A prospective study of fatal occupational accidents – relationship to sleeping difficulties and occupational factors


T. Åkerstedt IPM/Karolinska Institute, Box 230, 17177 Stockholm, Sweden. Tel.: +46 8 7286945; fax: +46 8 320521; e-mail: torbjorn.akerstedt@ipm.ki.se


Very little is known about the association between sleep and (fatal) occupational accidents. This study investigated this relationship using register data of self-rated sleep difficulties, together with occupational and demographic characteristics. The variables were related to subsequent occupational fatal accidents. A national sample of 47 860 individuals was selected at regular intervals over a period of 20 years, and interviewed over the phone on issues related to work and health. The responses were linked to the cause of death register (suicides excluded) and the data set was subjected to a (multivariate) Cox regression survival analysis. One hundred and sixty six fatal occupational accidents occurred, and the significant predictors were: male vs. female: relative risk (RR)=2.30 with a 95% confidence interval (CI) of 1.56–3.38; difficulties in sleeping (past 2 weeks): RR=1.89 with CI=1.22–2.94; and non-day work: RR=1.63 with CI=1.09–2.45. No significant effect was seen for age, socio-economic group, hectic work, overtime (>50 h per week), or physically strenuous work. It was concluded that self-reported disturbed sleep is a predictor of accidental death at work, in addition to non-day work and male gender.


Fatal accidents can be related to a number of causes (Baker et al. 1992). One potential cause that has been relatively little explored is disturbed sleep. In a prospective study, Kripke et al. (1979) found that individuals with very short or very long sleep had an increased mortality in a number of diagnoses, including accidents. Martikainen et al. (1998) used a 5-year follow-up and found changes in sleepiness related to road accidents. Roth and Ancoli-Israel (1991) used cross-sectional data and demonstrated a connection between disturbed sleep and accidents.

Disturbed sleep because of night work or long hours has been implicated in several major accidents, such as the nuclear plant accidents at Three-Mile Island and Chernobyl, as well as in transport disasters such as the grounding of the oil tanker Exxon Valdez (Dinges 1995; Mitler et al. 1988). Similar conclusions, based on risk analysis, have been obtained in industrial shift work (Smith et al. 1994), and in studies of the pattern of sleep-related road accidents (Horne and Reyner 1999). Attempts at estimating the costs to society of accidents and other effects of disturbed sleep have produced figures around $50 billion per year (Leger 1994).

On the other hand, apart from the study by Kripke et al. (1979), there has been no prospective epidemiological study on sleep and fatal accidents. And there seems to be no study at all on disturbed sleep and fatal accidents at work. The present study sought to approach this latter issue by using the regularly occurring Swedish `Living Conditions Survey' (ULF) in combination with the official cause-of-death registers. It was of particular importance to evaluate potential effects of sleep against other possible contributors to accident risk at work, such as non-day work, overtime work, stress, physical work load, and being a blue- or white-collar worker. Age and gender were also used as predictors.


The design used was an open cohort study with repeated national cross-sectional surveys, focusing on living conditions. These cross-sectional surveys utilize data obtained from the National Survey of Living Conditions (ULF), conducted annually by Statistics Sweden. The ULF study comprises representative samples of the gainfully employed in Sweden. Each individual participated in a 1-h face-to-face interview. The study base comprises all gainfully employed males and females in Sweden. Each survey includes 4773–11 783 respondents. The non-response rates vary between 14 and 23%, depending on the year.

Eleven surveys were selected for this study: 1977, 1979–81, 1987–89, 1994–96. These constitute all the surveys that included the items of interest for the present study. Altogether, 47 860 responders participated. The surveys were subsequently linked with the Swedish Cause-of-Death register. The 1977 survey was followed up until 1987 and the rest were followed up until 1996.

The outcome measures were: death from injury, underlying cause-of-death, chapter XVII (N800-959) according to the 8th and 9th revisions of the International Classification of Diseases (ICD). Cases of suicide and poisoning were excluded. Exposure was estimated through the items: `Have you had difficulties in sleeping during the last two weeks?' (Yes/No). The number of exposed individuals was 5659.

In addition, the following socio-demographic variables were included: gender (male vs. female), age (16–29, 30–49 vs. 50+ years), socio-economic group (unskilled workers, skilled workers, lower non-manual workers, and intermediate manual workers vs. high level non-manual workers). The work environment variables were: more than 50 work hours weekly (vs. less), physically strenuous work (at least one of: `repeated and one-sided movements', `twisted work position', `breaking into sweat each day', `shaking or vibrations', or `heavy lifting', vs. none), `hectic work' (vs. not) and `non-daytime work' (either of shift, night, early morning or evening work vs. daytime work).

Survival analyses were performed by Cox regression using PROC PHREG in SAS, version 6.12 (SAS/STAT Software, Changes and Enhancements through Release 6.12, SAS Institute AND Cary, USA, 1997). The major advantage with this method it that it allows data to be either left- or right-censored. This means that it permits a person to enter a study after it has been implemented and to leave before it is completed (Cox and Oakes 1984).


The number of fatal accidents was 166 (0.3%), and the number of individuals not having had such an accident was 47 694 (99.7%). The total sum was made up of 124 men and 42 women. Table 1 shows the prevalence of variables representing exposure.

Table 1.    Prevalence for each variable as well as (mutually adjusted) relative risk (RR) and 95% confidence interval (CI) for predictors of fatal accidents Thumbnail image of

The crude risk ratio showed a clear effect for difficulties in sleeping (relative risk=1.76 with a 95% confidence interval of 1.18–2.63). Table 1 shows the multivariate risk ratio with confidence intervals for all variables. The analysis showed a significantly increased risk for sleep problems, male gender, and non-daytime work.


`Difficulties in sleeping during the last two weeks' was associated with an increased risk of having a fatal occupational accident, as was being male and having non-daytime work. The contribution by gender to accident mortality is well established (Baker et al. 1992). Men, as a rule, have a higher accident risk.

The clear predictive power of sleep problems agrees with the early study by Kripke et al. (1979), although, that study used as a predictor the amount of sleep, not the reported sleep disturbance. That study also focused on overall accidental death, whereas the present study adds the dimension of death at work. The present study attempted to control for a number of occupational variables likely to be related to accidents. The results suggest that the effect of disturbed sleep seems to be independent of other factors, such as work hours, stress, etc.

Presumably, the mechanism behind the effect of disturbed sleep involves a lack of sufficient alertness at work or during leisure time, as lack of sleep is a powerful source of sleepiness (Dinges et al. 1997). However, sleepiness was not measured in the present study, and we did not include other mental or somatic states that may be related to disturbed sleep. It is possible, for example, that depression, anxiety, or similar states may have been involved. It is also a possibility that consumption of hypnotics may have contributed to our findings, even though drug related deaths (including suicides) were excluded. The present data do not permit testing of such a hypothesis.

The observation that non-day-work hours are associated with fatal accidents agrees with other studies of increased accidents at night. For example, non-fatal accidents among shift-workers in automotive plants (Smith et al. 1994) and road accidents with injuries (Horne and Reyner 1999). However, the present data demonstrate a difference between day and shift workers, with socio-economic group and, specifically, hectic work and physical workload, all controlled for. Presumably, the reason for the increased risk is the excessive tiredness associated with such work hours (Åkerstedt et al. 1983; Coleman and Dement 1986; Kogi and Ohta 1975; Martikainen et al. 1998; Prokop and Prokop 1955). The effects of sleep loss should have been accounted for by our survey item relating to disturbed sleep. However, a substantial proportion of night-work related tiredness/sleepiness may well be because of working at the circadian nadir (Dijk and Czeisler 1995).

It should be emphasized that the present study made use of rather crude indicators of disturbed sleep and that it only represented the occurrence of symptoms during the immediately preceding 2 weeks. This makes it all the more surprising that a predictive ability could be demonstrated, and the observation warrants further studies into the mechanism behind sleep and fatal accidents.


This work was supported by the Swedish Work Environment Fund.