key: cord-0543209-0yf3j6ye authors: Hosseini, Seyedmajid; Katragadda, Satya; Bhupatiraju, Ravi Teja; Ashkar, Ziad; Borst, Christoph W.; Cochran, Kenneth; Gottumukkala, Raju title: A multi-modal sensor dataset for continuous stress detection of nurses in a hospital date: 2021-07-25 journal: nan DOI: nan sha: 6a5a8bff5f31be4bd4159b9a257f92f2ef8ec641 doc_id: 543209 cord_uid: 0yf3j6ye Advances in wearable technologies provide the opportunity to continuously monitor many physiological variables. Stress detection has gained increased attention in recent years, especially because early stress detection can help individuals better manage health to minimize the negative impacts of long-term stress exposure. This paper provides a unique stress detection dataset that was created in a natural working environment in a hospital. This dataset is a collection of biometric data of nurses during the COVID-19 outbreak. Studying stress"in the wild"in a work environment is complex due to the influence of many social, cultural and individuals experience in dealing with stressful conditions. In order to address these concerns, we captured both the physiological data and associated context pertaining to the stress events. Specific physiological variables that were monitored included electrodermal activity, heart rate, skin temperature, and accelerometer data of the nurse subjects. A periodic smartphone-administered survey also captured the contributing factors for the detected stress events. A database containing the signals, stress events, and survey responses is available upon request. Prolonged exposure to stress factors such as high workload, lack of autonomy, and long hours can negatively impact employee health. Many studies point out that prolonged exposure to stress leads to chronic conditions such as obesity 1 or hypertension 2 , and it may exacerbate conditions such as type-II diabetes 3 . Monitoring and understanding stress in workplaces is important, especially in professions where there is increased exposure to stress, often leading to burnout and increased turn over 4 . Our work was inspired by previous work on wearables to monitor physiological signals related to stress. The AffectiveRoad dataset 5 used Empatica E4 and Zephyr Bioharness 3 to study the effect of driving conditions on stress of 10 drivers, for which each driver took a 1h26m driving test. The WESAD data-set 6 used RespiBAN and Empatica E4 to study stress of 15 students watching a movie and taking a TSST test 7 . The SWELL dataset 8 used Mobi, uLog, video, and Kinect to study stress and the associated postures and facial expressions of 25 students. MDPSD 9 is a comprehensive multi-modal stress detection dataset on university students using EDA and PPG signals while performing different tests(e.g., TSST 10 and color-word tests 11 ). The dataset offers physiological stress signals collected using signal streams from Empatica E4 during the COVID-19 outbreak in a nursing population. Our primary motivation to create this dataset was to conduct a stress study under real-world work conditions. The following are the key differentiators of this dataset: • Activities such as walking, running, and working on a computer have an effect on physiological signals (i.e. EDA, heart rate, and skin temperature). Unlike existing studies that were done in sitting conditions with tethered electronics, this paper presents a study under actual working conditions to allow researchers to link stress signals with activity information extracted from accelerometers. • This study was conducted on nurses in a hospital, which offers a unique perspective, unlike several prior studies, which were done with students. • Nursing is a stressful profession, and prior literature identified several factors that contribute to stress. The combination of wearable data and end-of-session surveys offers complementary advantages. • Moreover, the study was conducted during the second wave of the COVID-19 outbreak, and all the nurses were dealing with the influx of COVID-19 patients, which made it an event-rich environment. The following is the organization of our paper. Section 3 and 4 provide the data acquisition apparatus, data records, and data description. In Section 5, we discuss Qualitative Validation of Quantitative Stress Signals, significance tests, and potential applications. Section 6 describes limitations of stress data collection in real-world scenarios. Finally, we provide the code repository availability in Section 7. This section describes the data collection hardware, software used for the survey, and the validation platform. The data was gathered for approximately one week from 15 female nurses working regular shifts at a hospital. The age of the nurses ranged from 30 to 55 years. In total, 1,252 hours worth of data was collected in Apr-May and Nov-Dec of 2020, in two study sessions. 137.23 hours of data was detected as stressful. Table 3 shows the data description. The exclusion criteria were pregnancy, heavy smoking, mental disorders, and chronic or cardiovascular diseases. Table 1 presents the details captured from individual participants that can be used for analysis of stress for various nurses. In this study we aimed to detect stress of nurses during their daily routine using only wearables. Wearable devices are capable of being worn during the regular shifts and have the ability to continuously monitor the physiological signals with minimum intrusion. The subjects recruited in the study were nurses working on their usual schedule. An Empatica E4 was worn on the wrist of the dominant arm. During these experiments, we collected the physiological data from the nurses continuously from the start of their shift to the end of the shift. We detected stress events while monitoring the biometric signal streams of nurses. The nurses were then asked to validate the detected events with a survey response. The data set is anonymized and will be made available upon request to the corresponding author. Figure 1 presents the framework for the overall data collection. A detailed explanation of each of the components is presented below: E4 Wristband: An E4 wristband device (Empatica Inc., Milano, Italy) that collects physiological data such as EDA, heart rate, skin temperature, and accelerometer data from the right hand of the subject. EDA is measured via E4's silver (Ag) electrode (valid range [0.01-100] µS), while heart rate is measured via E4's photoplethysmographic (PPG) sensor. The E4 wristband is powered by a rechargeable lithium battery and transmits data to the subject's smartphone, using bluetooth, in near-real-time. All the data collected from the E4 wristband and the sampling frequencies are presented in Table 1 . The physiological data is then transmitted to the data collection app on a nurse's phone in near-real-time. The nurses can also tag the data using the tag button on the e4 device to indicate an undetected stress experience, and this is also transmitted to the data collection app. Data Collection App: The data collection app is a mobile app for both iOS and Android. It connects to the E4 wristband through bluetooth. The physiological signals are collected in near-real-time. These signals and any tags input by the nurses are transmitted to the machine learning based stress detection model. Machine Learning and Stress Detection Model: The machine learning based stress detection model ingests the physiological and accelerometer data gathered from the E4 wristbands in near-real-time. Stress signals are detected from these physiological signals over time using a stress detection algorithm. The skin temperature and lag-based features were used in the stress detection model. The longest stressful events were selected and up to 6 of them were presented to the users per day. More information about the stress detection algorithm is provided in on GitHub 1 . Once a stress signal is detected from the subject, a notification is sent to the participant inviting them to complete a survey identifying the causes of stress and the severity of the stress. A stress survey is also sent for all the stress tags received from the nurses as well. A reminder is sent to the nurses at the end of their shift to complete the surveys. Survey: The nurses were invited to complete a questionnaire about their experience during the detected stress events to identify the cause of the stress. The survey itself was not administered during the event in order to not add to the stress. Instead, the survey was administered at limited intervals, not too long after the stress. The questions in the questionnaire were selected based on a review of literature studying stress on nurses in hospital environment, as well as from our discussions with nurses. A list of questions in the survey is presented in table 2. The dataset contains 15 excel (.xslx) files with 11 pre-processed physiological signals extracted from Empatica E4 wristbands. The corresponding stress survey responses from 15 nurses who wore the wristband is provided in Survey.xlsx file. Both these files are linked by the Nurse identifier and the date-time field. The four signals: heart rate, skin temperature, EDA, and blood volume pulse have different frequencies. The frequency of these signals range from 1 Hz for the heart rate to 72Hz for the blood volume pulse. The EDA and skin temperature are about 4 Hz and 10 Hz each. For our stress detection, we use a frequency of 4 Hz to minimize information loss. The data points for the three signals are interpolated to 4 Hz . The code associated with the signal interpolation, cleaning, and pre-processing is provided along with the code repository. In addition, the raw signals with their original frequencies from Empatica E4 are also provided. The data-set is available upon request. The following is a description of directories and file in the dataset. Physiological.zip The zip file holds the data of 15 participants as different folders. Each folder contains raw data signals in csv format in a sub-folder. A raw data folder consists of 6 different csv files, namely EDA.csv HR.csv, TEMP.csv, IBI.csv, BVP.csv, ACC.csv showing electrodermal activity, heart rate, skin temperature, inter-beat intervals, blood volume pulse, and accelerometer signal, respectively. Additionally, one combined excel sheets file (e.g. DF.xlsx 187.36 MB) contains all the biometric signals as a single file with the following sheets: • Sheet A: ID Anonymized Id of the user. • Sheet B: datetime the datetime float number that contains the time that signal was generated using the internal clock of the wristband. • Sheet C: HR Heart rate generated from BVP and IBI sensors. • Sheet D: EDA Electrodermal activity of the skin, measuring the skin's electrical conductivity. • Sheet E: ST Skin temperature in Celsius. • Sheet F: IBI Inter-beat interval or beat-to-beat interval, being the time interval between individual beats of an individual's heart. • Sheet G: BVP Blood volume pulse is a method of measuring the heart rate. • Column C: CR COVID related. • Column D: PC Patient in crisis. • Column E: POPF Patient or patient's family. • Column F: DOC Doctors or colleagues. • Column G: ALPRAS Administration, lab, pharmacy, radiology, or other ancillary services. • Column H: IW Increased Workload. • Column I: TRS Technology related stress. • Column J: LOS Lack of supplies. • Column K: D Documentation. • Column L: CRS Competency related stress. • Column M: SPT Safety (physical or physiological threats). • Column N: WE Work Environment -Physical or others: work processes or procedures Methods for evaluating psychological stress detection include self-report questionnaires, and interviews. While stress surveys are considered sufficiently reliable, and widely adopted, they offer insights mainly on the moments of their administration. The responses are coarse designations of stress and so are unable to detect subtler shifts in stress over time 12 . Still, they constitute an important tool in the arsenal of stress evaluation because they provide the subject's subjective perception of stress, which is not available from the monitoring of the physiological signals. In this paper, we used near-real-time stress evaluation surveys, in conjunction with our bio-metric stress detection, to minimize the biases of recall. The dataset provides more than 1,252 hours of activity and physiological signals collected from 15 nurses during their daily routine responsibilities. 83 hours of data are labelled with stress descriptors based on the validated stressful events by nurses. We included the unlabeled signals since we expect the unlabeled signals to have predictive value in anticipating stress events. The No Stress events (level 0) should be interpreted with care. These are false positives detected by our early model, that was confirmed as unstressed epochs. Figure 3 shows temporal plots of data collection and stress event distribution. The box plots ( Figure 5 to Figure 8) show four individual physiological signals recorded by the wearable namely the heart rate, EDA, skin temperature, blood volume pulse, and the reported stress levels from survey. The normal skin temperature for a person is about 33°C or 91°F. Skin temperature varies for various activities due to skin blood temperature and its flow and is normally within the range of 33.5 to 36.9 13 . However, this can vary quite widely based on the type and length of activity and indoor room temperature. Given the open-ended nature of the experiment, there are some anomalies in the data. Out of 15 subjects, 11 subjects had higher skin temperature in high stress events, compared to medium-stress or low-stress situations. The relationship between skin temperature and stress has been discussed by Herborn et al 14 . The heart rates for a normal person despite the sex ranges from 60 to 100 while in resting state. However, this changes due to different activities and the duration of the activity. Given that the subjects are doing different activities, one can observe high variation in heart rates. Figure 5 shows the distribution of heart rate and associated stress level for all the subjects. Stress did not have strong correlation with the heart rate itself. However, there are lots of claims that heart rate variability is an important signal in stress detection. Moreover, average of heart rate is higher in stressful situations for certain participants. Figure 6 shows the distribution of electrodermal activity and associated stress level for all the subjects. Stress has strong positive correlation with the electrodermal activity for 10 out of 15 participants. Average of EDA is higher in stressful situations for some participants. However, for some participants, the EDA is not a good indicator of stress because EDA does not vary or it is not positively correlated. There is high variability in EDA signal for various subjects in stressful events. The normal range for humans is from 1 to 20 microsiemens We observe that the average skin's electrodermal activity for all the participants when there is no stress reported is below 5, and the range for low stress is the same as stress free situations. However, EDA can go up to 60 in stressful situations based on the activity participants perform. Figure 8 shows the stress contributors. In this study, treating a COVID-19 patient was reported as the most biggest contributor of high stress. However, there was only one claim about COVID-19 or fear from COVID-19 itself. Lack of supplies was a major problem in this period; however, it was not mentioned as a contributor of acute stress. Table 3 . Survey questions and their categories We performed the analysis of the means for each of the biometric signals and Tukey's test was used to assess the differences between the pairs of stress groups. For Heart Rate, a one-way ANOVA revealed that there was a statistically significant difference in the mean BVP levels between at least two stress groups (F(3, 5410605) = 148.2, p = 4.960409e-96). Tukey's HSD Test for multiple comparisons found that the mean stress level was significantly different between High stress and other groups. Additionally, the mean medium stress was statistically different from mean HR for No Stress epochs. The tukey values for different comparisons are presented in Table 4 . Table 4 . HR For EDA, as with the HR, For Heart Rate, a one-way ANOVA revealed that there was a statistically significant difference in the mean BVP levels between at least two stress groups (F(3, 21407757) = 283442.8, p = 0.0). Tukey's HSD Test for multiple comparisons found that the mean stress level was significantly different for high stress compared to all other groups. Additionally, the mean medium stress was statistically different from mean EDA for No Stress epochs. The results are presented in For Blood Volume Pulse, a one-way ANOVA revealed that there was no statistically significant difference in the mean BVP levels between at least two stress groups (F(3, 281865708) = 0.0049, p = 0.99). Tukey's HSD Test for multiple comparisons found that the mean stress levels were not significantly different across groups as shown in Table 7 Human well-being is an important consideration both for individuals and organizations. As such, organizations need mechanisms to carefully monitor and manage high-stress environments such as hospitals to improve both the employee well being and patient satisfaction. We believe this study can be useful for researchers in many domains. First, researchers in signal processing and machine learning might use the dataset to develop new machine learning models that improve stress detection performance. Second, the accelerometer signals can be used to associate basic activities such as walking, running, sitting with stress to understand the relationship between physical activity and stress. Finally, researchers in human resources / human factors / organizational psychology would find the survey dataset along with biometric signals useful, because it is a unique dataset that makes association between biometric signals and stress related factors during the COVID-19 outbreak. The dataset, even without the signal data, is additionally useful to understand the differential distribution of various work related stressors during the pandemic. We conceptualized the study before the pandemic. The original design had an on-site observer making independent assessments of tasks and stress, alongside nurses' and system stress assessments. The outbreak prevented us from locating an investigator on site, but provided a unique dataset in return. The nurses were busier than usual and we had to ensure that we were not interrupting them too often. Not all of the dataset is covered with stress labels. Unlabelled data does not necessarily imply a lack of stress; it just means that we did not detect stress using our signal streams and that the subjects did not independently report it as a stressful period. We did not insist on complete coverage of labels, as the most important priority of nurses is taking care of patients. We provide the unlabelled data because we suspect that it may contain predictive markers of stress that future analyses may reveal. • Compared to the laboratory scenarios where the undivided attention of the subject is available, in high impact scenarios, the subjects may not be distracted or interrupted frequently from their professional tasks for labeling. • We only validated stressful events because of our focus on acute stress detection. Because nursing was stressful during the peaks of the COVID-19 pandemic, we provided no more than 6 events a day for the nurses. • Biometric signals cannot detect chronic stress, just acute stress. • Social distancing requirements upended a prior experimental design that included onsite investigator data collection describing nurse tasks in conjunction to stress signals. • Although the surveys sent to the nurses at the end of the day. While we have immediate stress detection, we opted to delay the surveys to the end of the day in order to not interrupt the work of the nurses. While the latency may have some degree recall bias, it is an improvement over traditional surveys which do not specify precise time of stress events. Finally, unlike laboratory studies that are typically conducted in a controlled environment, stress detection in a natural environment is more complex due to the influence of many social, cultural and individual factors. While we have attempted to provide some context to stress by conducting a survey based on available literature, we feel that there are several social, cultural and individual variables we did not consider in our survey. The code and data are available upon request to the corresponding author. A presence-based context-aware chronic stress recognition system Stress in the workplace: A general overview of the causes, the effects, and the solutions Inflammation, stress, and diabetes Workload and burnout in nurses Affectiveroad system and database to assess driver's attention Introducing wesad, a multimodal dataset for wearable stress and affect detection The 'trier social stress test'-a tool for investigating psychobiological stress responses in a laboratory setting Stress detection in working people Introducing mdpsd, a multimodal dataset for psychological stress detection The trier social stress test protocol for inducing psychological stress The stroop color and word test Towards an automatic early stress recognition system for office environments based on multimodal measurements: A review The use of infrared thermography to detect the skin temperature response to physical activity Skin temperature reveals the intensity of acute stress