Objectives Shift work, like chronic jet lag, is known to disrupt workers’ normal circadian rhythms and social life, and to be associated with increased health problems (eg, ulcers, cardiovascular disease, metabolic syndrome, breast cancer, reproductive difficulties) and with acute effects on safety and productivity. However, very little is known about the long-term consequences of shift work on cognitive abilities. The aim of this study was to assess the chronicity and reversibility of the effects of shift work on cognition.
Methods We conducted a prospective cohort study of 3232 employed and retired workers (participation rate: 76%) who were 32, 42, 52 and 62 years old at the time of the first measurement (t1, 1996), and who were seen again 5 (t2) and 10 (t3) years later. 1484 of them had shift work experience at baseline (current or past) and 1635 had not. The main outcome measures were tests of speed and memory, assessed at all three measurement times.
Results Shift work was associated with impaired cognition. The association was stronger for exposure durations exceeding 10 years (dose effect; cognitive loss equivalent to 6.5 years of age-related decline in the current cohort). The recovery of cognitive functioning after having left shift work took at least 5 years (reversibility).
Conclusions Shift work chronically impairs cognition, with potentially important safety consequences not only for the individuals concerned, but also for society.
Statistics from Altmetric.com
What this paper adds
Shift work, like chronic jet lag, is known to disrupt workers’ normal circadian rhythms and social life, and to be associated with increased health problems and with acute effects on safety and productivity. However, very little is known about the long-term consequences of shift work on cognitive abilities.
Our prospective study shows an association between shift work and chronic cognitive impairment that is a function of length of exposure. We also show that recovery of cognitive function occurs some years after having ceased any form of shift work.
Measures should be considered that mitigate the impact that prolonged exposure to shift work has on cognitive abilities, including switching to normal day work.
Several studies have demonstrated the acute deleterious effects that non-standard working hours have on alertness and cognitive efficiency during night shifts and the following days.1–7 However, only four studies have examined whether there may also be a chronic impact of abnormal work schedules on cognitive abilities (ie, effects that last for several weeks, months or years). Cho et al8 showed cognitive performance deficits and higher cortisol levels in airline cabin crew who had experienced repeated exposure to jet lag for more than 3 years, compared with ground crew working for the same company. There were no such effects in aircrew who had been exposed for 3 years or less. Subsequently, Cho9 found that chronic exposure to short recovery periods (≤5 days) from jet lag was associated with lower cognitive performance, higher salivary cortisol and a smaller volume of the right temporal lobe. These findings were interpreted as showing a cumulative effect of chronic exposure to circadian disruption on cerebral structures and cognitive function.
A subsequent cross-sectional study10 also revealed cognitive deficits in male industrial workers who had been exposed to shift work relative to those who had not, and a decrease in memory performance with increasing exposure to shift work. These effects were independent of age and self-reported sleep quality and are similar to those of Cho and colleagues,8 ,9 in that they appear to reflect chronic exposure to circadian disturbances.
Most recently, a prospective cohort study of nurses found limited evidence of cognitive impairment in later life (≥70 years of age) being associated with a history of exposure to rotating night-shift work, as reported in midlife (ie, at the age of 58–68 years).11 Participants with ≥20 years exposure demonstrated modest impairments in a test of general cognition. However, there were no associations between shift work history and composite measures of general cognition and verbal memory, or between shift work history and cognitive decline.
Given the potentially detrimental impact of shift work-related cognitive decline on job performance and quality of life, the current study examines the effect of shift work on cognition in a large sample of workers followed over 10 years. The first set of analyses seeks to determine whether having experience of shift work affects cognition, by comparing workers who are either currently working shifts or who have prior experience of shift working with workers who have never worked shifts. The second set of analyses examines the effects of duration of exposure to shift work, by comparing three groups of workers: those with no experience of shift work, those with up to 10 years of exposure and those with more than 10 years of exposure. The third set of analyses examines whether there is a chronic effect of shift work that persists after exiting shift work, by comparing four groups of workers: those currently working shifts, former shift workers who left shift work within the previous 5 years, former shift workers who left shift work more than 5 years previously and those who have never worked shifts. No previously published studies have examined whether such chronic effects of shift work on cognition are reversed following the cessation of shift work.
Participants and procedure
The data were taken from the VISAT study.12 The initial sample was composed of 3232 present and former wage earners covering a wide range of occupations and economic sectors. The overall distribution by gender and socioeconomic position was very close to that observed at the national level by the French National Institute for Statistics and Economic Studies (INSEE). Participants were exactly 32, 42, 52 and 62 years old at the time of the first data collection (1996, t1). In the older age cohort, 83% were retirees at t1. Participants were randomly drawn from the patient list of 94 occupational physicians in three southern regions of France and were volunteers (participation rate: 76%). These lists comprised all the salaried workers in the region, as all workers in France have a mandatory annual medical assessment of their aptitude to work. Data were collected through questionnaires and clinical examination by occupational physicians especially trained for the purpose of the study during this annual assessment. Two subsequent data collections took place in 2001 (t2) and 2006 (t3). All those who participated at t1 were invited to participate at t2 and again at t3, irrespective of whether they were still in work. Data for the current study were available for 3119 participants at t1. Of these, 2183 were seen again at t2, and 1253 at t3. A total of 1197 were seen on all three occasions (56 participants who were not seen again at t2 were seen again at t3).
The shift work measures used in the present study were taken from the ESTEV cohort study13 and correspond to accepted national and international definitions of night work.14 At each measurement occasion, the participants were asked whether, for more than 50 days/year, their work schedule (1) involved rotating shift work (eg, alternating morning, afternoon and night shifts), or (2) did not allow them to go to bed before midnight, or (3) resulted in them having to get up before 5:00, or (4) prevented them sleeping during the night (night work). Possible responses for each question were: ‘yes, currently’ (current), ‘not now, but yes in the past’ (past), or ‘never’ (never). Note that this operationalisation means that the ‘never’ category, which serves as a control group in subsequent analyses, may include participants with a small amount of shift work exposure (≤50 days/year). Thus, any observed effect of shift work on cognition runs the risk of being underestimated in this context. At baseline, the percentage of each response category (current, past, never) was, respectively, 18.5%, 17.9%, 63.6% for rotating shift work; 8.0%, 12.5%, 79.5% for the second work schedule (‘midnight’); 11.8%, 15.4%, 72.8% for the third one (‘5:00’); and 7.2%, 11.5%, 81.3% for the fourth one (‘night’). For each of the four questions, information was collected on cumulative exposure duration and how many years had elapsed since they had stopped working such a schedule. In the present paper, those who answered ‘yes, currently’ or ‘in the past’ to any of these four questions at t1 were considered to be working, or to have previously been working, on some form of shift system at t1. Those with no experience of shift work are defined as ‘day workers’. We could not calculate the total exposure to all types of atypical work schedules by summing the exposures to each, since they were not mutually exclusive. For example, if a participant reported 1 year of exposure to each type of atypical work schedule, their total exposure to any type of atypical work schedule could be anything between 1 and 4 years, depending on the degree of overlap. For this reason, analyses of exposure duration were confined to rotating shift work (which was the most commonly reported of the four types of atypical work schedule).
Participants undertook three sets of cognitive tests on each measurement occasion: (1) a verbal episodic memory test adapted from the Rey Verbal Learning Test,15 including immediate and delayed retrieval tests. The participant was read the same list of 16 words three times, and was immediately asked to recall the words after each time. After a delay of 15 min, filled with other tests, the participant undertook a delayed free recall test, followed by a delayed recognition test. In the latter, the participant was required to locate the 16 previously learnt words that were randomly mixed in with 32 new words. Five memory measures were thus recorded: three immediate free recalls, one delayed free recall and one delayed recognition measure; (2) the Digit-Symbol Substitution subtest of the Wechsler Adult Intelligence Scale,16 a test mainly reflecting processing speed17; and (3) a selective attention test derived from the Sternberg test18 which was composed of two subtests. The first was a task consisting of looking as quickly as possible through a line of 58 alphabetic characters to find a target letter shown in the margin. This task was repeated six times, on six lines with a different target each time. The second subtest also had six lines of 58 alphabetic characters, but this time the memory load was greater because the target to be located was one of four letters shown in the margin. No time constraint was imposed for memory tests, while for the other tests, participants were instructed that speed was an important aspect of the task.
In order to summarise information from the eight cognitive tests, a principal component analysis (PCA) was performed. The first two axes accounted for 53.4% and 14.8% of the total variance, respectively. The first axis was a general performance axis that separated participants who obtained high scores in every test from participants who obtained low scores. The second axis ranked individuals in terms of the difference between their scores on the five memory orientated tests and their scores on the three speed orientated tests. The remaining five axes obtained with the PCA had no straightforward interpretation in terms of type of tests or ability, and were thus deemed to be of little interest.
A performance variable was constructed from this PCA, based on the factorial scores on the first axis, and was treated as a global cognitive performance score. Given the structure of the second axis, we decided to examine the possible differential impact of shift work on memory and speed performances. As it was not possible to extract memory and speed scores directly from axis 2 of the PCA, we performed two ancillary PCAs based respectively on the five memory oriented tests and on the three speed oriented tests. The first axis of the memory PCA accounted for 72.0% of the total variance and the first axis of the speed PCA accounted for 61.0% of the total variance. The factorial scores of the first axis of the memory PCA were thus used to create a memory performance variable, while those of the first axis of the speed PCA were used to create a speed performance variable. We used factorial scores instead of (standardised) means in order to maximise the variance summarised by the factorial axes.
The three cognitive variables (global cognitive performance, memory performance and speed performance) were continuous and normally distributed and could thus be used as dependent variables in linear mixed models. In order to ease interpretation of the results, all three variables were transformed so as to have scores in the range 0–100, with 100 indicating higher performance.
Several potential confounds of the relationship between shift work and cognition were statistically controlled in the analyses. Participants had to rate on a four-point scale (never, seldom, sometimes, often) the frequency in the past month of five symptoms associated with sleep problems: (1) difficulty falling asleep, (2) difficulty maintaining sleep, (3) difficulty getting back to sleep, (4) premature awakening and (5) hypnotic medication use. A sleep difficulty score was computed by summing the ratings (range 5–20, with 20 indicating the highest sleep difficulties; Cronbach's α=0.74). Perceived stress during the past month was assessed by means of the perceived-stress scale of Cohen et al19 (score range 4–20, with 20 indicating maximum stress; Cronbach's α=0.70). Other variables that were used as covariates were: age (at t1), gender, socioeconomic position (executive, ie, executives and high rank intellectual occupations, technicians and supervisors, vs non-executive, ie, office staff and blue-collar workers), alcohol use (every day vs not every day) and tobacco intake (current or in the past vs never). Measurement occasion (t1, t2, t3) was also incorporated in the statistical models. No attempt was made to control for retirement status, as this was highly positively correlated with age.
Mixed linear models were used to analyse the data in this study.20 A correction for the regression to the mean was applied to the cognitive scores at t1 (for the method and the rationale21). Statistical analyses were performed using Stata V.11.2.
Three sets of analyses were conducted which examined the effects on cognitive performance at all three measurement occasions (t1, t2, t3) of (1) shift work, (2) shift work exposure duration and (3) time elapsed (at t1) since having left shift work. Each analysis comprised two stages. In the first stage, an initial model was implemented with the shift work variable, measurement occasion and the following covariates: age, gender, socioeconomic position, sleep problem score, alcohol use, tobacco use, perceived stress. In the second stage, a full model was implemented that incorporated the significant predictors identified in the first stage, together with a set of interaction terms based on combinations of the significant predictors identified in the first stage that were relevant for the purposes of the current enquiry. At both stages, backward selection was used to identify the significant predictors of cognitive performance.
Our first set of analyses examined whether experience of any type of atypical work schedule (ie, shift work) at t1 (never vs current or past) affected global cognitive performance scores at t1, t2 and t3. The preliminary model (main effects) indicated that lower scores were predicted by ‘current or past’ shift work experience. Four significant interactions were kept in the final model (with interactions), but none involved shift work. Poorer global cognitive performance scores were again observed for ‘current or past’ shift workers as compared with those who had only ever worked as day workers (β=−1.62±0.367, p<0.0001; see figure 1). The effect of shift work can be compared with the differences in global cognitive performance scores observed at baseline between the age cohorts. In the final model, the effect of age was β=−11.40 (p<0.0001) for the 62-year cohort when compared with the 32-year cohort (ie, a decline in global cognitive performance score of 0.38 for each year). Thus, the cognitive impairment due to shift work was equivalent to 4.3 years of age-related cognitive decline, based on the comparison between 62-year-olds and 32-year-olds. The lack of a significant interaction involving shift work suggests that the effects of shift work were not influenced by any of the covariates or by measurement occasion. The same analyses were also conducted for the memory and speed subscores of performance and revealed the same result: poorer scores for ‘current or past’ shift workers as compared with those who had only ever worked as day workers (β=−1.33±0.37, p<0.0001 and β=−1.36±0.30, p<0.0001, respectively, for memory and speed performance). No interactions involving shift work were observed. It should be noted that since some individuals who had ‘never’ worked shift work at t1 may have become ‘current’ and then perhaps even ‘past’ shift workers at t2 and t3 (7 possible scenarios over the 3 measurement occasions), our results may have underestimated the magnitude of the performance deficits.
Our next analyses focused on rotating shift work (see Shift work subsection of Methods section for more details) and were conducted to examine the effects of shift work exposure duration. Participants were classified into three levels of exposure at baseline, namely: never worked rotating shifts (‘no exposure’ group); 10 years or less (‘≤10-year exposure’ group); more than 10 years (‘>10-year exposure’ group). The preliminary mixed linear analysis indicated that lower global cognitive performance was predicted by rotating shift work duration. The full model revealed three significant interactions, but again none involved shift work duration. Compared with those who were never exposed, rotating shift workers with over 10 years of experience had poorer cognitive scores (β=−2.46±0.51, p<0.0001), while those with 10 years or less of shift work experience showed the same, but non-significant, trend (β=−0.91±0.49, p=0.06; see figure 2). This was equivalent to 6.5 years of age-related cognitive decline for >10 years of exposure. Analyses of the subdimensions of the global score revealed the same result for memory scores, with poorer scores in the >10-year exposure group as compared with the no exposure group (β=−2.12±0.52, p<0.0001), while the difference between the ≤10-year exposure group and the no exposure group was not significant (β=−0.82±0.50, p=0.102, respectively). For speed scores, no significant difference was observed between the no exposure group and either the >10-year exposure group (β=−0.68±0.56, p=0.22) or the ≤10-year exposure group (β=0.09±0.55, p=0.87). However, there was an interaction between shift work exposure and the socioeconomic position indicating greater differences for the executive participants than for the non-executive participants in the comparisons between participants with no exposure and participants with >10 years exposure (β=−1.75±0.86, p<0.05), and between those with no exposure and those with ≤10 years exposure (β=−1.71±0.80, p<0.05). An interaction between exposure duration and the measurement occasion was also found, with participants who had >10 years exposure showing a significant decrease in the speed score between t1 and t2 (β=−1.81±0.51, p<0.0001), while no such decline was observed in the other two exposure groups.
Reversibility and recency
We then examined the possible reversibility of the chronic effect of shift work on cognition, by comparing performance differences between participants who had the following shift work statuses at t1: currently working rotating shifts (n=568); former shift worker, having left rotating shift work within the past 5 years (n=176); former shift worker, having left rotating shift work more than 5 years previously (n=350); and never worked any sort of shift system (n=1635). This is subsequently referred to as the effect of ‘recency’. The first model revealed a significant effect of recency of rotating shift work. In the subsequent model, three significant interactions were found, but none involved recency. Compared with participants who had never worked any sort of shift system, significantly poorer global cognitive performance scores were exhibited by those who were currently working rotating shifts (β=−2.30±0.50, p<0.0001) and by those who had left rotating shift work within the past 5 years (β=−2.74±0.80, p<0.001). The loss was equivalent to 5.8 years of age-related cognitive decline in our model for the current shift workers, and to 6.9 years for those who had left rotating shift work within the past 5 years. In contrast, those who had left rotating shift work more than 5 years previously did not differ significantly from those who had never worked any sort of shift system (β=−0.42±0.60, p=0.48). The same results were obtained in analyses based on the memory scores, with β coefficients in the same range as for the global cognitive performance score: β=−2.02±0.51, p<0.0001 for those currently working rotating shifts; β=−2.68±0.82, p<0.001 for those who had left shift work within the past 5 years; and β=−0.17±0.62, p=0.79 for those who had left shift work more than 5 years previously. For the speed score, results showed the same trend, though they did not reach significance: β=−0.97±0.50, p=0.054 for the current rotating shift workers; β=−1.56±0.85, p=0.07 for those who had left shift work within the past 5 years; and β=0.62±0.66, p=0.35 for those who had left shift work more than 5 years previously. An interaction was observed between recency and the socioeconomic position. Non-executive participants showed no effects of recency. However, among the executive participants, lower speed scores (relative to those who had never worked shifts) were observed among both current rotating shift workers (β=−1.88±0.87, p<0.05) and those who had left rotating shift work more than 5 years previously (β=−2.72±0.98, p<0.01).
As a check on the generality of the effects of recency of rotating shift work, we examined the effect of recency with respect to any of the four types of atypical work schedule, that is, participants were grouped with respect to the length of time that had elapsed since they ceased working any sort of shift system. The first model indicated that those who were currently working shifts (β=−2.67±0.58, p<0.0001) or who had left shift work within the past 5 years (β=−2.11±0.55, p<0.0001) exhibited significantly lower global cognitive performance scores, compared with those who had never worked shifts (see figure 3). The deficit was equivalent to 6.7 years of age-related decline for the current shift workers and to 5.3 years for those who had left shift work within the past 5 years. Again, those who had left shift work more than 5 years previously did not differ significantly from those who had never worked any sort of shift system (β=0.04±0.64, p=0.95), and there were no significant interactions involving recency. Similar results were obtained from the analyses based on memory scores: β=−2.29±0.60, p<0.0001 for the current shift workers; β=−1.93±0.57, p<0.001 for those who had left shift work within the past 5 years; and β=0.10±0.66, p=0.89 for those who had left shift work more than 5 years previously. For the speed score, a main effect of recency was not observed: β=−0.88±0.57, p=0.12 for those who had left shift work within the past 5 years; and β=0.84±0.72, p=0.24 for those who had left shift work more than 5 years previously, as compared with those who had never worked any sort of shift system. Only current shift workers significantly differed from those who had never worked any sort of shift system (β=−1.22±0.59, p<0.05). An interaction was found for speed scores between recency and socioeconomic position. While there were no effects of recency among non-executive participants, executive participants who were either currently exposed (β=−2.23±1.02, p<05) or who had left shift work more than 5 years previously (β=−2.25±1.05, p<05) had lower scores compared with those who had never been exposed.
Finally, in order to assess the possible influence of sample attrition during the course of the study, we compared (Fisher or Kruskal-Wallis test) the global cognitive performance scores at baseline of those individuals who were only present at t1 (ie, dropouts) with those who were also included in the analyses at t2 or t3. This was conducted separately for all relevant groups used in the analyses reported above (‘never’ and current or past shift work, ≤10 years and >10 years exposure duration, current or ≤5 years recency and >5 years recency). In all groups, those who had participated only at t1 showed systematically lower cognitive performance than those who were seen again at t2 or t3 (p range from 0.02 to 0.11), mainly because the dropouts were also a little older and less educated (p<0.0001). It thus seems unlikely that attrition biased our conclusions since the dropout effect impacted in the same direction all the groups that were compared with each other.
The current results indicated that (1) exposure to shift work was associated with a chronic impairment of cognition, (2) the association was highly significant for exposures to rotating shiftwork exceeding 10 years (with the exception of the speed scores among non-executive participants) and (3) the recovery of cognitive functioning after having ceased any form of shift work took at least 5 years (with the exception of speed scores).
The findings may reflect the disruption of the individuals’ circadian rhythms resulting in physiological stress, which has been shown to have an impact on brain structures involved in cognition and mental health over the lifespan.22 The apparent reversibility of the cognitive impairment found in the present study is consistent with the ‘stress—cortisol—atrophy of the hippocampus—cognitive impairment’ pattern observed in people submitted to repeated jet lag, because the hippocampus is a brain structure whose tissues seem to be able to regenerate through neurogenesis.23 Greater evidence was obtained of the effects of shift work in the memory scores than in the speed scores, especially in the analyses examining the persistence of the shift work effect after leaving shift work. This also provides support for the hippocampus hypothesis, as the hippocampus is known to be highly involved in memory processes. An alternative interpretation of the present results reflects the fact that shift workers show an increased incidence of metabolic syndrome,24 which has in turn been associated with impaired cognitive functioning.25 The current study lacked statistical power to satisfactorily assess the possible mediating role of the metabolic syndrome in the observed effects on performance. It has also been suggested that shift workers may be more prone to vitamin D deficiency because of their reduced exposure to daylight, and vitamin D deficiency has also been linked to impaired cognitive functioning.26
Unlike the study of acute effects, the direct study of the long-term consequences of atypical work schedules on the brain and cognitive functioning is complicated because of the great variability in the worker's history of atypical work schedules (possible multiple changes during the occupational life over a wide set of different shift systems). Hence, one limitation of the current study was that we were unable to conduct separate analyses on each type of atypical work schedule, thus obliging us to group them in some of the analyses. Thus, it was not possible to isolate which aspects of the atypical schedules were driving the observed effects on cognition. Conversely, the analysis of exposure duration focused exclusively on rotating shift work, and hence those results cannot necessarily be extrapolated to other forms of shift work. Another limitation was that some participants in the ‘never’ control group might have had minimal experience of shift work, insofar as the threshold for declaring experience of shift work had been placed at 50 days/year. However, if our control group was possibly slightly contaminated by shift work, this would suggest that, if anything, the current study underestimated the effects of shift work. Finally, although a causal effect of shift work on cognition seems highly plausible in the light of the long-term effects already observed on a variety of biological parameters, the reverse causal relationship cannot be excluded, at least for some participants. Indeed, it may be that those who quit shift work a long time ago may have had higher cognitive abilities and were thus better able to move into non-shift working jobs at an earlier stage in their career.
The cognitive impairment observed in the present study may have important safety consequences not only for the individuals concerned, but also for society as a whole given the increasing number of jobs in high hazard situations that are performed at night. It may also affect shift workers’ quality of life, with respect to daily life activities that are highly dependent on the availability of cognitive resources. The current findings highlight the importance of maintaining a medical surveillance of shift workers, especially of those who have remained in shift work for 10 years or more.
The authors are grateful to the occupational physicians and other researchers of the VISAT group who contributed to the VISAT programme.
Contributors J-CM designed the VISAT programme, supervised the data collection and managed the database. He contributed to the conception and design of the current study, advised on statistical aspects, interpreted the data and wrote several parts of the manuscript. PT conducted the preliminary analyses, advised on statistical aspects and contributed to the writing of the Methods section, to the interpretation of the data and to the preparation of the manuscript. SF contributed to the analysis and interpretation of the results, and to the preparation of the manuscript for publication. CG performed statistical analyses and participated in the writing of the Methods section. DA participated in the data collection and analyses, and to the formatting of the article. All authors had full access to all of the data in the study and can take responsibility for the integrity of the data and the accuracy of the data analysis. All authors gave final approval of the version to be published.
Funding This work was funded by grants from the French Agence Nationale de la Recherche (ANR 2006 SEST 04 101), the Institute of Occupational Safety & Health (IOSH, UK), and the Direction Générale de la Santé (DGS), la Caisse Nationale d'Assurance Maladie des Travailleurs Salariés (CNAMTS), du Régime Social des Indépendants (RSI) et de la Caisse Nationale de Solidarité pour l'Autonomie (CNSA), as part of the call for projects launched by Institut de Recherche en Santé Publique (IReSP) in 2009. All authors were independent from the funders.
Competing interests None.
Ethics approval The VISAT project was approved by the French National Committee for Computer Data and Individual Liberties, and all procedures were compatible with the international standards pertaining to human research.
Provenance and peer review Not commissioned; externally peer reviewed.
↵i Tables of the full set of results relating to all independent variables are available on request from the first author.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.