Analysis was restricted to deaths with recorded occupations (1,564,981 from 1979-90 and 1,199,234 from 1991-2000). These represented 91.4% of all deaths in the relevant age group during the first period, and 94.5% during the second period. Associations between job groups and causes of death were characterised by proportional mortality ratios (PMRs), standardised for age (in five-year bands) and social class (in six main strata)[1,4]. For the Armed Forces, which formed a separate social class category, standardisation was only for age. As a default, PMRs were calculated for ages 20-74 years, but for injury and poisoning, and for pneumonia in jobs involving exposure to metal fume, analysis was limited to deaths at normal working ages (20-64 years) since any effects of occupation on mortality were expected to be fairly immediate [5,6]. Confidence intervals (CIs) for PMRs were based on the Poisson distribution.

At the outset, from our prior knowledge, we identified 59 causes of death (26 diseases and 33 categories of injury and poisoning) that were well-established occupational hazards. A few causes of death were considered uniquely occupational (asbestosis, coal workers’ pneumoconiosis, silicosis, other pneumoconiosis, farmers’ lung disease and byssinosis), and all deaths from these causes were counted as attributable to work. For each of the other causes of death, we made an a priori listing of job groups in which clear excess mortality, if it occurred, could reasonably be ascribed to occupational activities or exposures. From these lists, we retained those job groups for which the PMR in either of the two study periods was greater than 1.2 with a lower 90% confidence limit greater than 1.0. Where a job group met this criterion for either of pleural cancer or peritoneal cancer, we retained it for both diseases. In total, 125 of the 475 originally specified combinations of job group and cause of death were discarded because they did not satisfy the conditions for retention. The remainder are listed in Supplementary Tables 1-3. For two causes of death (cancer of the oesophagus and poisoning by gas and other domestic fuels), none of the originally listed job groups met the criterion for retention, and these causes of death were therefore excluded from further analysis. For each of the retained combinations of cause of death and occupation, mortality attributable to work was calculated as the difference between the observed and expected numbers of deaths from which the corresponding PMR was derived (where the difference for one of the time periods was negative, it was counted as zero).

To explore trends over time in the burden of mortality attributable to work, we compared the excess frequency of deaths in the two time periods, 1979-80/1982-90 and 1991-2000. Because the first period covered 11 years and the other 10 years, this was expressed as excess deaths attributable to work per year.

Mortality attributable to other occupational diseases was lower, with 71.2 excess deaths per year during 1970-1990 and 62.7 during 1990-2000 (Table 2). By far the largest contribution to the excess was from alcohol-related disorders (in publicans and bar staff). However, there was also elevated mortality from viral hepatitis (in doctors), non-melanoma skin cancer (in various outdoor occupations), urothelial cancer (in chemical workers) and scrotal cancer (in turners and machine tool setters and operators). The excess of deaths from urothelial cancer in chemical workers was higher in 1991-2000 (2.4 deaths per year) than in 1979-90 (0.6 deaths per year).

Table 3 shows mortality from different categories of work-related injury and poisoning. During both of the time periods that were examined, the largest excesses of deaths were from motor vehicle traffic accidents, especially in lorry drivers, who had PMRs of 1.60 (45.3 excess deaths per year) in 1979-90 and 1.75 (38.0 excess deaths per year) in 1991-2000. Lorry drivers also had elevated mortality from off-road motor vehicle traffic accidents (4.5 excess deaths per year in 1979-90 and 3.6 in 1991-2000). The other categories of trauma that contributed importantly to work-related mortality were injury by machinery (particularly in farmers, production fitters, and other transport and machinery operatives), falls from buildings and from ladders or scaffolding (in various construction trades), and water transport accidents (in seafarers, fishermen and dock workers). The total excess mortality attributable to work-related injury and poisoning was substantially lower in 1991-2000 (138.5 deaths per year) than in the earlier period (246.3 deaths per year).

More detailed versions of Tables 1-​-3,3, showing observed and expected deaths by cause for each of the job groups that contributed to the overall excesses can be found in Supplementary Tables 1-3.

Table 4 lists the job groups that had the largest excesses of work-related mortality during 1991-2000. The biggest excess (105.8 deaths per year) was in “other coal miners” (principally from COPD and pneumoconiosis), followed by publicans and bar staff (alcohol-related diseases), lorry drivers (mostly motor vehicle traffic accidents) and farmers (mainly work-related injury and poisoning).

The method by which we assessed the burden of mortality that could be ascribed to work had various limitations, although robust conclusions are possible provided these shortcomings are recognised and taken into account when results are interpreted.

Because no satisfactory data were available on the populations at risk in each job group, we were obliged to base our calculations on proportional mortality. Thus, risk estimates may have been distorted if there were unusually low or high total death rates in occupations of interest. And further error could have occurred through confounding by non-occupational risk factors such as smoking, although this should have been reduced by standardisation of PMRs for social class.

Another possible source of bias was selective recruitment into certain jobs. In particular, the high mortality from alcohol-related diseases among publicans and bar staff, while partly a consequence of their ready access to alcohol at work, is likely to have resulted also from a propensity for heavier drinkers to seek work in bars [7-9]. Similarly, the high PMR for viral hepatitis in doctors may have occurred because an unusually high proportion of doctors were immigrants from countries with a high prevalence of hepatitis B infection in infancy, and it does not necessarily reflect infection acquired occupationally through sharps injuries. During 1982-90, only five of the 12 male doctors who died from hepatitis in England and Wales had been born in the UK. Of the other seven, four had been born in Africa [1].

Other shortcomings may have caused risks to be underestimated. Occupational data were limited to the decedent’s last full-time job, but for diseases that develop only after a long induction period (e.g. cancers), or that follow a prolonged clinical course before causing death (e.g. COPD), causally related occupations will not always be the last that were held before death. It is probably for this reason that some deaths from farmer’s lung disease were recorded in non-farmers, and some deaths from coal workers’ pneumoconiosis occurred in jobs unrelated to coal mining.

Furthermore, deaths could only be ascribed to occupation where the link could be made with reasonable confidence. Where a causal exposure was not typical of the job group to which an individual belonged, its effect may not have been recognised. For example, work-related death from injury by fire could occur in any occupation, but from the information that was supplied to us, there was no way of distinguishing fires in the workplace from those occurring in other circumstances. Thus, only in fire service personnel did it seem reasonable to attribute excess mortality from injury by fire to work.

In addition, errors are known to occur in the reporting of both occupation and cause of death on death certificates [10,11]. In general, such errors would be expected to obscure true associations between occupations and mortality.

There are also statistical uncertainties in our risk estimates, particularly where they were based on only small numbers of observed and expected deaths. Our decision to count the attributable number of deaths as zero where the observed minus expected number was negative, will have introduced a small inflationary bias in estimates of the total burden of mortality attributable to work. However, in practice this situation rarely occurred, and the effect will therefore have been minimal.

Because of these limitations, for some causes of death (e.g. many categories of acute injury), other sources of data such as reporting schemes are likely to give more reliable estimates of attributable numbers. For example, notifications to the Health and Safety Executive under the Reporting of Injuries, Diseases and Dangerous Occurrences Regulations (RIDDOR) [12], indicated some 275 deaths per year from fatal occupational injuries in all workers in Great Britain during 1992-2000. Even allowing for the fact that this statistic included deaths in women as well as men, and in Scotland as well as England and Wales, it indicates a larger problem than our figure of 138.9 excess deaths per year during 1991-2000. But for causes of death that cannot confidently be ascribed to occupation in the individual case because their relation to work is not sufficiently specific (e.g. COPD in coal miners or lobar pneumonia in jobs involving exposure metal fume), reporting schemes are unsatisfactory, and the approach that we adopted is the best available.

Our method is likely to have been most reliable in the information that it provided about trends over time, since the biases that have been described can be expected to apply in much the same way in different time periods. In support of this, the fall that we observed in excess deaths from occupational injuries is consistent with a reduction in such deaths reported through RIDDOR [12]. The analysis for 1979-90 included a small number of deaths (121 in total) from uniquely occupational diseases in men with known employment but inadequately described occupations, whereas during 1991-2000 men with inadequately described jobs were grouped with those who had no recorded occupation, and excluded from the analysis. Nevertheless, it seems clear that there was a substantial decline over the period of study in work-related mortality overall, and in that from most specific diseases and injuries. This is likely to reflect both improvements in working conditions and methods, and also a reduction in the number of men employed in more hazardous jobs. For example, the major fall in deaths from COPD attributable to coal mine dust was driven more by lower total numbers of deaths in coal miners than by a reduction in their PMR for the disease, suggesting that contraction of the industry had a bigger effect than improvements in working conditions.

It is notable that the decline in work-attributable mortality did not extend to asbestos-related disease. This accords with earlier analyses of national trends in mortality from mesothelioma [13,14], but it is of concern that there was also no reduction in mortality from asbestosis over the study period. While mesothelioma typically occurs with a long induction period from first exposure to asbestos, deaths from asbestosis are likely to be influenced also by more recent exposures. Our findings therefore reinforce the importance of continuing efforts to ensure that exposures to asbestos are properly controlled. Similarly, the absence of any clear reduction in excess mortality from sino-nasal cancer among wood-working occupations is an indication for further checks on the adequacy control measures, particularly in furniture manufacture.

Our analysis also highlights a substantial and continuing excess mortality from motor vehicle accidents (principally on-road but also off-road) in lorry drivers. During 1991-2000 this amounted to an average of 38 extra deaths per year with a PMR of 1.75 for on-road accidents. In the UK, prevention of road traffic accidents is the responsibility of the Department for Transport rather than the Health and Safety Executive, and data on work-related road traffic injuries are not therefore included in routine statistics of occupational injuries. The risk that we have identified is unlikely to be an artefact of our analytical method, and should be a spur to analysis of other, more detailed data on accidents involving lorries, as a basis for enhanced, targeted preventive strategies.

Notable also, is the high absolute risk of work-related death in certain occupations. In particular, during 1991-2000, excess mortality from COPD, pneumoconiosis and other work-related causes accounted for 4.3% of deaths at ages 20-74 years in “other coal miners”, while 4% of deaths among aircraft flight deck operators were from air traffic accidents. The risks of respiratory disease associated with coal mining are decreasing and this can be expected to continue as a consequence of better control of dust levels in recent decades. However, the high risk of accidental death in pilots, which has been noted before [1], and which is likely to result largely from accidents involving smaller aircraft, is a continuing concern. As already discussed, the high risk of alcohol-related deaths in publicans and bar staff is probably determined in part by selective recruitment of drinkers to work in bars, and therefore to over-represent the risks from such work. In many of the other job groups with apparently high proportional excess mortality, the excess was driven largely by cancer of the bronchus, and may in part have been confounded by smoking.

Supp Tables