Do Shift Work and Long Hours in Pregnancy Affect Birth Outcomes?

Cai et al. (2019) screened 3305 unique citations and pooled 62 observational studies covering 196,989 women who engaged in paid work during pregnancy. Three exposures were compared against two reference groups. Rotating shift work meant shifts that change according to a set schedule, and a fixed night shift meant a typical working period between 11:00 pm and 11:00 am; both were compared against a fixed day shift, defined as a typical working period between 8:00 am and 6:00 pm. Longer working hours meant more than 40 hours per week, compared against standard hours of 40 hours per week or fewer. Every figure below is an odds ratio against those reference groups.

For rotating shift work, the review reported higher odds of gestational hypertension (OR 1.19, 95% CI 1.10 to 1.29, I-squared 0%) and of small for gestational age (OR 1.18, 95% CI 1.01 to 1.38, I-squared 0%). It also reported preterm delivery at OR 1.13 (95% CI 1.00 to 1.28, I-squared 31%), an interval whose lower bound sits exactly on 1.00, and preeclampsia at OR 1.75 (95% CI 1.01 to 3.01, I-squared 75%), an interval wide enough to run from almost no association to a near tripling of the odds. The authors attached a low to very low certainty grade to this rotating-shift block, so the qualifier belongs in the same sentence as the numbers rather than in a footnote.

For fixed night shifts against fixed day shifts, two pooled estimates were reported: preterm delivery (OR 1.21, 95% CI 1.03 to 1.42, I-squared 36%) and miscarriage (OR 1.23, 95% CI 1.03 to 1.47, I-squared 37%). Note that the preterm figure of 1.21 appears twice in this paper with different intervals, once here for fixed nights and once for long hours, so the two should never be blended or quoted with each other's confidence interval.

The working-hours comparison produced the tightest and most internally consistent set of estimates in the review. Against 40 hours per week or fewer, more than 40 hours per week was associated with higher odds of preterm delivery (OR 1.21, 95% CI 1.11 to 1.33, I-squared 30%), of low birthweight under 2500 g (OR 1.43, 95% CI 1.11 to 1.84, I-squared 0%), and of miscarriage (OR 1.38, 95% CI 1.08 to 1.77, I-squared 73%). Small for gestational age sits in the same block (OR 1.16, 95% CI 1.00 to 1.36, I-squared 57%), but its lower bound rests on 1.00, so it does not belong in a list of settled findings. The low-birthweight estimate is notable for showing no measurable heterogeneity between studies while still excluding the null.

The authors also fitted a dose-response analysis, which showed that women working more than 55.5 hours per week compared with 40 hours per week had a 10% increase in the odds of having a preterm delivery. Treat that as a modeled relationship rather than an observed comparison between two staffed groups, and note that the published abstract reports it without a confidence interval, so its precision cannot be checked from the same source as the pooled estimates.

The two null-touching estimates are worth returning to, because they are the ones most likely to be quoted as though they had settled something. Rotating shift work and preterm delivery (OR 1.13, 95% CI 1.00 to 1.28) and long hours and small for gestational age (OR 1.16, 95% CI 1.00 to 1.36) are each consistent with a small increase in odds and equally consistent with no increase at all. That is a weaker statement than saying the exposure raises the outcome, and pooling more studies did not turn it into a stronger one.

Heterogeneity splits the review into two different kinds of finding. The preeclampsia estimate is the weakest of the set: its interval runs from 1.01 to 3.01, nearly touching the null at one end and more than tripling the odds at the other, and I-squared of 75% means the included studies disagree markedly with each other. The long-hours miscarriage estimate carries similar disagreement at I-squared 73%. Contrast those with gestational hypertension and low birthweight, both at I-squared 0%, where the pooled studies line up. One review can contain both kinds of finding, and the difference matters more than the point estimates.

Scale is the other place these numbers get mangled. Every estimate here is an odds ratio against a named reference group, so the gestational hypertension figure of 1.19 belongs in a sentence about the odds of that outcome among exposed women, not in one about how many extra cases a roster should expect. An odds ratio only approximates a risk ratio under conditions the abstract does not report, and the review does not supply the absolute rates that would let a reader convert one into the other, so restating these figures as changes in risk goes past what the paper says.

All 62 studies were observational. There are no randomized trials, quasi-experiments, or policy evaluations in the pool, so nothing here establishes that a schedule change would alter a birth outcome. Pooling does not remove confounding by occupation type, socioeconomic status, or physical workload, and it does not correct the healthy-worker effect, where people who stay in demanding jobs differ systematically from those who leave. The authors' own conclusion is that pregnant women who work rotating shifts, fixed night shifts, or longer hours have an increased risk of adverse pregnancy outcomes, which describes the exposed group rather than asserting a cause.

Three limits are worth stating plainly before anyone cites this review. The population is women in paid employment during pregnancy across a wide mix of countries, industries, and decades, so the findings do not transfer to non-working pregnant women and do not license a claim about any single occupation or sector. Stillbirth, gestational diabetes mellitus, and intrauterine growth restriction were among the pre-specified outcomes, but no pooled estimates for them are available in the published abstract, so no result should be asserted for those three. And the literature search closed on March 15, 2019, so anything published since then is outside this review, even though it remains the standard reference on the question.