Layover start timing predicts layover sleep quantity and timing on long-range and ultra-long-range trips

Abstract

Study Objectives

Airline transport pilot sleep during layover is an important factor for alertness on subsequent flights. Assessing pilots’ sleep on layover is an important first step in helping them obtain the most recuperative sleep possible on layover. Here, we investigate the quantity and timing of sleep during layovers and determine predictors for layover sleep.

Methods

Sleep was assessed in 256 pilots flying a total of 473 long-range (LR; flight time 12–16 hours) or ultra-long-range (ULR; flight time > 16 hours) trips. Sleep was assessed using actigraphy. We employed linear mixed-effects models with layover sleep characteristics as the outcomes. The predictor variables included operational factors and sleep history.

Results

Overall, pilots averaged 7.2 hours of sleep per 24 hours of layover, which was significantly less than their daily sleep before or after the trip. Layover start time (relative to home base time) was the most salient predictor of sleep timing and quantity during both shorter (∼24-hour) and the first 24 hours of longer (∼48-hour) layovers. During the last 24 hours of longer layovers, crew type predicted sleep quantity.

Conclusions

Although average sleep quantity during layovers was within the margins of recommended sleep duration, it was still less than pre- and post-trip sleep duration, suggesting modest sleep loss on layovers. Layover start timing was the strongest predictor of layover sleep quantity and timing and, thus, may be a modifiable factor to protect circadian-aligned sleep opportunities during layover.

Statement of Significance.

Airline transport long-haul pilots may have difficulty obtaining recuperative sleep during layovers due to multiple factors such as time zone changes and limited sleep opportunities. Assessing pilots’ sleep on layovers is an important first step toward ensuring that pilots are obtaining recuperative sleep on layovers to support the maintenance of alertness on subsequent flights. Previous studies on layover sleep may not reflect broad trends since typically only one or two routes are studied and these are from a single carrier. Here, using data from 473 trips, across 9 routes from 2 different carriers, we found that the timing of the layover start relative to home base time most reliably predicts layover sleep quantity and timing.

Introduction

Airline transport pilots navigate a demanding operational landscape in which it is often difficult to obtain adequate sleep to ensure they are well-rested and alert. Prescriptive flight, duty, and rest time regulations are intended to minimize fatigue and provide adequate recovery opportunities [1]. However, many factors can contribute to poor sleep during designated rest periods such as the timing of rest opportunities, the sleep environment, and jet lag, which can ultimately lead to degraded alertness and cognitive performance [2, 3].

Sleep research in aviation has largely focused on optimizing sleep opportunities to maintain or increase cognitive performance [4, 5]. Due to unconventional work schedules [6] and frequent travel across time zones [7] in long-haul operations, it can be challenging for pilots to obtain sufficient quantity and quality of sleep while traveling. Pilots flying routes that are long-range (LR; 12–16 flight hours), or ultra-long-range (ULR; > 16 flight hours), routinely suffer from fatigue due to sleep disruption [8–10], extended duty periods, and circadian misalignment [6]. The irregular schedules and physiological challenges of frequent time zone changes inherent in airline transport piloting underscore the need for a comprehensive understanding of sleep during layovers. This would inform our understanding of fatigue on the return or subsequent flights and can then be used to implement preventative measures and countermeasures in real-time operations.

The timing, quantity, and quality of sleep obtained during a layover is subject to various factors. The duration of layovers, the time of day they occur, relative to both local and home base time (HBT), and the amenities available at layover destinations all have the potential to influence the sleep opportunities available to pilots [11]. Perhaps unsurprisingly, pilots with layovers shorter than 40 hours obtain less total sleep during layover due to fewer sleep periods, relative to pilots with longer (∼62-hour) layovers [12]. The timing of sleep during a layover, relative to habitual sleep timing at home, may further exacerbate sleep loss, as sleep opportunities during the biological day have been shown to yield truncated sleep and higher subsequent sleepiness ratings compared to sleep consolidated during the night [13–15]. Several previous studies have reported that, on longer layovers (∼48 hours or longer), pilot sleep during layover tends to have a main sleep period in line with the local (destination) night, or shifting toward the local night [9, 16–20]. The number of hours of a layover period occurring at local or biological night has also been shown to impact variations in layover sleep [21]. A primary factor influencing the recovery value of a layover is the quantity and quality of sleep pilots are able to obtain during the layover [21], which may vary depending on the sleep strategies different pilots employ [9]. Therefore, understanding the nuances of these factors is pivotal for devising strategies to optimize recuperative sleep during layovers. Here, we explore how the quantity and timing of layover sleep depend on several predictor variables using data from two different carriers and nine different international routes. Our goal was to find broad trends that are not specific to a particular route or carrier.

Methods

Participants and data collection

Two airlines were included in this analysis. The study for Airline 1 was approved by the Washington State University Institutional Review Board (STUDY 14214). Recruitment was conducted via email to potential participants. Prior to participating, interested pilots contacted the Occupational Sleep Medicine Group at Washington State University for study information and signed a consent form. The study for Airline 2 was approved by the NASA Institutional Review Board (STUDY00000416). Recruitment was conducted via email to potential participants. Interested volunteers provided written informed consent to one of the study researchers. All individual data remained confidential and de-identified for both airlines.

Data collection occurred from 3 days before each trip, during outbound (OB) flights, layover, and inbound (IB) flights, and through 3 days after each trip. Prior to the study, participants were sent a sleep/work logbook and an actigraphy device (Philips Respironics, Bend, OR, USA; Models: Actiwatch Spectrum and Spectrum Plus), which monitors human rest/activity cycles to quantify sleep [22], followed by a training call on how to use the study materials. All actigraphy data were collected using a 1-minute epoch length. Using Philips Actiware 5 or 6 software, wake threshold was set to medium (40 activity counts per epoch). The sleep interval detection algorithm was used for sleep onset (10 immobile minutes) and sleep end (10 immobile minutes). Sleep start and wake times, self-reported in the logbook and confirmed by event markers pressed by participants on the actigraph device, were used to verify that the algorithm correctly captured all rest periods. Incorrect auto-detected rest periods were manually adjusted according to internal standard operating procedures. Sleep quantity is the variable “Sleep Time (minutes)” in the actigraphy output, which is equivalent to “%Sleep” multiplied by (SleepEnd-SleepStart). Sleep timing is defined as the percentage of sleep during the layover that occurred during the pilots’ HBT night (defined as 2300-0700). For sleep efficiency, we used the Efficiency variable in the actigraphy output, which is calculated as sleep quantity divided by the full rest period, multiplied by 100. Wake after sleep onset (WASO) was calculated using the variable “Wake Time (minutes)” for each sleep period. Note this only included wakefulness during the sleep period and not the full rest period, so it does not include “Snooze” time or sleep onset latency. Sleep onset latency (SOL) was calculated as the difference between the start of the sleep period and the start of the rest period (i.e., when it was indicated that the participant attempted to sleep). The data were imported, processed, visualized, and analyzed using the statistical programming language R version 4.3.2 (https://www.R-project.org).

Analysis

The dataset began with 691 trips. In total, 218 trips were removed as shown in the Consort diagram (Figure 1), leaving 473 trips in the final dataset. A trip was only included in the analysis if it had at least two consecutive pre-duty days with at least 3 hours of recorded sleep per day. This filtering step removed 18 trips that were cases of device non-wear for the entire pre-trip period and another 74 trips where only some pre-trip data were missing. Both the 18 trips with no pre-trip data and the 74 trips with partially missing data were part of the filtering step “technical issues with actigraphy causing missing data” shown in Figure 1. The remaining 12 trips in this category were removed due to ambiguity between hand-written logbook data and actigraphy data such that an accurate record of sleep could not be determined. The final dataset for all analyses in this study consisted of 473 trips. Each of the trips that remained in the analysis sample contained complete data for the pre-trip timeframe, the outbound segment of the trip, the layover, and the inbound segment.

Figure 1.

Consort diagram for excluded data. For the second to the last exclusion step, “shorter” and “longer” timeframes are described in the “Analysis” portion of the “Methods” section. Also see Supplementary Figure S1. Before carrying out the final exclusion step, we had two additional HBT bins, each containing fewer than 10 trips (7 in one, 5 in the other), while the other bins contained 50–142 data points. To avoid making comparisons between groups with such vastly different sample sizes, we removed trips that fell into these circadian bins. See Figures 3–5 for the remaining sample sizes. HBT, home base time.

We defined layover length as the scheduled inbound report time minus actual outbound landing time. We used this method, rather than the actual layover duration (actual inbound report time—actual outbound landing time) because the scheduled timing of the inbound report time likely plays a role in how pilots sleep during the layover. We expect pilots to sleep during layover in such a way as to prepare for the expected departure time of the inbound flight rather than the actual departure time. Last-minute delays to the inbound flight may mean that it takes off at a time that the pilots had not prepared for during the layover.

For statistical analyses, the data were grouped into trips consisting of “shorter” layovers (layover duration between 22 and 27 hours) and “longer” layovers (layover duration between 42 and 51 hours). These categories were determined by visual inspection of a histogram of layover duration that showed two distinct groups (a histogram is shown in Supplementary Figure S1).

For both shorter and longer layovers, the first 24 hours refers to the time frame beginning at the start of the layover and going forward 24 hours. The last 24 hours refers to counting back 24 hours from the end of the layover. For layovers equal to or greater than 48 hours, these are distinct time frames; for layovers shorter than 48 hours, these time frames overlap. For instance, for a 25-hour layover, the first 24 hours would be hours 1–24 and the last 24 hours would be hours 2–25. The distinction of the first 24 hours and the last 24 hours of shorter layovers was used to compare summary statistics to prior studies [23].

For the analysis of daily sleep quantity across pre-trip, layover, and post-trip days, we used one-way ANOVAs followed by Tukey HSD post hoc tests, if the omnibus ANOVA was significant. All p values were adjusted for multiple comparisons using the Benjamini–Hochberg method [24].

Initial predictor models

We were interested in determining which factors drive the timing and quantity of pilot sleep during layovers. We used a variable selection approach using the “kitchen sink” model. Each linear mixed-effects model contained the following five predictor variables: (1) the quantity of in-flight sleep obtained on the OB flight, (2) the timing of the in-flight sleep on the OB flight (calculated as the duration, in minutes, from the end of the last in-flight sleep episode to the start of the layover), (3) the timing of the layover start in HBT, (4) crew type (command crew, including pilot flying and pilot monitoring vs. relief crew) on the IB flight, (5) the direction of travel of the outbound flight (eastbound or westbound). Descriptive statistics about the five predictor variables can be found inTable 1. Subject was included as a random factor with a random intercept and fixed slope. The timing of the layover start was grouped into the following six HBT bins: >0200 and ≤ 0600, >0600 and ≤ 1000, >1000 and ≤ 1400, >1400 and ≤ 1800, >1800 and ≤ 2200, >2200 and ≤ 0200. This predictor variable was included as a factor variable referenced against 0200-0600. This bin is common in sleep and aviation research as it contains the circadian nadir of most people [8, 25, 26].

Table 1.

Descriptive statistics for the five predictor variables

Layover length	Quantity of Inflight sleep on OB flight (min)	Timing of sleep on OB flight (min)	Timing of layover start in HBT	Crew type	Direction of travel of OB flight
Shorter	243.6 ± 70.1	308.6 ± 198.8	0200–0600: 20% 1000–1400: 28% 1400–1800: 52%	Command: 43% Relief: 57%	Eastbound: 4% Westbound: 96%
Longer	260.4 ± 81.7	233.1 ± 159.9	0200–0600: 26% 1400–1800: 74%	Command: 49% Relief: 51%	Eastbound: 0%* Westbound: 100%

Sleep values are mean ± standard deviation. Timing of sleep on OB flight: duration, in minutes, from the end of the last in-flight sleep episode to the start of the layover. Note that this distribution had two peaks, one near 2 hours and one near 7 hours.

^*Trips with longer layovers were all in the Westbound direction, so this was not included as a predictor in models for longer layovers.

OB, outbound; HBT, home base time.

The quantity of in-flight sleep on the OB flight was included as a predictor because we expected that pilots who obtained more sleep during the OB flight may not need as much sleep during layover or sleep as early in the layover. We included the timing of the in-flight sleep on the OB portion because, particularly for shorter layovers, we expected that sleep later in the flight may lead to a delay in the initiation or reduction in the quantity of sleep during layover. We included the timing of the layover start as a factor because we expected pilots’ body clocks to be a large driver in timing and quantity of sleep. Crew type was included because crew types typically have different opportunities for sleep in-flight and we suspected that they may prepare differently during layover for the IB flight. The direction of travel of the OB flight was included because we expected different outcomes based on whether the local destination time was advanced or delayed relative to HBT, although we recognize the shift can be reversed or not detected when eight or more time zones are crossed [27–30].

Five different layover sleep outcomes were used for the mixed-effects models: (1) sleep quantity, (2) sleep timing, (3) sleep efficiency, (4) wake after sleep onset (WASO), and (5) sleep onset latency. To account for the variability in layover duration when quantifying sleep during layover, we standardized total sleep to the number of 24-hour periods in the layover. For instance, for a 25-hour layover, the total sleep quantity per 24 hours of layover was calculated as follows: Total Layover Sleep × (24/25). To quantify the timing of sleep during layover we calculated the percentage of layover sleep that occurred during the pilot’s HBT night (defined as 2300-0700). While both shorter and longer layovers were segmented into first and last 24-hour blocks for daily sleep quantity, only the first 24 h of shorter layovers were used in the statistical models. In total, we ran 15 linear mixed-effects models, since there are 3 types of layovers (shorter, first 24 of longer, last 24 of longer) and 5 types of outcomes (sleep quantity, sleep timing, sleep efficiency, WASO, and sleep onset latency).

Each of these models used the same five predictor variables described above, except that the direction of travel was removed from longer layover models since these trips all had the same OB flight direction.

For the three models with layover sleep quantity as the outcome variable, the outcome variable was normally distributed and the residuals were normally distributed, so we used the command lmer from the lme4 R package. For the models that used sleep timing, sleep efficiency, sleep latency, or WASO as outcome, the outcome variable was not normally distributed, so we used a generalized linear mixed-effect model: a binomial model for sleep timing, and a gamma model for sleep efficiency, sleep latency, and WASO. The best distributions were found using the fitdistrplus package in R. The models were run using glmer from the lme4 R package. For each mixed-effects model, we calculated the Variance Inflation Factor (VIF) to check for collinearity. VIF was below five for all models, indicating little collinearity between predictor variables [31]. See Supplementary Tables S1–S15 for the estimates, confidence intervals, and p values for each of these models.

Final models

To quantify the relationship between each outcome variable and its predictor variable, non-significant predictors were removed from the models described above. The predictors and outcomes of the final six models (those that contained at least one significant predictor variable) are shown in Table 2.

Table 2.

Final statistical models

Model	Predictor	Outcome
1	Layover start timing	Sleep quantity during shorter layovers
2	Layover start timing	WASO during shorter layovers
3	Layover start timing	Sleep quantity during the first 24 h of longer layovers
4	Layover start timing	Sleep timing during the first 24 h of longer layovers
5	Crew type	Sleep quantity during the last 24 h of longer layovers
6	Layover start timing	Sleep timing during the last 24 h of longer layovers

WASO, wake after sleep onset.

Results

Data were collected from 256 airline transport pilots flying 473 total trips across 8 westbound routes and 1 eastbound route ranging from 11 to 17 hours in scheduled flight duration. All routes originated in and returned to the United States. Scheduled layover duration ranged from 24 to 50 hours. Data were collected between July 2014 and April 2023 across various seasons. Participant demographics are shown in Table 3.

Table 3.

Demographics

		Value	N total	N prov
Airline 1	Age (years)	54.8 ± 5.9	220	144
	Gender	92% male	220	136
	BMI (kg/m2)	26.2 ± 3.3	220	143
Airline 2	Age (years)	57.3 ± 4.8	36	19
	Gender	100% male	36	19
	BMI (kg/m2)	-	-	-
All	Age (years)	55.1 ± 5.8	256	163
	Gender	93% male	256	155
	BMI (kg/m2)	-	-	-

Values are mean ± standard deviation. Airline 2 did not collect BMI information. Given the sensitive nature of the study, demographics questions were optional and missing demographic data does not reflect on the quality nor completeness of the sleep data. This table represents all of the participants who remained in the study after trips were excluded as shown in Figure 1.

BMI, body mass index; N prov, the number of participants who provided that demographic variable; N total, the total number of pilot participants included in the study.

Sleep across the trip

Table 4 shows the layover sleep quantity and schedule information for each route. In the “route” column, USPT refers to a city in Pacific Time, USA, USET refers to a city in Eastern Time, USA, and USCT refers to a city in Central Time, USA. Overall, layover sleep per 24 hours ranged from 6 to 8 hours and flight duty periods ranged from 13 to 16 hours. Before the trip, across all pilots and routes, 83% of sleep occurred during the HBT night, defined as 2300-0700.

Table 4.

Summary of flight time and layover sleep across routes

Route	Number of trips	Layover sleep (per 24 h) (h)	Layover sleep during the first 24 h (h)	Layover sleep during the last 24 h   (h)	Layover duration (h)	Scheduled OB FDP (h)	Scheduled IB FDP (h)	OB scheduled report time range (hhmm)	IB scheduled report time range* (hhmm)	Time difference (h)**
USPT-MEL	79	7.5 ± 1.9	7.6 ± 1.8	7.6 ± 1.8	24.4 ± 1.0	15.92	14.17	2005–2135	1500–1545	−5
USPT-PVG	55	6.5 ± 1.7	6.7 ± 1.8	6.6 ± 1.8	24.7 ± 1.0	13.83	11.67	1100–1225	0340	−8
USET-JNB	12	8.0 ± 2.6	8.3 ± 2.6	8.3 ± 2.6	24.9 ± 0.8	13.83	16.00	1915–2040	1130–1215	+7
USPT-SYD	34	7.6 ± 1.6	8.0 ± 1.7	7.9 ± 1.7	25.1 ± 0.6	15.25	13.58	2105–2230	1500	−5
USPT-SYD	101	7.3 ± 1.7	7.9 ± 1.9	7.6 ± 1.8	26.2 ± 0.6	15.17	13.58	2105–2225	1600	−5
USPT-PVG	8	6.6 ± 1.4	5.6 ± 1.8	6.0 ± 1.7	38.7 ± 0.7	13.17	11.17	0910–1315	0800	−8
USCT-HKG	42	7.1 ± 1.6	6.8 ± 1.6	7.1 ± 1.9	45.2 ± 1.1	17.30	17.35	1035–1250	1310–1545	−10
USPT-SIN	92	7.2 ± 1.5	7.9 ± 1.9	6.7 ± 2.0	48.8 ± 0.7	17.00	14.75	2050–2305	1630–1745	−8
USCT-SYD	50	7.3 ± 1.5	7.9 ± 1.8	7.0 ± 1.7	50.1 ± 1.5	17.67	15.58	1830–2115	1615–1720	−7
Average	-	7.2 ± 2.0	7.6 ± 2.0	7.2 ± 2.0	34.4 ± 11	15.89	14.25	-	-	-

All times are in home base time (HBT). Values are mean ± standard deviation. 

FDP, flight duty period; h, hours; HKG, Hong Kong, Hong Kong; IB, inbound flight; JNB, Johannesburg, South Africa; MEL, Melbourne, Australia; OB, Outbound flight; PVG, Shanghai, China; SIN, Singapore, Singapore; USET, a city in Eastern Time, USA; USPT, a city in Pacific Time, USA; SYD, Sydney, Australia; USCT: a city in Central Time, USA.

^*Inbound scheduled report time was not avilable for all flights.

^**Since the study was conducted over multiple seasons, the time zone differences between local and destination time varied by up to 2 hours based on relative daylight saving time.

Figure 2, A is a visualization of the sleep patterns of all pilot participants on one route (USPT-MEL) including three nights pre-trip, the OB flight, the layover, the IB flight, and three nights post-trip. During the layover, 45% of pilots on this route took a nap early in the layover and 86% of pilots had a long sleep episode that overlapped both the HBT night and the local night, as indicated by the gray and purple background shading, respectively. Generally speaking, after the IB flight, beginning with the first HBT night at home, pilots on this route returned to their pre-trip sleep schedule. Figure 2, B shows the proportion of pilots sleeping at each 30-minute bin of the layover, indicating that the overlap of the HBT and local nights was the time with the highest percentage of pilots sleeping. The colors in each bar represent the duration of each sleep episode. Sleep episodes near the beginning of the layover tended to be shorter as indicated by the darker color. See Supplementary Figures S2–S4 for equivalent visualizations of other routes.  

Figure 2.

Visualization of sleep and flight data for all participants flying an example route (USPT-MEL). Time is shown on the horizontal axis and includes three nights pre-trip, the outbound flight, the layover, the inbound flight, and three nights post-trip. The rows in panel A represent individual trips. Trips are grouped by crew type (command, relief). The horizontal axis has units of hours relative to midnight home base time (HBT) before the outbound flight. Blue bars (darkest bars) represent sleep episodes on the ground. Gold bars (lightest bars, only appearing in columns 4 and 6) represent in-flight sleep. Gray bars represent flights (darker shade in columns 4 and 6). Gray background shading (lighter shaded vertical columns) represents the HBT night (2300-0700) and the purple background shading (darker shaded vertical columns) represents local night in the layover destination. Panel B is a visualization of the proportion of pilots sleeping in each 30-min bin of the layover. The colors of the vertical bars represent the duration of the sleep episodes, with darker shades representing shorter sleep duration. See Supplementary Figures S2–S5 for other layover examples,includnig longer layovers.

Across all routes, pilots slept an average of 7.2 hours (± 1.7 hours standard deviation, SD) per 24 hours of layover. Table 5 shows the average sleep quantity per 24 hours (“daily sleep”) pre-trip, during layover, and post-trip. With all routes combined, pilots had less daily sleep during the layover than pre-trip or post-trip, and less sleep pre-trip compared to post-trip (repeated measures ANOVA F(2,510) = 15.12, p < .001; partial η² = 0.06; post hoc tests: Pre vs Layover: p = 3e⁻³, Cohen’s d = 0.14; Post vs Layover: p < .001, Cohen’s d = 0.25; Post vs Pre: p < .001, Cohen’s d = 0.18; all post hoc p values adjusted using Benjamini–Hochberg). For shorter layovers, there was no significant difference in daily sleep quantity across the trip days (repeated measures ANOVA F(3,486) = 2.56, p = 0.054). For longer layovers, pilots slept less during the last 24 hours of the layover than they did pre-trip, during the first 24 hours of the layover, or post-trip (repeated measures ANOVA F(3,381) = 14.43, p < .001; partial η² = 0.10; post hoc tests: Last24 vs Pre: p < .001, Cohen’s d = 0.41; Last24 vs First24: p < .001, Cohen’s d = 0.32; Last24 vs Post: p < .001, Cohen’s d = 0.43; all post hoc p-values adjusted using Benjamini–Hochberg). There was no significant difference in sleep quantity between the first 24 hours of short layovers and the first 24 hours of long layovers (one-way ANOVA F(1,471) = 0.024, p = 0.88). Compared to the last 24 hours of shorter layovers, pilots slept less during the last 24 hours of a longer layover (One-way ANOVA F(1,471) = 13.66, p < .001, ω² = 0.03). Finally, when standardized by the length of the layover, there was not a significant difference in sleep quantity between shorter layovers and longer layovers (one-way ANOVA F(1,471) = 0.11, p = 0.74. We did not compare sleep quantities between the first 24 hours and last 24 hours of shorter layovers since there was too much overlap in these time frames.

Table 5.

Daily sleep quantity across trip days

Layover length	Pre-trip	Layover		Post-trip
		First 24 hours	Last 24 hours
Shorter	7.37 ± 1.14	7.59 ± 1.93	7.47* ± 1.88	7.65 ± 1.17
Longer	7.64† ± 1.00	7.57‡ ± 1.93	6.81*†‡§ ± 1.91	7.77§ ± 1.26
All	7.48¶ ± 1.09	7.22‖  ¶± 1.71		7.70‖  ¶± 1.21

Values listed are hours of sleep per 24 hours (“daily sleep”) (mean ± standard deviation).

^*Denotes a significant difference in sleep quantity between the last 24 hours of shorter layovers and the last 24 hours of longer layovers.

^†Denotes a significant difference between pre-trip and the last 24 hours of longer layovers.

^‡Denotes a significant difference between the first 24 hours and the last 24 hours of longer layovers.

^§Denotes a significant difference between the last 24 hours and post-trip of longer layovers.  

^‖Denotes a significant difference between pre-trip days and layover days.

^¶Denotes a significant difference between layover days and post-trip days. See text for statistical values.

Sleep on shorter layovers

There was a moderate effect of the layover start timing on sleep quantity (Model 1; marginal R² = 0.08, conditional R² = 0.52) and WASO (Model 2; marginal R² = 0.03, conditional R² = 0.37) during layover (see Table 6). Specifically, compared to layovers starting at 0200-0600 HBT, layovers that started at 1000-1400 HBT (β = 1.506, 95% confidence interval (CI): 0.941 to 2.071, p < 0.001) and 1400-1800 HBT (β = 1.221, 95% CI: 0.743 to 1.700, p < .001) were associated with significantly more sleep (87 minutes more and 61 minutes more, respectively, which are increases of 22% and 15%, respectively; see Figure 3). For WASO, compared to layovers starting at 0200-0600 HBT, layovers that started at 1000-1400 HBT had on average 12 more minutes of WASO (β = 0.374, 95% CI: 0.030 to 0.718, p = 0.033) (see Figure 3, B). Layover start timing was not a significant predictor for the remaining outcome variables (sleep timing, sleep efficiency, sleep onset latency). Sleep quantity and timing on the outbound flight, crew type, and direction of travel were not significant predictors for any outcome variables.

Table 6.

Final model predictors for shorter layovers

Outcome	Predictor	Estimate (β)	95% CI	P
Sleep Quantity	Layover Start Timing (1000–1400 HBT)	1.506	0.941 - 2.071	<0.001
Sleep Quantity	Layover Start Timing (1400–1800 HBT)	1.221	0.743 - 1.700	<0.001
WASO	Layover Start Timing (1000–1400 HBT)	0.374	0.030 - 0.718	0.033

Significant effects from all linear mixed effects models for shorter layovers. See text for descriptions of effect sizes. CI, confidence interval; HBT, home base time; WASO, wake after sleep onset.

Figure 3.

Predictors of outcomes on shorter layovers. For shorter layovers, layover start timing (relative to home base time, HBT) predicted sleep quantity (panel A) and wake after sleep onset (WASO) (panel B) during the layover. The line inside the box represents the median. The lower and upper edges on each box correspond to the first and third quartiles, respectively. The upper whisker extends from the upper edge to the largest value no further than 1.5 × IQR from the edge of the box (where IQR is the inter-quartile range, the distance between the first and third quartiles). The lower whisker extends from the lower edge to the smallest value at most 1.5 × IQR of the lower edge. Individual data points displayed as dots. *p < 0.05, **p < 0.01, and ***p < 0.001.

Sleep on longer layovers

For longer layovers, there was an effect of the layover start timing on sleep quantity during the first 24 hours of layover (Model 3; marginal R² = 0.11, conditional R² = 0.59; Table 7 and Figure 4A). On average, pilots obtained an additional 80 minutes (an increase of 20%) of sleep during the first 24 hours of the layover when the layover began in the afternoon HBT (1400-1800) compared to layovers that began in the early morning HBT (0200-0600) (β = 1.443, 95% CI: 0.833 to 2.053, p < 0.001). There was a significant effect of layover start timing on the timing of sleep during the first 24 hours of longer layovers (Model 4; marginal R² = 0.25, conditional R² = 0.82; Figure 4, B). Longer layovers that began between 1400 and 1800 HBT were associated with an average of 5% more sleep during the HBT night during the first 24 hours of layover compared to layovers that began between 0200-0600 HBT (β = 4.878, 95% CI: 1.302 to 8.455, p = 0.008). Layover start timing was not a significant predictor of any other outcomes.

Table 7.

Final model predictors for longer layovers

	Outcome	Predictor	Estimate (β)	95% CI	P
First 24 hours	Sleep Quantity	Layover Start Timing (1400–1800 HBT)	1.443	0.833 – 2.053	<0.001
	Sleep Timing	Layover Start Timing (1400–1800 HBT)	4.878	1.302 – 8.455	0.008
Last 24 hours	Sleep Quantity	Crew Type	0.722	0.178 – 1.267	0.010
	Sleep Timing	Layover Start Timing (1400–1800 HBT)	2.039	0.830 – 3.248	0.001

Significant effects from all linear mixed effects models for longer layovers. See text for explanations of effect sizes. CI, confidence interval; HBT, home base time.

Figure 4.

Predictors of outcomes for the first 24 hours of longer layovers. During the first 24 hours of longer layovers, layover start timing (relative to home base time, HBT) predicted sleep quantity (panel A) and sleep timing (panel B). The line inside the box represents the median. The lower and upper edges of each box correspond to the first and third quartiles. The upper whisker extends from the upper edge to the largest value no further than 1.5 × IQR from the edge of the box (where IQR is the inter-quartile range, the distance between the first and third quartiles). The lower whisker extends from the lower edge to the smallest value at most 1.5 × IQR of the lower edge. Individual data points displayed as dots. *p < 0.05, **p < 0.01, ***p < 0.001.

We observed that relief crew pilots obtained, on average, 55 minutes (13%) less sleep in the last 24 hours of the layover than command crew pilots (Model 5; marginal R² = 0.04, conditional R² = 0.33; Figure 5, A, β = 0.722, 95% CI: 0.178 to 1.267, p = 0.010). Crew type was not a significant predictor of any other outcomes.

Figure 5.

Predictors of outcomes for the last 24 hours of longer layovers. During the last 24 hours of longer layovers, relief crew pilots slept less than command crew pilots (panel A). Pilots with layovers that began between 1400 and 1800 home base time (HBT) obtained more sleep during their HBT night in the last 24 hours of layover compared to pilots whose layovers began between 0200 and 0600 (panel B). The line inside the box represents the median. The lower and upper edges of each box correspond to the first and third quartiles. The upper whisker extends from the upper edge to the largest value no further than 1.5 × IQR from the edge of the box (where IQR is the inter-quartile range, the distance between the first and third quartiles). The lower whisker extends from the lower edge to the smallest value at most 1.5 × IQR of the lower edge. Individual data points displayed as dots. *p < 0.05, **p < 0.01, ***p < 0.001.

There was a significant effect of the layover start timing on the timing of sleep during the last 24 hours of longer layovers (Model 6; marginal R² = 0.13, conditional R² = 0.45; Figure 5, B). On average, when the layover began between 1400 and 1800 HBT, pilots obtained 33% of their sleep during HBT night as compared to 24% for layovers that began between 0200 and 0600 HBT (β = 2.039, 95% CI: 0.830 to 3.248, p = 0.001). On average, pilots slept 6.8 hours in the last 24 hours of longer layovers (see Table 5); therefore, this additional 9% of sleep translates to 37 additional minutes of sleep during the HBT night.

One of the main purposes of this study was to find commonalities across routes and airline carries, rather than peculiarities of specific routes. However, as shown in Table 4, sleep quantities during layover vary by route and, in particular, PVG appears to have lower amounts of layover sleep compared to the other routes. To determine if layover location was a significant factor in layover sleep quantities, we performed a secondary analysis comparing layover sleep quantity across routes (separated by shorter and longer layovers). For routes with shorter layovers, a 1-way ANOVA showed a significant effect of layover location (F(4) = 4.08, p = 0.003). Tukey post hoc pairwise comparisons showed that pilots flying USPT-PVG had less layover sleep per 24 h compared to each of the other routes in the shorter category except USPT-SYD (USPT-PVG vs USPT-MEL: p = 0.009, 95% CI = 0.18 to 1.91; USPT-PVG vs. USET-JNB: p = 0.044, 95% CI :  0.02 to 3.18; USPT-PVG vs USPT-SYD: p = 0.023, 95% CI :  0.1 to 1.18). None of the other pairwise comparisons (i.e., those that did not involve PVG) were significant. For longer layovers, there were no significant differences in layover sleep quantities across routes.

Discussion

We investigated factors influencing sleep quantity and timing during layovers on LR and ULR routes. Across both shorter and longer layovers, our findings indicate that layover start timing relative to a pilot’s HBT was the most common predictor of both sleep quantity and sleep timing during layover. Layover sleep quantity was also related to crew type on the inbound flight, with command crew obtaining 55 minutes more sleep than the relief crew in the last 24 hours of longer layovers. On average, pilots obtained at least 7 hours of sleep across all phases of a trip (pre-trip, layover, and post-trip), except for the last 24 hours of longer layovers. There was no difference in layover sleep quantity per 24 hours between shorter and longer layovers. These findings provide important insights for schedule design to help provide pilots with adequate opportunities for sleep on LR and ULR layovers.

On shorter layovers, pilots obtained significantly more sleep when the layover began during their home base daytime (1000-1800 HBT) compared to in the early morning hours (0200-0600 HBT). Among regular, aligned sleepers, 0200-0600 HBT is the approximate time in 24 hours when the circadian clock is at its nadir of promoting alertness (sometimes referred to as the Window of Circadian Low; WOCL) [32]. Therefore, landing at this time reduces the potential for pilots to obtain sleep during this period. For example, if the aircraft lands at or after 0200, it is unlikely that the pilot would be able to obtain sleep during the WOCL (i.e. sleep aligned with the biological sleep drive) due to post-flight duties, commute time (from the airport to the hotel), the potential need to find and consume food, and time needed to wind down prior to falling asleep.

Pilots had significantly more WASO on shorter layovers when the layover began between 1000 and 1800 HBT compared to 0200–0600 HBT. There were no significant findings for sleep onset latency or sleep efficiency outcomes. Given that pilots had a higher sleep quantity with this layover start time, these results suggest that pilots also stayed in bed longer when landing between 1000 and 1800 HBT. This may be due to pilots lengthening their sleep opportunity to include the local night as well as the HBT night when they are not landing in or near their WOCL.

On longer layovers, pilots obtained significantly more sleep during the first 24 hours of layover when the layover began 1400–1800 HBT compared to 0200-0600 HBT, similar to shorter layovers. This suggests that arriving during the WOCL is disadvantageous for maximizing sleep quantity on layover, regardless of layover length. Furthermore, a study by Sallinen et al. [33] found that flights ending near the WOCL led to a significantly higher probability of being more fatigued at top-of-descent. These combined results suggest that the timing of arrivals is a key component impacting fatigue during the final critical phase of flight and subsequent layover sleep, with greater degradation when occurring within the HBT WOCL.

On longer layovers only, pilots obtained significantly more sleep during their HBT night when the aircraft landed at 1400–1800 HBT, compared to 0200-0600 HBT, for both the first 24 hours and the last 24 hours of longer layovers. Landing in the afternoon or early evening relative to HBT may allow the pilot to anchor their sleep near or in the HBT WOCL in the first 24 hours and then maintain this pattern in the second 24 hours. Despite pilots having more flexibility to plan their sleep on longer layovers, layover start timing still significantly influenced sleep timing. Thus, layover start timing is clearly a strong predictor of both sleep quantity and, especially for longer layovers, sleep timing.

Interestingly, on longer layovers, sleep quantity during the last 24 hours of layover was predicted by crew type on the inbound flight. A potential reason the two crew types have different sleep quantities ahead of the inbound flight is that the crews may prepare differently for the inbound flight. For instance, if the relief crew pilots know that they will take the first in-flight rest break during the inbound flight [5], they may choose to avoid taking a pre-flight nap or may shift their main sleep timing further from the scheduled inbound report time at the end of the layover [34]. In this scenario, command crew pilots may be more likely to increase their sleep quantity pre-flight since they are unlikely to take the first in-flight rest break. In fact, across all trips with longer layovers, 96% of relief pilots took the first break on the inbound flight and none of the command pilots took the first break. This type of scenario could explain why the command crew obtained more sleep within the last 24 hours of the layover compared to the relief crew and may also explain why there is significantly less sleep quantity in the last 24 hours of longer layovers. Given that this relationship with crew type only exists for longer layovers, it is likely that on shorter layovers there is not the opportunity for such flexibility of sleep strategies; rather, sleep is obtained whenever possible.

Neither sleep quantity nor sleep timing during the outbound flight significantly predicted any of the outcome variables during layovers. In-flight sleep has been shown to be important for fatigue and performance levels at top-of-descent [4, 25, 26, 35]. We had predicted that the timing of sleep on the outbound flight may have influenced the timing of subsequent layover sleep, but it appears that any effects were small, especially relative to layover start timing. It is important to note that factors such as pilots’ individual layover strategies (e.g. attempting to stay on HBT to minimize jet lag) and/or social factors (e.g. sightseeing, meal timing) may also contribute to layover sleep timing, but were not included in our models. A post hoc analysis suggested by a reviewer showed that the total amount of sleep in the 24 hours prior to the outbound flight was also not a significant predictor for our models. This finding is in line with prior literature that showed napping before the outbound flight did not change the quantity or quality of inflight sleep during the outbound flight [36].

Our result that pilots obtained more sleep in the last 24 hours of shorter layovers compared to the last 24 hours of longer layovers is consistent with results from a similar study by Gander et al. [23]. They found a difference of approximately 2.5 hours, whereas we observed a smaller difference of approximately 40 minutes. Lamond et al. [12] reported no change in sleep quantity, but they compared much shorter (9 hours) and much longer (62 hours) layovers. Thus, our results only hold true for similar-length layovers (i.e. 1 day vs 2 days).

We observed that pilots slept 49 minutes less during the last 24 hours of longer layovers compared to the pre-trip 24-hour average, and 45 minutes less compared to the first 24 hours of layover. This is similar to a previous study of a 48-hour layover that showed pilots obtained 35 minutes less sleep in the last 24 hours compared to pre-trip and 66 minutes less than the first 24 hours of layover [36]. We did not see a difference between pre-trip sleep and the first 24 hours of layover sleep, while Signal et al. reported 33 minutes more sleep in the first 24 hours of layover compared to pre-trip sleep. This could be due to differences in the home base timing of the layovers between studies, especially given we have shown the importance of layover timing on layover sleep outcomes.

It is important to highlight that, except for the last 24 hours of longer layovers, sleep quantity per 24 hours did not differ across trip phases or layover length [6]. Moreover, pilots slept at least 7 hours per 24 hours across these phases of the trip. The joint consensus statement from the American Academy of Sleep Medicine and Sleep Research Society recommends seven or more hours of sleep at night to maintain optimal health, with sleep less than 7 hours associated with impaired performance, increased errors, and risk of accidents [37]. These results suggest that pilots can obtain the minimum recommended amount of sleep during layover. However, considering individual differences in sleep need, some pilots may not be obtaining their optimal sleep quantity. In addition, while we assessed crude proxies for sleep quality from actigraphy, more robust sleep measures are needed to determine the restorative quality of the sleep obtained during layover.

We did not include layover location as a predictor in our statistical models since we were focused on operational factors and we were looking for model inputs that were predictive across routes. While a one-way ANOVA comparing layover sleep quantities did show a significant effect of route for shorter (but not longer) layovers, post hoc tests showed that this was due to one route only. The outlier route, USPT-PVG, may have peculiarities specific to that layover location that result in less layover sleep.

Limitations

Our actigraphic sleep study of two airline carriers including 256 airline transport pilots operating 473 trips across nine unique LR and ULR routes is not without limitation. First, we did not account for fatigue mitigation strategies, social influences, and other factors that may have impacted layover sleep quantity, timing, and quality (e.g. caffeine use, crew plans, sight-seeing, hotel-related disturbances, availability of food and facilities). A study focusing on the individual, qualitative, and cultural aspects of layover sleep is needed to address these remaining gaps in the literature.

Although we have a relatively large sample size for this type of study, we did not have sufficient data from trips with layovers beginning across the whole 24-hour period. Since the layover start timing was the most compelling predictor of sleep quantity and sleep timing for both shorter and longer layovers, our study would have been enhanced if we had further diversity in the dataset (i.e., a wider variety of outbound flight landing times). Although our study results tend to align with prior research from other regions, our data may not generalize beyond the two US-based carriers observed. For example, it may be that some of the features of our dataset are somewhat contingent on a particular pilot culture within these two airlines. A dataset with more diversity, both in airline carrier and country of origin for LR and ULR flight operations, is needed to confirm the novel aspects of our findings.

We also acknowledge that our analysis used data from predominantly westbound operations. The circadian effects related to eastbound versus westbound travel have been well-documented elsewhere and may have an influence on our findings [19, 38]. To that end, as a large field study, we did not directly measure circadian phase and used HBT as a proxy with the caveat that estimating circadian phase in trans-meridian shift workers is notoriously difficult.

As part of the process of quantifying the timing of sleep, we defined HBT night as 2300-0700. Many of the original two-process model papers use this timeframe (see, e.g., Ref. [39]), as well as some of the seminal papers related to desynchrony [40]. Also, an 8-hour window aligns with the Sleep Research Society and American Academy of Sleep Medicine joint study that contains a consensus recommendation of 7–9 hours of sleep per night to support optimal health in adults [37]. However, this choice does not account for individual differences such as chronotype or sleep needs. Future studies could collect these data and investigate them as potential predictors of sleep quantity and timing during layovers.

One of the two outputs in our models was sleep timing: the percentage of sleep that occurred during HBT night (defined as 2300-0700). We were curious if pilots tend to stay on their HBT or attempted to switch/successfully switched their sleep schedules to local time during layovers. This metric may not account for important information, that is, when exactly the sleep occurs during HBT night. In our analysis, a short sleep episode during the beginning of the HBT night is counted the same as a short episode during the end of the HBT night. While we agree that certain qualities of sleep vary by the time of night, we did not measure circadian phase nor any metric of sleep depth or quality. Our approach, although not perfect, is better than a binary metric that categorizes each sleep episode as happening during HBT night or not. Future research could use circadian markers along with the other input variables to predict not just the proportion of sleep that happens during HBT night, but also when that sleep happens relative to circadian phase.

In this study, we did not have an individualized estimate of sleep timing pre-trip. This pilot-specific metric could be helpful in making tailored predictions of layover sleep, but this presents several challenges that are particular to this population. First of all, it is difficult to define and estimate habitual, aligned at-home sleep timing in airline pilots based on pre-trip data. For example, a pilot may change their sleep patterns pre-trip to prepare for the trip, or pre-trip sleep may have been impacted by recovery from a previous trip. To overcome these difficulties, future studies should collect measures of circadian timing as well as qualitative data about preparation for the trip and sleep choices during the layover.

Conclusion

Sleep quantity and quality during layover are important for alertness and cognitive performance on the inbound flight. To better understand the sleep quantity and quality pilots are obtaining on layovers, we measured sleep duration and timing, with timing impacting sleep quantity and quality. Our findings suggest that while pilots generally obtain at least 7 hours of sleep per 24 hours across a trip, scheduling factors and individual strategies can impact sleep quantity and timing (e.g., different crew types prepare differently for the inbound flight). Layover start timing significantly influences the quantity and timing of layover sleep with flights landing in the WOCL associated with reduced sleep. Furthermore, on longer layovers, crews may be using different strategies in the last 24 hours of the layover to prepare for inflight sleep opportunities on the inbound flight. These results reveal the importance of scheduling for sleep opportunities on layover and highlight the need to better understand the layover sleep strategies adopted by pilots so that education on optimal sleep strategies can be tailored to acknowledge factors beyond the two-process model of sleep.

More studies are needed to further tease apart the multi-factorial questions about layover sleep. Future studies should assess whether pilots have specific strategies for layover sleep and the motivations behind these strategies, making sure to include both junior and senior pilots. These findings can then be connected to safety performance indicators prior to and during the inbound flights to better understand the impact of layover duration and timing on LR and ULR flight operations.

Funding

Funding for a portion of this study was provided by United Airlines and by the NASA Airspace Operations and Safety Program, System-Wide Safety Project.

Disclosure Statement

Financial disclosure: None. Non-financial disclosure: None.

Author Contributions

Michael Rempe (Conceptualization [Equal], Formal analysis [Lead], Investigation [Equal], Software [Lead], Visualization [Lead], Writing—original draft [Lead], Writing—review & editing [Lead]), Ian Rasmussen (Data curation [Lead], Formal analysis [Supporting], Investigation [Equal], Methodology [Equal], Validation [Equal], Writing—original draft [Supporting], Writing—review & editing [Supporting]), Kevin Gregory (Data curation [Equal], Methodology [Equal], Software [Supporting], Validation [Equal], Writing—original draft [Supporting], Writing—review & editing [Supporting]), Cheyenne Johnson (Visualization [Supporting], Writing—review & editing [Supporting]), Matthew Hsin (Writing—original draft [Supporting]), Erin Flynn-Evans (Conceptualization [Equal], Formal analysis [Equal], Funding acquisition [Lead], Methodology [Equal], Writing—original draft [Supporting], Writing—review & editing [Supporting]), Amanda Lamp (Conceptualization [Equal], Funding acquisition [Lead], Methodology [Equal], Supervision [Equal], Writing—original draft [Equal], Writing—review & editing [Equal]) and Cassie Hilditch (Conceptualization [Equal], Formal analysis [Equal], Methodology [Equal], Supervision [Equal], Writing—original draft [Equal], Writing—review & editing [Equal]).

Data Availability

The data underlying this article cannot be shared publicly due to the terms of a data-sharing agreement between Washington State University, airline pilots unions, and airline management.