Impaired self-monitoring ability on reaction times of psychomotor vigilance task of nurses after a night shift

ABSTRACT

The cognitive ability to self-monitor one’s current performance is important for hospital nurses to maintain safety and health. However, studies on the effects of rotating shift work on self-monitoring ability are insufficient. We examined the differences in self-monitoring accuracy across shifts in a rotating three-shift system among 30 female ward nurses (mean age 28.2 years). Their self-monitoring ability was calculated by subtracting the predicted reaction times of the psychomotor vigilance task performed just before exiting the workplace from the actual reaction times. A mixed-effect model was employed to assess the effects of shift, awake hours, and prior sleep duration on self-monitoring ability. We observed impaired self-monitoring ability in nurses, particularly after the night shift. Although actual performance remained high across all shifts, their self-predictions on reaction times became pessimistic in the night shift, resulting in a difference of approximately−100 msec. The effect of the shift on self-monitoring was obvious even after adjusting for sleep duration and hours awake. Our findings indicate that the misalignment between their working hours and circadian rhythms may affect even professional nurses. Occupational management that emphasizes maintaining circadian rhythms will improve the safety and health of nurses.

KEYWORDS:

• Self-awareness
• alertness
• performance
• shift work schedule
• rotating shifts
• health care worker
• field survey

Introduction

As health problems follow no set schedule, healthcare services need to be provided 24 hours a day, 7 days a week. Thus, healthcare institutions depend heavily on nurses, particularly those who work the night shift. However, problems linked to work performance and health are more prevalent in night shift workers than in day shift workers owing to the deviation between their work schedule and circadian rhythm (Vedaa et al. 2016; Zion and Shochat 2018). In nurses, impaired ability to perform tasks and a high risk of accidents are obvious risk factors for safety and health (Caldwell et al. 2019; Dorrian et al. 2000; Härmä 2006). Ensuring the safety and health of nurses and patients also depends on their capacity to accurately self-monitor their performance (Barbe et al. 2018). For instance, to utilize compensatory behaviors for subpar performances, we must be aware of our status (Boardman et al. 2021). Reportedly, risk-coping and protective behaviors are closely correlated with accurate self-awareness of one’s performance (Slovic 1978; Stanton et al. 2001).

In general, self-monitoring ability refers to the discrepancies between the actual measured task performance and predicted performance prior to the task or self-rating after the task (Baranski and Pigeau 1997; Boardman et al. 2021). The majority of earlier research on self-monitoring has centered on the impact of sleep deprivation or restriction on self-monitoring ability (Aidman et al. 2019; Baranski and Pigeau 1997; Boardman et al. 2018; Dorrian et al. 2000), and these studies have discovered that sleep deprivation causes a synchronized decline in alertness and performance perception (Boardman et al. 2021). The relationship between rotating shifts and self-monitoring ability in actual shift workers has not been investigated, although a few self-monitoring studies exist in which simulated night shifts were performed (Dorrian et al. 2003; Morris et al. 2017).

Most Japanese ward nurses work rotating shifts, i.e., their reporting times to the workplace change daily (Kida and Takemura 2022). Additionally, workers encounter accidents and injuries more often during late night hours (Folkard et al. 2005). To maintain the safety and health of shift workers and patients, it is crucial to comprehend the decision-making process and behaviors that underlie self-monitoring.

The subjective cognition among nurses is linked to improved sleep in terms of both amount and quality (Veal et al. 2021). Nevertheless, although sleep and fatigue are significant elements that may affect self-monitoring ability, these issues have not been investigated because the majority of earlier studies have focused on the impact of sleep loss in lab settings. We intended to compensate for these variables and assess their effect on self-monitoring ability by incorporating the times spent sleeping and awake in the statistical model and enlisting the actual nurses as participants. Moreover, investigating workers in their actual workplace may yield valuable data that more closely reflect actual conditions than laboratory experiments performed on university students.

We selected the psychomotor vigilance task (PVT) for the performance measurement. The PVT is an extensively used vigilance measurement method and has high reliability, task simplicity, and robustness to the practice effect (Basner and Dinges 2011; Roach et al. 2006; Wilkinson and Houghton 1982). Furthermore, it was anticipated that the reaction time feedback (displayed for each button press during PVT) would provide valuable data for the forecast to be made in the actual millisecond format.

In this study, we examined the connection between circadian rhythm and the self-monitoring ability of hospital ward nurses working under a rotating three-shift system. Circadian misalignment issues have been documented in nurses who work rotating shifts (Chellappa et al. 2019; Smith and Eastman 2012). Moreover, one’s self-monitoring ability has been reported to be related to the circadian rhythm in a laboratory study that also recorded the body temperature and heart rate data (Morris et al. 2017). Each nurse works more day shifts than night or evening shifts in hospitals that use a rotating shift system (Kida and Takemura 2022), including the studied hospital. Moreover, nurses may be required to stay awake during the day for various reasons, such as social and family events (Berkman et al. 2015; Winwood et al. 2006). Hence, we considered a high adaptive load to evening, particularly for night shifts rather than day shifts, based on the assumption that their circadian rhythm might be aligned similarly to those without the night shifts, as stated in an earlier study (Costa et al. 1994).

Given the divergence from the circadian cycle, we hypothesized that self-monitoring ability would be most impaired after the night shift. We used shift (i.e., the timing for task operation), awake hours (including working hours), and sleep duration as proxies for indication of deviations from circadian rhythm, fatigue, and recovery, respectively.

Materials and methods

Participants

Thirty female nurses aged 22–46 years (M = 28.2, SD = 5.9) from a single ward (mainly in charge of cardiology and cardiac surgery) of a general hospital in Kanagawa Prefecture, Japan, participated in the study. The participants were oblivious to the research hypothesis, and we assured them that the collected data would be solely used for research and no other purposes, such as personnel evaluation. Their average work experience was 6.4 ± 5.6 years. Nurses in this hospital work under a backward rotating, three-shift system, comprising the day (scheduled work time: 8:15–17:00), evening (15:45–0:30), and night (0:00–8:45) shifts. We asked the participants to specify their chronotype by answering the following question: “Please indicate how active you are in the morning or evening” (1: obviously active in the morning, 2: somewhat active in the morning, 3: somewhat active in the evening, or 4: obviously active in the evening). The targeted ward did not receive any patients with COVID-19, and no infections were noted throughout the study period.

Research ethics

Before beginning the study, we received each participant’s signed, informed consent. Additionally, all participants received a debriefing after the study was completed. All procedures undertaken in this work were in accordance with the ethical standards of the institutional committees on human experimentation and the Helsinki Declaration of 1975, as revised in 2008. The ethics review committee of the National Institute of Occupational Safety and Health in Japan approved all procedures (2020N07).

Data collection

PVT task and performance prediction

From 2020/10/16 to 2020/11/06 (22 days), the participants were asked to perform a PVT after every workday (just before leaving the ward) using a PVT testing device (PVT-192, A.M.I., USA) that was positioned in the break room throughout the study period. The PVT is a widely employed psychomotor task designed to objectively measure one’s state of alertness (Basner and Dinges 2011; Wilkinson and Houghton 1982). To reduce the task load on the participants, we applied a shorter (5 min) version of the PVT (Roach et al. 2006). The lapse threshold was>500 ms, and the PVT’s interstimulus intervals ranged from 2000 to 10 000 ms (Cvirn et al. 2019). Before beginning the PVT, the participants were asked to use their smartphones to report their shift type (day, evening, or night shifts), subjective sleepiness, and performance prediction. Subjective sleepiness was conveyed on a 10-point Likert scale that ranged from 1 representing “not sleepy” to 10 representing “sleepy.” (Riegel et al. 2013). In the performance prediction section, the participants predicted their overall reaction time (RT, range: 200–1000 ms) of the PVT, which was conducted immediately after answering the questionnaire item “What do you estimate the average PVT reaction time (milliseconds indicated by the red number in the small window) will be? (0.1 seconds = 100 milliseconds).” We provided thorough instructions on the PVT and performance prediction before the commencement of the study via face-to-face orientation and video instructions. The first and second days of data collection for each participant were not included in the analysis. We asked the participants to predict the “average RT” to make predictions simpler, and then, we used the median reaction time (mRT; ms) to calculate the actual PVT performance to optimize the sensitivity while handling numbers in millisecond units (Basner and Dinges 2011). To assess the accuracy of performance prediction, we calculated the correctness of the prediction (Δ mRT) by subtracting the predicted RT from the actual performance (mRT) (Boardman et al. 2021). A Δ mRT close to 0 indicates a correct prediction, whereas a negative value denotes impaired self-monitoring by a prediction that is too pessimistic.

Sleep length and awake hours

The participants documented their sleep using a graphical self-written sleep daily with a time resolution of 30 minutes. They were asked to keep track of all periods of sleep, including naps. By adding all the sleep times initiated in the previous 24 hours leading up to check-in at work, entries in this sleep diary were used to calculate the sleep duration before each duty. Based on the diary, the awake hours from the nearest wake-up time until each PVT was calculated. These values were employed as a covariate in later analysis.

Statistical analysis

We used a linear mixed model (LMM, or multilevel analysis) with three fixed effects (shift, prior sleep, and prior awake hours) and one random effect (subject) to analyze the effects of shift, sleep, and awake hours on PVT measurements (except the number of lapses) and self-monitoring ability. We selected LMM to consider the unbalanced data structure from a different number of shift assignments. To test the model’s multicollinearity, the shift’s df-adjusted generalized variance inflation factor (GVIF) and GVIF for other variables were below 1.8. On the fitted models, the effect of the shift was tested using Wald’s F tests with Kenward – Roger’s degree of freedom (Luke 2017). As a post-hoc test, we performed pairwise comparisons of the estimated marginal means between shifts (Searle et al. 1980) using Holm’s method for multiple comparisons if the effect of the shift was significant (Holm 1979). Additionally, the impacts of sleep and awake hours on independent variables were assessed using t-tests on the estimated coefficients using Kenward – Roger’s degree of freedom.

Assuming the Poisson distribution of the errors, we utilized an extended linear mixed model with the same design as the LMM to examine the number of lapses. Lastly, as a supplementary study, Pearson’s correlation coefficients computed for pooled data of each shift were used to examine the link between anticipated RT and actual median RT. All data processing and statistical tests were performed in R version 4.1.3 (R Core Team 2021), running under Ubuntu 20.04.4. The tests were conducted using functions in the car (Fox and Weisberg 2019), effects (Fox 2003; Fox and Weisberg 2019), emmeans (Lenth et al. 2022), lmerTest (Kuznetsova et al. 2017), and multcomp (Hothorn et al. 2008) packages of R in addition to the basic R functions. A 5% level of significance was set for all tests.

Results

We examined 305 measurements from 30 participants during the study period. Of the participants, 19 (63.3%) identified their chronotype as “3: somewhat active in the evening.” Moreover, six (20%) participants were “4: obviously active in the evening,” four (13.3%) participants were “2: somewhat active in the morning,” and only one (3.3%) participant was “1: obviously active in the morning.” The descriptive data for the lengths of work and sleep for each shift are shown in Table 1. During the 22-day trial, the participants had, on average, 10.2 ± 2.0 days of duty, and they were most often assigned to the day shift. Consecutive two-night shifts happened 0–2 times during the study period (M = 1.03, SD = 0.72). There were no three or more consecutive night shifts during the study period. The increased need for overtime work resulted in more working hours on day shifts. A preparative sleep period is required before working in the night shift, as evidenced by the fact that the amount of sleep before the day shift was shorter than that before other shifts.

Table 1. Number of assignments, number of working hours, and preceding sleep durations for the day, evening, and night shifts.

	Shift (scheduled working time)
	Day (8:15–17:00)	Evening (15:45–0:30)	Night (0:00–8:45)
Number of assignments per participant^*1	6.0 ± 1.9	2.6 ± 1.0	2.3 ± 1.0
Hours of duty^*2	11.7 ± 0.7	10.5 ± 0.5	10.9 ± 0.7
Timing of PVT [hh:mm]	18:56 ± 0:27	1:09 ± 0:18	9:28 ± 0:21
Awake hours from last sleep to PVT	12.6 ± 1.1	14.4 ± 2.0	14.8 ± 3.3
Total hours of prior sleep^*3	5.1 ± 1.5	8.8 ± 2.6	8.3 ± 3.4

Summary of 305 measurements by 30 participants in 22 days. Mean ± SD. *1: Number of assigned shift types during the study period per participant. *2: Including legal break time (i.e., 1 h). *3: Total hours of sleep(s) started 24 h before the check-in time.

PVT: psychomotor vigilance task.

Median RT

Figure 1a compares the median RT values of the different nursing shifts and illustrates no significant effect of shift (F _{2, 189.7} = 0.25, p = 0.781). Awake hours and sleep duration displayed no effect on median RT (coefficient = −0.73, 95% CI [−2.39, 0.94], p = 0.397; coefficient = 3.08, 95% CI [−1.15, 7.26], p = 0.155).

Figure 1. Differences in psychomotor vigilance task measures and subjective sleepiness of day, evening, and night shifts. The fitted values for the day, evening, and night shifts are shown as filled circles, triangles, squares, and vertical lines along with their associated 95% confidence intervals. To determine the Δ reaction time, the anticipated RT was subtracted from the median RT.

Predicted RT

The shift had a substantial impact on predicted RT (F _{2, 191.8} = 8.97, p < 0.001), as shown in Figure 1b. Following a post-hoc analysis, it was determined that there were significant differences between the day and night shifts and the evening and night shifts (t _195.1 = −4.11, p < 0.001; t _188.3 = −2.97, p = 0.007, respectively). In contrast, there was no discernible difference between the predicted RTs for the day and evening shifts (t _192.4 = −1.36, p = 0.176). The most pessimistic predictions were made in the night shift condition. The predicted RT was not affected by the number of awake hours (coefficient = −1, 95% CI [−5.3, 2.93], p = 0.565). However, the sleep duration exerted a significant effect on the predicted RT (coefficient = 12.39, 95% CI [2, 22.64], p = 0.02), i.e., the prediction of RT became 12.4 ms slower for every one hour of prior sleep.

Self-monitoring (Δ mRT)

Self-monitoring ability, as determined by Δ mRT, differed significantly between shifts (F _{2, 195.6} = 7.29, p = 0.001, Figure 1c). Post-hoc analysis indicated significant differences between day and night shifts and between evening and night shifts (t _200.6 = 3.76, p = 0.001; and t _190.3 = 2.49, p = 0.027, respectively). The Δ mRT values between the day and night shift nurses were approximately−50 and−100 ms, respectively. There was no difference between the self-monitoring scores of the day and evening shifts (t _196.7 = 1.45, p = 0.149). Awake hours and sleep duration did not affect the self-monitoring ability (coefficient = 0.01, 95% CI [−4.43, 4.37], p = 0.995 and coefficient = −9.21, 95% CI [−20.02, 1.66], p = 0.101, respectively).

Subjective sleepiness

Figure 1d illustrates that the shift significantly affected subjective sleepiness (F _{2, 193.8} = 6.47, p = 0.002). Post-hoc analysis showed significant differences between day and night shifts and between evening and night shifts (_198.1 = −3.55, p = 0.001; and t _189.4 = −2.28, p = 0.048, respectively). Conversely, the difference between the day and evening shifts was insignificant (t _194.7 = −1.45, p = 0.148). The nurses in the night shift were the most sleepy (7.9 points), followed by those working in the evening and day shifts (7.0 and 6.7 points, respectively). The duration of awake hours and sleep did not affect subjective sleepiness (coefficient = 0.05, 95% CI [−0.06, 0.15], p = 0.382; coefficient = 0.03, 95% CI [−0.22, 0.28], p = 0.82, respectively).

Number of lapses

The number of lapses (RT > 500 ms) differed significantly between shifts (Chisq ₂ = 14.4, p < 0.001; see Supplemental Figure S1). Post-hoc analysis indicated significantly more lapses during the day shift (Z = 3.57, p = 0.001; and Z = 2.86, p = 0.009 for the day – evening and day – night pairs, respectively). The number of lapses did not differ significantly between evening and night shifts (Z = −0.540, p = 0.5894). Although the effect of the shift attained significance, the numbers of lapses in all shifts were less than two (number of lapses: 1.39 [95% confidence interval (95% CI): 0.87, 2.24], 0.65 [0.4, 1.1], and 0.63 [0.36, 1.12] times in the day, evening, and night shifts, respectively). Hence, the difference in raw lapse times between shifts was slight, suggesting that shifts had little effect on lapses. Both awake hours and sleep duration did not exert a significant effect on the number of lapses (coefficient = −0.005, 95% CI [−0.072, 0.058], p = 0.89; and coefficient = 0.03, 95% CI [−0.1, 0.16], p = 0.679, respectively).

Supplemental correlation analysis

Table 2 displays the outcomes of the supplemental correlation analysis between predicted and actual performance. The predicted RT and actual median RT were positively correlated in the day and evening shifts, whereas no correlation was observed in the night shift data.

Table 2. Relationships between predicted reaction time and actual median reaction time in each shift.

	df	R (95% CI)	R^2	P
Day shift	136	0.42 (0.27, 0.55)	0.17	<0.001
Evening shift	59	0.47 (0.24, 0.64)	0.22	<0.001
Night shift	44	0.20 (−0.09, 0.47)	0.04	0.175

Results of Pearson’s product – moment correlation tests conducted using consolidated data from each shift.

Discussion

In this study, we investigated how rotating shifts affected the self-monitoring capacity of ward nurses. In line with our theory, the ability was affected by rotating shifts and was most impaired in the night shift. Nevertheless, the nurses’ actual performance levels were not affected by shifts and remained high. This is the first study to test the capacity for self-monitoring using RT (ms) of PVT rather than the commonly used Likert scale or visual analogue scale (which requires z-transformations). It was feasible to immediately measure the impact of the shift on one’s capacity for self-monitoring by obtaining forecasts in raw RT (Boardman et al. 2021).

Shift work and self-monitoring ability

The nurses’ ability to accurately self-monitor decreased considerably when they worked in the night shift. This finding somewhat supports a previous finding that sleep loss resulted in an increased discrepancy between objective and (pretask) subjective PVT performance (Smith et al. 2016). However, another study has reported that sleep loss may lead to an overestimation of PVT performance (Baranski et al. 2002). The underestimation detected in this study was not due to impaired PVT performance (mRT) but because of pessimistic predictions of performance, which have rarely been documented in earlier sleep deprivation studies. Most earlier investigations on sleep deprivation have reported relatively accurate self-monitoring (Åkerstedt et al. 2014; Dorrian et al. 2000, 2003; Lee et al. 2016; Morris et al. 2017). Contrarily, we found that even while their actual performance was unaffected, their self-monitoring was impaired, which could be explained by gloomy forecasts. Thus, sleep deprivation and deviation from one’s circadian rhythm may affect self-monitoring capacity differently.

The nurses in the present study might have made negative forecasts as a result of their work. They are officially mandated to minimize the risk of accidents during work (Winwood et al. 2006), and healthcare workers tend to typically steer clear of risk in medical decision-making (Arrieta et al. 2017). These tendencies may lead nurses to underestimate their PVT performance even when their actual performance is not impaired. Such pessimistic self-monitoring may even be advantageous, but underestimating is problematic from a production standpoint. For instance, underestimated pretask predictions caused by lack of sleep have been stated in physical activity (Daviaux et al. 2014) and on an academic exam (Terlizzese et al. 2019). One’s work performance and overall productivity may suffer if these underestimates lead to task avoidance. On the contrary, negative forecasts made during the night shift may not be related to the circadian rhythm, but some differences exist among shifts, which we could not control for.

The homeostatic sleep – wake process (Process S) and the circadian process (Process C) are the two widely used process models of sleep regulation (Borbély 1980, 1982; Borbély et al. 2016). The night shift had the longest awake hours in the current investigation, but the measurements were unaffected. However, shift as a proxy of deviation from circadian rhythms significantly affected the self-monitoring capability. The current outcomes indicate a more potent effect of process C than process S on the self-monitoring ability of shift-working nurses. Nonetheless, in the work setting of the current study, a majority of these factors are standardized and have slight variations; thus, they should be examined further on a broader group of shift workers.

Strengths and limitations

To the best of our knowledge, this is the first study to investigate how shift work affects a person’s self-monitoring ability in an actual work setting. By conducting a field study, we gathered information from actual nurses who worked shift rotations and analyzed its impact. Our statistical model, which considered sleep length and awake hours, provided an in-depth understanding of the self-monitoring ability of professional healthcare workers. Additionally, we collected performance estimates in a manner that can be directly compared with the measured values and avoided the subjective rating scales extensively used previously (Boardman et al. 2021).

However, there are several limitations in this study. First, we used shift type as a proxy of deviation for one’s circadian rhythm; however, there is enormous ambiguity in this assumption. As cognitive performance and circadian rhythm are strongly linked (Blatter and Cajochen 2007), future research should objectively determine the participants’ circadian rhythm via physiological means, such as melatonin (Benloucif et al. 2008). Second, we did not consider the qualitative aspects of sleep, which may be equally crucial as the quantities. Many quality indicators can be used to evaluate sleep (Cudney et al. 2022). In addition, the sleep data were only acquired from the participants’ sleep diaries. Objective means to record sleep, such as an actimeter, should be considered in the future, although we assume high compliance of the participants based on their PVT results and interview at the debriefing session. Third, PVT is restricted to the evaluation of vigilance levels. Future research should specifically examine more complicated activities, such as those used in prior studies, even though vigilance is an important underlying aspect for more complex brain functions (Baranski and Pigeau 1997; Baranski et al. 2002; Sallinen et al. 2013). Finally, we did not perform a post-test evaluation of the PVT performance and solely considered the pretask forecasts. Further research on the capacity to self-perceive the performance of a task that has just been completed is warranted.

Future implications

As mentioned in the limitation section, objective measurement of the circadian phase is needed in the future. The subjective sleepiness of the nurses was highest during the night shift (answered at approximately 9:30 in the morning), which is in line with earlier studies on shift work (Folkard et al. 2005; Wilson et al. 2019). Subjective sleepiness/alertness is greatly impacted by circadian rhythms, particularly those relying on melatonin secretion (Arendt 2005; Cajochen et al. 2003). An earlier report also indicated the involvement of circadian rhythms in sleepiness and self-monitoring (Dorrian et al. 2003). Additionally, considering the effect of consecutive night shifts (Haidarimoghadam et al. 2017) may offer more profound insights.

Owing to the fact that shift work was related to impaired self-monitoring ability, occupational management aimed at maintaining circadian rhythms will be advantageous for the safety and health of the nurses. Implementing a forward-rotating shift system can alleviate the adverse effects of shift work (Garde et al. 2020; Sallinen and Kecklund 2010). Countermeasures premeditated to lessen sleepiness and fatigue during the night will also aid in managing the self-monitoring ability. For instance, taking a nap (Lee et al. 2021) or installing workplace lighting with a highly correlated color temperature (Cajochen 2007) may be beneficial. Nevertheless, its effect on circadian rhythm should be cautiously considered (Higuchi et al. 2021).

The current study’s outcomes may apply to workplaces employing rotating shift systems because they exhibited the negative effects of deviating from biological rhythms. However, our findings were obtained from a single ward in a Japanese hospital, which may necessitate caution in generalizing. For instance, diverse shift systems (van Amelsvoort et al. 2004) between hospitals and countries, various responsibilities in the hospital (Oyama et al. 2015), napping opportunity (Lee et al. 2021), and workers’ age structure (Philip et al. 2004) may pose difficulties in generalizing the current results.

Conclusions

We examined the self-monitoring capability of hospital ward nurses working under a rotating three-shift system by performing PVTs and requesting performance predictions. Consequently, we discovered an impaired self-monitoring ability, particularly after night shift duties. An overly pessimistic self-assessment influenced the result, although the actual performance remained high. The outcomes also suggested a more substantial impact of the circadian phase than homeostatic pressure on the self-monitoring ability. Thus, occupational management concentrating on maintaining circadian rhythms will be beneficial for the safety and health of the nurses.

Acknowledgments

The authors would like to thank Atsuko Tamaoki for her help in collecting and analyzing the data. We also wish to express our gratitude to all the personnel and participants for their unwavering commitment to the research during the COVID-19 pandemic. This work was supported by the Industrial Disease Clinical Research Grants from the Ministry of Health, Labour and Welfare, Government of Japan (grant numbers 150903-01, 180902-01) and Grant-in-Aid for Young Scientists (start-up) from the Japan Society for the Promotion of Science (grant number 19K24187).

Disclosure statement

No potential conflict of interest was reported by the authors.

Data availability statement

The data supporting this study’s findings are available from the corresponding author upon reasonable request.

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/07420528.2023.2193270

Additional information

Funding

The work was supported by the Japan Society for the Promotion of Science [19K24187]; Ministry of Health, Labour and Welfare [150903-01, 180902-01].