise 10- old c oss- alida ion p ocedu e [30]. Fo each o he 10 olds, he algo i hm was92
op imized using 90% o he cases, and he accu acy esul s we e ob ained om he emaining 10%93
( es old). This p ocedu e gua an eed ha he op imiza ion o he classi ie and he es ima ion94
o i s accu acy we e done on da a om sepa a e pa ien s, and ha he pe o mance was assessed95
using all a ailable da a.96
3.3. Case s udy: Re ospec i e iden i ica ion o pa ien s wi h ROSC97
Using he PR/PEA classi ie , a simple me hod was de eloped o au oma ically iden i y pa ien s98
wi h ROSC in a e ospec i e analysis o a se o OHCA episodes. This me hod may be used as99
an au oma ed ool o pos a es deb ie ing o anno a ion. Comple e episodes (un il EOE) we e100
p ocessed and he case was labelled as ROSC i om any h ee consecu i e segmen s a leas wo101
we e iden i ied as PR by he classi ie .102
Ou g ound u h was he ROSC ins an anno a ed by clinicians on scene, which disc imina ed103
he g oup o pa ien s wi h ROSC and pa ien s wi hou ROSC, and he de ec ion o episodes wi h104
ROSC was e alua ed using he es se s in he 10- old c oss alida ion p ocedu e.105
3.4. S a is ical analysis106
ME CO2dis ibu ions did no pass he Kolmogo o -Smi no no mali y es , and a e epo ed107
as median and in e qua ile ange (IQR). ME CO2dis ibu ions a di e en imes (wi hin108
ROSC cases) o be ween ROSC/no-ROSC cases we e compa ed using he Mann-Whi ney U es .109
Di e ences we e conside ed signi ican o p < 0.05.110
PR/PEA classi ica ion was e alua ed using Recei e Ope a ing Cha ac e is ic (ROC) cu es,111
and he a ea unde he cu e (AUC) was used as measu e o pe o mance [31]. The Youden index112
was used o de ine he op imal poin in he ROC cu e, which gi es equal impo ance o he113
sensi i i y (SE, o PR segmen s) and speci ici y (SP, o PEA segmen s) [32].114
When he classi ie was used as a e ospec i e ool o iden i y ROSC, SE and SP we e de ined115
as he p opo ion o co ec ly iden i ied ROSC and no-ROSC cases, espec i ely.116
4. Resul s117
The mean (s anda d de ia ion) du a ions we e 58 (23) min and 38 (11) min o he episodes wi h118
and wi hou ROSC, espec i ely. The comme cial AED algo i hm de ec ed 5098 segmen s wi h119
o ganized hy hms. A o al o 3639 PR segmen s we e ex ac ed om episodes wi h ROSC, and120
1459 PEA segmen s, 308 om episodes wi h ROSC and 1151 om episodes wi hou ROSC. Some121
examples o he ex ac ed ECG segmen s can be ound in Figu e 1 and Figu e 4. The median122
(IQR) en ila ion a e pe episode was 7.8 (5.7-10.5) min−1.123
The ME CO2 o ROSC cases we e s a is ically signi ican ly la ge han o no-ROSC cases a 124
all ime-s amps (Figu e 2). Ele a ed E CO2le els we e obse ed in pa ien s wi h ROSC, wi h an125
upwa d end om 41 mmHg (a 3 min be o e ROSC) o 57 mmHg close o ROSC onse (see Figu e126
2 a).127
Figu e 3 shows he ROC cu es o he RF classi ie o di e en ea u es se s. The cu es in128
panel (a) we e calcula ed using he whole da ase , while he cu es in panel (b) we e calcula ed129
excluding he PEA segmen s ex ac ed om pa ien s wi h ROSC, ha is he p e-ROSC PEA130
segmen s. The analysis o he ROC cu es is shown in Table 1. The ROC cu es showed ha he131
AUC o he PR/PEA classi ie inc eased as ea u es om di e en sou ces we e added. Including132
ME CO2in he classi ie inc eased he AUC o all ea u e combina ions, hanks o he added133
unco ela ed in o ma ion. Adding ME CO2 o an ECG-only and o an ECG+TI based classi ie s134
inc eased hei AUCs in 3 and 2-poin s, espec i ely. The bes classi ie combined all ea u es135
and p esen ed an AUC o 0.92 wi h a SE and SP o 84% and 86%, espec i ely (see Table 1).136
ME CO2alone was also a good classi ie (AUC a ound 0.76), he median ME CO2 alues we e 46137
(32-64) mmHg o PR and 20 (8-38) mmHg o PEA segmen s (p < 0.05).138
The accu acy o he classi ie s inc eased when PEAs ha ansi ioned o PR (episodes wi h139
ROSC) we e no included. The accu acy inc ease was on a e age 4-poin s o all classi ie s (see140
Table 1). Signi ican di e ences we e obse ed be ween PEAs in ROSC and no-ROSC cases.141
The ME CO2 alues o he PEA in he ROSC and no-ROSC cases we e 31 (20-44) mmHg and 16142
(7-35) mmHg (p < 0.05), espec i ely. The p obabili ies o being PR, pp , o he classi ie wi h all143
ea u es we e also signi ican ly di e en o hese wo subg oups o PEA, 0.09 (0.03-0.29) o he144
PEA om no-ROSC cases and 0.33 (0.11-0.58) o hose o he ROSC cases (p < 0.05).145
Figu e 4 shows he pe o mance o he PR/PEA classi ie wi h h ee consecu i e segmen s in146
[1] Pe kins GD, S ephenson B, Hulme J, Monsieu s KG. Bi mingham assessmen o b ea hing s udy (BABS).289
Resusci a ion 2005;64(1):109–113.290
[2] Ruppe M, Rei h MW, Widmann JH, e al. Checking o b ea hing: e alua ion o he diagnos ic capabili y291
o eme gency medical se ices pe sonnel, physicians, medical s uden s, and medical laype sons. Annals o 292
eme gency medicine 1999;34(6):720–729.293
[3] Nyman J, Sih onen M. Ca diopulmona y esusci a ion skills in nu ses and nu sing s uden s. Resusci a ion 2000;294
47(2):179–184.295
[4] Ebe le B, Dick W, Schneide T, Wisse G, Doe sch S, Tzano a I. Checking he ca o id pulse check: diagnos ic296
accu acy o i s esponde s in pa ien s wi h and wi hou a pulse. Resusci a ion 1996;33(2):107–116.297
[5] Ochoa FJ, Ramalle-Goma a E, Ca pin e o J, Ga cıa A, Sa alegui I. Compe ence o heal h p o essionals o check298
he ca o id pulse. Resusci a ion 1998;37(3):173–175.299
[6] Babbs CF. We s ill need a eal- ime hemodynamic moni o o CPR. Resusci a ion 2013;84(10):1297–1298.300
[7] Ch is enson J, And usiek D, E e son-S ewa S, e al. Ches comp ession ac ion de e mines su i al in pa ien s301
wi h ou -o -hospi al en icula ib illa ion. Ci cula ion 2009;120(13):1241–1247.302
[8] Be g RA, Sande s AB, Ke n KB, e al. Ad e se hemodynamic e ec s o in e up ing ches comp essions o 303
escue b ea hing du ing ca diopulmona y esusci a ion o en icula ib illa ion ca diac a es . Ci cula ion304
2001;104(20):2465–2470.305
[9] Soa J, Nolan JP, B¨o ige BW, e al. Eu opean esusci a ion council guidelines o esusci a ion 2015: sec ion306
3. Adul ad anced li e suppo . Resusci a ion 2015;95:100–147.307
[10] Ha mann SM, Fa is RW, Di Genna o JL, Robe s JS. Sys ema ic e iew and me a-analysis o end- idal ca bon308
dioxide alues associa ed wi h e u n o spon aneous ci cula ion du ing ca diopulmona y esusci a ion. Jou nal309
o in ensi e ca e medicine 2015;30(7):426–435.310
[11] Sheak KR, Wiebe DJ, Lea y M, e al. Quan i a i e ela ionship be ween end- idal ca bon dioxide and CPR311
quali y du ing bo h in-hospi al and ou -o -hospi al ca diac a es . Resusci a ion 2015;89:149–154.312
[12] Poko n´a M, Neˇcas E, K a och ´ıl J, Skˇ ipsk`y R, And l´ık M, F anˇek O. A sudden inc ease in pa ial p essu e313
end- idal ca bon dioxide (PETCO2) a he momen o e u n o spon aneous ci cula ion. The Jou nal o 314
eme gency medicine 2010;38(5):614–621.315
[13] Lui CT, Poon KM, Tsui KL. Ab up ise o end idal ca bon dioxide le el was a speci ic bu non-sensi i e316
ma ke o e u n o spon aneous ci cula ion in pa ien wi h ou -o -hospi al ca diac a es . Resusci a ion 2016;317
104:53–58.318
[14] Da is DP, Sell RE, Wilkes N, e al. Elec ical and mechanical eco e y o ca diac unc ion ollowing319
ou -o -hospi al ca diac a es . Resusci a ion 2013;84(1):25–30.320
[15] B ink ol P, Bo owski M, Me elmann C, Lukas RP, Pidde-K¨ullenbe g L, Bohn A. P edic ing ROSC in321
ou -o -hospi al ca diac a es using expi a o y ca bon dioxide concen a ion: Is end-de ec ion ins ead o 322
absolu e h eshold alues he key? Resusci a ion 2018;122:19–24.323
[16] Pellis T, Bise a J, Tang W, Weil MH. Expanding au oma ic ex e nal de ib illa o s o include au oma ed de ec ion324
o ca diac, espi a o y, and ca dio espi a o y a es . C i ical ca e medicine 2002;30(4):S176–S178.325
[17] C omie NA, Allen JD, Na a o C, Tu ne C, Ande son JM, Adgey AAJ. Assessmen o he impedance326
ca diog am eco ded by an au oma ed ex e nal de ib illa o du ing clinical ca diac a es . C i ical ca e medicine327
2010;38(2):510–517.328
[18] C omie NA, Allen JD, Tu ne C, Ande son JM, Adgey AAJ. The impedance ca diog am eco ded h ough wo329
elec oca diog am/de ib illa o pads as a de e minan o ca diac a es du ing expe imen al s udies. C i ical330
Ca e Medicine 2008;36(5):1578–1584.331
[19] Lose H, Risdal M, S e z F, e al. Tho acic-impedance changes measu ed ia de ib illa o pads can moni o 332
signs o ci cula ion. Resusci a ion 2007;73(2):221–228.333
[20] Risdal M, Aase SO, K ame -Johansen J, E esol T. Au oma ic iden i ica ion o e u n o spon aneous ci cula ion334
du ing ca diopulmona y esusci a ion. IEEE T ansac ions on Biomedical Enginee ing 2008;55(1):60–68.335
[21] Alonso E, A amendi E, Daya M, e al. Ci cula ion de ec ion using he elec oca diog am and he ho acic336
impedance acqui ed by de ib illa ion pads. Resusci a ion 2016;99:56–62.337
[22] Ruiz JM, de Gauna SR, Gonz´alez-O e o DM, e al. Ci cula ion assessmen by au oma ed ex e nal de ib illa o s338
du ing ca diopulmona y esusci a ion. Resusci a ion 2018;128:158–163.339
[23] A amendi E, Elola A, Alonso E, e al. Feasibili y o he capnog am o moni o en ila ion a e du ing340
ca diopulmona y esusci a ion. Resusci a ion 2017;110:162–168.341
[24] Ayala U, E es øl T, Alonso E, e al. Au oma ic de ec ion o ches comp essions o he assessmen o 342
CPR-quali y pa ame e s. Resusci a ion 2014;85(7):957–963.343
[25] Alonso E, Ruiz J, A amendi E, e al. Reliabili y and accu acy o he ho acic impedance signal o measu ing344
ca diopulmona y esusci a ion quali y me ics. Resusci a ion 2015;88:28–34.345
[26] I us a U, Ruiz J, A amendi E, de Gauna SR, Ayala U, Alonso E. A high- empo al esolu ion algo i hm o346
disc imina e shockable om nonshockable hy hms in adul s and child en. Resusci a ion 2012;83(9):1090–1097.347
[27] Elola A, A amendi E, I us a U, Del Se J, Alonso E, Daya M. ECG-based pulse de ec ion du ing ca diac a es 348
using andom o es classi ie . Medical & biological enginee ing & compu ing 2019;57(2):453–462.349
[28] Wei L, Chen G, Yang Z, Yu T, Quan W, Li Y. De ec ion o spon aneous pulse using he accele a ion signals350
acqui ed om CPR eedback senso in a po cine model o ca diac a es . PloS one 2017;12(12):e0189217.351
[29] B eiman L. Random o es s. Machine lea ning 2001;45(1):5–32.352
[30] S one M. C oss- alida o y choice and assessmen o s a is ical p edic ions. Jou nal o he oyal s a is ical socie y353
Se ies B (Me hodological) 1974;:111–147.354
[31] Zou KH, O’Malley AJ, Mau i L. Recei e -ope a ing cha ac e is ic analysis o e alua ing diagnos ic es s and355
p edic i e models. Ci cula ion 2007;115(5):654–657.356
[32] Ruopp MD, Pe kins NJ, Whi comb BW, Schis e man EF. Youden Index and op imal cu -poin es ima ed om357
obse a ions a ec ed by a lowe limi o de ec ion. Biome ical Jou nal: Jou nal o Ma hema ical Me hods in358
Biosciences 2008;50(3):419–430.359
[33] Kola M, K iˇzma i´c M, Klemen P, G mec ˇ
S. Pa ial p essu e o end- idal ca bon dioxide success ul p edic s360
ca diopulmona y esusci a ion in he ield: a p ospec i e obse a ional s udy. C i ical ca e 2008;12(5):R115.361
[34] Touma O, Da ies M. The p ognos ic alue o end idal ca bon dioxide du ing ca diac a es : a sys ema ic362
e iew. Resusci a ion 2013;84(11):1470–1479.363
[35] G mec ˇ
S, Klemen P. Does he end- idal ca bon dioxide (E CO2) concen a ion ha e p ognos ic alue du ing364
ou -o -hospi al ca diac a es ? Eu opean Jou nal o Eme gency Medicine 2001;8(4):263–269.365
[36] Weise C, Poppe M, S e z F, e al. Ini ial elec ical equency p edic s su i al and neu ological ou come in ou 366
o hospi al ca diac a es pa ien s wi h pulseless elec ical ac i i y. Resusci a ion 2018;125:34–38.367
[37] Skje lo GW, No dse h T, Loennechen JP, Be gum D, Skog oll E. ECG changes du ing esusci a ion o pa ien s368
wi h ini ial pulseless elec ical ac i i y a e associa ed wi h e u n o spon aneous ci cula ion. Resusci a ion369
2018;127:31–36.370
[38] Elola A, A amendi E, I us a U, e al. ECG cha ac e is ics o Pulseless Elec ical Ac i i y associa ed wi h Re u n371
o Spon aneous Ci cula ion in Ou -o -Hospi al Ca diac A es . Resusci a ion 2018;130:e54.372
[39] P osen G, K iˇzma i´c M, Za ˇsnik J, G mec ˇ
S. Impac o modi ied ea men in echoca diog aphically con i med373
pseudo-pulseless elec ical ac i i y in ou -o -hospi al ca diac a es pa ien s wi h cons an end- idal ca bon374
dioxide p essu e du ing comp ession pauses. Jou nal o In e na ional Medical Resea ch 2010;38(4):1458–1467.375
[40] Raimondi M, Sa as ano S, Pamploni G, Molina i S, Degani A, Bellia o M. End- idal ca bon dioxide moni o ing376
and load band de ice o mechanical ca dio-pulmona y esusci a ion: Ne e us he numbe s, belie e a he377
cu es. Resusci a ion 2016;103:e9–e10.378
[41] Ruiz J, Alonso E, A amendi E, e al. Reliable ex ac ion o he ci cula ion componen in he ho acic impedance379
measu ed by de ib illa ion pads. Resusci a ion 2013;84(10):1345–1352.380
Figu e Legends381
Figu e 1 ECG, Tho acic Impedance (TI) and capnog aphy signals o a pa ien 382
wi h ROSC, panel (a), and wi hou ROSC, panel (b). ROSC onse ,383
as anno a ed by a clinician on si e, is ep esen ed by a ed line in384
he i s example. The ex ac ed 3.2-s segmen s a e shaded in g ey385
and he ECG and TI (g een) a e zoomed in. Ches comp ession386
in e als a e depic ed abo e TI signal. In he ROSC case a PEA387
and a PR segmen s we e ex ac ed in he depic ed in e al, and wo388
PEA segmen s in he no-ROSC case. Ven ila ions we e au oma ically389
de ec ed in he CO2cu e, and he au oma ically measu ed E CO2
390
alue is highligh ed wi h ed do s. In he ROSC case a e pulse391
eco e y he ECG p esen s s able and no mal QRS complexes and392
hea a e, and ches comp essions a e s opped so he e is no ac i i y393
in he impedance.394
Figu e 2 Median E CO2(ME CO2) alues and hei in e qua ile anges o 395
cases wi h ROSC (le ) and no-ROSC ( igh ). Fo ROSC cases he396
in e al a ound ROSC onse is analysed, in he no-ROSC cases he397
5 min be o e he end o episode (EOE) a e shown. ME CO2was398
calcula ed as he median E CO2 alue o all en ila ions in a 1-minu e399
in e al be o e he indica ed ime-s amp.400
Figu e 3 ROC cu es o he RF classi ie o di e en ea u e se s. Panel (a)401
shows esul s o he whole da ase , while panel (b) shows he cu es402
a e excluding he PEAs om episodes wi h ROSC. The AUC alue403
o each classi ie is shown be ween pa en heses.404
Figu e 4 Examples o he case s udy. Panels (a), (b) and (c) show a405
co ec ly iden i ied pa ien wi h ROSC, a co ec ly iden i ied pa ien 406
wi hou ROSC, and a pa ien wi hou ROSC inco ec ly iden i ied,407
espec i ely. Each panel depic s he h ee consecu i e PEA/PR408
segmen s analysed. The ex on op o each segmen indica es i s ue409
label ollowed by he p edic ed label by he classi ie . The capnog am410
co esponds o he minu e be o e he onse o he segmen and he411
dashed ho izon al line ep esen s he ME CO2.412
Figu e 5 Time e olu ion o pp o he PEA segmen s as he pa ien s app oach413
ROSC. Blue do s indica e alues o each segmen , and he ed cu e414
is i ed o he median alues o pp e e y 2 minu es.415
15
-2
0
2
ROSC
ECG (mV)
-1000
0
1000
TI (m )
11:08:20 11:09:10 11:10:00 11:10:50 11:11:40 11:12:30 11:13:20
0
50
100
CO2 (mmHg)
(a) A case o a pa ien wi h ROSC
PEA PR
-2
0
2
ECG (mV)
-1000
0
1000
TI (m )
13:12:00 13:12:50 13:13:40 13:14:30 13:15:20 13:16:10 13:17:00
0
50
100
CO2 (mmHg)
(b) A case o a pa ien wi hou ROSC
PEA PEA
Figu e 1: ECG, Tho acic Impedance (TI) and capnog aphy signals o a pa ien wi h ROSC, panel (a), and wi hou
ROSC, panel (b). ROSC onse , as anno a ed by a clinician on si e, is ep esen ed by a ed line in he i s example.
The ex ac ed 3.2-s segmen s a e shaded in g ey and he ECG and TI (g een) a e zoomed in. Ches comp ession
in e als a e depic ed abo e TI signal. In he ROSC case a PEA and a PR segmen s we e ex ac ed in he depic ed
in e al, and wo PEA segmen s in he no-ROSC case. Ven ila ions we e au oma ically de ec ed in he CO2cu e,
and he au oma ically measu ed E CO2 alue is highligh ed wi h ed do s. In he ROSC case a e pulse eco e y
he ECG p esen s s able and no mal QRS complexes and hea a e, and ches comp essions a e s opped so he e is
no ac i i y in he impedance.
-3 min -2 min -1 min ROSC 1 min
ime (min)
10
20
30
40
50
60
70
80
ME CO2 (mmHg)
(a) ROSC cases
-4 min -3 min -2 min -1 min EOE
ime (min)
10
20
30
40
50
60
70
80
ME CO2 (mmHg)
(b) No-ROSC cases
Figu e 2: Median E CO2(ME CO2) alues and hei in e qua ile anges o cases wi h ROSC (le ) and no-ROSC
( igh ). Fo ROSC cases he in e al a ound ROSC onse is analysed, in he no-ROSC cases he 5 min be o e he end
o episode (EOE) a e shown. ME CO2was calcula ed as he median E CO2 alue o all en ila ions in a 1-minu e
in e al be o e he indica ed ime-s amp.
0 20 40 60 80 100
100-SP (%)
0
20
40
60
80
100
SE (%)
(a) All PR/PEA segmen s
ETCO2 (0.76)
ECG (0.88)
ECG+TI (0.90)
ECG+ETCO2 (0.91)
ECG+TI+ETCO2 (0.92)
0 20 40 60 80 100
100-SP (%)
0
20
40
60
80
100
SE (%)
(b) Excluding PEA segmen s om ROSC cases
ETCO2 (0.79)
ECG (0.93)
ECG+TI (0.94)
ECG+ETCO2 (0.95)
ECG+TI+ETCO2 (0.96)
Figu e 3: ROC cu es o he RF classi ie o di e en ea u e se s. Panel (a) shows esul s o he whole da ase ,
while panel (b) shows he cu es a e excluding he PEAs om episodes wi h ROSC. The AUC alue o each
classi ie is shown be ween pa en heses.
-1
0
1
ECG (mV)
PEA/PEA
996 997 998 999
Time (s)
-50
0
50
TI (m )
0
50
100
CO2 (mmHg)
PR/PR
1086 1087 1088
Time (s)
(a) Co ec ly iden i ied pa ien wi h ROSC
PR/PR
1106 1107 1108
Time (s)
-1
0
1
ECG (mV)
PEA/PEA
705 706 707 708
Time (s)
-50
0
50
TI (m )
0
50
100
CO2 (mmHg)
PEA/PEA
772 773 774
Time (s)
(b) Co ec ly iden i ied pa ien wi h no-ROSC
PEA/PEA
792 793 794
Time (s)
-1
0
1
ECG (mV)
PEA/PEA
1763 1764 1765 1766
Time (s)
-50
0
50
TI (m )
0
50
100
CO2 (mmHg)
PEA/PR
1915 1916 1917 1918
Time (s)
(c) Inco ec ly iden i ied pa ien wi h no-ROSC
PEA/PR
1935 1936 1937 1938
Time (s)
Figu e 4: Examples o he case s udy. Panels (a), (b) and (c) show a co ec ly iden i ied pa ien wi h ROSC, a
co ec ly iden i ied pa ien wi hou ROSC, and a pa ien wi hou ROSC inco ec ly iden i ied, espec i ely. Each
panel depic s he h ee consecu i e PEA/PR segmen s analysed. The ex on op o each segmen indica es i s ue
label ollowed by he p edic ed label by he classi ie . The capnog am co esponds o he minu e be o e he onse o
he segmen and he dashed ho izon al line ep esen s he ME CO2.
-10 -8 -6 -4 -2 ROSC
Time (min)
0
0.2
0.4
0.6
0.8
1
pp
Figu e 5: Time e olu ion o pp o he PEA segmen s as he pa ien s app oach ROSC. Blue do s indica e alues o
each segmen , and he ed cu e is i ed o he median alues o pp e e y 2 minu es.
Table Legends416
Table 1 ROC cu e analysis o he machine lea ning classi ie when he whole417
PR/PEA da ase is conside ed and when he PEAs om ROSC cases418
we e excluded. The SE and SP a e gi en o he op imal poin acco ding419
o he Youden index.420
21
