Quantifying Occupational Stress in Intensive Care UnitNurses: An Applied Naturalistic Study of CorrelationsAmong Stress, Heart Rate, Electrodermal Activity, andSkin TemperatureNima Ahmadi, Houston Methodist Hospital, Texas, USA, Farzan Sasangohar ,Houston Methodist Hospital, Texas, USA, and Texas A&M University, CollegeStation, USA, Tariq Nisar, Houston Methodist Hospital, Texas, USA,Valerie Danesh , University of Texas at Austin, USA, Ethan Larsen, HoustonMethodist Hospital, Texas, USA, Ineen Sultana, Texas A&M University, CollegeStation, USA, and Rita Bosetti, Houston Methodist Hospital, Texas, USAObjective: To identify physiological correlates tostress in intensive care unit nurses.Background: Most research on stress correlates aredone in laboratory environments; naturalistic investigationof stress remains a general gap.Method: Electrodermal activity, heart rate, and skintemperatures were recorded continuously for 12 - hrnursing shifts (23 participants) using a wrist- worn wearable technology (Empatica E4).Results: Positive correlations included stress andheart rate (ρ .35, p .001), stress and skin temperature(ρ .49, p .05), and heart rate and skin temperatures (ρ .54, p .0008).Discussion: The presence and direction of somecorrelations found in this study differ from those anticipated from prior literature, illustrating the importanceof complementing laboratory research with naturalisticstudies. Further work is warranted to recognize nursingactivities associated with a high level of stress and the underlying reasons associated with changes in physiologicalresponses.Application: Heart rate and skin temperature maybe used for real- time detection of stress, but more workis needed to validate such surrogate measures.Keywords: critical care, job stress, physiologicalmeasurement, nursing and nursing systems, naturalisticstudyAddress correspondence to Farzan Sasangohar, Industrialand Systems Engineering, Texas A&M University, CollegeStation, TX 77843, USA; e-mail: s asangohar@ tamu. eduHUMAN FACTORSVol. 00, No. 0, Month XXXX, pp. 1-14 DOI: 10. 1177/ 0018 7208 2110 40889Article reuse guidelines: sagepub. com/ journals- permissionsCopyright 2021, Human Factors and Ergonomics Society.INTRODUCTIONIntensive care unit (ICU) personnel including physicians, respiratory therapists, pharmacists, and case managers (Mealer et al., 2012;Pastores et al., 2019), and particularly nurses(Donchin & Seagull, 2002), experience highlevels of stress. Fundamental aspects of nursingcare in critical care environments include initiating and facilitating ongoing dialog with clinicians and family members related to life supportinterventions, palliative care, and end- of- lifediscussions. Furthermore, ICU nurses repeatedly carry responsibility for rapid interventionfor severe physiologic deterioration events, cardiopulmonary resuscitation events, and interactions with families and others in high- stresssituations (Mealer et al., 2012). Cumulativeexposure to acute stress may lead to psychological distress and burnout syndrome—a stateof fatigue, frustration, and diminished interest(Embriaco et al., 2007; Mealer et al., 2017). Innursing, burnout is associated with an increasein absenteeism and turnover rates (Aiken et al.,2002; Scanlan & Still, 2019; Van der Heijdenet al., 2019; Wang et al., 2020). Turnover aloneis costing the U.S. economy an estimated 2.1 million each year (Heinrich, 2001).Traditionally, occupational stress has beenmeasured using self- reported instruments(Augusto Landa et al., 2008; Farquharson et al.,2012; Rodriguez- Paras et al., 2018; Sveinsdóttiret al., 2006). However, self- reported instruments suffer from subjects’ biases, retrospective

2memory retrieval, and study administrators’influence on responses (Johnston et al., 2016). Toaddress these drawbacks, direct measurements ofphysiological responses such as heart rate (HR; orpulse rate [PR] when measured using photoplethysmography [PPG]), heart rate variability (HRV),electrodermal activity (EDA), skin temperature(ST), cortisol levels, blood pressure, and pupildiameter data have been investigated for assessment of stress (Greene et al., 2016) and otherrelated constructs such as workload and arousal(Hancock & Matthews, 2019; Matthews et al.,2015). The recent evolution and availability ofHR, EDA, and ST sensor technology has enabledcontinuous collection of physiological responsesin real- world settings and has empowered naturalistic research (Healey & Picard, 2005; Jenkset al., 2020; Mehler et al., 2012; Rodrigues et al.,2015; Schneegass et al., 2013). While naturalistic studies of stress using wearable sensors(e.g., Chen et al., 2014; Giakoumis et al., 2012;Giannakakis et al., 2017; Gjoreski et al., 2017;McDuff et al., 2014) have shown promise in complementing participant self- report and simulatedlaboratory- based measurement, the correlationsbetween several psycho- physiological responsesused to assess stress have not been documentedin unconstrained and naturalistic environments.Understanding correlations between stress andphysiological responses in a complex dynamicwork environment such as an ICU with long shiftdurations, as well as fluctuations in time pressure,workload, and emotional strain is necessary toassess convergence with the published evidenceof such correlations, which are mostly drawn fromlab studies. This knowledge is necessary and willhave implications for the design and developmentof effective stress monitoring technologies.Our objective is to compare and contrast correlations between EDA, HR (estimated usingPPG), and ST responses with stress betweena naturalistic study of ICU nurses and the lab- derived correlations. We first provide a reviewof documented correlations (the following subsection) before evaluating such correlations inour own exploratory study (remaining sections).Month XXXX - Human FactorsPhysiological Responses to StressMost HR and skin- based physiological measures are controlled through the activation ofsympathetic nervous system (SNS) and/or theparasympathetic nervous system (PNS), whichmake up the autonomic nervous system (ANS).SNS activates in stressful situations to preparethe body for fight- or- flight responses, and thePNS triggers to restore body homeostasis viarelaxation responses (Andreassi, 2013).Electrodermal activity and stress. EDA is ameasure of the subject’s electrical skin conductance (Greene et al., 2016) and is solely controlled by the SNS (Boucsein, 2012; Cacioppoet al., 2007). Skin conductivity is influenced bysurface sweat and sweat glands triggered bystress as well as positive and negative emotions(Kreibig, 2010). Time series of skin conductance has two main features: tonic and phasicresponses. Tonic response, or skin conductancelevel (SCL), is the overall long- term trend of theelectrodermal time series, and phasic, skin conductance responses (SCRs), are the acute spikesin the SCL (Benedek & Kaernbach, 2010). Inaddition to SCR amplitude, other well- knowntemporal features of SCR are latency and rise andrecovery time (Dawson et al., 2017). An SCR iscalled a specific response once it is triggered bya stimulus. A nonspecific response, conversely,is spontaneous (Benedek & Kaernbach, 2010).Electrodermal activities, as a prime indicator of stress and mental workload, have beenmeasured in different domains such as healthcare (Ritz et al., 2000), transportation (Mehleret al., 2012; Ruscio et al., 2017; Schneegasset al., 2013), psychology (Blechert et al., 2006;Nomikos et al., 1968), and engineering (Acerbiet al., 2017). Most of the previous researchestablishes that phasic EDA increases significantly in a stressful or demanding situation(Giannakakis et al., 2019). Schneegass et al.(2013) collected subjective and objective datafrom drivers to assess the potential of physiological responses to detect automotive drivers’workload in different road types (e.g., freewayand highway), finding that SCR and ST weresensitive to different road types. Ruscio et al.(2017) studied the drivers’ mental workload innormal driving versus assisted- driving tasks and

Physiological Correlates of Stressassociated the elevated SCR of assisted- drivingtasks with the higher mental workload of drivers. Hernandez et al. (2011) trained a machinelearning model on SCR data collected from acall center’s employees and predicted stressfulcalls with an accuracy of 73.4%.Heart rate and stress. Unlike EDA, HRis controlled by either SNS, PNS, or both(Berntson et al., 1994). In an acute crisis, HRincreases and the pattern of blood distributionchanges (Schneiderman et al., 2005). Whenthe stressful situation is alleviated, PNS triggers to slow down the HR (Choi & Gutierrez- Osuna, 2009; Greene et al., 2016) and restoresthe body to normal condition. de Looff et al.(2018) reviewed 38 articles related to occupational stress and reported a positive correlation between HR and stress; however, it wasrecommended that results across studies becompared with caution, as both HR (Bextonet al., 1986) and EDA (Hot et al., 1999, 2005)display diurnal variation. A recent study onemergency medicine residents’ HR data usingwearable technology revealed that most participants experienced at least one episode ofvery high level of stress in their shifts, whichwas associated with maximum HR data (Jenkset al., 2020). HR has also been studied in thecontext of low arousal (Borghini et al., 2014).For example, a study of prolonged night drivinghas shown the HR decline due to fatigue anddrowsiness (Riemersma et al., 1977).Electrodermal activity, heart rate, andstress. Early stress studies were conducted toassess the effect of stress on individuals using acombination of EDA and HR (Boucsein, 2012;Folkins, 1970; Gatchel et al., 1977; Niemelä,1975; Nomikos et al., 1968). In addition tostress, heightened EDA and HR have been usedas indices for mental workload (Hart & Hauser,1987; Ruscio et al., 2017; Schneegass et al.,2013); arousal (Boucsein, 2012); and anxiety,anger, and fear (Kreibig, 2010). One study(Nomikos et al., 1968) investigated the effect ofthe anticipatory stress on the viewer of scenesof industrial accidents and noticed that SCLincreased significantly but not HR. Folkins(1970) studied the responses of individuals whoreceived electric shocks delivered in differenttime intervals from 5 s to 20 min; the SCL and3HR were increased significantly in time intervals up to 1 min.Recent studies (Acerbi et al., 2017;Hernandez et al., 2011; Mehler et al., 2012;Ruscio et al., 2017; Schneegass et al., 2013)were particularly interested in evaluating thepotential of monitoring stress using physiological responses. Acerbi et al. (2017) analyzedphysiological responses collected by wearabletechnology and self- reported instruments todetect stress status in a laboratory experiment.The developed classifier was able to identifystress status; elevated levels of HR and SCRwere reported in the stress period. Mehler et al.(2012) assessed the potential of physiologicalresponses to develop workload detection systems for drivers. HR and SCL were collected ina naturalistic driving experiment with (imposedby secondary verbal tasks) and without mental workload. Both HR and SCL were indicative of different levels of mental workload. Inhealthcare, Ritz et al. (2000) examined a claimthat stress and emotion cause respiratory resistance in asthmatic patients. Emotions and stresswere induced by watching short video clips andstress tasks. Changes in HR and SCL causedby the stress- induced task were consistent withprevious research.Skin temperature and stress. Body temperature, as a coping mechanism, fluctuatesin response to stress (Kleitman & Jackson,1950; Marazziti et al., 1992; Vinkers et al.,2010, 2013) and anxiety (McFarland, 1985).When SNS activates, blood distribution in thebody changes such that venous blood flow isreduced, and arterial blood flow is increased tosupport fight- or- flight responses. These bloodflow changes raise the core temperature—aphenomenon referred to as stress hyperthermia(Marks et al., 2009). However, this reallocationof arterial blood flow to stress response organsis balanced with a corresponding restrictions ofarterial blood flow in the extremities includingthe wrists and hands—a phenomenon referredto as peripheral vasoconstriction (Herbornet al., 2015). The decrease in blood flow inwrists and hands results in cooling of skin, achange that can be detected using peripheralelectrodermal wrist sensors (Gjoreski et al.,2017).

4Month XXXX - Human FactorsST in the wrist and hand has been evaluatedto assess changes due to stress. In a relaxationinstruction and a therapy procedure to treatheadache, an increase in finger temperature inthe relaxation period and decrease in the stressperiod was observed (Boudewyns, 1976). Linet al. (2011) studied the effect of stress anddepression on physiological responses andreported lower finger temperature in depressedindividuals. One study reported a nonsignificantreduction in fingertip ST in female undergraduates when they watched fearful movie scenesand performed cognitive tasks (Rimm- Kaufman& Kagan, 1996). Vinkers et al. (2013) investigated the effect of stress on different regions ofST in a laboratory experiment in which participants were given a stressful task or nonstressful control. They found significant reduction intemperature at the fingertip, finger base, andhand palm, and nonsignificant reduction at thewrist during stressful tasks. Engert et al. (2014)investigated the application of thermal infraredimaging in stress detection using a standardized stress test. A decrement in temperature wasnoted at forehead, finger, and chin.METHODSA naturalistic study of ICU nurses was conducted to evaluate correlations between stressand its previously known correlates: HR, EDA,and ST.ParticipantsA total of 28 participants were recruited, viaemail or during morning rounds, from RegisteredNurses working in a 40- bed CardiovascularICU (CVICU) of a 900- bed tertiary care facilityin the Southwestern United States. All CVICUnurses were eligible for inclusion in the studyexcept those who wore eyeglasses, due to astudy component involving eye- tracking equipment (to be detailed in a later report) conductedin parallel with the work reported here. Due todata loss associated with broken EDA and STsensors, physiological data of 23 participantswere analyzed. The majority of CVICU nurseswere female (78.3%), and the average age ofnurses was 35.5 ( 8.5) years (Table 1).ToolsThe Empatica E4 (Empatica Inc., Cambridge,MA; Garbarino et al., 2014), a lightweight andunobtrusive watch- like technology, was usedto log physiological data including HR estimation, EDA, and ST (measured from dorsal wrist)continuously. The battery life of Empatica E4(more than 36 hr in the recording mode) aswell as the wrist- watch form (with no display)makes Empatica a suitable apparatus for continuous measurement of bio- signals in nursingwork environments. The E4 device has beenused in various published work, and its HR andEDA sensors have been validated for accuracy(Ollander et al., 2016; Schuurmans et al., 2020;van Lier et al., 2020). For each physiologicalresponse, this device generated time- series datain the .csv format. The HR, EDA, and ST weresampled at 1, 4, and 4 Hz, respectively.DesignThis was a prospective naturalistic studyin which physiological measures were collected during the entire 12 - hr shift of CVICUnurse participants. Data collection was part ofa broader research project that collected eye- tracking data (ETD) as well as two additionalshifts for longitudinal analysis. However, thispaper only documents the data from the firstshift for present purposes of correlational analysis. All aspects of this study were approved bythe Houston Methodist IRB (Pro00019025), andparticipants provided written informed consent.ProceduresThe research team met with participantsbefore the start of working shifts to explain thestudy objectives, procedures, and apparatus, andto collect informed consent. Following devicetraining on the Empatica E4, participants placedthe device on the wrist of the self- reported nondominant hand. Data collection began at the startof each shift. A member of the research team waspresent in the central nursing station during theentire shift to address any potential issues. At theend of the 12 - h shift, the research team met thenurses at the nursing station, collected the recording device, and downloaded the data.

Physiological Correlates of Stress5TABLE 1: Demographic Information of Cardiovascular Intensive Care Unit NursesDemographic InformationN (%)National Average* N (%)Sex Male5 (21.7%)1,148 (15.9%) Female18 (78.3%)6,071 (84.4%)1(4.3%)21 (0.3%)Race/Ethnicity American Indian or Alaska Native Asian5 (21.7%)742 (10.3%) Black1 (4.3%)400 (5.6%) Native Hawaiian or Other Pacific Islander3 (13.0%)43 (0.6%)1 (4.3%)457 (6.6%) Single, Never Married7 (30.4%)n/a Married or Domestic Partnership15 (65.2%)n/a0 (0.0%)n/a White (non- Hispanic) Hispanic or LatinoMarital Status Widowed Divorced1 (4.3%)n/a Separated0 (0.0%)n/aAge, years (mean Std. Dev.)35.5 8.544ICU experience (mean Std. Dev.)8.4 6.7n/aNote. *Acute Care/Critical Care and Emergency/Trauma (National Nursing Workforce Study, personalcommunication, August 10, 2020)Data AnalysisArtifacts were removed from the collecteddata at four different phases. First, we definedcut- off values for each physiological response.The HR data over 200 bpm, ST more than 44 Celsius, and SCR amplitude greater than 10micro- Siemens (µS) were removed from thedataset. Second, Empatica E4 calculated andoutput HR and the inter- beat interval (IBI) fromblood volume pulse (BVP) signals (capturedby PPG sensor) and employed an algorithm toremove the incorrect peaks due to noise in theBVP signal (Empatica, 2020). Third, a visualinspection of IBI time series data was conducted based on recommendations providedby Empatica (2020). Finally, to derive stressindex (SI) from IBI signal, the Kubios V3.3.1was employed for two reasons: first, to apply anartifact correction algorithm (Tarvainen et al.,2009) on IBI time- series as a measure to dealwith missing data; and second, to measure theSI (Tarvainen et al., 2009). Since the samplingrates (4 Hz) of EDA and ST were different fromthe number of derived HR per second (1 Hz),physiological data were synchronized using aPython script and stored in a database for further analysis.The SI characterizes the activity of the sympathetic part of an autonomic nervous systembased on HRV; this measure was calculated foreach 1 - min interval using Kubios, which isequal to the square root of Baevsky’s SI (conventional unit), which utilizes the distributionof IBI as: AMo(2Mo) (MxDMn) (1)where Mo and AMo, respectively, are modeand amplitude of the most frequent of the IBI,and MxDMn is the range of IBI values thatis indicative of variability of IBI (Baevsky &Berseneva, 2009). Baevsky’s SI between 50 and

6150 c.u. is considered normal; the square rootof this range was used to define the intensity ofsympathetic cardiac activation—stress zone;SI 12.2 c.u. represents reduced activation ornormal stress zone, and SI 12.2 c.u. indicateshigh activation or high- stress zone. SI—whichwas measured based on HRV parameters—andHR reflect activation of SNS and sympatheticarousal is expected to result in both increasedmean HR and high SI. However, this positivecorrelation has not been documented in a naturalistic environment.Raw EDA data were processed and analyzed using LEDALAB V3.4.9 (Benedek &Kaernbach, 2010; Ledalab, n.d.). The continuous decomposition analysis approach was usedto extract phasic responses and their amplitude;the amplitude threshold was set 0.1 µS. Next,averages per minute of SCR amplitude, HR,and ST were computed.Statistical analyses were performed usingR V3.6.3 (R foundation, Project for statistical computation and graphics). Continuousvariables are presented as means and standarddeviations. Histograms were plotted to understand the distribution of the continuous variables, and Shapiro- Wilk and Anderson Darlingtests were used to determine the normality ofthe variables. Repeated measurement correlation by using the “rmcorr” package in R wasused to establish the correlation between normally distributed physiological parameters(Bakdash & Marusich, 2017). Spearman correlation was used to establish correlation fornonnormalized physiological parameters. Wereport the correlation coefficient (ρ), p value(p), and 95% confidence intervals (CIs) of thephysiological measures. Associations betweennormal and high stress were examined usingbivariable and multivariable logistic regressionmodels with generalized estimating equation(GEE) accounting for repeated measurements.Multivariable logistic regression included variables that were statistically significant in thebivariable logistic regressions. The statistical models were adjusted for age, marital status, and ICU work experience of participants.A sided α of 0.05 was used to determine statistical significance. We report odds ratio (OR)and 95% CI.Month XXXX - Human FactorsRESULTSIn addition to the expected positive correlation between the SI and HR (ρ .35; 95% CI[0.34, 0.36]; p .001), our results showed a significant positive correlation between SI with ST(ρ .49, p .05), and HR with ST (ρ .54, p .05). However, phasic EDA (SCR amplitude)was not correlated with SI, HR, or ST. Figure 1illustrates the resulting correlations betweenvarious parameters.Multiple logistic regression models afteradjusting for demographics, marital status,number of children, and years of nursing experience in the ICU, showed that the ratio of theprobability of experiencing high stress to theprobability of not experiencing high stress is1.10 times higher with an increase of one unit inHR (OR 1.10; 95% CI [1.08, 1.11]; p .001;degrees of freedom [DF] 12274). Additionally,with every one- unit increase in ST, the ratio ofthe probability of being in a high- stress zone tothe probability of not being in that zone are 1.20times higher (OR 1.20; 95% CI [1.09, 1.31]; p .001; DF 14784). However, we did notfind any significant association between phasicEDA (OR 1.05; 95% CI [0.90 – 1.23]; p .53;DF 8001) and stress zones.DISCUSSIONGiven the increasing trend in utilization ofwearable sensors and tools in naturalistic stressmonitoring research, the gap in knowledge regarding the correlation between stress and various biomarkers needs further investigation. In this study,we reviewed correlations between increased stressand its biomarkers including HR, phasic EDA,and ST in a naturalistic study involving ICUnurses. Prior results indicate that EDA (Acerbiet al., 2017; McDuff et al., 2014; Ruscio et al.,2017; Schneegass et al., 2013) and HR (Finsenet al., 2001; Lackner et al., 2011; Lundberg et al.,1994; Reinhardt et al., 2012; Ritz et al., 2000;Steptoe et al., 2001) increase in healthy individuals and ST slightly but not significantly reduce atthe wrist (Vinkers et al., 2013). The discriminativepower of physiological responses has been studied as well: a higher level of HR and SCR amplitude (Ruscio et al., 2017; Schneegass et al., 2013)was observed in demanding tasks, and ST was

Physiological Correlates of Stress7Figure 1. Correlations between stress index, heart rate, electrodermal activity, and skin temperature; stress andSCR amplitude (a), stress and heart rate (b), stress and skin temperature (c), heart rate and SCR Amplitude (d),skin temperature and SCR amplitude (e), and heart rate and skin temperature (f).indicative of workload (Schneegass et al., 2013).Correlations between stress, HR, EDA, and SCRwere therefore expected to be positive. In our findings, however, only HR was indicative of stressintensity, not EDA. Our findings also showed thatthe correlation between stress and ST is positive(Table 2).Although the literature has found a positive association between phasic EDA and HR(Kettunen et al., 1998), we found no such correlation. Correlation between phasic EDA andST (dorsal wrist) was previously reported as positive (Khan et al., 2019; Lobstein & Cort, 1978);however, this study found no correlation betweenSCR and ST. While the correlation between HRand ST in stressful events remains a research gap(Neves et al., 2016), given the known positive correlation between stress and HR (e.g., Acerbi et al.,2017; Folkins, 1970; Mehler et al., 2012; Ritz et al.,2000) and evidence of no correlations betweenstress and ST (at wrist; Vinkers et al., 2013), weaknegative or no correlation between HR and ST wasexpected. However, our results show a significantpositive correlation between these parameters. Toour knowledge, this is the first empirically deriveddocumentation of correlations between HR and ST.

8Month XXXX - Human FactorsTABLE 2: Comparison of Correlations Documented in the Literature Versus Those Observed in ThisStudy’s FindingsFeatureStress and SCR amplitudeStress and HRStress and skin temperature (dorsal wrist)SCR amplitude and HRSCR amplitude and skin temperature (dorsal wrist)HR and skin temperatureKnown CorrelationsObserved Correlations No correlation No correlation No correlation No correlationNot studied Note. HR heart rate; SCR skin conductance response.Discrepancies between the current and previous results could be due to the differences betweenthe research designs, environments, tasks, andtype of stressors in these studies. While most ofthe previous research was conducted in a laboratory setting utilizing induced stress, the currentresearch was a naturalistic study performed in acomplex work environment. In addition, the previously documented correlations are in responseto a wide range of mental, psychological, andphysical stressors and include both healthy andnonhealthy individuals. Finally, to our knowledgethis is the first study of stress correlates conductedin the ICU setting covering the entire shift of thenurse population. These factors are discussed inmore details below.Duration of stressed states. In laboratory- induced stress research, typically stress tasksdo not last more than 30 min. For example, thewidely used Trier Social Stress Test takes about15 min per participant (Vinkers et al., 2013). In areal- world setting where data are captured acrossan extended period (e.g., 12 - hr ICU shifts), participants may be exposed to longer periods ofstress due to the workload, time pressure, anddemanding tasks. If that is the case, the inhibitivecoping mechanism and habituation to resist prolonged activation of repeated stress (Grissom &Bhatnagar, 2009) might have affected the physiological responses. Furthermore, in long workingshifts, nurses must deal with fatigue and drowsiness, which could also change the intensity ofphysiological responses. For instance, drowsiness has been associated with a reduction in HR(Riemersma et al., 1977).Type of stress. In a laboratory setting study, participants receive given stress tasks one at a time.This procedure allows researchers to examinethe effect of a specific stress task on physiological responses. For instance, Vinkers et al. (2013)used Trier Social Stress Test (Chen et al., 2014)to emulate mental and psychological stress. Inthe current study, CVICU nurses carried out several tasks including medication- related activities,administrative and clinical documentation tasks,and patient care with various complexity levelsand cognitive loads which might have affectedphysiological variables differently.Physical activity. In a laboratory- inducedexperiment, participants are asked to remain idle(often seated) while performing stress tasks. Incontrast, in a naturalistic study nurses may experience different levels of physical activity throughout a working shift. Physical activity is a knownsource of noise in physiological data and mayaffect responses in unexpected ways (Martinez- Nicolas et al., 2013; Sun et al., 2012; Van Steenis& Tulen, 1997; Wilhelm et al., 2006).Unconstrained environmental factors. Lackof control over the research environment in thecurrent study may have resulted in several environmental factors with impacts on physiologicalvariables of interest. For instance, in ICUs, nurseswear gloves and perform frequent hand hygiene(i.e., hand washing with soap and water, hand sanitizer). These practices could affect the measurement of physiological responses. For instance,evaporation of water or alcohol reduces ST, whilewearing disposable hospital gloves could act asinsulation and disrupt heat dissipation, thereby

Physiological Correlates of Stressincreasing ST. Hence, while blood circulationreduces, peripheral body temperature may rise bywearing gloves inside ICU rooms and decline bytaking off gloves and washing hands with alcohol- based solutions.Electrodermal lability. Individual trait differences in EDA or EDA lability have been shownto impact the rate of nonspecific response andhabituation. Evidence suggests that stable or labileindividuals demonstrate different electrodermalbehavior (Sarchiapone et al., 2018). Future workcan examine individual traits to further evaluateEDA responses to stress. In the laboratory environment, nonspecific responses could be detectedas the onset of stimuli is known, while in the current naturalistic stress assessment, it is not possibleto distinguish specific responses from nonspecificresponses.LimitationsThis study offers data from more than 300 hrof physiological monitoring of ICU nurses in12 - hour day and night shifts. Several technicaland methodological limitations are noteworthy:Hardware reliability. In our study, we sufferedfrom data loss, which was associated with theEmpatica E4 hardware malfunctions. For example, in some cases, data were captured with briefperiods of disruption, and some devices had defective or broken EDA and ST sensors. Additionally,we noticed that the EDA sensor was sensitive towater, and nursing tasks require frequent handhygie

address these drawbacks, direct measurements of physiological responses such as heart rate (HR; or pulse rate [PR] when measured using photopleth-ysmography [PPG]), heart rate variability (HRV), electrodermal activity (EDA), skin temperature (ST), cortisol levels, blood pressure, a