Exploratory Analysis of Ventilation Signals from Resuscitation Data of

Newborns

Huyen Vu

, Trygve Eftestøl

, Kjersti Engan

, Joar Eilevstjønn

, Jørgen E. Linde

3,5

and Hege Ersdal

4,5

Department of Electrical Engineering and Computer Science, University of Stavanger, Stavanger, Norway

Strategic Research, Laerdal Medical AS, Stavanger, Norway

Department of Pediatrics, Stavanger University Hospital, Stavanger, Norway

Department of Anesthesiology and Intensive Care, Stavanger University Hospital, Stavanger, Norway

Department of Health Sciences, University of Stavanger, Stavanger, Norway

Keywords:

Ventilation Signal Processing, Therapeutic Parameters, Newborns, Statistical Signiﬁcance Test, P-value,

Exploratory Data Analysis Framework.

Abstract:

Prevention of neonatal mortality and morbidity because of birth asphyxia is still a major challenge. In a

non-breathing baby, resuscitation including manual ventilation should start within one minute after birth. In-

formation extracted from ventilation signals might give a good indication of the effectiveness of therapy. A

framework for exploratory data analysis was developed facilitating the development of signal parameters to

identify the relationships between certain signal characteristics and various outcome groups. Low p-values

found for some ventilation parameters indicates that the method presented could be useful in discovering fac-

tors and parameters that might be important for the outcome of ventilation therapy and for guiding further

treatment.

1 INTRODUCTION

In an asphyxiated infant, positive pressure ventilation

should commence within 1 minute after birth accord-

ing to the International Liaison Committee on Resus-

citation guidelines for neonatal resuscitation (ILCOR,

2010). Ventilation of newborns is challenging be-

cause it is time critical and involving complicated in-

teractions between newborn pathophysiology and the

clinical treatment. Determining beneﬁcial character-

istics of different ventilation parameters is necessary

for clinicians in order to improve neonatal treatment

and survival.

Five ventilation parameters were derived from bag

mask ventilation pressure and ﬂow signals. The hy-

pothesis is that certain characteristics of these ven-

tilation parameters could relate to speciﬁc neonatal

conditions and improvement in outcome and thus be

useful for guiding further treatment. “Apgar score”

is an universal scoring system to evaluate the clini-

cal status of the baby after birth, usually performed

at 1, 5, 10 and 20 minutes. The evaluation is based

on ﬁve subjective factors: Appearance (skin color),

Pulse (heart rate), Grimace response (or reﬂex irri-

tability), Activity and muscle tone, and Respiration

(breathing rate and effort). Each factor is scored on a

scale from 0 to 2 (the best score). Thus, the total score

is from 0 to 10. The 1-minute score reﬂects the im-

mediate condition of the baby after birth, whereas the

5-minute score shows how the cardio-respiratory tran-

sition from intrauterine to extrauterine life progresses.

In this exploratory study, the change in Apgar score

from 1 minute to 5 minutes is used as the measure of

effectiveness of ventilation. An improved score for

an asphyxiated baby, implies a positive response to

the ventilation given and vice versa.

The International Liaison Committee on Resus-

citation and the World Health Organization provide

healthcare workers with guidelines for neonatal re-

suscitation (ILCOR, 2010; WHO, 2011). However,

the optimal values of pressure, volume and frequency

during positive pressure ventilation of newborns are

still unclear although research is growing in this ﬁeld

(Perlman et al., 2012). The interaction between me-

chanical ventilation parameters and the physiology of

the lungs has been investigated, concluding that it was

difﬁcult to select an effective respirator setting be-

cause the results were highly dependent on the char-

Vu H., Eftestøl T., Engan K., Eilevstjønn J., E. Linde J. and Ersdal H..

Exploratory Analysis of Ventilation Signals from Resuscitation Data of Newborns.

DOI: 10.5220/0005176800120020

In Proceedings of the International Conference on Bio-inspired Systems and Signal Processing (BIOSIGNALS-2015), pages 12-20

ISBN: 978-989-758-069-7

 2015 SCITEPRESS (Science and Technology Publications, Lda.)

acteristics of the newborns population (Ramsden and

Reynolds, 1987). Software systems to extract ventila-

tion parameters and to provide decision support dur-

ing mechanical ventilation has been previously stud-

ied (Ciurea et al., 2011; Tehrani, 2011; Schulze et al.,

1984). However, very little research is done to un-

derstand the different initial situation of manual posi-

tive pressure ventilation during resuscitation at birth.

In this previous group work (Vu et al., ), we have

analysed ECG signal from the novel Laerdal New-

born Resuscitation Monitor (LNRM) developed by

Laerdal Global Health to investigate effects of venti-

lation parameters during initial resuscitation on heart

rate changes.

In this work, by using signal analysis of wave-

forms in the recorded biophysical signals, we detect

and parameterize events deﬁned as inﬂations from the

start of pressing the bag to the end of exhalation. Fur-

thermore we propose a data explorative approach to

identify ventilation parameters which might be deter-

minant for beneﬁcial neonatal outcome.

2 MATERIALS AND METHODS

2.1 Dataset

This exploratory analysis is based on data from the

“Safer Births” project at Haydom Lutheran Hospital

in Northern Tanzania. Haydom is a resource lim-

ited rural hospital with a great shortage in health care

staff. During the study period, basic newborn re-

suscitations (i.e. stimulation, suction, and bag mask

ventilation) and Apgar scoring were predominantly

conducted by midwives, always observed by trained

research assistants recording the ﬁndings on a data

collection form. The implementation of the research

project was approved by National Institute for Medi-

cal Research (NIMR) in Tanzania and Regional Com-

mittee for Medical and Health Research Ethics (REK)

in Norway.

The Laerdal Newborn Resuscitation Monitor

(LNRM) is a resuscitation monitor designed for re-

search use in low resource settings where newborn

resuscitations usually are performed by a single care

provider. The whole set up is presented in ﬁgure 1.

LRNMs were employed in the labour ward of

Haydom to measure various physiological data such

as ECG signals through dry-electrode ECG measured

on the thorax, CO

concentration, airway pressure

and ﬂow signals. ECG signal was sampled at 500

Hz, CO

signal was sampled at 20 Hz, pressure and

ﬂow signals were sampled at 100 Hz. A ﬂow-sensor

(Acutronic Medical Systems AG) is arranged between

the face mask and the resuscitator bag. The air-

way adapter also connects two plastic tubes with the

LNRM: one tube draws a small sample of exhaled air

(50 ml/min) for standard CO

measurement (Masimo

Sweden AB), and one tube is used for standard pres-

sure measurement.

The dataset contains recording of 218 infants col-

lected between July 2013 to June 2014 with complete

signals. Quality control and management of all re-

search data were performed on a daily basis by local

research staff.

2.2 Processing and Parameterization of

Ventilation Signals

To characterize the ventilations given by healthcare

workers, we detected bag-mask ventilation events by

using two signals: the airway pressure and the ﬂow

signals from the ventilation sensors. We deﬁne ﬁve

ventilation parameters: average ventilation frequency,

average peak inspiratory pressure (PIP), average ex-

pired volume, initial peak inspiratory pressure, and

ventilation time percentage (the percentage of time of

ventilation sequences in the total time of ventilation

including pauses).

Airway Pressure Signal

In this paper, the term “ventilation event” corresponds

to pressing the ventilation bag. The start of one ven-

tilation event is detected when the value of pressure

increases from baseline then exceeds a threshold of 5

mbar. The PIP of each ventilation event is the maxi-

mum value of the pressure signal as illustrated in ﬁg-

ure 2.

A “pause” is deﬁned as the period of time when

the pressure signal value is lower than 5 mbar for

more than 3 seconds. A “ventilation sequence” rep-

resents several continuous ventilation events without

pauses. This is explained in the ﬁgures 3 and 4. The

“total ventilation time” is the time from the ﬁrst to

the last ventilation event including pauses as shown

in ﬁgure 5. Among the ﬁve ventilation parameters,

four of them are derived from the pressure signal and

can be described as follows:

• The average ventilation frequency ( f

) is the ra-

tio of total number of ventilation events (n

) over

the sum of duration of each ventilation sequence

). This parameter represents how fast healthcare

workers press the ventilation bag.

∑

(1)

ExploratoryAnalysisofVentilationSignalsfromResuscitationDataofNewborns

Figure 1: Laerdal Newborn Resuscitation Monitor.

Figure 2: Pressure signal - Detection of ventilation event and PIP.

• The average PIP (PIP

) is the “weighted aver-

age” of mean values of PIPs of ventilation se-

quences (PIP) where the weight is the duration of

each ventilation sequence (t

), which means that

the long sequences dominate to the PIP

than the short sequences. This parameter shows

the average peak inspiratory pressure applied for

ventilation.

PIP

∑

PIP

∗t

∑

(2)

• Initial peak inspiratory pressure (PIP

init

) is the

ﬁrst average PIP value of the ﬁrst ventilation se-

quence that has the duration longer than 0.5 sec-

ond. This parameter represents the initial peak in-

spiratory pressure to open the lung.

• Ventilation time percentage (V T

PRC

) is the per-

centage of sum of duration of all ventilation se-

quences in the total ventilation time (T

). This pa-

rameter shows the percentage of time the rescuer

spending on ventilation during the whole venti-

lation procedure. For example, V T

PRC

= 60 %

means that 60 % of the total ventilation time is

spent for ventilation effort and 40 % of time could

be used for other resuscitation methods.

V T

PRC

∑

∗ 100% (3)

Volume Waveform

The volume waveform is integrated from the ﬂow sig-

nal which is measured by the hot-wire ﬂow sensor

(Acutronic Medical Systems AG). The expired vol-

ume is the amount of air going back through the ﬂow

sensor after one inﬂation. The expired volume is the

volume drop from the maximum value to zero or to a

non zero value when there is mask leakage. The aver-

age expired volume (ExV

) is the mean of all expired

volume values. Figure 6 shows one inﬂation cycle.

2.3 Statistical Approach

We want to investigate if certain patterns of these pa-

rameters show association with speciﬁc neonatal out-

comes (e.g.,improved Apgar score from 1 to 5 min-

utes) when comparing different groups of babies. P-

value method for statistical signiﬁcance test among

groups of babies were calculated by using Wilcoxon

rank-sum testing in Matlab which is a non-parametric

method for non-normal distributed data. P-values

from the signiﬁcance tests are used to represent the

discriminative capability of different ventilation pa-

rameters. The criteria for grouping patients is based

on delta Apgar score (∆AP) which is the change from

Apgar score at 1 minute (AP1) to Apgar score at 5

BIOSIGNALS2015-InternationalConferenceonBio-inspiredSystemsandSignalProcessing

Figure 3: Pressure signal - Two consecutive ventilation sequences.

Figure 4: Pressure signal - One continuous ventilation sequence.

Figure 5: Total ventilation time is the duration from the ﬁrst ventilation event to the last one.

minutes (AP5) deﬁned in equation 4.

∆AP = AP5 − AP1 (4)

Group 1 has ∆AP <= threshold and group 2 has ∆AP

> threshold. Therefore, group 2 is considered to be

the improved group (or the group with positive out-

come). For example, if the Apgar score at 1 minute

and Apgar score at 5 minute of a baby are 3 and 7 re-

spectively, then the ∆AP is 4. And if the threshold is

2, then the baby is categorized into group 2. A low

p-value implies the difference in medians of the two

groups and that the corresponding ventilation param-

eter could have important effect on the result of the

treatment (i.e. improvement in Apgar scoring from 1

minute to 5 minutes).

2.4 Proposed Framework for Data

Exploration

Evaluation Criteria for the Effectiveness of

Ventilation during Resuscitation

We use ∆AP (equation 4) as a criteria to evaluate the

effectiveness of the ventilation effort. For a low p-

value, ∆AP above a threshold might indicate an ap-

propriate resuscitation whereas ∆AP equal or below a

threshold could imply an ineffective one.

Grid of P-values

To see how the medians of ventilation parameters af-

fect various subsets of patients with different starting

conditions, the data were segmented based on Apgar

score at 1 minute. A grid of p-values is deﬁned as a

plot where x (Apgar score at 1 minute lower bound-

ary) and y (Apgar score at 1 minute upper boundary)

axes deﬁne the range of the Apgar score at 1 minute

and colors represent different p-values. Each point

on the grid corresponds to a subset of patients having

the Apgar score at 1 minute in the range indicated by

the two axes of the grid (Apgar score at 1 min lower

boundary, Apgar score at 1 min upper boundary). For

example, the coordinate (x = 3, y = 8) includes a sub-

set of babies having Apgar score 1 minute in the range

from 1 (lower boundary) to 8 (upper boundary). This

subset of patients is categorized into two groups ac-

cording to their ∆AP values. We compute the p-value

of each subset to test the difference in medians of the

ExploratoryAnalysisofVentilationSignalsfromResuscitationDataofNewborns

Figure 6: Volume waveform of one inﬂation cycle.

two groups deﬁned by a threshold value of ∆AP. The

p-value is assigned a color value by using a color map.

The low p-values (< 0.05) are illustrated as triangles

otherwise as circles. Group 2 is the improved group

with higher ∆AP, thus we illustrate the relative differ-

ence between the median value of group 2 (the group

with the positive outcome) compared to group 1 (the

group with the negative outcome) by using upward

and downward triangles. The upward triangles repre-

sent the higher median value of group 2 compared to

group 1 and vice versa. For example, if the p-value is

low (< 0.05) and group 2 has a higher median value

than group 1, the corresponding point in the grid is

represented by an upward triangle, otherwise a down-

ward triangle. The size of each point on the grid is

proportional to the number of patients in the smallest

group, thus illustrating the size of data used for statis-

tical testing. For example, group 1 has 34 patients and

group 2 has 45 patients, then the size of the circle or

the triangle is proportional to the size of group 1 that

has the smaller number of patients.

3 EXPERIMENT AND RESULTS

To ﬁnd an appropriate threshold value of ∆AP for our

experiment, a histogram representing number of pa-

tients with different Apgar score at 1 minute and Ap-

gar score at 5 minutes is shown in ﬁgure 7.

The blue diagonal line on the histogram represents

the ∆AP equal to 0 since the Apgar score at 5 minutes

is equal to Apgar score at 1 minute. The line also sep-

arates the histogram into two parts: the left one with

∆AP > 0 (Apgar score at 5 minutes > Apgar score

at 1 minute) and the right one with ∆AP < 0 (Apgar

score at 5 minutes < Apgar score at 1 minute). The

lines parallel with the main diagonal represent differ-

ent ∆AP thresholds and this is a visualization of how

threshold values separate our data into groups with

positive outcome and negative outcome. We experi-

mented with ∆AP = 2 and ∆AP =3 to ensure that the

number of babies in each group was sufﬁciently large.

Each ventilation parameter has two p-value grids cor-

responding to these different threshold values of ∆AP.

Figure 8a is the p-value grid of PIP

using ∆AP = 2

and ﬁgure 8b is the corresponding box plots of the

red-circled point in the grid (the color bar next to the

p-value grid represents ranges of p-values). The num-

ber on each box plot represents the number of babies

in each group. The box plots show the difference in

medians of two outcome groups, speciﬁcally, group 2

has the lower median value of PIP

Figures 9 and 10 show p-value grids for other ven-

tilation parameters with ∆AP = 2 and ∆AP = 3 re-

spectively.

4 DISCUSSION

P-values below signiﬁcance level (0.05) were found

for some ventilation parameters: average peak in-

spiratory pressure (PIP

), average expired volume

(ExV

), and ventilation time percentage (V T

PRC

)

when comparing different thresholds of ∆AP scores.

In the p-value grids, points in the top left corner usu-

ally have no low p-values. There could be two rea-

sons: one is that for these points we include a wide

range of patients or the diversity of the samples are

high therefore there is no big difference in medians.

The other reason may be that babies born with a high

Apgar score will improve independently of perfor-

mance of ventilation. For points in the left bottom of

the grid that correspond to severely asphyxiated ba-

bies at birth, there are no low p-values either. Effec-

BIOSIGNALS2015-InternationalConferenceonBio-inspiredSystemsandSignalProcessing

Figure 7: Histogram of distribution of data for Apgar score at 1 minute and 5 minutes.

(a) (b)

Figure 8: (a) P-value grid for average PIP (PIP

), ∆AP threshold = 2. The low p-values (< 0.05) are illustrated as triangles

otherwise as circles. The upward-pointing triangles indicate the higher medians and the downward-pointing triangles indicate

the lower medians of group 2 (the improved group with higher ∆AP) in comparison with group 1. The size of each point

in the grid is proportional to the minimum number of patients between two groups. (b) Box plot for the red-circled point

corresponding to a subset of patients having Apgar score 1 minute in the range indicated by the coordinate of that point in the

grid (Apgar score 1 minute ranging from 0 to 6). P-value obtained from statistical signiﬁcance test on PIP

for this subset

is 0.011. The median value of group 2 (including infants in that subset having ∆AP > 2) is 30.68, which is lower than the

median value 34.51 of group 1 (infants with ∆AP <= 2), thus the downward-pointing triangle is used.

tive bag mask ventilation (opening of functional resid-

ual capacity) maybe more difﬁcult in these babies and

ventilations alone might not be sufﬁcient to improve

outcome. In this low-resourced setting, cardiac com-

pression, intubation, and intravenous medication are

not performed. In our exploratory data, we ﬁnd low p-

values indicating different medians of ventilation pa-

rameters more often in the middle range (not very low

and not very high) of Apgar score at 1 minute.

This exploratory test suggests that infants with a

low Apgar at 1 minute who improved to a higher Ap-

gar score at 5 minutes have a lower median value of

ventilation time percentage (V T

PRC

). This could pos-

sibly be explained by that the importance of stimula-

tions was not included in the analysis. Therefore, this

interpretation needs to be further explored by integrat-

ing information about stimulation effort into V T

PRC

parameter in future work.

The explorative nature of this analysis imposes a

limitation that should be noted. As the number of

ExploratoryAnalysisofVentilationSignalsfromResuscitationDataofNewborns

(a) (b)

Figure 9: ∆AP = 2. (a) P-value grids for initial PIP (PIP

init

). (b) P-value grids for ventilation frequency ( f

). (c) P-value

grids for average expired volume (ExV

). (d) P-value grids for ventilation time percentage (V T

PRC

groups tested are exhaustive, it is expected that some

of the low p-values might have appeared by chance.

The intent of the analysis is to identify ventilation pa-

rameters to investigate further in future studies.

5 CONCLUSIONS AND FUTURE

WORK

In this paper, we illustrate a way to process ventila-

tion signals from bag-mask ventilations recorded in

a Newborn Resuscitation Monitor, deﬁne ventilation

parameters and propose a framework for data explo-

ration. We reveal the possibility of using information

extracted from ventilation signals related to improved

Apgar scoring to potentially describe beneﬁcial value-

ranges of different ventilation parameters.

Some ventilation parameters such as average peak

inspiratory pressure (PIP

), average expired vol-

ume (ExV

), and ventilation time percentage (V T

PRC

)

were identiﬁed by the analysis. These ﬁndings might

indicate that these parameters could be determinant

factors for beneﬁcial positive pressure ventilation. As

this is a study using exploratory data analysis, the

ﬁndings needs to be further investigated by using new

data.

For future work, this framework of data analysis

could be extended to test the discriminative capability

of different ventilation parameters with other outcome

deﬁnitions as well. We might also count the stimula-

tion time as the ventilation time when we derive the

ventilation time percentage (V T

PRC

) parameter. There

is also a possibility to combine some ventilation pa-

rameters to investigate their effect altogether on the

resuscitation outcome.

REFERENCES

Ciurea, B. M., Palade, D., and Kostrakievici, S. (2011).

Lung ventilators parameters determination. U.P.B.

Sci. Bull.

ILCOR (2010). Neonatal resuscitation: 2010 interna-

BIOSIGNALS2015-InternationalConferenceonBio-inspiredSystemsandSignalProcessing

(a) (b)

(e)

Figure 10: ∆AP = 3. (a) P-value grids for initial PIP (PIP

init

). (b) P-value grids for ventilation frequency ( f

). (c) P-value

grids for average expired volume (ExV

). (d) P-value grids for ventilation time percentage (V T

PRC

). (e) P-value grids for

average PIP (PIP

tional consensus on cardiopulmonary resuscitation

and emergency cardiovascular care science with treat-

ment recommendations. Resuscitation.

Perlman, J., Kattwinkel, J., and Wyllie, J. (2012). Neonatal

resuscitation: in pursuit of evidence gaps in knowl-

edge. Resuscitation.

Ramsden, C. A. and Reynolds, E. O. (1987). Ventilator

settings for newborn infants. Archives of Disease in

Childhood.

Schulze, K., Stefanski, M., Soulos, T., Masterson, J., Kim,

Y. I., and Rey, H. R. (1984). Computer analysis of

ventilatory parameters for neonates on assisted venti-

lation. Engineering in Medicine and Biology Maga-

zine, IEEE, 3(3):31–33.

ExploratoryAnalysisofVentilationSignalsfromResuscitationDataofNewborns

Tehrani, F. (2011). Efﬁcient Decision Support Systems -

Practice and Challenges in Biomedical Related Do-

main, chapter Computerized Decision Support Sys-

tems for Mechanical Ventilation. ISBN: 978-953-307-

258-6.

Vu, H., Eftestøl, T., Engan, K., Eilevstjønn, J., Linde, J.,

and Ersdal, H. Analysis of heart rate changes in new-

borns to investigate the effectiveness of bag-mask ven-

tilation. In Proceedings of Computing in Cardiology

conference 2014 (To be published).

WHO (2011). Guidelines on basic newborn resuscitation.

WHO Library Cataloguing-in-Publication Data.

BIOSIGNALS2015-InternationalConferenceonBio-inspiredSystemsandSignalProcessing