• Epidemiology
• mortality studies

In their article, Coggon and colleagues1 report on work-related mortality in England and Wales over the period 1979–2000. Occupational mortality studies are one of the oldest approaches in epidemiology, including the decennial reports on patterns of mortality for occupational groups in England and Wales that have been published since the 19th century. These studies have made use of routinely collected data (death certificates), to study cause-specific mortality patterns by occupation and socioeconomic status. Even today, few other study designs, if any, can provide such a wide range of information on the occupational health status of a population, for so little cost.

Nevertheless, occupational mortality studies are currently not held in high regard. This perhaps reflects the low status of descriptive epidemiology- and hypothesis-generating studies in general,2 but there are additional specific methodological concerns regarding occupational mortality studies. In particular, most occupational mortality studies have used census data to produce standardised mortality ratios (SMRs). Using such external denominator data risks the introduction of a numerator-denominator bias, as the occupation registered on the census (usually the self-reported occupation of that person), is not fully comparable with the occupation registered on the death certificate (usually the last held occupation of the deceased as reported by the next of kin). There therefore has been ‘a certain reluctance in accepting mortality excesses registered for specific occupational groups in these studies’.3 Other criticisms of occupational mortality studies include that the method cannot monitor occupational diseases that are not fatal; that information on confounders such as smoking is not available; that only one occupation per individual is available; and that they are susceptible to chance findings due to the multiple comparisons made.

How can then the current place of occupational mortality studies within the field of occupational epidemiology be characterised? It is probably fair to say that they have a reputation of being simple and crude, descriptive, uncreative, and susceptible to unreliable results. They still enjoy some local interest as a surveillance tool, but are by and large seen as not worthy of publication in international journals. There is perhaps a general feeling that surely by now we should have moved away from such primitive methods that were already in place more than 100 years ago.

In their article Coggon and colleagues give us a good illustration of the continuing value of occupational mortality studies. Because satisfactory denominator data were not available, Coggon and colleagues chose for a proportional mortality ratio (PMR) approach, where the proportion of deaths from a specific cause within a specific occupation is compared with the proportion of deaths from that cause among all deaths in the study, standardising for age and social class. This approach, although having certain disadvantages, eliminates the numerator-denominator bias that occupational mortality studies using the SMR approach have been criticised for. Furthermore, it makes the approach even simpler and more accessible for the many countries that do not have any denominator data (ie, census data) available, including developing countries where occupational mortality studies have never been carried out.

Occupational mortality data can be used more creatively than has been done before. Historically, occupational mortality studies were mainly seen as a hypothesis-generating tool, producing lists of occupations that show an excess in mortality for certain causes. Coggon and colleagues instead aimed to quantify the number of deaths due to known occupational risk factors and studied this pattern over time. This showed a clear decline in excess mortality attributable to work, but it also indicated which occupations and which occupational diseases contributed most to this decline. This is an interesting alternative to the attributable fraction approach that is most often used to quantify the burden of work-related mortality of a certain population. The attributable fraction approach is, however, not as flexible in detecting changes over time and within specific occupations, and often relies on attributable fraction estimates derived from other populations than the one under study, making it less population-specific than the method used by Coggon and colleagues.

The fact that occupational mortality studies have only very crude occupational information is not necessarily a disadvantage. Certainly, the field of occupational epidemiology has been working hard to improve the exposure data used in its studies, but this important development should not imply that there is no longer a place for studies that are based on occupation alone, particularly those using routinely collected data. The use of ‘occupation’ as unit of analysis has its own merits as illustrated by Coggon and colleagues. Their findings, for example, indicate that the reduction in excess work-related deaths for certain occupational groups is in large part due to the contraction of the industry, while the impact of improvements of working conditions within these occupational groups is only moderate.

The criticism that occupational mortality studies produce unreliable results is not fully justified. In fact, many occupational carcinogens were first discovered in such studies, with (some of) the hypotheses subsequently being tested and confirmed in analytical studies. For example, the first evidence of an increased risk of cancer in farmers came from occupational mortality studies, which led to a large number of cohort and case-control studies linking pesticides to increased risks of various types of cancer.4 False-positive findings will occur, but as for any other study design each association should be evaluated with respect to its biological plausibility, the likelihood that it could be explained by confounding or biases, and its consistency with other epidemiological studies. Because occupational mortality studies can be repeated over time, and compared between different populations, this type of study is in fact ideally placed to detect consistency in findings.

There is however one aspect of occupational mortality studies that should have provoked more criticism rather than less. These studies have, although claiming to be ‘nation-wide’, historically excluded half of the nation's working population: women workers. The fact that these early studies (and many other occupational epidemiological studies for that matter) only included men, is still having its effects into today's world where most hypotheses have been based on male populations, thus underestimating the magnitude of women's occupational health issues. The practical reasons for not including women in occupational mortality studies (ie, the fact that occupation was often not recorded on death certificates for women), have largely disappeared, making the exclusion of women no longer justifiable.

To conclude, in the 21st century, it is time for a rebirth of our old friend the occupational mortality study. They play an important role not only in disease monitoring and surveillance, but also as part of the scientific cycle of hypothesis generation and testing. They provide information that no other study design will be able to give for a relatively small cost. However, to get the maximum value from them, we need them to be published in scientific journals, not just in low-circulation technical reports. To accomplish this, let's approach them more creatively. Let's do them in more countries, particularly developing countries. Let's do them repeatedly over time. Let's finally do them in women workers. Let's just do them.