A new way to track how lack of sleep increases risk in emergency medicine

The monocentric study REST (Recovery optimization in Emergency medicine addressing Stress adaptation Techniques) was approved by the Ethics Committee for the Protection of Individuals (2021-A03171-40) and was preregistered on ClinicalTrials.gov↗ (Identifier: NCT05251246↗). REST was designed to measure the impact of a recovery intervention on EHP and data used for this naturalistic observational study was derived from the baseline measures. Data were collected for 2 weeks from staff working rotating shifts in two emergency departments of a university hospital of Lyon, France. The shift system requires nurses to rotate 12-h shifts (7:00 to 19:00 the day shift and 19:00 to 7:00 the night shift) and physicians to rotate between 10.5-h day shifts (8:00 to 18:30) and 14-h night shifts (18:30 to 8:30).

In total, 72 volunteer EHPs were enrolled. Exclusion criteria were non-full-time staff (work time less than 80%) or staff that had worked less than a year in an emergency department. Participants were fully informed about the study and provided their voluntary, written informed consent to participate.

Between February and November 2022, we collected both schedule data and objective sleep data, with shift rosters for SLI computation provided by emergency department and ambulance administration. The study was conducted in 2 phases: In the first phase (February–July 2022), we tested our index in a preliminary group of participants (early-stage subset) by comparing it to established risk measures (theoretical validation). In the second phase, following the completion of participant enrollment in November 2022 (REST protocol), we assessed the index’s relationship with objective sleep data collected at the pre-intervention stage for the full study sample (behavioral validation). Participants completed questionnaires at baseline (study inclusion) and again at the end of the sleep monitoring period. Demographic variables captured included sex, age, BMI, relationship status, presence of young children (≤ 5 years), shift-work experience, and commute time.

Diurnal preference was determined by the Morningness Evenigness Questionnaire MEQ (26). This validated questionnaire is comprised of 19 items. The total score ranges from 16 to 86 and defines the chronotype: scores ranging from 16 to 30 indicate extreme eveningness preference (E*); scores from 31 to 41 indicate evening preference (E); scores from 42 to 58 are categorized as neutral (N); scores from 59 to 69 indicate morning preference (M); and scores from 70 to 86 indicate extreme morning preference (M*).

The SLI is a validated predictive scoring system originally developed by Gander et al. (13, 14) to assess, based on planned work schedules, the imposed risk of limited recovery opportunities. For the present study, we adapted the SLI to the French shiftwork system and extended it by including a specific rating for the degree to which work hours conflict with typical social hours (evenings, weekends) recognizing that such conflicts may increase social pressure and further reduce sleep opportunities (27). These adaptations are detailed in a publicly available preprint (28). The SLI score generated for a work schedule of one week comprises 8 items, (Table 1); each with predefined cut-off scores to define 3 different risk levels: 0 (low risk), 1 (medium) or 2 (high). The higher the shift load, the greater the misalignment between work schedules and biological or social rhythms. Accordingly, higher risk levels represent less opportunity for recovery and restorative sleep. The total SLI score (ranging from 0 to 16) is obtained by summing the risk levels across all items. If possible, we also recommend including a ninth item in an extended version of the index to account for cumulative fatigue arising from the previous week. In the theoretical validation phase, the SLI score was computed based on two-week work schedules, allowing the use of the extended version rated on all nine items. In contrast, for the behavioral validation phase, the score was calculated from one-week work schedules to match the available behavioral data, and thus was rated on eight items only. Ratings were generated for each day using a dynamic sliding window approach, whereby each day’s score reflects recovery opportunities within the preceding seven-day period. This method enables a continuous and temporally sensitive assessment of recovery risk.

To validate the adapted SLI against an established biomathematical model commonly used in 24/7 industries, we included FAID Quantum scores for cross-measure comparison (29). The FAID Quantum scores are widely used biomathematical models in 24/7 industries. FAID Quantum estimates fatigue and alertness based on schedules derived from work / non-work hours and provides 2 primary metrics: (i) FAID-FATIGUE Score: Predicts on-duty physical fatigue based on the Fatigue Audit InterDyne algorithms; (0–100+; higher scores indicate higher predicted fatigue). Scores above 80 represent an established risk threshold validated by FAID Quantum; for comparison, a typical office week yields a score of approximately 41; (ii) FAID-ALERTNESS Score: Predicts on-duty sleepiness based on the Karolinska Sleepiness Scale (KSS) integrating the Three-Process Model of Alertness (1–9; higher scores indicate higher predicted sleepiness). The scores were obtained for work shifts only and not for days off.

Objective sleep data were obtained using an actigraphy device (Actiwatch 2; Philips - Respironics, Murrysville PA, USA). Participants wore the wrist device for 3 weeks and were asked to press a button to indicate bed time, defined as the moment when they attempted to fall asleep and again in the morning to indicate when they got out of bed (get up time). Sleep onset and Sleep Offset were defined the moment the subject fell asleep and woke up. Actigraphy monitors the activity level based on an accelerometer with data being collected at a frequency of 32 Hz and segmented to 30-s epochs. Data was analyzed using the Philips Actiware software version 6.0.1. Sleep regularity (SRI), total sleep time (TST) and time in bed (TIB) were measured whenever the sleep–wake pattern of an EHP included seven consecutive days of recording; all sleep types were cumulated (nighttime sleep, daytime sleep, sleep obtained at work, and naps, regardless of timing or setting). TST refers to the time asleep between the moment the EHP has fallen asleep (sleep onset) until the time they wake up (sleep offset), so any disrupted sleep (nocturnal awakenings) decreases the sleep quantity. TIB refers to the time the participant spends in bed from bedtime to get up time and represents the overall sleep opportunity.

SRI quantifies the regularity of sleep–wake patterns by comparing sleep/wake patterns over consecutive days (25, 30). It expresses the likelihood that any two time-points 24 h apart are in the same sleep or wake state. The SRI ranges from 0 (random pattern) to 100 (perfectly regularity) and is sensitive to naps, sleep fragmentation and non-consecutive data. Established cut-offs from larger cohorts exist for interpreting SRI scores as irregular, intermediate, or regular sleep patterns, typically based on the lowest and highest quintiles of the score distribution (30, 31). The SRI is computed with the following formula (Equation 1):

where M is the number of daily epochs, N is the number of days, s_i,j = 0 for sleep and s_i,j = 1 for wake, and δ(s_i,j, s_{i + 1,j}) = 1 if s_i,j = s_{i + 1,j} (meaning the epochs separated by 24 h are in the same state) and 0 otherwise.

First, during the theoretical testing phase, the extended 9-item Shift Load Index (SLI) was compared to FAID Quantum scores, as FAID Quantum is also typically applied to two-week schedule data for comprehensive fatigue risk profiling. A generalized linear mixed-effects model (GLMM) was used, controlling for shift type and accounting for clustering by subject as a random effect (32–34).

Subsequently in the behavioral validation phase, the basic 8-item SLI was applied to objective sleep data over 1 week. Demographic group comparisons for sleep variables (SRI, TST, TIB) were conducted using t-tests or ANOVA, as appropriate. Sleep–wake patterns over time were visualized using locally estimated scatterplot smoothing (Loess). The predictability of sleep variables based on SLI scores was further tested with GLMMs, applying a sliding window approach with day-level data clustered by subject. Models were adjusted for demographic covariates, commute time, chronotype, and shift work experience. Model residuals were visually inspected using probability plots to assess normal distribution. A significance level of 0.05 was applied throughout.

Finally, to assess how sleep regularity and duration jointly predict shift-load scores, we performed an exploratory multivariable mixed-effects regression controlling for subject with daily SLI as the dependent variable and SRI and TST entered simultaneously as predictors.

In the theoretical testing phase, schedule analysis included 140 shifts worked by 55 EHPs (Figure 1, flow chart), with each participant providing repeated measures (two to three shifts each). Night shifts accounted for 51% of the shifts. The risk distribution across the extended Shift Load Index (SLI) with its 9 items is illustrated in Figure 2. While most SLI items exhibited varied risk levels, those related to rest opportunities generated minimal risk within this sample.

A total of 21% of shifts (n = 29) yielded SLI scores above 9, indicating elevated risk for insufficient recovery. The average FAID-FATIGUE score was 56.4 (± 24.3), with 19.5% of shifts demonstrating a peak score above the established threshold of 80. The average FAID-ALERTNESS score was 6.3 (± 1.8; corresponds to “some signs of sleepiness”), with 20.3% of shifts showing a peak KSS score of 8 or above (“sleepy, but some effort to keep awake”).

Given the variability of these outcomes across shift types, shift type was included as a random effect in the GLMM.

The GLMM showed that SLI was a significant predictor of both FAID-FATIGUE (β = 3.91, p < 0.001) and FAID-ALERTNESS (β = 0.02, p < 0.001), highlighting the impact of shift load on predicted fatigue and alertness levels. When SLI items were analyzed individually in separate models, the explanatory power increased substantially by adding shift type (R² = 70 ➔ 88% for FAID-FATIGUE; R² = 51 ➔ 99% for FAID-ALERTNESS). Detailed item-level findings are provided in the preprint (28).

The SLI in its eight item version was further assessed using objective sleep data from a subsample of 35 EHP (21 nurses, 14 physicians) each providing at least 7 consecutive 24 h recordings (totaling 360 observations; Figure 1). As shown in Table 2, participants SLI, TIB, TST, and SRI along with group comparisons by sex, profession, partner status, presence of young children, and chronotype are summarized. Descriptively, Figure 3 shows that TIB and TST declined with increasing SLI-leveling off at moderate loads before dropping further at high loads. SRI was stable at low SLI, dipped mid-range, then rebounded at high SLI, forming an inverted-U.

To assess the influence of shift scheduling on sleep outcomes, GLMMs were constructed with SLI as a predictor and subject as a random effect (Figure 4). SLI significantly predicted TIB and TST (respectively β = −0.53, p < 0.001, marginal R² = 5% and β = −0.29, p < 0.001, marginal R² = 2%). indicating that for each one-point increase in SLI, TIB decreased by approximately 32 min and TST by 17 min. SLI also significantly predicted lower sleep regularity (β = −1.06, p < 0.001), with a marginal R² of 5%. With every increase of one point in the risk index, SRI decreased by approximately 1%. As a control, GLMMs with subject as a random effect showed that SRI exhibited only a non-significant trend with TIB (β = −0.24, p = 0.053; marginal R² = 2%) and no association with TST (β = 0.02, p = 0.894), confirming that sleep regularity and sleep duration were not meaningfully associated.

The associations between shift load and individual sleep remained robust after adjusting for sex, age, BMI, weekly sport hours, young children, couple status, profession, shift experience, commute time and chronotype. In the adjusted models, the effect of SLI on TIB and on TST remained significant (respectively β = −33.19, p < 0.001, marginal R² = 15% and β = −18.30, p < 0.001, marginal R² = 20%), as did the effect of SLI on SRI (β = −1.06, p < 0.001; R² of 21%). Further, none of the covariates independently predicted these outcomes.

Following the predictive value of SLI on behavioral sleep parameters, we plotted SRI against TST, with point color indicating the observed SLI (Figure 5). The exploratory multivariable regression of SLI on SRI and TST yielded a fitted prediction surface:

GLMM revealed that SRI and TST were both significantly associated with SLI, yielding a total explanatory R² of 48%, with fixed effects accounting for 18% of the variance. This interindividual variance in shift adaptation can be captured by calculating each individual’s residual (Observed SLI – Predicted SLI), we can identify those who fall outside the model’s confidence bounds (CI 95% [−0.516, 0.664]), highlighting participants who significantly outperform or underperform relative to their peers. For example, a positive residual signifies that a person handled a higher shift load than their sleep patterns would predict, indicating better adaptation to demanding schedules. Approximately 42% had residuals below −0.516, signifying worse adaptation and heightened sleep vulnerability.

The Shift Load Index (SLI) was developed by adapting a pre-existing risk assessment matrix to align with French work regulations, specifically targeting recovery opportunities within emergency department shift schedules. The SLI incorporates key factors such as working hours, shift length, night shifts, and the loss of biologically and socially meaningful time. In doing so, the present work extends the original Australasian scheme by calibrating it to European regulations and incorporating a social-rhythm component, thereby capturing aspects of circadian and social disruption not accounted for in earlier models. Interestingly, recovery periods contributed minimally to risk scores in this cohort, likely due to strict regulatory constraints. During the theoretical validation phase, the SLI demonstrated strong predictive validity for fatigue, aligning with results from two established biomathematical models: the Three Process Model of Alertness and the FAID model. While the former focuses on immediate sleep–wake patterns and daytime sleep, the latter accounts for cumulative fatigue over a week, integrating both physiological sleep pressure and social pressures. The SLI’s alignment with both models reinforces its ecological validity, particularly in identifying high-risk shifts for fatigue accumulation. Importantly, unlike the original scheme, which relied exclusively on self-reported fatigue and questionnaire data, our study addresses the missing field validation of this type of streamlined index against objective sleep metrics. Across models, every fifth shift in our cohort presented a heightened risk in terms of recovery and alertness. In comparison, a recent, larger study of British anesthesia residents reported only 12.7% of shifts exceeding elevated FAID fatigue thresholds, compared to our 19.5%, suggesting that French emergency department working hours may pose a higher burden (38). However, it should be noted that the Roche et al. study included residents and a much larger number of shifts, while our analysis focused only on full-time nurses and physicians, which may affect direct comparability.

Our findings indicate that maintaining adequate sleep becomes increasingly challenging as shift load intensifies, with SLI scores of 6 or above associated with significant sleep disruption. Given its robust associations with objective sleep metrics, SLI appears to be a reliable and practical rating system offering practical advantages for hospital management, risk assessment and application in real-world field settings.

Our proposed framework on shift adaptation lays the groundwork for a more nuanced understanding of the interplay between work schedules and individual differences in sleep response. While the SLI effectively captured sleep deficits associated with shift demands, our analysis revealed that approximately one-third of the variance in sleep adaptation was attributable to individual factors. This underscores the existence, within this challenge, of a modifiable margin for individual adaptation or vulnerability to shift schedules. Sleep behavior is highly individual and varies across schedules, suggesting that while work schedules have a tangible impact, adaptation can vary significantly depending on the risk profile, for instance morning type, age and family obligations intervene with shift adaptation (35). Recognizing this potential for adaptation supports the development of more personalized strategies for shift scheduling and fatigue management.

Building on this, the shift adaptation framework provides a conceptual basis for translating SLI outputs into actionable strategies. In practice, it could help identify vulnerable populations for targeted fatigue management, inform schedule design by highlighting high-risk sequences, and support occupational health monitoring at both the individual and organizational level. At present, the framework remains conceptual, and future studies should evaluate its value in diverse hospital systems while examining whether integrating SLI-based adaptations can reduce fatigue, improve sleep efficiency, and enhance clinical performance. The personalized shift adaptation framework seeks to pave the way for substantial improvements in clinical performance and organizational health.

Sleep regularity is increasingly recognized as a critical determinant of physical and mental health (23, 24). It is closely modulated by lifestyle and environmental factors such as light exposure, food intake, and behavioral routines (39–41). In the context of emergency departments, exposure to artificial brightly light during night shifts may further disrupt circadian rhythms, complicating efforts to maintain consistent sleep regulation (42, 43). Although the SLI effectively captured overall sleep disruption, the specific analysis of sleep regularity revealed a U-shaped curve, largely driven by the frequency of night shifts, consistent with previous simulation studies (25). However, high SRI scores under high shift load may not necessarily reflect healthy sleep behavior, but rather stable, biologically inverted, and ultimately unhealthy patterns. Our findings are not surprising in that they confirm, both theoretically and in the field, that shift work produces irregular sleep patterns (25, 44). Critically, however, this is the first study in emergency healthcare professionals to quantify the extent of reduced sleep regularity with the SRI: the majority of participants fall below the 20% lowest-quintile cutoff established in non–shift-working populations. This stark shift relative to normal cohorts highlights the substantial health constraints imposed by these demanding schedules.

Our findings suggest that sleep deficits among EHPs stem more from reduced sleep efficiency than insufficient time in bed, emphasizing the need for interventions that enhance both sleep quantity and quality (45, 46). Beyond work schedules, environmental factors are increasingly recognized as key determinants of sleep health. Variations in thermal comfort, noise, light exposure, and air pollution can compromise sleep initiation and maintenance and may differentially increase vulnerability, particularly in urban environments such as in our study sample (47, 48). Our observations align with Borbély’s two-process model of sleep regulation, which highlights the importance of addressing both sleep duration and regularity (49, 50). In high-stress environments like emergency departments, chronic sleep disruption may exacerbate compassion fatigue and burnout as well as general health, further underscoring the need for targeted sleep interventions (51, 52).

The consequences of sleep deprivation extend beyond individual health, affecting both care quality and institutional costs. Fatigue-related medical errors alone cost health care systems billions annually (53), while high turnover rates in emergency services, driven by shift work and workplace stressors, cost an estimated 6,000 euros per departure (54). Institutional strategies must therefore move beyond mere scheduling adjustments to implement systemic workload monitoring, optimize staffing models, and foster a culture that proactively addressed fatigue risks (44, 55). Addressing critical incidents, workplace violence, and understaffing may further alleviate chronic stress and prevent burnout (56) as health problem (57).

Despite well-documented risks associated with shift work-related sleep disruption, evidence-based interventions remain underexplored. Our findings suggest that individual variability may outweigh shift schedule impacts, highlighting the need for targeted strategies on both organizational and individual levels (9, 58). Importantly, emerging evidence points to a genetic basis for individual vulnerability to sleep loss. Specific polymorphisms have been associated with differential responses to sleep deprivation in terms of cognitive performance, alertness, and subjective fatigue (59, 60).

At the organizational level, implementing programs that include sleep hygiene education, stress management training, and light exposure protocols can help mitigate sleep disruption and circadian misalignment (61). Additionally, tailored guidance on sleep practices, coping strategies, and shift-specific recovery techniques on an individual and chronotype level may significantly improve sleep quality and reduce fatigue, especially in high-stress settings like the emergency departments (62, 63). Orchestrated efforts have the potential to reduce sleep deprivation in healthcare settings.

While our study provides valuable insights, several limitations should be acknowledged. First, our data did not include cumulative shift burden from the previous week, limiting our ability to assess extended recovery deficits. Second, the behavioral validation was based on a modest sample drawn from two emergency departments within a single European city, limiting generalizability to other emergency departments or shift-working contexts. The modest sample size may also reduce statistical power, even though the repeated-measures design increases robustness. Third, the risk value distribution was skewed, generally reflecting lower-risk shifts, thereby constraining the analysis of those at higher risk levels. Fourth, our methodological choice to use a sliding window approach, while valuable for capturing time-series data, may have inadvertently introduced data redundancy.

Additionally, potential multicollinearity among SLI items raises the question of whether further simplification of the index may enhance utility without compromising predictive accuracy. Moreover, the current SLI does not account for excessive sleep, which may reflect compensatory behaviors and could carry its own risks. Lastly, our SRI metric did not account for sleepiness during activity intervals, overlooking diurnal sleepiness and nocturnal wakefulness patterns.

Future studies should address these gaps by incorporating more granular assessments of sleep–wake dynamics, particularly in diverse shift work settings. Expanding the dataset and applying the SLI in varied contexts will be crucial to validating and extending the framework’s applicability.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.