Does Night Shift Work Cause Breast Cancer?

Travis et al. (2016) pooled the three biggest UK prospective cohorts with data on night work: 522,246 Million Women Study participants, 22,559 in EPIC-Oxford, and 251,045 in UK Biobank, 795,850 women in total. Among women who reported night shift work, 673, 28, and 67 developed breast cancer, and the relative risks for any versus no night shift work were 1.00 (95% CI 0.92 to 1.08), 1.07 (95% CI 0.71 to 1.62), and 0.78 (95% CI 0.61 to 1.00) respectively. The team then meta-analyzed all 10 prospective studies published at the time, 1.4 million women with 4,660 breast cancers among those reporting night work, and found a combined relative risk of 0.99 (95% CI 0.95 to 1.03) for any night shift work.

Duration did not change the picture. The combined relative risk was 1.01 (95% CI 0.93 to 1.10) for 20 or more years of night shift work and 1.00 (95% CI 0.87 to 1.14) for 30 or more years. Inside the Million Women Study the estimates were similarly flat: 0.93 (95% CI 0.83 to 1.03) below 10 years, 1.14 (95% CI 0.96 to 1.35) at 10 to 19 years, 1.00 (95% CI 0.81 to 1.23) at 20 or more years, 0.98 (95% CI 0.69 to 1.39) at 30 or more years, and 1.10 (95% CI 0.94 to 1.30) for night work within the last 10 years, with a trend test at P = .68. Among long-serving nurses, the most exposed and most studied group, ever versus never night work gave 0.96 (95% CI 0.75 to 1.23) and 20 or more years gave 0.88 (95% CI 0.62 to 1.25). Note which number is which: 1.01 belongs to 20 or more years of exposure, and the headline figure for any night shift work is 0.99. Swapping the two is the easiest error to make with this paper.

Two features of the data cut against a clean reading. The UK Biobank point estimate of 0.78 sits below the null, which would imply night work protects against breast cancer; that is not biologically credible and points instead to healthy-worker selection and UK Biobank's documented volunteer bias, so it should never be presented as evidence of benefit. Travis et al. (2016) also describe the 20-or-more-years meta-analysis as showing no strong heterogeneity at P = 0.011, but by conventional thresholds P = 0.011 is significant heterogeneity, and the any-exposure figure at P = .052 is borderline. If you lean on how narrow these confidence intervals are, that qualification has to travel with them.

The exposure measurement carries its own limits. Night work was self-reported, the Million Women Study counted it as midnight to 06:00 on at least three nights a month, and definitions differ substantially across the studies pooled. The exposed group was also dominated by one occupation: 45.0% of all night shift workers and 61.1% of those reporting 20 or more years had worked as a nurse for at least 10 years. Every estimate here is for women, all three cohorts are UK based, and the paper predates both the IARC re-evaluation and the newer dose-response work. The authors concluded that the prospective evidence shows night shift work has "little or no effect" on breast cancer incidence and that classifying it as a probable human carcinogen was "no longer justified". That second claim is the contested one.

Three years after that paper, a Working Group of 27 independent international experts from 16 countries met in Lyon from 4 to 11 June 2019 to re-evaluate night shift work for the IARC Monographs. Their conclusion had four parts: limited evidence in humans, with positive associations observed for cancers of the breast, prostate, colon, and rectum; sufficient evidence in experimental animals for the carcinogenicity of alteration in the light-dark schedule; strong mechanistic evidence in experimental systems, based on effects consistent with immunosuppression, chronic inflammation, and cell proliferation; and an overall evaluation of "probably carcinogenic to humans (Group 2A)" (International Agency for Research on Cancer, 2020).

The Working Group did not overlook the null cohorts. It weighed them and decided which design to trust more. In its own account, the case-control studies were given greater prominence because of their stronger exposure assessments, the largest and highest-quality of those studies observed positive associations, and results were inconsistent among cohort studies, which is why the human evidence was graded limited rather than sufficient. Limited, in IARC's vocabulary, means a positive association has been observed but chance, bias, or confounding could not be ruled out. The classification is therefore a judgment about which exposure measurements deserve more weight, not a claim that the cohorts miscounted.

It would be easy to read this as one large study overturned by a committee that had not read it, and that is not what happened. The Volume 124 List of Participants and the Working Group's own signed summary both record Ruth Travis of the University of Oxford as a member, alongside Johnni Hansen and Eva Schernhammer (IARC Monographs Vol 124 Group, 2019). Hansen and Schernhammer had each contested the 2016 analysis in a signed letter to the Journal of the National Cancer Institute the year after it appeared, so the lead author of the null prospective study and two of its published critics sat on the same panel that graded the evidence.

Group 2A states how strong the evidence is that an exposure can cause cancer. It says nothing about how much cancer the exposure causes. IARC's own account of its Monographs program is that it identifies cancer hazards, meaning the potential for an exposure to cause cancer, and that it does not indicate the level of cancer risk associated with exposure at different levels. Two agents can sit in the same group and carry risks that differ enormously, so ranking night shift work against other Group 2A entries as though the label implied comparable danger is a category error, and converting the label into a personal risk percentage is not something the classification supports at all.

The distinction matters at scale, because IARC estimates that about 20% of workers in North America, Europe, and elsewhere are employed outside the standard daytime work shift. The animal evidence does not close the gap either: sufficient evidence in experimental animals is a causal statement about rodents kept on altered light-dark schedules, and it does not convert into a number for humans. Every human figure quoted on either side comes from workers whose rosters were set by employers rather than by researchers, so all of it stays associational. Workers who report night shift work also report certain cancer outcomes at different rates than workers who do not, and that is as far as the human data can carry anyone.

Moon et al. (2024) fitted a two-stage dose-response model separately by design, using 10 cohort reports (15,953 cases across 6,812,138 person-years) and 11 case-control reports (9,196 cases and 12,210 controls), with a literature search running to 30 December 2022. The fitted cohort curve gave pooled risk ratios that rose with cumulative exposure: 1.04 (95% CI 1.01 to 1.07) at 10 years, 1.09 (95% CI 1.03 to 1.15) at 20 years, and 1.13 (95% CI 1.04 to 1.23) at 30 years, about 13% higher risk after three decades of night work. These are predictions from a curve rather than observed pooled estimates, so the fourth-decimal precision in the published table should not travel with them.

From the case-control reports the same model gave fitted odds ratios of 1.23 (95% CI 1.11 to 1.37) at 10 years, 1.52 (95% CI 1.24 to 1.88) at 20 years, and 1.88 (95% CI 1.38 to 2.57) at 30 years, about 88% higher odds after the same three decades. Read the scales exactly as written: the cohort figures are risk ratios and the case-control figures are odds ratios, and higher odds are not the same thing as that many more cases. The usual explanation for the gap is recall bias inflating case-control estimates while exposure misclassification dilutes cohort estimates, but Moon et al. (2024) do not adjudicate it and say the reason should be investigated further. Report both lines side by side; averaging the designs or quoting only the larger figure would misrepresent the field.

Two cautions belong with those numbers. The one-year cohort estimate of 1.0042 (95% CI 1.0014 to 1.0070) excludes 1 only because a dose-response curve was fitted across the whole exposure range, which makes it a modeling artifact rather than evidence that a single year of night work measurably changes anything. And Moon et al. (2024) claim no causality: they list survivorship bias, residual variation between studies, and unmeasured confounders including environmental estrogens, type 2 diabetes, obesity, and hormone-disrupting personal care products. Their analysis also graded the Million Women Study as reliable and included it as a cohort input, so this is not a rival body of evidence so much as a different way of pooling much of the same evidence.

For anyone running a night roster, three findings sit side by side and do not collapse into one sentence. The largest prospective evidence shows little or no association between night shift work and breast cancer incidence. A standing international expert evaluation says the hazard evidence is limited but sufficient to keep the probable-carcinogen label, on the strength of the studies with better exposure measurement. And the two study designs feeding both conclusions disagree by a factor nobody has yet explained. Nothing on this page reflects evidence published after December 2022.