Content uploaded by René Riedl
Author content
All content in this area was uploaded by René Riedl on Aug 30, 2020
Content may be subject to copyright.
Content uploaded by Thomas Fischer
Author content
All content in this area was uploaded by Thomas Fischer on Nov 04, 2018
Content may be subject to copyright.
Analysis of Heart Rate Variability (HRV)
Feature Robustness for Measuring
Technostress
David Baumgartner, Thomas Fischer, René Riedl and Stephan Dreiseitl
Abstract Technostress has become an important topic in the scientific literature,
particularly in Information Systems (IS) research. Heart rate variability (HRV) has
been proposed as a measure of (techno)stress and is widely used in scientific inves-
tigations. The objective of the pilot study reported in this paper is to showcase how
the preprocessing/cleaning of captured data can influence the results and their inter-
pretation, when compared to self-report data. The evidence reported in this paper
supports the notion that NeuroIS scholars have to deliberately make methodological
decisions such as those related to preprocessing of physiological data. It is therefore
crucial that methodological details are presented in NeuroIS papers in order to create
abetterunderstandingofthestudyresultsandtheirimplications.
Keywords Data preprocessing ·Heart rate variability (HRV)
NeuroIS research methodology ·Technostress ·Signal feature ·Stress
1 Introduction/Motivation/Related Work
In recent years, technostress has become an important concept in NeuroIS research
(e.g., [1,2]). It has been argued and demonstrated that mixed methods research,
D. Baumgartner ·S. Dreiseitl
University of Applied Sciences Upper Austria, Hagenberg, Austria
e-mail: david.baumgartner@fh-hagenberg.at
S. Dreiseitl
e-mail: stephan.dreiseitl@fh-hagenberg.at
T. Fischer (B)·R. Riedl
University of Applied Sciences Upper Austria, Steyr, Austria
e-mail: thomas.fischer@fh-steyr.at
R. Riedl
e-mail: rene.riedl@fh-steyr.at
R. Riedl
Johannes Kepler University, Linz, Austria
©SpringerNatureSwitzerlandAG2019
F. D. Davis et al. (eds.), Information Systems and Neuroscience,
Lecture Notes in Information Systems and Organisation 29,
https://doi.org/10.1007/978-3-030-01087- 4_27
221
rene.riedl@jku.at
222 D. Baumgartner et al.
particularly involving physiological measures, is crucial for this research domain
[2–6]. Amongst other methods, the collection of heart rate data and the calculation
of heart rate variability (HRV) as a measure of stress have been proposed as viable
additions, particularly to field studies investigating technostress [7,8].
As part of a larger project, we are currently investigating the potential of sev-
eral data collection methods for technostress research in a longitudinal field study.
This includes heart rate data, which we collect using consumer level devices. In this
paper, we report on a pilot study in which we tested the feasibility of letting individ-
uals track their own heart rate during their working hours using a chest belt with a
heart rate sensor. The objective of this pilot study is to assess the quality of the data
generated in this manner. We showcase how the preprocessing/cleaning of captured
data can influence the results and their interpretation when compared to self-report
data, which we obtained by having study participants fill out a technostress question-
naire. By showing how data collected using consumer-grade devices can be useful to
assess individual stress levels, we also specifically seek to support the call for further
technostress research in field settings [2,9].
In particular, we are interested in how data cleaning influences the correlation
coefficient between HRV data and self-report data. While the HRV analysis methods
investigated here are in no way particular to technostress research, they are never-
theless a necessary first step in every data processing endeavor.
In Sect. 2, we present an overview of the data collection procedures that were
applied, the heart rate features that were extracted and the how the features were
preprocessed. Then, in Sect. 3, we present our results regarding the influence of
preprocessing methods on the results and their interpretation. Finally, in Sect. 4,we
discuss our findings and present some recommendations for future research utilizing
heart rate data in technostress research.
2 Materials and Methods
2.1 Data Collection
We coll e c ted our dat a i n t he week of 11 / 2 7/2017– 1 2 /01/201 7 u s i ng a Polar H7 c h e st
belt in combination with a smartphone app1that collected the captured data. Fifteen
employees (12 female, 3 male) of a publishing company with its headquarters close to
Salzburg, Austria, participated in the study. They were instructed to put on the chest
belt after they had arrived at work and to start the data collection on the smartphone
app. They stopped the data collection upon leaving the workplace. After removing
unusable data, we retained 44 samples from 11 study participants from one week
of data collection. At the end of the workweek, which for most of the participants
was Friday, the participants were also invited to take part in an online survey, which
1https://itunes.apple.com/at/app/heart-rate-variability-logger/id683984776?mt=8 [03/05/2017].
rene.riedl@jku.at
Analysis of Heart Rate Variability (HRV) Feature … 223
included questions on their technostress level using a German version [10]ofthe
“Technostress Creators” questionnaire by Ragu-Nathan et al. [11]. The questionnaire
was deliberately handed out after the workweek as we wanted to avoid any changes
in perception (e.g., situations that are pointed out as stressful in the questionnaire are
then seen as more stressful due to heightened awareness) that could bias our results.
2.2 Heart Rate Features
The literature reports on a variety of indicators extracted from electrocardiography
(ECG) recordings that can help in assessing HRV [12,13]. In general, most of these
indicators are calculated either in the time-domain or the frequency-domain of the
signal. When analyzing the signal in the time-domain, the most relevant aspect is the
time interval between subsequent peaks in the QRS complex, i.e., the time duration
of one heartbeat (for details, see for example [12]). This interval is known as the RR
interval; RR intervals of normal signals are known as NN intervals. In this work, we
focus on the following five indicators, which we will call features of the signal. The
first three are from the time domain, measured in milliseconds, the last two from the
frequency domain, measured in squared milliseconds [13]:
SDNN: The standard deviation of NN intervals in the signal.
SDANN: The standard deviation of the averages (taken over five minute segments)
of the NN intervals in the signal.
RMSSD: This feature depends on the differences of subsequent NN intervals.
Square these differences, then take the square root of the arithmetic mean
of these squares.
LF: The power of the signal in the low-frequency spectrum (0.04–0.15 Hz).
HF: The power of the signal in the high-frequency spectrum (0.15–0.5 Hz).
2.3 Data Preprocessing/Cleaning
Preprocessing and cleaning of data is a necessary initial step in most data analysis
tasks, particularly in analyses of physiological data. We chose the Kubios HRV
software, a state of the art tool for studying the variability of heart-rate intervals,
mainly for its data-cleaning functionality, ease of use, and the wide range of HRV
features it calculates. Kubios thus fits our requirements, and can also be considered
the standard software in this application domain (e.g., [14,15]). It provides tools
for artifact removal (missed or spurious beats), analysis methods in the time and
frequency domains, as well as the ability to calculate a number of less frequently
used features (such as entropy measures, or measures calculated from a recurrence
plot).
rene.riedl@jku.at
224 D. Baumgartner et al.
Obtaining artifact-free raw data samples from a consumer-grade device is almost
impossible in a real-world setting. In this study, the quality of the data obtained from
the Polar H7 chest belt depends mostly on its correct placement. Most HRV features
and metrics are highly influenced by noisy data [16]; the degree of preprocessing and
noise removal thus has a direct influence on the features reported by HRV software
tools.
It has to be noted, though, that using consumer-grade devices in this context is
already a very important deliberate choice, with several associated benefits and chal-
lenges (e.g., low intrusiveness, but particular need for data cleaning). Nonetheless,
despite the particular obstacles, Schellhammer and colleagues [7]havedemonstrated
that consumer-grade devices are a valuable addition to technostress research in the
field, particularly if one is interested in heart rate measures. Yet, just as technostress
studies in the field are still scarce [2], so are technostress studies that have applied
heart rate measures and reported their data cleaning procedures, aside from simply
referencing the tool they used (e.g., AcqKnowledge in [17]). We therefore focus
on this particular step in the research process, but want to highlight that there are
several other challenges generated by the selection of a specific research design and
measurement method in NeuroIS [1].
Kubios HRV allows threshold-based artifact correction, with automatic correction
available in some versions [18]. In this preprocessing step, every RR interval value
is compared to a simple moving median over a time window. An RR interval value
is considered an artifact if it differs more than a specified threshold from this local
median. Five equidistant threshold values are available in Kubios HRV, corresponding
to increasingly lenient requirements for values to be considered artifacts: On the
lowest level, only those values further than 0.45 s from the median are removed; on
the highest level, all values further than 0.05 s from the median are removed (these
values, given here for 60 bpm, are adjusted for heart rate). The correction uses cubic
spline interpolation, which can lead to unusual beats if too many beats are corrected.
3 Results
Below, we present two sets of numerical results: one for the effects of data prepro-
cessing on the HRV features, and one for the effects of data preprocessing on the
correlation of HRV feature values with self-reported technostress levels.
The effect of five artifact correction thresholds on the features described in
Sect. 2.2 is measured by average percentage change over all data sets. For each
feature, its baseline value was calculated on the raw unchanged data samples. Com-
pared to this baseline, the percentage change turned out to be quite high for most
features. The results of applying five different artifact correction levels to five fea-
tures is shown in Fig. 1.Inthisfigure,wecanobservethatthecorrectionatthe
threshold level “very low” already resulted in a percentage difference ranging from
2.6% to 20.9% on the different features. At this level, the average percentage of
corrected RR intervals is only 0.46%. It is surprising that there is such a noticeable
rene.riedl@jku.at
Analysis of Heart Rate Variability (HRV) Feature … 225
Fig. 1 Average percentage difference between baseline (raw feature values) and feature values
obtained at different artifact correction levels
difference, given this small amount of correction. The percentages of corrected RR
intervals were, in increasing order of artifact correction levels, 0.79%, 1.26%, 3.08%
and 25.03%, respectively. The LF, HF, RMSSD and SDNN features especially show
large percentage changes, and thus appear to be highly influenced by the degree of
artifact correction.
It can be observed that the SDANN feature is much more robust than the other
features. This can be attributed to the fact that the SDANN feature is defined as the
standard deviation of the average RR interval calculated over 5 min periods [13]; this
additional averaging smooths out the effect of the artifact corrections.
Figure 2illustrates the influence of the artifact correction level on the correlation
coefficient between feature values and the self-reported technostress levels. We can
observe that changing the correction level results, for two features, in a change of
sign of the correlation coefficient between raw and corrected data. As in Fig. 1,the
SDANN feature is the most robust feature, with the other features showing small
negative correlations with self-reported technostress levels. This concurs with the
literature, where high self-reported stress levels are reported to be correlated with
low HRV standard deviations and low LF/HF ratios [8].
rene.riedl@jku.at
226 D. Baumgartner et al.
Fig. 2 Correlation of the questionnaire technostress level results and feature values obtained at
different artifact correction levels
4Discussion
Our calculations suggest that analyzing an entire data sample may detect too many
artifacts, and thus result in a biased overcorrection. According to the state of the art in
HRV artifact removal tools [18,19], there is currently no possibility to fully automate
the detection, correction and analysis process. Therefore, a human investigator is
necessary to select threshold levels appropriate for the particular data analysis task.
Hence, in the context of the research reported here, determining how well self-report
data correlates with HRV data requires human intervention in the data-preprocessing
stage.
Previous reviews of the literature have shown that there is still a need for field
studies applying neurophysiological measures in technostress research [2,9]. Yet,
current developments in the area of consumer-grade devices may allow for inves-
tigations using this approach to be conducted more frequently in the future [7]. In
order to support individual researchers interested in such settings, who then have to
preprocess the generated data, in this paper we present a concrete example based on
real world data. We show that in NeuroIS research, scholars have to make multiple
decisions with respect to data preprocessing and analysis and present some potential
effects of these decisions (e.g., on the correlation between measures).
As indicated by Riedl and colleagues in a paper on NeuroIS research methodology
([1], specifically refer to Section "Objectivity" starting on page xix), such decisions
“may affect the corroboration and/or rejection of the research hypothesis [and there-
fore] it is important that authors report details related to study design, data collection,
rene.riedl@jku.at
Analysis of Heart Rate Variability (HRV) Feature … 227
preprocessing, and analysis in their papers” (p. 26). Transparency in this regard may
also help to foster the creation of more robust results which can then, for example,
be building blocks for automated approaches to the analysis and use of data (e.g.,
when creating stress-sensitive adaptive enterprise systems [20]).
Based on the evidence reported in this paper, we make a renewed call for delib-
erately making methodological decisions (such as those related to preprocessing of
physiological data) and presenting methodological details in NeuroIS papers.
Acknowledgements This research was funded by the Upper Austrian Government as part of the
Ph.D. program “Digital Business International”, a joint initiative between the University of Applied
Sciences Upper Austria and the University of Linz, and as part of the project “Digitaler Stress in
Unternehmen” (Basisfinanzierungsprojekt) at the University of Applied Sciences Upper Austria.
References
1. Riedl, R., Davis, F.D., Hevner, A.R.: Towards a NeuroIS research methodology: intensifying
the discussion on methods, tools, and measurement. J. Assoc. Inf. Systems 15,i–xxxv(2014)
2. Fischer, T., Riedl, R.: Technostress research: a nurturing ground for measurement pluralism?
Commun. Assoc. Inf. Systems 40,375–401(2017)
3. Tams, S., Hill, K., de Guinea, A.O., Thatcher, J., Grover, V.: NeuroIS—alternative or com-
plement to existing methods? illustrating the holistic effects of neuroscience and self-reported
data in the context of technostress research. J. Assoc. Inf. Systems 15,723–753(2014)
4. Riedl, R.: On the biology of technostress: literature review and research agenda. DATA BASE
Adv. Inf. Systems 44,18–55(2013)
5. Riedl, R., Kindermann, H., Auinger, A., Javor, A.: Technostress from a neurobiological per-
spective—system breakdown increases the stress hormone cortisol in computer users. Bus. Inf.
Systems Eng. 4,61–69(2012)
6. Riedl, R., Kindermann, H., Auinger, A., Javor, A.: Computer breakdown as a stress factor during
task completion under time pressure: identifying gender differences based on skin conductance.
Adv. Hum. Comput. Interact. 1–8 (2013)
7. Schellhammer, S., Haines, R., Klein, S.: Investigating technostress in situ: understanding the
day and the life of a knowledge worker using heart rate variability. In: IEEE Proceedings of
HICSS 2013, pp. 430–439 (2013)
8. Hjortskov, N., Rissén, D., Blangsted, A.K., Fallentin, N., Lundberg, U., Søgaard, K.: The effect
of mental stress on heart rate variability and blood pressure during computer work. Eur. J. Appl.
Physiol. 92,84–89(2004)
9. Fischer, T., Riedl, R.: Theorizing technostress in organizations: a cybernetic approach.
In: Thomas, O., Teuteberg, F. (eds.) Proceedings of the 12th International Conference on
Wirtschaftsinformatik, pp. 14 53–1467 (2015)
10. Maier, C., Laumer, S., Weinert, C., Weitzel, T.: The effects of technostress and switching stress
on discontinued use of social networking services: a study of facebook use. Inf. Systems J. 25,
275–308 (2015)
11. Ragu-Nathan, T.S., Tarafdar, M., Ragu-Nathan, B.S., Tu, Q.: The consequences of technostress
for end users in organizations: conceptual development and empirical validation. Inf. Systems
Res. 19,417–433(2008)
12. Xhyheri, B., Manfrini, O., Mazzolini, M., Pizzi, C., Bugiardini, R.: Heart rate variability today.
Prog. Cardiovasc. Dis. 55,321–331(2012)
13. Task Force of the European Society of Cardiology, and the North American Society of Pac-
ing and Electrophysiology: Heart rate variability: Standards of measurement, physiological
interpretation, and clinical use. Eur. Heart J. 17,354–381(1996)
rene.riedl@jku.at
228 D. Baumgartner et al.
14. Giles, D., Draper, N., Neil, W.: Validity of the polar V800 heart rate monitor to measure RR
intervals at rest. Eur. J. Appl. Physiol. 116,563–571(2016)
15. Wijaya, A.I., Prihatmanto, A.S., Wijaya, R.: Shesop Healthcare: Android Application to Mon-
itor Heart Rate Variance, Display Influenza and Stress Condition Using Polar H7. Unpublished
(2016)
16. Shaffer, F., Ginsberg, J.P.: An overview of heart rate variability metrics and norms. Frontiers
Public Health 5,1–17(2017)
17. Neben, T., Schneider, C.: Ad intrusiveness, loss of control, and stress: a psychophysiological
study. In: AIS Proceedings of ICIS 2015 (2015)
18. Tarvainen, M.P., Niskanen, J.-P., Lipponen, J.A., Ranta-Aho, P.O., Karjalainen, P.A.: Kubios
HRV—heart rate variability analysis software. Comput. Methods Programs Biomed. 113,
210–220 (2014)
19. Kaufmann, T., Sütterlin, S., Schulz, S.M., Vögele, C.: ARTiiFACT—a tool for heart rate artifact
processing and heart rate variability analysis. Behav. Res. Methods 43,1161–1170(2011)
20. Adam, M.T.P., Gimpel, H., Maedche, A., Riedl, R.: Design blueprint for stress-sensitive adap-
tive enterprise systems. Bus. Inf. Systems Eng. 59,277–291(2017)
rene.riedl@jku.at
