Does Overtime Cause Injuries, or Do Hazardous Jobs Use More of It?

Dembe and colleagues followed 10,793 participants in the US National Longitudinal Survey of Youth from 1987 to 2000, covering 110,236 job records and 89,729 person-years of accumulated working time. Respondents reported 2,799 injuries or illnesses in jobs that carried one of the long-hours exposures and 2,339 in jobs that did not.

In the adjusted models, built on 109,087 job records and controlled for age, gender, occupation, industry, and region, overtime carried a hazard ratio of 1.61 (95% CI 1.43 to 1.79). Shifts of at least 12 hours a day came in at HR 1.37 (95% CI 1.16 to 1.59), and weeks of at least 60 hours at HR 1.23 (95% CI 1.05 to 1.45). Any of those exposures pooled together gave HR 1.38 (95% CI 1.25 to 1.51).

That adjustment is aimed at one objection: that long-hours jobs look dangerous because they cluster in dangerous industries and occupations. The authors took on a second objection too, that people who work more hours also spend more total time exposed. Their rates are expressed per 100 accumulated worker-years, and on that basis they concluded that the excess is not explained by more time at risk. The crude gap behind that reasoning is 7.49 injuries per 100 worker-years in jobs with overtime against 4.06 in jobs without, 84% higher before any adjustment. The caveat a reader should carry is that calendar person-time does not fully equalize the hours at risk inside a 60-hour week and a 40-hour week, so the objection is answered by the authors rather than closed by the data.

On the confounding question, the authors concluded that jobs with long hours were not riskier merely because they cluster in inherently hazardous industries or occupations. They were equally explicit about the limits of that conclusion, stating that they "cannot be certain of a causal connection" from this study alone (Dembe et al., 2005). It is an adjusted cohort answer to two objections, not an experiment.

Matre and colleagues screened 10,072 records through December 2020 and pooled 22 observational studies with populations ranging from 97 to 150,438 workers. Twelve studies were rated moderate risk of bias and 10 high risk, and heterogeneity across analyses reached an I-squared of 92.80%.

The figures the published abstract headlines come from the restricted analysis of moderate-risk-of-bias studies. Days of more than 12 hours pooled at RR 1.24 (95% CI 1.11 to 1.40), weeks of more than 55 hours at RR 1.24 (95% CI 0.98 to 1.57), days of more than 8 hours at RR 0.93 (95% CI 0.72 to 1.19), overtime versus no overtime at RR 1.08 (95% CI 0.75 to 1.55), 41 to 48 hour weeks at RR 1.02 (95% CI 0.92 to 1.13), and 49 to 54 hour weeks at RR 1.02 (95% CI 0.97 to 1.07). Only the first of those excludes no effect.

Pool all the studies instead, and two of those results move in the opposite direction. Days of more than 12 hours become non-significant across 8 studies (RR 1.24, 95% CI 0.85 to 1.81), while weeks of more than 55 hours become significant across the five studies with usable risk ratios (RR 1.42, 95% CI 1.06 to 1.91); a sixth study reported that contrast as cubic regression coefficients and was left out of the pool. Quoting either set as the answer is a decision about which studies to trust rather than a reading of the literature, and Matre et al. (2021) graded certainty as low for every contrast and very low for 41 to 48 hour weeks.

Two details inside the review are worth holding onto anyway, as long as you track where each one comes from. One included study (Weaver) tested dose-response inside the long-week range and reported RR 1.41 (95% CI 1.22 to 1.64) for 70 to 80 hours a week against 70 or fewer, and RR 1.78 (95% CI 1.53 to 2.07) for more than 80 hours against 80 or fewer. Those are one study's estimates against its own reference weeks, not pooled figures against a normal-length week. Separately, very long shifts of more than 20 to 24 hours, drawn from two studies in the same population, were associated with percutaneous injury at RR 1.61 (95% CI 1.46 to 1.78), vehicle crashes at RR 2.30 (95% CI 1.60 to 3.30), and near misses at RR 5.81 (95% CI 5.32 to 6.19).

Harma and colleagues approached the same question from a different angle, using a case-crossover design on 18,700 occupational injuries among hospital employees across 11 towns and 6 hospital districts. Each injured worker's payroll record in the 37 days before the injury was compared against that same worker's earlier days, so stable differences between people and between jobs drop out of the comparison by construction.

In that design, the length of the work shift was associated with injury (OR 1.22, 95% CI 1.06 to 1.42) while the length of weekly working hours was not associated with increased risk. The number of preceding night shifts and quick returns showed no increase either, although risk did rise after a week containing 5 or more morning shifts or 3 or more evening shifts.

The largest odds in that study concerned timing rather than volume. Workdays with evening shifts carried OR 1.09 (95% CI 1.03 to 1.14) and workdays following night shifts OR 1.33 (95% CI 1.17 to 1.52). Excluding commuting injuries raised both, to OR 1.15 (95% CI 1.09 to 1.23) and OR 1.44 (95% CI 1.24 to 1.69). These are odds ratios, so they describe higher odds of an injury day, not a count of extra injuries.

Read together, the three studies agree more about long individual shifts than about long weeks. Dembe et al. (2005) found a larger adjusted hazard for days of 12 hours or more than for weeks of 60 hours or more. The most stable positive estimate in Matre et al. (2021) is also the daily one. Harma et al. (2020) found shift length associated with injury while weekly hours were not. Three different designs converging is meaningful, and all three are still observational.

None of the three studies here evaluated an hours cap, and none observed injuries before and after a reduction in hours, so nothing here shows what happens when overtime is taken away. In Dembe et al. (2005) both the exposure and the injury were self-reported at interview with one-year or two-year recall and no external validation, and education, income, family composition, and health status were not adjusted for. There is no information on time of day, job activity, or specific cause of injury, and controls at the occupation and industry level can hide differences within a category.

The cohort is a single US birth cohort, Americans aged 14 to 22 in 1979, so participants were roughly 22 to 43 years old across the study window rather than a full-age-range workforce. The data end in 2000. Injury rates across exposure categories fell 54% to 69% between 1988 and 2000, a reminder that the underlying risk environment was moving quickly even inside the study period.

The wider evidence base skews to healthcare. Most studies included in Matre et al. (2021) covered healthcare workers, and 15 of the 22 mixed day and non-daytime workers, which confounds hours with time of day. Harma et al. (2020) covers hospital employees only. One null result travels with all of this as a useful check: extended commutes of more than 2 hours showed no association with injury in Dembe et al. (2005), at HR 0.87 (95% CI 0.59 to 1.23), so a long day away from home is not by itself the exposure that these data flag.