The authors have declared that no competing interests exist.
Auditioning is at the very center of educational and professional life in music and is associated with significant psychophysical demands. Knowledge of how these demands affect cardiovascular responses to psychosocial pressure is essential for developing strategies to both manage stress and understand optimal performance states. To this end, we recorded the electrocardiograms (ECGs) of 16 musicians (11 violinists and 5 flutists) before and during performances in both low- and high-stress conditions: with no audience and in front of an audition panel, respectively. The analysis consisted of the detection of R-peaks in the ECGs to extract heart rate variability (HRV) from the notoriously noisy real-world ECGs. Our data analysis approach spanned both standard (temporal and spectral) and advanced (structural complexity) techniques. The complexity science approaches—namely, multiscale sample entropy and multiscale fuzzy entropy—indicated a statistically significant decrease in structural complexity in HRV from the low- to the high-stress condition and an increase in structural complexity from the pre-performance to performance period, thus confirming the complexity loss theory and a loss in degrees of freedom due to stress. Results from the spectral analyses also suggest that the stress responses in the female participants were more parasympathetically driven than those of the male participants. In conclusion, our findings suggest that interventions to manage stress are best targeted at the sensitive pre-performance period, before an audition begins.
The first attempt to introduce a taxonomy of stress dates back to Hans Selye in 1936, who defined stress as a “non-specific endocrine response” [
Music performance is a particularly apt domain for studying ANS reactivity to stress [
The majority of stress research into music performance has focused on the psychological construct of performance anxiety using questionnaires, while neglecting the objective assessment of corresponding physiological components. A notable exception is the work of Craske and Craig [
The analysis of HRV in the time, frequency, and non-linear domains is now widely used to assess the biomarkers of stress. In particular, the high frequency (HF) power in HRV is considered to reflect PNS activity influenced by vagal control, while the low frequency (LF) power is multifaceted and was previously believed to reflect SNS activity. The ratio of the power in the LF to HF frequency bands (so-called LF/HF ratio) was long thought to indicate the degree of sympathovagal balance; the higher the ratio, the greater the dominance of SNS, while a lower ratio was thought to suggest the dominance of PNS activity [
Although the analysis of HRV in the frequency domain can identify and capture changes in stress, nonlinear analysis in the form of structural complexity has recently been used to quantify degrees of determinism versus randomness in signals and has become prevalent [
Most studies of complexity loss in HRV have been conducted in the context of understanding cardiovascular diseases. Our recent study on psychosocial stress in public performance [
The aim of the present study, therefore, is to establish a systematic approach to the examination of physiological stress in music performance contexts. The evolution of stress responses to performance was modeled over a cohort of 16 musicians whose electrocardiograms (ECGs) were recorded for 5 min prior to and 5 min during two performances: (i) a low-stress condition with no audience present and (ii) a high-stress condition in front of an audition panel. An audition was deemed particularly well suited for the high-stress scenario, owing to the scrutiny under which musicians are placed. Auditions also allow enhanced experimental control and maintain high ecological validity through the assignment of appropriate pieces to be played and the possibility of demarcating precise timings before and during performance. We used modern wearable sensing devices for the collection of ECG data and advanced analysis techniques to capture the signature of stress. Specifically, multiscale sample entropy (MSE) [
Eleven violinists from the Royal College of Music (RCM) and five flutists from the Conservatory of Southern Switzerland (CSI) participated in the study. The cohort consisted of healthy male (n = 9) and female participants (n = 7) with a mean age of 23.12±2.42 years (range 19–27), all of whom were advanced music students with at least 10 years of public performance experience. Recruitment at the RCM took place from October 2012 to March 2013, with data collection in March 2013, while recruitment at the CSI took place from March to April 2011, followed by data collection in May 2011. Participants were assigned to perform individually the
For the violinists, ECG was recorded using the Bioharness, a physiological monitoring device from ZephyrTM which has been validated in a similar scenario by Johnstone and colleagues [
Prior to each performance, participants completed Form Y1 of the State-Trait Anxiety Inventory (STAI) [
Before conducting the experiment, every participant attended a 20-minute induction session and confirmed their willingness to deliver multiple polished performances (on separate days) of either the
The low- and high-stress performances were scheduled on separate days, and the order was counterbalanced across participants. The musicians were asked to arrive 30 minutes before the pre-performance period for the attachment of the ECG recording devices and also for usual performance preparation (e.g. warming-up, tuning, and rehearsing). Stage calls were given at 20 minutes and 10 minutes before performance by a member of the research team acting as the “backstage manager”. At 5 minutes before performance, participants were brought to a backstage area and asked to complete Form Y1 of the STAI. The backstage manager then gave a confirmation signal and allowed the participant to enter the performance room; this period is referred to as
The timeline was designed for collecting physiological data from the participants experiencing the low- and high-stress conditions.
The R-peaks in the recorded ECGs were detected using a combination of matched filtering and Hilbert transform algorithms [
MF-HT stands for the combination of matched filter and Hilbert transform algorithms, while PP and P respectively designate the periods pre- and during performance.
HRV signal of a participant, interpolated from the NN intervals in the low- and high-stress conditions.
The analyses were conducted in the time, frequency, and nonlinear domains. The averages of NN intervals (AVNN), the standard deviation of all NN intervals (SDNN), the square root of the mean of the squares of the differences between adjacent NN intervals (rMSSD), and the percentage of differences between adjacent NN intervals (pNN50) were used as standard time-domain metrics [
For the nonlinear analysis, the MSE and MFE were computed to quantify degrees of uncertainty of HRV over ten scales. The selected parameters for estimating SampEnt were: embedding dimension = 2, tolerance = 0.15 times the standard deviation of the data, time lag = 1 [
The t-test was used to examine statistical differences in each time, frequency and nonlinear metric according to four possible comparisons: (1) high-stress: pre-performance (PP) vs performance (P), (2) low-stress: pre-performance (PP) vs performance (P), (3) pre-performance (PP): low- vs high-stress, and (4) performance (P): low- vs high-stress (see
Metrics | Descriptive statistics | Statistical comparisons | |||||||
---|---|---|---|---|---|---|---|---|---|
Low- stress: PP |
Low- stress: P |
High-stress: PP |
High- stress: P |
Low- stress: |
High- stress: |
PP: |
P: |
||
Time-domain | |||||||||
AVNN | 624.0±78.6 | 637.2±108.2 | 557.5±91.9 | 555.3±115.0 | -0.52, 0.61 | 0.07, 0.95 | 3.00, 0.009 | ||
SDNN | 51.4±20.6 | 42.4±17.5 | 48.7±21.6 | 38.6±18.8 | 1.77, 0.10 | 1.30, 0.21 | 0.49, 0.63 | 0.97, 0.35 | |
rMSSD | 22.5±10.9 | 23.8±12.4 | 16.2±7.7 | 16.4±6.4 | -0.43, 0.67 | -0.07, 0.95 | 2.70, 0.02 | 2.97, 0.01 | |
pNN50 | 2.7±3.5 | 2.8±4.3 | 1.3±1.5 | 0.7±0.9 | -0.06, 0.96 | 1.26, 0.23 | 1.98, 0.07 | 2.02, 0.06 | |
Frequency-domain | |||||||||
TOTPWR | 51.79±14.37 | 36.50±12.35 | 41.42±16.78 | 26.41±12.16 | 2.89, 0.01 | ||||
VLF | 51.61±14.28 | 36.41±12.30 | 41.28±16.70 | 26.35±12.13 | 2.90, 0.011 | ||||
LF | 0.168±0.126 | 0.087±0.080 | 0.146±0.117 | 0.062±0.070 | 2.22, 0.04 | 0.61, 0.55 | 1.12, 0.28 | ||
HF | 0.037±0.037 | 0.026±0.027 | 0.020±0.018 | 0.013±0.010 | 1.18, 0.26 | 1.56, 0.14 | 2.26, 0.04 | 2.55, 0.02 | |
LF/HF | 5.51±2.82 | 3.78±1.93 | 7.31±3.69 | 5.92±9.71 | 2.24, 0.04 | 0.5, 0.62 | -2.00, 0.06 | -1.02, 0.32 | |
Nonlinear complexity | |||||||||
MSE | 9.54±1.56 | 11.50±1.85 | 8.00±1.41 | 10.18±1.91 | 1.97, 0.07 | ||||
MFE | 11.70±1.64 | 13.61±1.90 | 10.00±1.64 | 12.39±1.84 | 1.79, 0.09 |
The unit of AVNN, SDNN, rMSSD, and pNN50 is millisecond (msec) and the unit of the TOTPWR, VLF, LF, and HF is s2/Hz.
LS and HS are abbreviations for the low- and high-stress conditions; PP and P are abbreviations for the pre-performance and performance periods, respectively. Note that only the low- and high-stress entropies for the performance period (bottom right) are not statistically different (
LS and HS are abbreviations for the low- and high-stress conditions; PP and P are abbreviations for the pre-performance and performance periods, respectively. Note that only the low- and high-stress entropies for the performance period (bottom right) are not statistically different (
Analyses of the normalized LF and HF powers revealed differences in the responses of the 9 male and 7 female participants in the high-stress condition. Using a significance level of
The reported state anxiety of the musicians was significantly higher in the high-stress condition (mean = 39.12±7.04) than in the low-stress condition (mean = 34.37±7.88) t15 = -2.594,
We have examined the cardiovascular reactivity of musicians experiencing low- and high-stress performance conditions within the framework of complexity loss theory, quantified using the multiscale entropy (MSE) and multiscale fuzzy entropy (MFE) algorithms. Unlike standard questionnaire-based anxiety assessments, this has been achieved through a suite of objective stress measures based on physiological responses to stress in two scenarios, low- versus high- and before- versus during-performance. Advanced signal processing algorithms for R-peak detection and HRV extraction have been employed to deal with noisy cardiac data in real-life scenarios, while state-of-the-art data analysis techniques in the time, frequency and nonlinear complexity domains have been used to quantify the signatures in HRV related to stress in performance. The analysis has also revealed that currently used spectral analyses of HRV may be inadequate for detecting stress reactivity, as exemplified by the statistically non-significant findings reported in
The time-domain analysis based on the AVNN and rMMSD metrics has suggested that, in high-stress conditions, the heart rates of the participants were higher than in low-stress conditions. However, these higher heart rates were accompanied by a smaller difference in heart rate variability when comparing the pre-performance and performance periods, in all time-domain metrics. The standard HRV frequency analysis showed decreases in the LF and HF powers from the pre-performance to performance period, and from the low- to high-stress condition, suggesting a shift of vagal activity from the HF to LF band [
The analysis of the normalized LF and HF powers from the pre-performance to the performance period has also revealed a significant decrease in the proportion of LF power, accompanied by a significant increase in the proportion of HF power among the female participants in the high-stress condition. Therefore, assuming a low proportion of HF power to be indicative of low parasympathetic tone, these results suggest that the female participants experienced a more pronounced activation of their parasympathetic nervous systems from the pre-performance to the performance period. This finding has been alluded to in several other studies, with Kattimani [
Our nonlinear analyses have shown that the MSE and MFE approaches have achieved robust discrimination of the underlying features related to the dynamics of the heart, regulated by the autonomic nervous system. Based on the complexity loss theory, both MSE and MFE have shown that the transitions from pre-performance to performance corresponds to a lowering of stress levels in the musicians. The same complexity pattern was presented in the discrimination from the high-stress condition to the low-stress condition. For rigor, these
The well-documented lack of suitable data acquisition devices (robust to musicians’ movement and motion artefacts, unobtrusive, discreet and comfortable) and a shortage of signal processing algorithms for real-world wearable applications have so far been prohibitive to larger-scale studies of stress experiences in human performance. In this study, we have used both wearable and stationary physiological recording devices and have addressed the imperfections and artefacts in such real-world data through advanced data analysis methods. Our study has focused on combining physiological and psychological measures, analyzed within the framework of the complexity loss theory, to analyze data from a number of performers and to extend a previous single-person study, to address a more general issue of musicians’ emotional and physiological adaptability to psychosocial stressors. However, our study still has limitations, such as: (1). the relatively low sample size, which is due to the constraints of recruiting musicians who were willing and able to provide multiple, polished performances of challenging repertoire and (2) the lack of resting state (baseline) and post-performance cardiographic data, as the study was designed to compare only stress reactivity in pre-performance and performance periods.
Subsequent work will consider joint analysis of multivariate physiological data, such as HRV, respiration rate and skin conductance. The collection, analysis and examination of multivariate data, in relation to strategies for managing stress and enhancing performance quality, promises to offer personally and professionally significant advancements in musicians’ training and skill development, particularly if targeted at the the sensitive period before performance, as shown in this study, and employed in a range of performance contexts.
The research was granted ethical approval by the Conservatoires UK Research Ethics Committee and was conducted according to ethical guidelines of the British Psychological Society. Written informed consent was obtained from all participants.
(DOCX)
The ZIP folder consists of 16 Matlab files (.mat) and one readme.txt file.
(ZIP)
We wish to thank the musicians and the adjudicators who took part in this study. The research reported in this article was supported by grants from the UK’s Arts and Humanities Research Council (grant ref. AH/K002287/1), the Peter Sowerby Foundation, EPSRC grant EP/K025643/1, EPSRC Pathways to Impact Grant PSA256, and EPSRC Multidisciplinary University Research Initiative EP/P008461.