Does Shift Work Disrupt the Menstrual Cycle?

Hu et al. (2023) searched PubMed, Embase, Cochrane and Web of Science through December 2022 and pooled 21 studies covering 195,538 women, 16 cross-sectional and 5 cohort, always comparing shift workers with fixed-day-shift workers. Across the 16 studies reporting menstrual disorders, shift workers had 30% higher odds of irregular menstruation (OR 1.30, 95% CI 1.23-1.36, I-squared 41.9%, fixed-effects model). Odds are not risk, and the defensible sentence is 30% higher odds of reporting an irregular cycle rather than 30% more cases or 1.3 times the risk.

Two further outcomes were pooled. Dysmenorrhea, from the 8 articles the paper names as reporting it, reached OR 1.35 (95% CI 1.04-1.75) in the pooled multivariable analysis, with high heterogeneity at I-squared 73.0% and a random-effects model. Early menopause, from 2 cohort studies of women over 45 years old using Cox proportional hazards models, gave HR 1.09 (95% CI 1.04-1.14, I-squared 0.0%). That last figure is a 9% higher hazard resting on two studies in an older population, on a third scale again, and it carries nothing about workers in their twenties or thirties.

One warning about repeating these figures. The discussion section of the same paper restates the three results as likelihoods increasing by 1.30, 1.35 and 1.09 times, which is not what an odds ratio or a hazard ratio means. Anyone quoting this meta-analysis should work from the abstract and the subgroup tables rather than that sentence, because a hazard ratio restated as a likelihood multiplier overstates what the pooled data show.

The irregular-menstruation estimate carries a publication-bias signal that the authors report openly. The Egger test was significant (t = 2.63, P < 0.05) while the Begg test was not (z = 0.81, P > 0.05), and the authors treat the Egger result as the more informative one when the two conflict. Trim-and-fill added five virtual studies and returned OR 1.29 (95% CI 1.16-1.44). The point estimate barely moves while the interval roughly doubles in width, so the honest summary is around 1.3 with an interval of 1.16 to 1.44 once likely missing studies are imputed.

The counts come before the pattern, because the paper does not reconcile with itself here. Its selection text names 8 articles reporting dysmenorrhea, while its subgroup table splits 5 high-quality studies from 4 moderate-quality ones, one study more than that, and the design split behaves the same way with 8 cross-sectional studies plus a single cohort. The discrepancy sits in the source rather than in the transcription, so quote each subgroup row with the count printed beside it instead of describing the subgroups as a re-sorting of the same 8 studies.

With that stated, the dysmenorrhea estimate is the weakest number in this literature, and its quality gradient runs the wrong way. Among the 5 high-quality studies the pooled result was not significant (OR 1.15, 95% CI 0.65-2.02, P > 0.05), while only the 4 moderate-quality studies reached significance (OR 1.55, 95% CI 1.19-2.03). The single cohort estimate was also non-significant (OR 1.35, 95% CI 0.74-2.46), as were the rotating-shift-work subgroup (OR 1.26, 95% CI 0.78-2.05) and women aged 30 or under (OR 1.19, 95% CI 0.61-2.32). When the better-conducted studies show less, that is a reason for caution rather than confidence.

The irregular-menstruation subgroups hold together better. Every one of them was significant, including rotating shift work (OR 1.23, 95% CI 1.14-1.33), rotating night shifts (OR 1.34, 95% CI 1.16-1.55), the 14 cross-sectional studies (OR 1.34, 95% CI 1.26-1.44) and the 2 cohort studies (OR 1.24, 95% CI 1.15-1.34). A milder version of the same quality gradient still appears, with high-quality studies at OR 1.26 (95% CI 1.20-1.33) and moderate-quality studies at OR 1.71 (95% CI 1.44-2.03). Workers aged 30 or under showed a larger association (OR 1.70, 95% CI 1.23-2.34) on only 5 studies, which is a signal to watch rather than a finding to act on.

Hu et al. (2023) also report, in their discussion, an earlier pooled analysis of 12 studies by Chang and Chang (2021) that found OR 1.35 (95% CI 1.28-1.42) for irregular menstruation and OR 1.66 (95% CI 1.13-2.44) under age 30. Those figures are recorded here as they appear in that discussion rather than checked against the original paper, and overlapping reviews of overlapping literatures agree partly because they pool many of the same primary studies.

Stocker et al. (2014) pooled 16 independent cohorts from 15 studies covering 123,403 women, defining shift work as work outside 8:00 AM to 6:00 PM and menstrual disruption as cycles shorter than 25 days or longer than 31 days. Unadjusted, 16.05% of shift workers (2,207 of 13,749) met that definition against 13.05% of non-shift workers (7,561 of 57,932), giving OR 1.22 (95% CI 1.15-1.29, I-squared 0%) across 71,681 women. After confounder adjustment the association held but attenuated to OR 1.15 (95% CI 1.01-1.31), with the lower bound sitting on 1.01.

The other outcomes are where this review earns its credibility, because two of them failed. Infertility, defined as time to pregnancy beyond 12 months, looked large unadjusted at OR 1.80 (95% CI 1.01-3.20) with I-squared 94%, then fell to non-significance after adjustment (OR 1.11, 95% CI 0.86-1.44). Early spontaneous pregnancy loss, defined as loss before 25 weeks, showed no association with shift work overall (OR 0.96, 95% CI 0.88-1.05, I-squared 0%, across 23,604 women). Night shifts specifically did show one, OR 1.29 (95% CI 1.11-1.50) unadjusted and OR 1.41 (95% CI 1.22-1.63) adjusted.

The authors then draw the practical line themselves. Their conclusion reports an association between shift work and early reproductive outcomes and states that there is currently "insufficient evidence for clinicians to advise restricting shift work in women of reproductive age." The larger and more recent meta-analysis reaches a compatible directional conclusion, that shift workers have significantly higher odds of menstrual disorders, dysmenorrhea and early menopause, and likewise proposes no rule about who should be assigned to which shift.

Neither review pools a randomized trial, a quasi-experiment or a law evaluation, so the direction of the association is not established. Fourteen of the 16 studies behind the irregular-menstruation estimate were cross-sectional, capturing schedule and cycle at the same moment and mostly by structured questionnaire. Women with difficult cycles may leave shift work or avoid taking it in the first place, which would produce the same correlation with nothing running from schedule to physiology.

Confounding is acknowledged as unresolved. The included studies commonly adjusted for age, BMI, smoking, drinking, education, physical activity and sleep quality, while past disease history, work pressure and caffeine intake were usually ignored, so Hu et al. (2023) state that unmeasured confounding cannot be eliminated. They also flag a likely healthy worker effect, recall bias from self-reported schedules, follow-up bias, and two included studies with 100 or fewer participants. They describe the etiology of menstrual dysfunction as multifaceted and the association as potentially mediated by social, psychological, physiological and environmental factors.

There is also no usable dose-response curve. Two included studies suggested an exposure-response relationship between shift frequency and cycle length, but Hu et al. (2023) state they could not quantify it given the small number of studies and poorly defined cycle-length measures, so nobody can say how many nights corresponds to how much change. Coverage is English-language only, with 8 of the 21 studies from developing countries and the rest from developed ones, and the authors note cultural and geographic variation in how menstruation is reported.