Subjective and objective fatigue dynamics in air traffic control

Abstract

Fatigue is a longstanding issue in air traffic control (ATC), closely associated with shift work and time-related factors. However, the dynamics of fatigue across morning, evening, and night shifts in an area control center (ACC) remain largely underexplored. This study examined sleep duration and fatigue progression across different shift types. Both objective (three-minute Psychomotor Vigilance Task, PVT-B) and subjective (Stanford Sleepiness Scale, SSS) measures were conducted at the beginning, middle, and end of each shift. Results indicated that pre-shift sleep duration was shortest before night shifts, likely increasing sleep pressure and reducing alertness during the window of circadian low (WOCL). Subjective fatigue remained stable throughout morning shifts but increased towards the end of evening shifts, reflecting circadian influences. Night shifts exhibited peak fatigue during the WOCL, driven primarily by circadian rhythms rather than task load. Objective measures revealed a mid-shift decline in performance, with only partial recovery in the latter half of night shifts. Compared to day shifts, night shifts resulted in significantly higher fatigue levels, underscoring the critical role of circadian rhythms in fatigue dynamics. These findings highlight the need for targeted fatigue mitigation strategies that address circadian vulnerabilities and irregular sleep patterns in ATC shift systems.

Introduction

Across many domains, continuous 24/7 operations necessitate shiftwork to maintain uninterrupted service. However, shiftwork imposes considerable demands on operators, especially when their schedules vary drastically from week to week. This unpredictability, especially with night shifts, is highly associated with increased fatigue levels¹⁾, which can compromise individual performance and, in turn, degrade the overall safety and effectiveness of socio-technical systems²⁾. Understanding the impact of shiftwork on fatigue is essential, particularly in safety-critical domains such as air traffic control (ATC). ATC, a core component of air traffic management (ATM), involves the dynamic, integrated management of air traffic and airspace, ensuring the continuous and safe provision of air traffic services, as described by the International Civil Aviation Organization³⁾. ATC serves three essential functions: ensuring safe separation between aircraft, optimizing airspace usage to increase traffic throughput, and providing additional services to pilots, such as weather and navigation assistance. At the core of these functions are air traffic controllers (ATCOs), whose cognitive performance is critical to maintaining system safety and efficiency. Given the high demands of shiftwork, fatigue among ATCOs has become a growing concern⁴⁾. Supporting cognitive performance and managing fatigue is essential, as fatigue can degrade attention, decision-making, and reaction times, which is potentially compromising safety⁵⁾.

The interaction between fatigue and shift types in a shift system in ATC will be systematically addressed. Fatigue in ATC is investigated by presenting results from five field studies that provide insights into fatigue progression over time and across different shift types. These findings offer a deeper understanding of fatigue in ATC, supporting the development of effective countermeasures and management strategies to enhance system safety.

Fatigue as a concept

The concept of (mental) fatigue is widely recognized but lacks a universally accepted, sharp definition. Generally described as a state of profound tiredness or weariness, fatigue arises from insufficient sleep quality or quantity, extended mental exertion, monotony and stress^{2, 6, 7)}. Tailored to the aviation context, ICAO⁸⁾ defines fatigue as “a physiological state of reduced mental or physical performance capability resulting from sleep loss or extended wakefulness, circadian phase, or workload (mental and/or physical activity) that can impair a crew member’s alertness and ability to safely operate an aircraft or perform safety-related duties”. Williamson et al.²⁾ present a model of fatigue, that, similar to the ICAO definition, identifies time of day, time awake and task-related factors as key contributors to fatigue, which can impair performance and lead to safety-critical situations. In this model, fatigue is seen as a biological drive for recuperative rest, typically through sleep. Fatigue is also conceptualized as a broad phenomenon encompassing various forms, such as physical fatigue and sleepiness, the latter of which is considered the most relevant form in transportation sectors²⁾. Sleepiness, often defined as the tendency or proneness to fall asleep, also referred to as sleep propensity^{7, 9)} is closely linked to sleep-related factors, including sleep quality, sleep quantity, and time awake^{10, 11)}. It is also influenced by circadian rhythms, particularly the time of day and the window of circadian low (WOCL), which spans from approximately 2:00 a.m. to 6:00 a.m.^12,13,14). In this work, sleepiness is understood as a related term to the overarching concept of fatigue. The two-process model of sleep regulation¹²⁾ defines fatigue as the result of two interconnected mechanisms: a homeostatic sleep drive and a circadian process. Fatigue arises when the homeostatic drive exceeds a threshold, while wakefulness occurs when it falls below another¹¹⁾. The circadian process regulates these thresholds in a daily oscillatory pattern, promoting wakefulness. The precise timing of the circadian process can vary among individuals due to differences in their circadian rhythms.

Consequences of fatigue in ATC

Fatigue is increasingly recognized as a safety concern in ATC^14,15,16). In the past decade, research into the contributing factors and effects of fatigue in ATC has grown, reflecting its importance in maintaining safe and effective operations. Fatigue is broadly viewed as a detrimental factor that affects various cognitive and emotional functions¹⁷⁾. Additionally, fatigue is understood both as a subjective experience of tiredness and an objective decrement in performance¹⁸⁾, making it an issue that warrants careful examination. Thomas et al.¹⁹⁾ describe the consequences of increased fatigue in transportation sectors. The authors describe basic physiological changes (such as altered brain activity), primary cognitive effects (such as slower cognitive processing speed and decreased attention), higher-order cognitive effects (such as worsened memory processes and reduced decision-making quality) and psycho-social impacts (such as lowered mood and decreased communication abilities). Taken together, these effects may have adverse consequences on performance and safety, as all are linked to higher error rates and microsleep propensity, as well as a reduced risk avoidance¹⁹⁾.

In ATC, fatigue has also a significant impact on ATCOs’ cognitive and work performance²⁰⁾. Recent studies have shown that crucial cognitive functions required for effective task performance in ATC are impaired under fatigue¹⁵⁾. Evidence indicates that fatigue can lead to reduced sustained attention, less situational awareness, lower accuracy, attentional lapses, slower reaction times, loss of detail focus, and diminished perceptual and planning abilities^{15, 21)}. Such impairments support the view of fatigue as a major safety risk²²⁾, although its full impact often goes underrecognized²³⁾. One reason for this underestimation is that the specific deficits associated with fatigue are difficult to predict, and fatigue-related incidents are rarely straightforwardly identifiable. However, increased fatigue correlates with greater performance variability²⁴⁾. Humans are generally not adept at intuitively assessing risks²⁵⁾, particularly as fatigue often degrades gradually, similar to (visual) change blindness²⁶⁾. This gradual decline makes it perhaps difficult for ATCOs to detect subtle changes in their cognitive functioning.

Shiftwork in ATC

An appropriate shift system and well-planned shift design is the basis for aviation safety. Working conditions and schedules significantly affect ATCO fatigue and thus performance. Shift systems refer to a range of work-hour arrangements and schedules⁴⁾ that are typically individualized and might vary between ATC organizations. Shiftwork describes any work outside conventional daytime schedules^{27, 28)}, with work being typically structured into two or three-shift systems⁴⁾, consisting of morning, evening and night shift types²⁹⁾. The impact of shiftwork is not trivial, as it is associated with overall impaired physical health^{30,31,32,33,34)} and mental health^35,36,37), as well as reduced cognitive performance capability^38,39,40,41). Non-standard work schedules, such as shift or night work have one aspect in common: a potential disruption of circadian processes⁴²⁾. This in turn is associated with overall disturbed sleep, increased sleepiness and fatigue^{9, 43,44,45)}. Work against the inner body clock, which happens usually during night time, seems to be associated with fatigue⁴⁶⁾. Night shifts per se are linked to elevated fatigue^{4, 15, 47,48,49,50,51)} and increased error rates, despite usually less traffic and task load⁵²⁾. During night time, especially the WOCL is linked to increased fatigue^{14, 50)}. Moreover, consecutive night shifts are associated with fatigue, which seems to increase with every extra night shift added^53,54,55). The relationship between shift length and fatigue remains inconclusive. While some authors favor 12-h shifts over eight-hour shifts because longer shifts are associated with longer recovery periods and greater job satisfaction⁵⁶⁾, fatigue and task performance appear to be associated with increasing shift duration. Time on task^{22, 57)} and time since last break⁵⁸⁾, as well as overall long shift durations^{53, 59,60,61,62,63)} were found to be associated with fatigue in ATC. Besides those shift-immanent factors, the scheduling or rostering of shifts is crucial as well. Fixed shift schedules dictate a clear and predictable pattern of shifts to work, such as two morning, two evening and two night shifts, with three days off. Individuals who work in a fixed shift system generally have good planning options with clear working hours and rest periods. Most Air Navigation Service Provider with fixed shifts have established a forward rotating shift pattern, some use backward rotation⁶⁴⁾. The length of shifts varies between organizations, typically falling within a range of eight to 12 h⁵⁸⁾. Opposed to fixed shift schedules, irregular shift schedules offer more variability in shift distribution over a time period. The irregular pattern might result in various shift combinations, but the effects of such a system are mixed. In these irregular shift systems, the shift types within a roster period can be distributed almost randomly, provided they comply with regulations. Forward and backward rotations are possible. According to the European Commission Directive 2003/88, average weekly working hours, including overtime, must not exceed 48 h over four months for ATCOs. Individuals are also entitled to 11 consecutive daily rest hours and 24 h of weekly rest within two weeks. While Sallinen and Kecklund⁴⁾ see no differences between fixed- and irregular schedules, recent studies are more critical. Increased recovery demand and fatigue¹⁾ as well as reduced mental health⁶⁵⁾ were found. Other findings indicated higher stress⁶⁶⁾, more endocrine disorders⁶⁷⁾, more sleep disturbances⁶⁸⁾, and obesity⁶⁹⁾ for workers in irregular shift systems. A recent review of atypical shift systems, such as irregular schedules, concluded wide disruptive effects on the circadian system and elevated sleep-wake disturbances⁷⁰⁾. However, irregular shift systems also offer flexibility, which significantly reduces work-family conflicts⁶⁸⁾.

Purpose and research questions

The previously presented evidence outlines the complex interplay between fatigue, sleep and shiftwork in ATC. It is known that aviation is in demand of appropriate fatigue management strategies. In order to develop effective countermeasures, knowledge on fatigue progression between and within shifts is necessary. The interaction between fatigue and shift types in an irregular shift system has not been satisfactorily addressed in a systematic way in ATC. Thus, this work aimed on investigating fatigue progression within shift and between shifts. We posed two questions: First, how does fatigue progress within shift types over time? In this analysis, the primary objective is to quantify the variance of fatigue within shift types over time. Second, how does fatigue vary between shift types depending on time on shift? The aim is to compare shift types and their sleep-break patterns. The analysis is based on objective and subjective data, involving sleep duration measurements, as sleep duration and quality seem to be relevant predictors for fatigue build-up. These two research questions are analyzed in relation to performance-related measures of fatigue (sustained attention) and self-reported fatigue (subjective sleepiness).

Methods

This work presents results of studies conducted at a Scandinavian ACC. We analyzed cumulative data from five field studies carried out between 2020 and 2023, all consistently applying the same methods and study protocol. The objective was to measure operational personnel at three key timepoints per shift over the course of three weeks, using performance-based and self-evaluation measures at each key timepoint. A field study was conducted to achieve high external validity.

Irregular shift system at the ACC

The study aimed to investigate fatigue within different shifts for operational-active professionals (ATCOs and Air Traffic Control Assistants, ATCAs). Operators work in an irregular shift system, resulting in various shift type combinations and varying rest periods over a roster period. The shift system consists of four shift types, early/morning (Early) and late/evening shifts (Late) as well as two mirrored night shifts (NA and NB). Early and Late shifts range from six to eight hours, with varying start and end times. Early shifts begin between 6:15 a.m. and 10:00 a.m. and end between 2:00 p.m. and 4:00 p.m. Late shifts start at 2:30 p.m. and end at 10:00 or 11:00 p.m. In contrast, the nine-hour night shifts have fixed starting times. The NA shift begins operational work at 10:00 p.m. and allows for a sleep phase while remaining available for sudden demands. At around 3 a.m., the shifts switch, with NB beginning operational work while NA has the opportunity to sleep but remains on-call. This results in an uneven distribution of operational and sleep phases: NA works for five hours and has a four-hour sleep opportunity (extendable), while NB has a five-hour sleep opportunity but must work operationally for four hours during the second half of the night. This split night shift arrangement was developed to maintain the overall day-night pattern by allowing for a continuous, albeit brief, sleep phase during the night shift. During the sleep opportunity, participants use separate dormitories equipped with standard-sized single beds and en-suite bathrooms. The rooms feature soundproofed windows and darkened curtains.

Instruments and software

The amount of sleep before the shifts was determined by a sleep diary question, integrated into the study software. Participants were asked about the duration of the major sleep episode in the last 24 h. On-shift sleep duration was measured using a question asked after their sleep opportunity at work.

Subjective sleepiness was measured using the Stanford Sleepiness Scale, SSS⁷¹⁾, a seven-point Likert scale where higher scores indicate higher sleepiness levels. The scale was integrated into the study software. The SSS ranges from 1 (“feeling active, vital, alert, or wide awake”) to 7 (“no longer fighting sleep, sleep onset soon; having dream-like thoughts”). In this work, sleepiness is seen as a related term to the overarching concept of fatigue (subjective fatigue).

Sustained attention was measured with a three-minute version of the Psychomotor Vigilance Test, PVT-B⁷²⁾, fully integrated into the study software. During the PVT-B, participants pressed the space bar with their dominant hand as quickly as possible after a red circle appeared on screen, with intervals between stimuli varying randomly between 2,000 ms and 7,000 ms. The main measure was mean reaction time (RT), with slower RTs indicating reduced attention. RTs below 100 ms were recorded as false starts, and RTs of 500 ms or more were recorded as lapses. The reciprocal RT, multiplied by 1,000 and referred to hereafter as “1/RT” for simplicity, was used for statistical analysis to maximize sensitivity⁷³⁾. In this work, sustained attention is seen as a performance outcome of fatigue and hence understood as a related but indirect measure of fatigue (objective fatigue).

Study software

To facilitate efficient objective and subjective measurements, a study software was developed and applied. The software was programmed in QT (version 6.0.4) and was designed self-explanatory as it had to guide participants through the required measurements.

Pre-questionnaire

Prior to data collection, a mandatory pre-questionnaire gathered socio-demographic and work experience information. This included the Morningness-Eveningness Questionnaire (MEQ)⁷⁴⁾, which assesses chronotype to determine whether individuals are definitive morning, moderate morning, intermediate, moderate evening or definitive evening types.

Study design and procedure

A safety assessment was conducted to identify and mitigate safety hazards and risks. The study approach adhered to the American Psychological Association (APA) principles of research ethics, including signed informed consent and anonymous, voluntary participation. Participants were granted anonymity through individual codes required for software login. Prior to the study, an introductory session familiarized participants with the study procedure and the PVT-B, aiming to minimize potential learning effects⁷⁵⁾. Testing occurred at three points: before the shift (Pre) as a baseline, in the middle of the shift (Mid), and after the shift (Post). Subjects were instructed in advance to complete any task approximately at the specified time. At each timepoint, participants completed the computer-based PVT-B and SSS, along with reporting their sleep duration.

Statistical analyses

Statistical analyses were performed using R, version 2024.04.2 Build 764⁷⁶⁾. A Welch t-test compared sleep duration before and during shifts between both shift types, followed by linear mixed models (LMMs) to account for variability by timepoints and shift types, including participant-by-shift interactions. The lme4 package⁷⁷⁾ was used for LMM analysis, employing restricted maximum likelihood (REML) estimation. In cases of singular fits for maximal models, model reduction was applied until a suitable model was identified. Inference was conducted using the Kenward-Roger approximation for degrees of freedom⁷⁸⁾, which has been shown to maintain nominal Type 1 error rates even for smaller samples⁷⁹⁾. The maximal model included fixed factors for shift type, timepoint, their interaction, and participant-specific random intercepts and coefficients. In lme4 syntax, the maximal model is specified as follows:

Response ~ Shifttype × Timepoint + (Shifttype × Timepoint | Participant)

Post-hoc tests were conducted using the emmeans package⁸⁰⁾, with p-values adjusted using the Holm–Bonferroni method. All graphs were created in R using the ggplot2 package⁸¹⁾

Results

Sample characteristics and measured shifts

The data analyzed in this study stems from five field studies. The cumulated sample comprised N=74 individuals (n=65 ATCOs, n=9 ATCAs) of which were 43 females and 31 males, ranging between 25 and 59 yr (M_age=46.0, SD_age=7.6, M_{work experience}=18.1, SD_{work experience}=8.6). The MEQ indicated n=31 with an intermediate chronotype, n=39 with a moderate morning type and n=4 with a moderate evening type. Across the five field studies, a total of 384 shifts were measured. The majority were Early (184), followed by Late (94). Night shifts were measured less frequently, with NA shifts recorded 53 times and NB shifts 55 times.

Sleep duration

The sleep durations prior to shift start indicated the longest sleep (in minutes) before Late (M=417, SD= 2.7), followed by NA (M=414, SD=69.7) and Early (M=391, SD=59.2). The shortest sleep was noted before NB (M=366, SD=113). Due to unequal variances, a Welch’s ANOVA was conducted. The Welch’s ANOVA indicated significant differences between the shift types, Welch’s F(3, 114.05)=5.263, p=0.002, partial η²=0.122. For the planned sleep episodes during shifts, the sleep duration was M=213 (SD=80.2) for NB at Mid, and M=321 (SD=123) for NA at Post. No outliers were detected.

Subjective sleepiness

The SSS values were compared within and between shift types. Figure 1(a) gives the mean SSS values over time and per shift type. We tested whether SSS was influenced by shift type or timepoint. Results indicated significant main effects of timepoint (F(2, 55.66)=32.411, p<0.001, partial η²=0.534) and shift type (F(3, 23.96)=30.995, p<0.001, partial η²=0.796), and an interaction between both predictors (F(6, 34.01)=11.264, p<0.001, partial η²=0.665). Within and between shift effects were post-hoc investigated by contrasts, the Holm–Bonferroni-adjusted results are given in Tables 1 and 2. No outliers were detected.

Fig. 1.

Mean SSS values (left) and 1/RT * 1,000 (right) as a function of shift type and time on shift.

Colors indicate shift types, with green=Early, orange=Late, purple=NA and blue=NB. Error bars indicate SD.

Table 1. Contrast table for post-hoc SSS comparisons of timepoints within shift types.

Shift type	Contrast	Estimate	df	t	p
Early	Pre-Mid	0.15	44.0	1.37	1.000
	Pre-Post	−0.13	48.1	−0.96	1.000
	Mid-Post	−0.28	37.4	−2.30	0.240
Late	Pre-Mid	−0.21	32.4	−1.49	1.000
	Pre-Post	−0.80	36.7	−4.51	0.003
	Mid-Post	−0.59	33.2	3.45	0.048
NA	Pre-Mid	−1.67	26.1	−7.86	<0.001
	Pre-Post	−1.13	29.5	−4.20	0.008
	Mid-Post	0.55	30.3	1.84	1.000
NB	Pre-Mid	−0.85	26.5	−3.96	0.019
	Pre-Post	−0.69	25.0	−3.08	0.132
	Mid-Post	0.16	23.8	0.81	1.000

Post hoc p-values were Holm–Bonferroni adjusted. NA: night A; NB: night B.

Table 2. Contrast table for post-hoc SSS comparisons of shift types per timepoint.

Timepoint	Contrast	Estimate	df	t	p
Pre	Early-Late	0.66	47.7	4.15	0.005
	Early-NA	0.26	43.4	1.42	1.000
	Early-NB	−0.60	26.3	−3.07	0.132
	Late-NA	−0.4065	45.1	−2.01	0.952
	Late-NB	−1.26	38.5	−5.65	<0.001
	NA-NB	−0.85	40.6	−3.62	0.026
Mid	Early-Late	0.30	41.5	2.06	0.952
	Early-NA	−1.57	40.0	−7.48	<0.001
	Early-NB	−1.60	34.1	−7.72	<0.001
	Late-NA	−1.87	43.6	−8.18	<0.001
	Late-NB	−1.90	44.3	−8.20	<0.001
	NA-NB	−0.03	34.4	−0.12	1.000
Post	Early-Late	−0.01	39.7	-0.06	1.000
	Early-NA	−0.74	33.2	−3.17	0.090
	Early-NB	−1.16	27.7	−5.37	<0.001
	Late-NA	−0.73	42.3	−2.84	0.164
	Late-NB	−1.15	37.7	−4.77	0.001
	NA-NB	−0.42	35.6	−1.52	1.000

Post hoc p-values were Holm-Bonferroni adjusted. NA: night A; NB: night B.

Reaction time

The 1/RT values were compared within and between shift types. Figure 1(b) gives the mean 1/RT values over time and per shift type. We tested whether 1/RT was influenced by shift type or timepoint. Results indicated significant main effects of timepoint (F(2, 54848.49)=262.123, p<0.001, partial η²=0.009) and shift type (F(3, 52366.80)=28.339, p<0.001, partial η²=0.002), and an interaction between both predictors (F(6, 54837.95)=22.544, p<0.001, partial η²=0.002). Within and between shift effects were post-hoc investigated by contrasts, the Holm–Bonferroni-adjusted results are given in Tables 3 and 4.

Table 3. Contrast table for post-hoc 1/RT comparisons of timepoints within shift types.

Shift type	Contrast	Estimate	df	t	p
Early	Pre-Mid	0.05	54846	11.22	<0.001
	Pre-Post	0.05	54855	10.16	<0.001
	Mid-Post	−0.00	54853	−0.70	1.000
Late	Pre-Mid	0.04	54862	5.90	<0.001
	Pre-Post	0.06	54873	7.98	<0.001
	Mid-Post	0.02	54858	2.45	0.199
NA	Pre-Mid	0.13	54836	14.29	<0.001
	Pre-Post	0.06	54841	6.07	<0.001
	Mid-Post	−0.07	54839	−8.02	<0.001
NB	Pre-Mid	0.10	54869	10.62	<0.001
	Pre-Post	0.12	54850	13.17	<0.001
	Mid-Post	0.02	54846	2.81	0.081

Post hoc p-values were Holm–Bonferroni adjusted. NA: night A; NB: night B.

Table 4. Contrast table for post-hoc 1/RT comparisons of shift types per timepoint.

Timepoint	Contrast	Estimate	df	t	p
Pre	Early-Late	−0.02	54899	−3.012	0.044
	Early-NA	0.01	54464	0.147	1.000
	Early-NB	−0.01	54565	−1.033	1.000
	Late-NA	0.02	54562	2.137	0.358
	Late-NB	0.01	54629	1.071	1.000
	NA-NB	−0.01	54663	−0.955	1.000
Mid	Early-Late	−0.03	54888	−5.128	<0.001
	Early-NA	0.08	54478	9.207	<0.001
	Early-NB	0.03	54398	3.918	0.002
	Late-NA	0.11	54571	11.893	<0.001
	Late-NB	0.06	54530	7.167	<0.001
	NA-NB	−0.05	54517	−4.672	<0.001
Post	Early-Late	−0.10	54885	−1.430	1.000
	Early-NA	0.01	54472	0.784	1.000
	Early-NB	0.01	54466	7.112	<0.001
	Late-NA	0.02	54657	1.711	0.784
	Late-NB	0.07	54650	7.358	<0.001
	NA-NB	0.05	54620	5.119	<0.001

Post hoc p-values were Holm-Bonferroni adjusted. NA: night A; NB: night B.

False starts and lapses

The number of false starts and lapses were compared within and between shift types. Figure gives the mean values over time and per shift type. We tested whether false starts and lapses were influenced by shift type or timepoint. Results for false starts and lapses indicated no significant main effects for timepoint and shift type.

Discussion

This paper investigated subjective and objective fatigue in a preference-based irregular shift system in ATC. Sleep duration differences and fatigue dynamics within shifts and between shift types were examined and are discussed in the following.

Sleep duration differences

Several subjective sleep durations were assessed at different times across study shifts. At the beginning of each shift, participants were asked about the duration of their main sleep episode in the previous 24 h. The shortest pre-shift sleep was observed in participants working NB, with an average of 6 h and 6 min. Both night shifts began at 10:00 p.m.; however, the counter-rotating NA shift indicated an average pre-shift sleep of 6 h and 54 min, about an hour longer than NB. This difference might reflect a compensation strategy among NB participants, who may have reduced their pre-shift sleep as a homeostatic strategy to not satiate sleep need ahead of the sleep window from 10 p.m. to 3 a.m.⁵⁰⁾. However, during the sleep window for NB, participants reported an average sleep duration of only 3 h and 33 min, within a theoretical window of 5 h. In contrast, during the sleep window for the NA, an average sleep duration of 5 h and 21 min was found. This suggests that most NA participants stayed beyond their required standby time, prioritizing sleep over minimum time at work. In contrast, no participants arrived Early for the NB shift to sleep beforehand, which may reflect the difficulty of advancing rather than delaying the sleep-wake rhythm, as earlier described⁸²⁾.

Participants working Early had the second-shortest sleep durations, averaging 6 h and 31 min, consistent with findings that sleep is typically shortest before morning shifts^{4, 83, 84)}. Early generally start between 06:00 a.m. and 08:00 a.m., which limits sleep duration due to early wake times. The longest pre-shift sleep was observed before Late, with an average of 6 h and 57 min. As discussed in literature, participants likely extended their sleep to delay subjective fatigue onset^{47, 85)}, especially given that Late can end as late as 10:00 p.m.

Subjective fatigue (subjective sleepiness)

The dynamics of subjective sleepiness were analyzed within and between shift types. Within shift types, no changes in subjective sleepiness were observed for participants on Early, indicating that their subjective sleepiness levels remained stable throughout the shift. However, for Late, subjective sleepiness increased at the end of the shift (Post) compared to the beginning (Pre) and midpoint (Mid). This suggests a circadian influence or an effect of sunset, as the shift ended at 10:00 p.m.¹²⁾. A time-on-task effect does not appear to apply here, as no similar change was found in Early indicating results contrary to previous findings^{22, 57)}. For Late shifts, the increase in sleepiness appeared only in the latter part of the shift, and no consistent linear increase in sleepiness was observed. Comparing both night shifts (NA and NB), those do not seem to vary. Subjective sleepiness was peaking at the midpoint (3 a.m.). This peak likely reflects the circadian drop in alertness around the WOCL^12,13,14), rather than the sleep-break behavior, as NA participants worked during the first half of the shift while NB participants worked in the second half. This supports the conclusion that the circadian system, which is mostly influenced by time of day, seems to be the main influence on subjective sleepiness, rather than actual behavioral factors on shift⁸⁶⁾ or traffic-related task load¹⁵⁾.

Comparisons between shift types at timepoints revealed higher subjective sleepiness before the shift (Pre) for Early and NB shifts compared to Late and NA shifts. This aligns with sleep duration findings, where Early and NB shifts indicated shortest pre-shift sleep. Mid-shift, night shifts (NA and NB) had higher subjective sleepiness than Early and Late, which can be attributed to the WOCL¹³⁾. At the end of the shift (Post), ATCOs on NB reported greater subjective sleepiness than those on Early or Late, indicating that the second part of the night shift is more susceptible to sleepiness compared to the second part of day shifts. No increased subjective sleepiness was observed for NA at Post, suggesting a benefit from the sleep received in the second part of the night shift^{22, 87)}.

Objective fatigue (sustained attention)

The dynamics of sustained attention across and within shift types was also examined. Over the course of each shift, RTs slowed significantly at Mid and Post compared to Pre for all shift types. However, only NA shifts showed faster RTs at Post compared to Mid. In general, this pattern reflects a “level-off” trend, where RTs slow down to Mid but do not continue to worsen toward the end of the shifts—except for NA shifts. NA shifts show a decrease in RT between Mid and Post, likely due to a recovery effect due to the sleep received in the second part of the night. This is in line with findings by Lim and Dinges⁸⁸⁾, indicating the sensitivity of the PVT to sleep deprivation.

Comparing shift types, fewer differences were identified. At Pre, Early showed slower RTs than Late. At Mid, however, more variation was observed: Early were slower than Late, and both NA and NB were slower than Late and Early shifts. Slower RTs at Mid during nighttime compared to daytime indicates a possible circadian influence. This is in line with previous results, indicating overall higher fatigue during night shifts^{15, 52, 58)}. Interestingly, NA’s slower RT performance at Mid compared to NB suggests a possible effect of extended wakefulness⁵⁸⁾, as NB participants slept between Pre and Mid, potentially buffering WOCL impact on RT⁸⁸⁾. At Post, NB shifts showed slower RTs than Early, Late, and NA shifts. This difference reflects heightened fatigue for those working the second part of the night, during the WOCL¹³⁾. The lack of RT differences between day shifts and NA, and the slower RTs for NB compared to NA, suggest that NA participants benefitted from sleep in the second part of the night.

No significant differences were found in lapses or false starts across shift types, which is contrary to expectations, as these measures are repeatably reported to be sensitive to fatigue^{73, 88)} This could be due to the relatively short task duration of the PVT-B test (three minutes), as most research employs a longer, original PVT version lasting up to ten minutes, potentially offering a more sensitive fatigue assessment^{72, 73)}.

Moreover, when comparing subjective and objective fatigue, different patterns can be observed. For instance, fewer differences in subjective fatigue were found over time compared to the progression of objective fatigue. This could indicate that self-evaluation capabilities are insufficient for detecting performance changes over time, aligning with earlier findings, which indicate reduced associations between self-assessed and performance-related fatigue^{18, 89)}.

Another important factor is the variation in traffic-associated taskload across shift types. Subjective evaluations of a given taskload are defined as workload and either work over- and underload should be avoided⁹⁰⁾. Typically, lower traffic during nighttime may reduce workload and also lead to boredom, potentially exacerbating fatigue⁹¹⁾. At the measured ACC, traffic patterns occur in waves (e.g., morning and afternoon rush), likely causing workload variations. Since ATCOs follow a one-hour-on/one-hour-off pattern during day shifts, their workload can vary significantly. Some may experience high traffic and elevated workload, while others may face calmer work sessions. This might have had an influence on subjective fatigue and performance⁹²⁾, however, as air traffic volumes were not systematically measured in this study, their influence is unclear.

Methodological limitations

This study’s design limitations meant exact sleep timing information was unavailable, a significant drawback given that specific timing would reveal more about participants’ sleep strategies. Furthermore, subjective sleep estimates are not always reliable^{93, 94)}, so using objective sleep measurement methods would have considerably enhanced data validity⁹⁵⁾. Operational effects due to varying traffic load, secondary tasks or the intake of psycho-active substances was not monitored. It is known, that fatigue is multifactorial caused¹⁶⁾ and only a fraction of potential confounding variables was monitored. Additionally, participation was voluntary, which may have introduced selection bias—a limitation that is difficult to entirely avoid⁹⁶⁾. It is possible that individuals with high fatigue awareness or those with effective fatigue management strategies were more motivated to participate, while those experiencing significant fatigue may have chosen not to participate due to concerns about worsening their fatigue. Moreover, since the software was utilized outside the operational environment, the walking time between the workplace and the testing devices may have influenced participants by temporarily increasing their activation levels. The SSS had perhaps drawbacks as well. Any self-evaluation can be biased, such as due to the optimism bias⁹⁷⁾.

Implications and conclusion

In this study, fatigue dynamics in ATC were investigated using a multivariate approach. Five field studies assessed operational professionals’ fatigue in operational shifts across four different shift types. Data collection spanned a three-week study period each, capturing self-evaluated sleep duration, subjective sleepiness and sustained attention at three key timepoints per shift.

The results of subjective sleepiness (SSS) and sustained attention (PVT-B) reveal distinct but interrelated patterns across shift types. For subjective sleepiness, a “peak pattern” was observed, particularly in night shifts (NA and NB), where sleepiness peaked at Mid (around 3 a.m.). This is likely due to circadian influences, specifically the WOCL, rather than operational or behavioral factors, as both NA and NB followed different on-shift routines but shared similar sleepiness peaks. Day shifts showed smaller changes, with Early maintaining steady sleepiness and Late increasing slightly by the shift end. Regarding sustained attention (PVT-B), RTs were generally slower at Mid and Post compared to Pre across all shifts, indicating a broad “level-off” effect. However, for NA, which included a sleep phase in the second part of the night, RTs improved by the end of the shift, unlike NB, where operators slept in the first part of the night. Night shifts were especially impacted by time-of-day effects, showing slower RTs than day shifts, mostly pronounced at Mid, presumably due to the WOCL. The recovery in RT for NA after a sleep period supports the potential restorative effect of sleep during shifts. In summary, while both SSS and PVT-B reveal increased fatigue at night, the sleep phase for NA may help mitigate performance deterioration in sustained attention, underscoring the benefits of scheduled sleep breaks during WOCL hours.

The results show the heterogeneity of fatigue, which is given due to its sensitivity to various factors. The multifaceted nature of fatigue has been recognized for some time, and these findings emphasize the significant influence of time of day and shift type. Additionally, as there is quite some difference between subjective sleepiness and performance-related fatigue (sustained attention). This raises the question of how working under fatigue affects the subjective perception of sleepiness⁹⁸⁾. This is particularly concerning, as many safety mechanisms depend on humans detecting and responding to potential risks. Relying solely on subjective fatigue evaluations appears insufficient and should always be complemented by objective measures, as is the case for data analyzed in this study. This is especially critical for designing shift systems or if allowing individuals to select their own shifts, as is also the case in this investigated Scandinavian ANSP. However, this approach could result in suboptimal shift selections, as operators choose rotations with increased time off-work, increasing fatigue-related risks⁹⁹⁾.

Shift systems across industries are highly heterogeneous, complicating the generalization of the findings to other systems. Additionally, ATCOs operate under specific conditions, meaning the transfer of evidence to other domains requires consideration of differing task characteristics. This highlights the need for further research, in particular to identify the primary factors influencing shift selection and to develop effective strategies for mitigating fatigue risks. Future studies can build on our findings regarding the influence of shift characteristics to enhance safety in ATC and other industries.

Funding

This research was funded by the Swedish Transport Administration (Trafikverket).

Conflict of Interest

The authors declare no conflicts of interest with respect to the research and/or publication of this article.