Global and regional burden of disease and injury in 2016 arising from occupational exposures: a systematic analysis for the Global Burden of Disease Study 2016

Objectives This study provides an overview of the influence of occupational risk factors on the global burden of disease as estimated by the occupational component of the Global Burden of Disease (GBD) 2016 study. Methods The GBD 2016 study estimated the burden in terms of deaths and disability-adjusted life years (DALYs) arising from the effects of occupational risk factors (carcinogens; asthmagens; particulate matter, gases and fumes (PMGF); secondhand smoke (SHS); noise; ergonomic risk factors for low back pain; risk factors for injury). A population attributable fraction (PAF) approach was used for most risk factors. Results In 2016, globally, an estimated 1.53 (95% uncertainty interval 1.39–1.68) million deaths and 76.1 (66.3–86.3) million DALYs were attributable to the included occupational risk factors, accounting for 2.8% of deaths and 3.2% of DALYs from all causes. Most deaths were attributable to PMGF, carcinogens (particularly asbestos), injury risk factors and SHS. Most DALYs were attributable to injury risk factors and ergonomic exposures. Men and persons 55 years or older were most affected. PAFs ranged from 26.8% for low back pain from ergonomic risk factors and 19.6% for hearing loss from noise to 3.4% for carcinogens. DALYs per capita were highest in Oceania, Southeast Asia and Central sub-Saharan Africa. On a per capita basis, between 1990 and 2016 there was an overall decrease of about 31% in deaths and 25% in DALYs. Conclusions Occupational exposures continue to cause an important health burden worldwide, justifying the need for ongoing prevention and control initiatives.

occupational turnover estimates (OTs) based on a risk-exposure period defined by cancer latency  years for solid tumours [long latency], 0-20 years for haematopoietic cancers [short latency]), annual staff turnover estimates, and normal life expectancy were developed and applied to the original prevalence data. Separate estimates were provided for men and for women, and for the solid tumours and for haematopoietic cancers, for 2016. Separate life tables (based on a representative country in each region) were used to estimate the OTs by region. Further detail on the approach used is available elsewhere 9 10 .
A different approach was taken to estimate the proportion of persons ever exposed to asbestos. Rates of malignant mesothelioma were used to estimate the past prevalence of exposure to asbestos in each country. This Asbestos Impact Ratio approach was analogous to the Smoking Impact Ratio approach described elsewhere 11 . The relevant companion paper provides more detail on this and the methods used for the occupational carcinogen analysis 9 . The comparison for each carcinogen-specific analysis was no exposure above background to the relevant carcinogen.

Particulate matter, gases and fumes, and chronic obstructive pulmonary disease
Industry was used as a proxy for exposure to particulate matter, gases and fumes (PMGF) because we identified no suitable and valid data sources at a country or global level of exposure to PMGF, either singly or to PMGF as a group. Current industry was used as the basis of exposure estimates, but the estimates of proportions exposed within each industry attempted to take into account past exposure (to estimate ever exposed), given that both past and current exposure appear to increase the risk of chronic obstructive pulmonary disease (COPD). Estimates of proportion exposed at lower and higher levels were based on sparse published data and expert opinion by GBD authors. Information on risk was obtained by conducting a systematic review of international literature and meta-analysis of relevant results. Relative risks in these studies were for COPD greater than or equal to GOLD stage II (defined as requiring non-reversibility after using bronchodilators for provocation, a forced expiratory volume in one second/forced vital capacity (FEV1/FVC) ratio of less than 0.70 and an FEV1 of less than 80% predicted) 12 . Relative risk estimates were used for an overall "lower" level and an overall "higher" level of exposure to the agents of concern. The

Supplementary material
Occup reference group was persons not working and persons working in trade, finance, or service industries.
Further information on the methods is available in the companion paper on airborne risk factors 13 .

Asthmagens and asthma
Exposure for asthmagens was based on the current occupation distribution in each country because there were no suitable and valid data sources at a country or global level describing exposure to the wide range of occupational asthmagens. All relative risk information except that for agricultural occupations came from a study by Karjalainen and coworkers, a comprehensive national population study of incident asthma 14 15 .
Relative risks for agricultural occupations were based on a separate study by Kogevinas and coworkers 16 , with an inverse variance-weighted estimate obtained using the separate estimates for "farmers" and "agricultural" workers provided in the paper. This information was used because the results were thought to be more generalizable to agriculture in the rest of the world, especially the LMI regions. Separate risks were available and used for males and females (except for agricultural operations), although the sexspecific risks were similar and within the limits of random variation. The same relative risks were used for all age groups. The referent group (used as the counterfactual) was persons not working and administrative workers. Further information on methods is available in the companion paper on airborne risk factors 13 .

Pneumoconioses
Pneumoconioses were estimated as part of the envelope (the total number of cases) of disease component of the GBD, rather than using the attributable fraction approach. The methods used are described elsewhere 17 . The attributable fraction is essentially 100% because virtually all pneumoconioses arise as a result of occupational exposure. Separate estimates were available for silicosis, asbestosis, and coal workers' pneumoconiosis (CWP), with the remaining cases grouped under an "other pneumoconiosis" category.

Supplementary material
Occup

Noise and noise-induced hearing loss
Current industry was used as a proxy for exposure to occupational noise because of insufficient useable exposure data at a country or global level. Australian national data on noise exposure in various industries (sampled across a range of tasks) provided the basis for the mean and standard deviation of noise exposure in each industry 18 . This information was in turn used to estimate the proportion of workers exposed at "low" levels of noise (85-<90dB) and "high" levels of noise (90dB or higher). For the presented analysis, no account was made of the use of hearing protection, except for the mining industry, where mean exposure levels were decreased by 3dB (based on expert opinion) to take this into account. The proportions were modified for LMI countries, to take into account the likely higher exposure levels due to the less extensive use of noise controls. For LMI countries, the mean exposure was estimated to be 3dB higher (over an entire shift -double the noise level of high-income countries) in 1990, and 1.8dB higher (50% increase in the noise level of developed countries) in 2016 (based on expert opinion).
The outcome factor was hearing loss at 41dB or greater. Relative risks for high and low exposure were obtained in a two-step method. First, the absolute excess proportions of persons with hearing loss at 41dB or greater due to occupational exposures at 85-90dB, and at greater than 90dB, was obtained, separately for different ages, from Nelson and coworkers 19 . Information on background prevalence of hearing loss at 41dB or greater was obtained from population surveys in the United Kingdom 20 and Australia 21 . The same excess and population prevalences were used for all countries. The relative risk (RR) was then estimated as RR = 1+ excess prevalence of hearing loss/background prevalence of hearing loss. This was calculated separately for the age ranges used in the relevant publications and then adjusted to fit the GBD age ranges.
For example, for ages 50 to 59 years, the absolute excess proportion of persons with hearing loss at 41dB or greater due to occupational noise exposure at greater than 90dB was 0.169. The background prevalence of hearing loss at 41dB or greater was 0.037. Therefore, the relative risk for this age group was 1+(0.169/0.037) = 5.6.

Supplementary material
Occup The referent group was persons exposed only at background noise levels. Further information on methods is available in our companion paper on occupational noise 22 .

Ergonomic risk factors and low back pain
Current occupation was used as a proxy for ergonomic risk factors associated with occupation, as described in detail elsewhere 23 . The occupations were grouped to be consistent with the exposure data in the studies that provided risk information. Information on the relative risk of low back pain was obtained from a systematic review of international literature and meta-analysis of relevant results, as described previously 23 . The same relative risk estimates were used for males and females and for all age groups. The referent group was persons not working and clerical workers.

Injury risk factors and injury
"Injury" was defined as any injury to a worker due to work-related exposures and which would have warranted some type of health care in a system with full access to health care 24 , excluding self-harm and injuries sustained driving between home and work or vice versa. The approach to estimating injuries did not use population attributable fraction methods (although we did calculate PAFs secondarily after having estimated the proportion of work-related injuries). Instead, the number of fatal occupational injuries was estimated by applying industry-specific fatal injury rates taken directly from the ILO database 25 (by country where available and modeled using covariates and rates in geographically and demographically similar countries when country-specific data were not available) to the estimated industry-specific population in each country. The same rates were used for all ages and both sexes, but age-and sex-specific PAFs were estimated by dividing the work-related fatal injuries by the total fatal injuries in the relevant age-sex group.
The injuries were distributed across different injury causes using a custom set of injuries that were selected for each industry a priori, based on previous published work 26 27 . Lacking suitable data, non-fatal injury PAFs could not be estimated directly. Instead, non-fatal injury PAFs were assumed to be the same as the fatal injury PAFs. The comparison was zero work-related injuries.

Supplementary material
Occup

Second-hand smoke
Second-hand smoke was included as a risk factor for several different types of outcomes, consistent with the approach taken in the GBD overall. The relevant outcomes were breast cancer, lung cancer, ischaemic heart disease, stroke, lower respiratory tract infection and diabetes mellitus. Information on exposure prevalence was as described above for carcinogens, except that separate "high" and "low" prevalences were not used for the non-cancer outcomes. Further information on the relative risk measures used is in the main GBD risk factors paper 28 .

Supplementary material
Occup