Global, regional, and national burdens of ischemic heart disease and stroke attributable to exposure to long working hours for 194 countries, 2000–2016: A systematic analysis from the WHO/ILO Joint Estimates of the Work-related Burden of Disease and Injury

1The protection and promotion of occupational and workers’ safety and health requires actions to prevent exposures to occupational risk factors. One such occupational risk factor is exposure to long working hours. The very first International Labour Standard, Hours of Work (Industry) Convention, 1919 (No. 1), and several other international labour standards adopted since have limited hours of work. In 2019, countries renewed their commitment to ensuring maximum limits on working time in the Centenary Declaration on the Future of Work.

2The Hours of Work (Industry) Convention provides that the working hours of employed persons shall not exceed 8 hours/day and 48 hours/week (with some exceptions). In some countries, however, the definition of long working hours depends on national regulations. Nevertheless, many countries define standard working hours as 35–40 hours/week and working ≥41 hours/week as overtime work. Occupational epidemiologists often categorize long working hours into the three analytical categories of 41–48, 49–54, and ≥55 hours/week, and compare these to standard working hours (35–40 hours/week). After average working time decreased steadily over the second half of the 20th century in most countries, this overall downward trend ceased and even began to reverse in some countries during the 21st Century. As new information and communication technologies revolutionize work, working time is predicted to further increase for some industries.

3Evidence from previous studies suggests working long hours can increase mortality and morbidity from ischemic heart disease and stroke through psychosocial stress. Two primary causal pathways are conceivable (Fig. 1). The first is through biological responses to psychosocial stress: release of excessive stress hormones due to working long hours may trigger functional dysregulations in the cardiovascular system and structural lesions. The second pathway is through behavioral responses to stress that are established cardiovascular risk factors, including tobacco use, harmful alcohol use, unhealthy diet, physical inactivity, and resultant impaired sleep.

4Some long working hours categories are associated with a higher risk of ischemic heart disease and stroke events. The World Health Organization (WHO) and the International Labour Organization (ILO), supported by large Working Groups of individual experts, have conducted systematic reviews and meta-analyses of the relative risks of ischemic heart disease and stroke among persons working 41–48, 49–54, and ≥55 hours/week, compared with persons working 35–40 hours/week (Table 1). Confounding was considered by, at least, sex, age, and an indicator of socioeconomic status (SES; e.g., income, education, or occupational grade). The Working Groups of individual experts judged there to be sufficient evidence that working ≥55 hours/week is associated with a higher risk of both ischemic heart disease and stroke, compared with working 35–40 hours/week (Table 1). To our knowledge, these are the first systematic reviews on these topics that are based on pre-published protocols; applied a dedicated systematic review framework (the Navigation Guide; ); and formally proceeded to assess quality and strength of evidence based on pre-published criteria and methods.

5Countries, along with relevant international and regional organizations, and social partners, would benefit from accurate and transparent estimates of long working hours exposure and the attributable burden of disease. These provide the evidence base for designing, developing, planning, costing, implementing, and evaluating legislation, regulations, policies, programmes, and interventions to mitigate occupational risk factor exposure and its attributable disease burden. This article presents WHO/ILO Joint Estimates of the Work-related Burden of Disease and Injury (WHO/ILO Joint Estimates). These are estimates of exposures to long working hours (≥55 hours/week), for 194 countries, and those of the burdens of ischemic heart disease and stroke attributable to these, for 183 countries. All estimates are reported at the global, regional (WHO regions), and national levels, by sex and age group, for the years 2000, 2010, and 2016.

6WHO and ILO produced the estimates as part of the global Comparative Risk Assessment (e.g., ), which systematically estimates health burden arising from exposure to risk factors at levels above a counterfactual distribution. We developed definitions of the occupational risk factor, risk factor levels, and theoretical minimum risk exposure level (Table S1 in Supplementary data 1). Estimates of attributable burden were based on combining the distribution of exposure in the population with relative risks of health outcomes due the effects of exposure. This combination is then used to calculate the population-attributable fractions (PAFs). PAFs quantify the proportion of death or disease that is attributable to a specific risk factor, and, therefore, the respective reduction that would be expected if exposure was removed or present at a reduced level. Estimates were produced and are reported for populations defined by country, sex (i.e., three categories: both sexes, female, and male), and age group (i.e., 18 categories: ≥15, 15–19, 20–24, …, 90–94, ≥95 years). The estimates are reported according to the Guidelines for Accurate and Transparent Health Estimates Reporting (GATHER). The complete script used to generate the estimates in the computer software R can be found in Supplementary data 2.

7The estimates were produced using six sets of input data. Details of these and their sources are shown in Table 2. An overview of how data sources, input data, and models were combined to produce model outputs and burden estimates is provided in Fig. S1 (Supplementary data 1). Additionally, Tables S2–S5 provide detailed descriptions of the databases compiled from surveys (see Supplementary data 1). Briefly, data on the proportions exposed were obtained from Input Data 1 and 2 (databases compiled from survey data); estimates of the total population (Input Data 3) and the probability of death (Input Data 4) came from the United Nations population prospect and WHO life tables, respectively; estimates of relative risks of the association between exposure to long working hours and the two health outcomes (Input Data 5) came from the WHO/ILO systematic reviews; and estimates of the total number of deaths and disability-adjusted life years (DALYs) for ischemic heart disease and stroke by country, sex, and age group for the years 2000, 2010, and 2016 (Input Data 6) were sourced from the WHO Global Health Estimates.

8The WHO/ILO systematic reviews generated estimates of relative risk of the effect of exposure to long working hours on both morbidity (disease incidence) and mortality, which were used to produce estimates of burden of disease (Fig. S1; Supplementary data 1). As in other burden of disease studies, we only proceeded to estimation if the effect estimate identified in the systematic review was statistically significant (P < 0.05) and the strength of evidence was rated as “sufficient evidence for harmfulness”. We applied the following a priori criteria to select the “best” effect estimate (relative risks for morbidity versus mortality) based on strength of evidence ratings: First, if there is any evidence for fatal or non-fatal events of the outcome that was rated as “sufficient evidence for harmfulness” (using standard Navigation Guide ratings as per ), proceed to selection of “best” estimate. Second, if there is such sufficient evidence:either only for fatal events or only for non-fatal events of the outcome (i.e. not both), then the relative risk for the event type with sufficient evidence is selected.in the event that both fatal and non-fatal events have sufficient evidence, estimates for fatal events are prioritized.

9either only for fatal events or only for non-fatal events of the outcome (i.e. not both), then the relative risk for the event type with sufficient evidence is selected.

10in the event that both fatal and non-fatal events have sufficient evidence, estimates for fatal events are prioritized.

11For exposure to ≥55 hours/week and ischemic heart disease, strength of evidence was rated “sufficient evidence of harmfulness” for mortality ( and Table 1), so the relative risk for fatal events (RR 1.17, 95% confidence interval 1.05–1.31) was used as the best estimate. For exposure to ≥55 hours/week and stroke, the strength of evidence was considered “inadequate evidence of harmfulness” for mortality, but “sufficient evidence for harmfulness” regarding morbidity ( and Table 1); the relative risk for non-fatal events (RR 1.35, 95% confidence interval 1.13–1.61) was therefore the best estimate we used.

12Because there was no evidence for effect modification by country, sex, and age group, neither with regard to ischemic heart disease nor with regard to stroke, the best estimate was assigned to all cohorts defined by country, sex, and age.

13The estimation strategy comprised modelling the input data in four distinct models that consecutively built on each other (Models 1–4). These are described below.

14For each year between 1980 and 2016, for each population defined by country, sex, and age group, we produced estimates of the proportion of the population in each of the six standard working hours categories. Because we had microdata on the number of hours worked (direct exposure data), we produced these estimates without needing to consider proxies (e.g., occupation and number of years of labour market activity in the population). As for other WHO estimates, we modelled this proportion based on Input Data 1 using an established multilevel model that predicts the prevalence over time and in the geographical region. This method is used by WHO to produce Sustainable Development Goal indicators (e.g., indicator 3.9.1; see https://unstats.un.org/sdgs/metadata/files/Metadata-03–09-01.pdf) and has therefore passed the approval of the United Nations Statistical Commission. It creates continuous estimates over the specified time period and estimates an average intercept and an average slope with residual variances across countries. Where there is reliable survey information of a specific country, the country curve will closely follow the data; where there are few country data points or high within-country variability, estimates will be close to survey data points, but the trend will tend to follow the overall mean.

15The model used is described below (Model 1):where proportioni is the proportion of the population in the working hours category i in a given population defined by country, sex, and age group; year is the survey year; and sex, age, and country are the sex, age group, and country of the population. Ai(sex, age, country) and Bi(sex, age, country) of the year, dependent on sex, age, and country, are estimated using a multilevel model with country as fixed effects, and with sex and age as random effects, nested in the country within the region (with regions treated independently). Because proportioni was strongly non-linearly dependent on age, we linearized age by 5th order orthogonal polynomials to prevent collinearity. For countries and years without any data on long working hours, estimates were produced using multilevel modelling based on data from countries with such data.

16For each population defined by country, sex, and age group, we estimated the probability of transitioning from one of the six working hours categories in yeart to one specific working hours category in yeart+1. The j denotes one of 36 possible transitions. We adopted the methods developed by Eurostat for calculating these transition probabilities. Using Input Data 2, we scaled the survey weights for the target year (yeart+1) to represent the correct labour market status by country, sex, and age group for the index year (yeart) and the target year and then adjusted the complete sample in the target year to match margins for labour market status in both years, using iterative raking by sex. We modelled the probability based on Input Data 2 using the following multinomial logit regression model (Model 2):where is the set of regression coefficients describing the longitudinal weights associated with the transition j; is a set of explanatory variables (sex, and age as a 2-degree fractional polynomial associated with the transition j); and the summation (index α) goes through all possible transitions of j (except the transition from i = 0 in yeart to i = 0 in yeart+1, which was chosen as a pivot outcome).

17With Model 2, WHO and ILO derived 15,900 transition probabilities for the 15 countries with data in Input Data 2. In addition, Eurostat derived 31,104 transition probabilities covering 27 countries based on sub-samples of the European Union Labour Force Surveys using Model 2, and it shared these transition probabilities with WHO and ILO. For populations defined by country, sex, and age group for whom probabilityi could not be calculated because the required longitudinal data were unavailable, probabilityi was imputed. The imputed probabilityi was the mean of all transition probabilities of the population defined by the same sex and age in the region, weighted by the number of observations contributing to the transition probabilities.

18For each population defined by country, sex, and age group, we estimated the proportion of the population in each working hours category (k) over the time window of exposure. We defined k as the highest working hours category i in any year in the time window. Our modelling assumptions and the evidence supporting these are presented in Table 3. The approach assumed a 10-year latency between exposure and occurrence of clinical disease, and it adopted a 10-year exposure time window (centred on exposure year) to identify the highest exposure category to assign to the relevant exposure year (Fig. S2; Supplementary data 1). For example, to estimate the burden in 2016, the relevant exposure year is 2006; we assume the exposure for a person is the highest exposure category to which this person belonged during the time window of exposure, 2001–2010. In effect, this considers occupational turnover, with someone exposed at any time during the 10-year window considered at-risk for 2006.

19We used microsimulation, a method for generating micro-level estimates by combining individual- and aggregate-level datasets. For each country, we initiated a synthetic population, using Input Data 3 to ensure a representative sex and age distribution at the first year of the time window (i.e. estimation year minus 15 years), and estimates outputted from Model 1 to probabilistically assign each individual to a working hours category (i) at the first year. Using transition probabilities outputted in Model 2, for each year over the entire time window, transitions from one working hours category to another were stochastically modelled to estimate each synthetic individual’s working hours category in each year. Using Input Data 4, at each year step from the first year of the time window to the estimation year, each individual is stochastically assigned to the states of “died” or “alive”. All synthetic individuals that reach the state “died” before the estimation year are censored. Other possible exits from the population, such as outmigration from the country, were not considered. Using this microsimulation method, is derived using the following model (Model 3):where the summation runs through individuals “alive” in the estimation year; is the Kronecker delta function; Sl is the sequence of the lth individual of all working hours categories (i) in each year in the time window; and max denotes that the highest i that the lth individual experiences in the sequence is assigned.

20The Comparative Risk Assessment framework is used to estimate the burden of disease attributable to exposure to long working hours. We estimate the proportional reduction in death or disease that would occur if exposure was reduced to a level with a minimum risk (i.e. working 35–40 hours/week), while other conditions remain unchanged. The population distribution of exposure to the risk factor is combined with the increased risk of acquiring the disease or dying fromit that was attributable to exposure to the risk factor.

21Using estimates outputted from Model 3 and Input Data 5, we calculated the population-attributable fraction (PAF), the proportion of the health loss seen in a given population that can be attributed to exposure to the specific occupational risk factor, using Model 4:where Pk is the proportion of the population in a working hours category k; RRk is the relative risk for the working hours category k; and n is the total number of long working hours categories. Applying this fraction to the total disease burden (Input Data 6), gives the population attributable disease burden. To be specific, for each cause and risk factor category, the PAF was calculated for each cohort defined by country, sex, and age group; this was then applied to the total disease burden envelope for the cause also by country, sex, and age cohort to get the respective burden for this specific cohort. As there is no evidence to suggest that the relative risk for mortality and morbidity is different, we assume the PAFs are the same for non-fatal and fatal health events, and apply the same individual cohort-level PAFs to generate numbers of deaths and DALYs; however, once individual cohorts are combined (e.g., to give regional, global, all ages or both sexes), resulting PAFs for deaths and DALYs will differ due to the different envelopes.

22When calculating burden of disease, several sources of uncertainty may exist, including selection bias, confounding, and statistical error. Consistent with previous global health estimates, for all estimates (i.e., exposure, PAF, death, and DALYs), we used bootstrapping to derive uncertainty (or prediction) ranges at the 2.5th and 97.5th percentiles of the resulting random deviates. Uncertainty in input parameters was propagated across all four models. The relative risks were log normally distributed and modelled using this distribution; all other input parameters were assumed to be normally distributed and modelled accordingly. Uncertainty ranges of estimates for combined cohorts (e.g., regional, global, all age groups, and both sexes) were produced using the method described in, as done in previous official estimations. For each estimate, we report the point estimate and its 95% uncertainty range (UR).

23To test the assumptions made for year 2016 (i.e., lag time of 10 years; time window of 10 years from 2001 to 2010; and assignment of the highest exposure category over the time window), we conducted the following sensitivity analyses:Reduced the lag time to 8 years (2003–2012);Increased the lag time to 12 years (1999–2008);Reduced the time window for the exposure to 8 years (2002–2009);Increased the time window to 12 years (2000–2011); andAssigned the long working hours category with the largest number of years in the time window (censoring years spent in labour market inactivity).

24Reduced the lag time to 8 years (2003–2012);

25Increased the lag time to 12 years (1999–2008);

26Reduced the time window for the exposure to 8 years (2002–2009);

27Increased the time window to 12 years (2000–2011); and

28Assigned the long working hours category with the largest number of years in the time window (censoring years spent in labour market inactivity).

29Global health estimates for 2016 are summarized in this article. Estimates for 2000 and 2010 are found in Table 4, Table 5, regional estimates are displayed in Tables S6–S10 in Supplementary data 1, and country estimates are available on www.who.int/teams/environment-climate-change-and-health/monitoring/who-ilo-joint-estimates/ and http://www.ilo.org/global/topics/safety-and-health-at-work/. Here, regional estimates are reported for the six WHO regions (Africa, Americas, Eastern Mediterranean, Europe, South-East Asia, and Western Pacific); however, with estimates for each country available, estimates could be based on any regional clustering, including the five ILO regions (Africa, Americas, Arab States, Asia and the Pacific, and Europe and Central Asia).

30Globally in 2016, 488 million people (UR 472–503), or 8.9% of the population (UR 8.6–9.1), worked ≥55 hours/week (Table 4). A full set of estimates can be found elsewhere (www.who.int/groups/who-ilo-joint-estimates and http://www.ilo.org/global/topics/safety-and-health-at-work/). Males and adults of early middle-age were more commonly exposed (Fig. 2). Between 2000 and 2016, the global prevalence of this exposure category increased by 9.3% (UR 4.3–14.6) (Table 4).

31In 2016, regional exposure prevalence was largest for South-East Asia (11.7%, UR 10.8–12.5), and lowest in Europe (3.5%, UR 3.5–3.6) (Table S6; Supplementary data 1). Over the 2000–2016 period, the Western Pacific had the largest regional increase; prevalence decreased most in Africa. A map of the proportion of population exposed by country is shown in Fig. 3.

32In total, an estimated 745,194 deaths (UR 705,786–784,601) and 23.3 million DALYs (UR 22.2–24.4) from ischemic heart disease and stroke combined were attributable to long working hours exposure globally in 2016. This was roughly equal between the two causes, with ischemic heart disease and stroke accounting for 46.5% and 53.5% of estimated deaths, respectively. Fig. 4, Fig. 5 present maps of the rates of deaths and DALYs from ischemic heart disease and stroke that are attributable to exposure to long working hours by country.

33Globally in 2016, an estimated total of 9,401,800 ischemic heart disease deaths occurred. Of these, 346,753 (UR 319,658–373,848) were attributable to exposure to long working hours (Table 5). Thus, the PAF is 3.7% (UR 3.4–4.0). Males carried a larger burden, and numbers and rates of deaths increased with age up to 70 years (Fig. 6). Between 2000 and 2016, the numbers of ischemic heart disease deaths attributable to exposure to long working hours increased substantially by 41.5% (UR 27.9–56.5) (Table 5). The PAF was 3.5% (UR 3.3–3.7), 3.6% (UR 3.4–3.9), and 3.7% (UR 3.4–4.0), in 2000, 2010, and 2016, respectively. The upward trend in number of these attributable ischemic heart disease deaths was therefore driven by the increase in the total disease burden envelope, rather than an increase in the exposure.

34Of all WHO regions in 2016, South-East Asia had the largest number of ischemic heart disease deaths attributable to exposure to long working hours (159,832 deaths, UR 135,442–184,242; Table S8 in Supplementary data file 1). Africa had the lowest (16,942 deaths, UR 15,878–18,005).

35Globally in 2016, of the 202.8 million DALYs lost from ischemic heart disease in total, 10.7 million DALYs (UR 9.9–11.4) were attributable to exposure to long working hours (Table 5). Between 2000 and 2016, attributable DALYs increased by 41.2% (UR 28.5–54.6). In 2000, 2010, and 2016, the PAF was 4.8 (UR 4.5–5.1), 5.1 (UR 4.7–5.4), and 5.3 (UR 4.9–5.6), respectively (Table 6).

36Globally, stroke caused an estimated 5,747,289 deaths in 2016. Of these, 398,441 (UR 369,826–427,056) were attributable to exposure to long working hours (Table 5). Thus, the PAF was 6.9% (UR 6.4–7.5). Both in absolute and relative terms, males and older age groups (60–74 years) carried a larger burden (Fig. 6). Over 2000–2016, the total number of stroke deaths attributable to exposure to long working hours increased by 19.0% (UR 7.8–31.1) (Table 5). The PAF was 6.5% (UR 6.1–7.0), 6.8% (UR 6.3–7.3), and 6.9% (UR 6.4–7.5), in 2000, 2010, and 2016, respectively (Table 6). The upward trend in number of attributable stroke deaths was therefore also primarily driven by the increase in the total disease burden envelope.

37The largest number of deaths regionally was estimated for South-East Asia (158,987 deaths, UR 141,968–176,006; Table S10 in Supplementary file 1). The lowest number was for the Americas (18,285 deaths, UR 17,162–19,409).

38Globally in 2016, of the 135.9 million DALYs lost from stroke, 12.6 million (UR 11.8–13.4) were attributable to exposure to long working hours (Table 5), up by 21.7% from 2000 (UR 11.8–32.4). In 2000, 2010, and 2016, the PAF was 8.6 (UR 8.1–9.1), 9.0 (UR 8.5–9.5), and 9.3% (UR 8.7–9.9), respectively (Table 6).

39Results from sensitivity analyses showed that despite some variation in estimates, number of deaths and DALYs remained appreciable when the assumed lag time was reduced to 8 years and increased to 12 years and when the time window was reduced to 8 years and increased to 12 years (Table S11; Supplementary file 1). Assigning the most common exposure category, rather than the highest one, however, reduced the deaths and DALYs substantially.

40This article presented WHO/ILO Joint Estimates of exposure to long working hours and the attributable burdens of ischemic heart disease and stroke. In summary, in 2016, 8.9% of the global population were exposed to working ≥55 hours/week. An estimated 745,194 deaths and 23.3 million DALYs from ischemic heart disease and stroke combined were attributable to this occupational risk factor. Of all deaths from ischemic heart disease and stroke in 2016, 3.7% and 6.9% were attributable to exposure to working long hours; as were 5.3% and 9.3% of all DALYs from ischemic heart disease and stroke. The disease burdens were disproportionately higher in the South-East Asian and Western Pacific regions, men, and people of middle to older working age. Between the years 2000 and 2016, the exposed population increased by 9.3%, and the attributable burdens of deaths from ischemic heart disease and stroke increased by 41.5% and 19.0%, respectively.

41The distribution in the population and trends over time in the WHO/ILO Joint Estimates of exposure to long working hours are consistent with recent ILO analyses of population distributions and time trends observed in official survey data on working time. To our knowledge, there are no prior estimates of burden of disease attributable to exposure to long working hours that we could compare these first WHO/ILO Joint Estimates against.

42These WHO/ILO Joint Estimates demonstrate that the disease burden attributable to exposure to long working hours is the largest of any occupational risk factor calculated to date, compared with those attributable to other occupational risk factors included in global Comparative Risk Assessments.

43The population prevalence of exposure to long working hours increased substantially between 2010 and 2016. If this trend continues, it is likely that the population exposed to this occupational risk factor will expand further. Potential reasons for this include expansion of the gig economy, the uncertainty introduced, and new working-time arrangements (e.g., on-call work, telework, and the platform economy). Past experience has shown that working hours increased after previous economic recessions; such increases may also be associated with the COVID-19 pandemic. Estimated increases in exposure have been largest in South-East Asia and the Western Pacific. Furthermore, the total envelopes of the burdens of ischemic heart disease and stroke are also increasing rapidly. As both exposed population and total disease burden expand, the burdens of ischemic heart disease and stroke that can be attributed to exposure to working long hours may therefore also be expected to increase.

44WHO and ILO have produced a detailed set of estimates of the burdens of ischemic heart disease and stroke attributable to exposure to long working hours. These estimates were based on systematic reviews and meta-analyses of evidence to date. Multiple data sources provided large samples from all regions to calculate estimates.

45However, the study has some limitations, which should be considered when interpreting the findings. First, the systematic reviews and meta-analyses estimated relatively small increases in risk, even where statistical significance is reached. Evidence comes from observational studies, for which residual confounding cannot be ruled out, and we cannot be certain that a causal association exists; however most of the high-quality evidence came from prospective cohort studies which took steps to reduce confounding and will be more representative of population risk. Taking this into consideration, the Working Group of individual experts rated the evidence as of “moderate quality” and as “sufficient evidence for harmfulness” for the exposure category ≥55 hours/week for both ischemic heart disease and stroke. Other researchers have disagreed with the rating of the Working Group that there is “sufficient evidence for harmfulness” of long working hours with regard to ischemic heart disease. The Working Group has acknowledged this disagreement and has elaborated why the assigned rating is supported by the evidence.

46Second, it has been argued that, at least with regard to ischemic heart disease, not only length of working hours, but also the quality of the work in which persons spend their hours, may be of importance. Although presented data and the Working Group of individual experts conducted analyses stratified by SES that suggested a trend for stronger associations between exposure to long working hours and risk of ischemic heart disease among individuals of lower SES, these were limited to studies from high-income countries and one region only (Europe). The Working Group deemed the evidence for a possible effect modification of SES in the association between long working hours and ischemic heart disease as inconclusive for a global study. Currently, evidence does not support producing such global health estimates disaggregated by SES; however more research is needed in additional and more diverse countries and regions on the roles of SES and work quality for the association of long working hours with health outcomes.

47Third, quality of data on both long working hours and burdens of ischemic heart disease and stroke will vary by source. Most data regarding long working hours were obtained from national statistics offices with established, official data collection protocols (e.g., statistical standards), but variation can still be expected. All surveys used self-reported data on working hours. Several studies showed both reliability and validity of self-reported hours, compared with administrative records; however, this may vary. This could lead to under- or over-estimations of the burden, depending on the direction of the error.

48Fourth, several assumptions were made during modelling. While they were based on current knowledge and transparently described in detail (Table 3), these may need review as more evidence becomes available. In addition, as noted above, we provide policy makers with alternative scenarios through sensitivity analyses. While changes to the assumed lag time and time window, in the sensitivity analyses, resulted in some variation from the main results, a relatively large reduction in attributable deaths and DALYs were seen when the most common exposure category was used, rather than the highest. As described in Table 3, the highest exposure category was used as risks from exposures can remain, even when exposure levels are reduced, for diseases with long latency periods. However, this may have resulted in overestimation of health loss attributable to exposure to long working hours. Assignment of the highest exposure category (as in our main analysis) is the standard practice in burden of disease estimation studies; however, by providing additional information for policy making, sensitivity analyses of estimates with the most common exposure category assigned, we show that if this exposure assignment was to be found to be evidence-based, then this would be the estimated disease burden, adding further transparency through adding another scenario.

49Fifth, there are additional factors likely to affect both working hours and disease burdens, which were not considered in these analyses (e.g., shift work), but could be considered in future cycles of the WHO/ILO Joint Estimates. Seasonal variations could be a factor for some occupations (e.g., agricultural workers). For some countries longitudinal survey data were available, for which quarterly surveys took data from an individual over multiple time points. When these data were available, we took an average of the data which should mitigate some of the seasonal differences; however, the majority of the data used in the assessment of long working hours exposure were cross-sectional, so this is a limitation. There are also several potential mediators (Fig. 1), which are challenging to address. Sixth, there may be competing risks; for example, people exposed to long working hours may die of other causes or migrate before reaching the typical ages when cardiovascular disease events occur.

50WHO and ILO estimate that exposure to long working hours (≥55 hours/week) is a prevalent occupational risk factor, attributable for a large number of deaths and DALYs due to ischemic heart disease and stroke. In the global Comparative Risk Assessment, it is currently the occupational risk factor with the largest attributable disease burden. These first WHO/ILO Joint Estimates of the Work-related Burden of Disease and Injury provide the basis for actions to prevent exposure to hazardous long working hours and thereby reduce the attributable burden of ischemic heart disease and stroke, at the global, regional, and national levels, across the health and labour sectors. This includes implementation of the international labour standards on working time, such as setting standards on working time limits. Legislation, regulations, policies, programmes, and interventions on working time arrangements must ensure that the setting, monitoring, and enforcement of hours of work and the number of additional hours performed by workers occur within a framework that does not harm human health.

51The authors alone are responsible for the views expressed in this article and they do not necessarily represent the views, decisions or policies of the institutions with which they are affiliated. The views expressed herein can in no way be taken to reflect the official opinion of the European Union.

52The designations employed and the presentation of the material in this publication do not imply the expression of any opinion whatsoever on the part of WHO concerning the legal status of any country, territory, city or area or of its authorities, or concerning the delimitation of its frontiers or boundaries. Dotted and dashed lines on maps represent approximate border lines for which there may not yet be full agreement.

53Terminology used to refer to countries, territories, and areas as well as representation of countries, territories, and areas, including delimitation of frontiers or boundaries, and any direct or indirect attribution of status in this publication follow exclusively the institutional style and practice of WHO. All reasonable precautions have been taken by WHO to verify the information contained in this publication.

54However, the published material is being distributed without warranty of any kind, either expressed or implied. The responsibility for the interpretation and use of the material lies with the reader. In no event shall WHO be liable for damages arising from its use.

55This study was prepared with financial support to WHO from: the National Institute for Occupational Safety and Health of the Centres for Disease Control and Prevention of the United States of America (Grant 1E11OH0010676-02; Grant 6NE11OH010461-02-01; and Grant 5NE11OH010461-03-00); the German Federal Ministry of Health (BMG Germany) under the BMG-WHO Collaboration Programme 2020–2023 (WHO specified award ref. 70672); and the Spanish Agency for International Cooperation (AECID) (WHO specified award ref. 71208). ILO acknowledges financial support from the Vision Zero Fund (VZF) project on filling data and knowledge gaps on occupational safety and health in global supply chains funded by the European Union, which is implemented within the framework of the ILO Flagship Programme Safety + Health for All. The funders had no role in study design, data collection, and analysis, decision to publish, or preparation of the manuscript

56Frank Pega: Conceptualization, Data curation, Methodology, Formal analysis, Visualization, Writing - original draft, Supervision, Writing - review & editing. Bálint Náfrádi: Data curation, Methodology, Formal analysis, Visualization, Writing - review & editing. Natalie C. Momen: Visualization, Writing - review & editing. Yuka Ujita: Writing - review & editing. Kai N. Streicher: Formal analysis, Visualization, Writing - review & editing. Annette M. Prüss-Üstün: Methodology, Formal analysis, Supervision, Writing - review & editing. Technical Advisory Group: Methodology, Writing - review & editing.

57The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: Professor Alexis Descatha reports personal fees from University of Angers, personal fees from the Angers University Hospital Center, personal fees from Elsevier, and grants from Regional found (Pays de la Loire/Angers Loire Metropole), outside the submitted work.