How Common Is Shift Work Sleep Disorder, and Who Develops It?

The two figures most often quoted are not competing estimates of the same quantity. Pallesen et al. (2021) pooled 29 prevalence studies and reported 26.5%, 95% CI 21.0 to 32.8, which is raw symptom prevalence: the share of shift workers who meet criteria at all. Drake et al. (2004) reported approximately 10% of the night and rotating shift work population, an excess figure that accounts for the insomnia and sleepiness day workers report anyway.

One dataset can produce both numbers. A later clinical review co-authored by that survey's first author gives the breakdown by shift type: 14.1% to 32% of night workers (n = 174) and 8.1% to 26% of rotating shift workers (n = 360) met DSM-IV criteria (Wickwire et al., 2017). The top of each range is raw symptom prevalence and the bottom is what remains after subtracting the rate day workers reported, so the same 2,570-person survey supports either quotation depending on which end is taken.

Quoting the two side by side without that distinction reads as a contradiction. If the question is how many people on a night roster would screen positive, the raw figure is the relevant one. If the question is how much of that burden sits above the background rate in the adult population, the excess figure is.

The pooled 26.5% should be read as a central tendency across very different studies rather than a population rate. Cochran Q was 1,845.4 (df = 28, p < 0.001) and I-squared was 98.5%, which the authors describe as very high heterogeneity. Part of that spread is explained by method: studies applying ICSD-3 criteria and studies with larger samples each reported systematically lower prevalence (Pallesen et al., 2021).

Kalmbach et al. (2015) followed 96 normal sleeping non-shift workers, mean age 47.9, 62.5% female, all screened to exclude a lifetime history of insomnia or baseline excessive daytime sleepiness, and all of whom transitioned into rotating shift work over the following year. Eighteen new cases appeared, an incidence of 18.8% over the year in the overall sample. Because the group was selected to be low risk at the start, that figure reads as a floor rather than a typical incidence.

The characteristic that separated the two groups was sleep reactivity, the tendency for sleep to break down under stress, measured at baseline with the Ford Insomnia Response to Stress Test. Among those scoring 16 or higher, 30.8% (16 of 52) developed the disorder, against 4.5% (2 of 44) of the low reactivity group. Adjusted for gender and baseline sleepiness, the odds ratio was 5.59, 95% CI 1.08 to 28.97, p = .04, and 88.9% (16 of 18) of those who developed it had been classified as highly reactive at baseline (Kalmbach et al., 2015).

The interval matters more here than the point estimate. It barely excludes 1 and runs to nearly 29, which is what 18 cases among 96 people buys, and the authors flag their limited sample size as a constraint on statistical power. Quoting 5.59 without the interval attached overstates what was found. The direction repeats on the continuous measure, where each 1-point increase carried an odds ratio of 1.18, 95% CI 1.04 to 1.33, p < .01, but the size of the association is not pinned down by this study.

Baseline daytime sleepiness was the other pre-transition signal. Each 1-point increase on the Epworth Sleepiness Scale carried an odds ratio of 1.87, 95% CI 1.18 to 2.95, p < .01, in the covariate screening model, and 1.49, p = .02, in the final model, in a sample that had already excluded anyone with clinically excessive sleepiness before the transition. These are odds of onset among people who were already sleeping normally, and nobody in the study was assigned to a schedule.

A year into rotation, workers who had developed the disorder differed from asymptomatic rotating workers on every sleep measure reported: weekday total sleep time of 345 versus 408 minutes (d = .74), sleep onset latency of 54 versus 27 minutes (d = .87), wake after sleep onset of 42 versus 16 minutes (d = .98), and Epworth sleepiness of 8.06 versus 5.53 (d = .90). These are between-group differences at follow-up inside an observational cohort, not the effect of an assigned roster.

Mood separated by a similar margin. Depression scores on the QIDS-SR16 with sleep items removed were 7.27 versus 3.79 (d = 1.04) and Beck Anxiety Inventory scores were 10.67 versus 3.79 (d = .95), both at p < .001 (Kalmbach et al., 2015). A mediation analysis put the indirect path from sleep reactivity to depression through the disorder at 0.13 with a 95% CI of 0.002 to 0.322, and to anxiety at 0.40 with a 95% CI of 0.043 to 0.891. A mediation model assumes the ordering it decomposes; in an observational cohort it describes how the variables covary, not a demonstrated pathway. Both intervals exclude zero, though the depression interval barely does.

Drake et al. (2004) compared shift workers who met criteria against shift workers who did not, and reported ulcers at an odds ratio of 4.18, 95% CI 2.00 to 8.72. The comparator is the part that gets lost in retelling: this is a contrast against other shift workers, not against day workers or the general population, and the ulcers were self-reported rather than chart-confirmed. Sleepiness-related accidents, absenteeism, depression, and missed family and social activities were also more frequent in the affected group at P < .05, with no effect sizes given in the abstract.

That study also reported that in most cases the morbidity associated with shift work sleep disorder was significantly greater than that experienced by day workers with identical symptoms. That comparison is the paper's actual argument: the label appears to pick out something beyond the ordinary poor sleep that a share of day workers report too.

Christopher Drake is an author on the 2004 survey, on the 2015 cohort, and on the 2017 clinical review, and all three list the Henry Ford Hospital Sleep Disorders and Research Center in Detroit as their home. Wickwire et al. (2017) contributes no new prevalence data; it restates the 2004 survey's own figures by shift type, so two of the four citations on this page trace back to one telephone survey. Pallesen et al. (2021) shares no authors with the other three, and its pooled estimate corroborates the broad prevalence picture without resolving the definitional split.

Drake et al. (2004) surveyed one US metro area by random-digit dialing in the early 2000s, with 360 people working rotating shifts, 174 working nights, and 2,036 working days, using DSM-IV and ICSD era criteria. ICSD-3 later tightened the definition, and Pallesen et al. (2021) found ICSD-3 studies report systematically lower prevalence, so the older figures may overstate what current diagnostic standards would return. Occupational mix and scheduling practice have moved on as well.

Cross-sectional prevalence is also pushed downward by who is left to survey, because workers who tolerate shifts badly tend to leave shift work. Running the other way, Kalmbach et al. (2015) determined the disorder by self-report rather than clinical interview, collected no circadian biomarkers, and called for replication with clinician-diagnosed cases. Its one-year change in weekday sleep time did not reach significance, so that particular comparison should not be reported as a finding at all.

Nothing in this evidence identifies a safer rotation. Kalmbach et al. (2015) recorded neither shift timing nor rotation frequency, and the authors limit their conclusions to rotating shifts, stating that generalization to early morning, split, extended, or night shifts is untested. Anyone using these numbers to argue for one roster pattern over another is going past what the studies measured.