Usefulness of occupation and industry information in mortality data in South Africa from 2006 to 2015

Deaths from 2006 to 2015 were used to determine mortality rates, coding of occupation and industry and reporting of occupation. In the coding of the South African mortality data, similar jobs were grouped into ten main groups (Table 1), but many of these occupations while similar have different exposures. Thus, the value of the ten occupational groups is limited for longitudinal and investigative analysis. For example, elementary worker – the largest employment group - encompasses street workers, cleaners, food preparation assistants, mining and building labourers. This limits the usefulness of mortality data for occupational disease surveillance and investigations. The years from 2013 to 2015 contain more disaggregated occupational information in two extra variables, sub occupation and minor occupation allowing for surveillance investigations. Industry was coded in 9 groups until 2010 when more disaggregated industries were coded. These are useful particularly in conjunction with occupation providing useful information on workplace exposures. Due to the coding of occupation before 2013 not containing the three digits codes, only 2013–2015 data which contained the more detailed information were used to investigate the association between cause of death, usual occupation and industry. It is accepted that South African mortality data are of mediocre quality due to a range of factors including limited resources [14, 15]. The variables used in the analysis were cleaned by recoding any unknown values to missing in age year, marital status, education and sex; so that the numbers used to code this information are not included in calculations. All deaths aged 14 and below were removed from the data set as this analysis focuses on occupational causes of disease (15 years and above is considered working age). Marital status was condensed into ever married and never married. Minor occupation (coded to three digits) was used, but where there was only one death due to pneumoconiosis in the three years, the occupation was combined with a similar one. Legislators and senior officials were combined with managing directors and CEOs due to the similar nature of their occupations in terms of hazards, building frame workers and building finishers were combined. Combining these occupational and industry groups involved adding the cases and non-cases from each group together to create one new group. A similar exercise was performed with industry where the central government was combined with other services, and non-specified retail was combined with other retail. These changes reduced wide confidence intervals by increasing numbers in each group, but the trends remained the same.

Mortality odds ratios were chosen over standardised mortality rates as the number of workers in each occupation for each year in South Africa is mostly unknown. Also due to the poor completion of the occupation question in the death certificate, it is not possible to determine the proportion of workers who worked in each occupation or industry who had their information recorded on the death notification. Data analysis was conducted using STATA software version 14.2 and Microsoft Excel spreadsheet 2010. Reported usual job and industry in deaths over age fifteen years were used to determine MOR’s. Trends in reporting of occupation and industry were created and illustrated in graphs. The 2010 SA NBD list, which comprises 140 specific causes of death was used to group pneumoconiosis deaths, these are ICD 10 codes J60 to J65. Due to the small numbers of deaths due to pneumoconiosis data from the years 2013–2015 were combined. They are the only years available with coding to three digits for occupation. Combining years will improve the analysis by increasing the power and remove spurious associations.

MORs were calculated using logistic regression of deaths due to pneumoconiosis against deaths due to all other causes. The outcome chosen here, pneumoconiosis, has a low event rate. However, we could not apply Firth’s penalised maximum likelihood estimation for the regression analysis logistic regression although it is the recommended method for analysing rare events in large datasets to reduce bias as we did not have the required computing power [16]. The MOR’s determined were adjusted for age, sex, smoking and death year all significant factors associated with annual mortality. We compared all occupation minor-groups to persons not economically active to adjust for socioeconomic deprivation.

Mortality in South Africa from 2006 to 2015 showed an absolute decrease from 621,378 deaths in 2006 to 473,938 deaths in 2015 while the overall crude rates also decreased (Fig. 1).

Only a small proportion of death notification forms (mean 16.1%) of those of working age (15–65) were completed with usual occupation and industry information (Fig. 2), while 43.1% of the same age population were employed at the time [17]. Risk of death may not be the same between employed and unemployed. The labour force surveys (LFS) were started by Statistics South Africa in 2008. Therefore information for 2006 and 2007 is not available.

The proportion of those completing the occupation question has increased by 5% over the past eight years while those reporting actual employment has changed little with an average of 16% of deaths between the ages 20 and 65 containing information of usual employment from 2008 to 2015 (Table 2). The increase in the proportion reporting some employment increased significantly by 0.5% from 2008 to 2015 (proportions test p = 0.0337), but the increase consisted mostly of people reporting being unemployed.

The ten categories used to group occupations have remained the same over these ten years although the location of the deaths of people who worked for the armed forces moved within categories. They were placed in group 0 for 2006–2012 excluding 2008 where they had their own group and then placed in group 5 for 2013 to 2015. This makes an analysis of the trends in cause of death by occupational group difficult. The subcategories and minor occupation groups provide more useful data for analysis as different occupations are clearly defined [16].

The most recent three years of mortality data were combined in a total of 1,363,392 deaths which provided 450 deaths due to pneumoconiosis from 2013 to 2015. A proportion of only 26.3% of the pneumoconiosis deaths were in females.

The most common reported pneumoconiosis was Unspecified although this group appears to contain some misclassified deaths as the proportion of female deaths is much higher than in the other types and the main province of death was not a mining centre nor a well-recognised labour-sending area of South Africa [18]. Those with pneumoconiosis and TB appeared to die at a younger age while those with asbestos and mineral fibres exposure died at a later stage as seen in Table 3.

There are a number of significant sociodemographic and occupation differences observed between those with pneumoconiosis and those with other causes of death, using chi-square analysis (Table 4). These are likely to be factors associated with working in a dusty job or industry rather than pneumoconiosis itself.

These differences were then adjusted for in a logistic regression to identify those industries with increased odds of dying of pneumoconiosis (Table 5).

Only three industries showed significantly increased MORs for pneumoconiosis: Mining, manufacturing and maintenance of vehicles (Table 5). No industries show a protective effect although there were a number of industries with no cases of pneumoconiosis suggesting they are protective (not shown here). Mining accounted for the largest number of pneumoconiosis deaths where industry was specified.

Nine occupational groups had significantly increased odds of pneumoconiosis death (Table 6). When industry was analysed most of these were related to the mining or manufacturing industries despite their job titles (Fig. 3). A number of occupations had non-significantly increased odds. Some of these may be due to small numbers such as mining and construction managers.

The combination of occupation and industry provides some useful information such as the number of non-core occupations at risk within an industry, and the possible high-risk occupations within the industry (Fig. 3). In a number of deaths, information is only provided on one question either occupation or industry but not both. Combining them increases the information available. Some occupations are industry-specific while others are not and providing the industry those jobs were associated with, provides exposure information. For example, vehicle repairs and mechanical maintenance jobs can be found in many industries. However, only in mining and manufacturing are they at risk of pneumoconiosis. One pneumoconiosis death reported a nursing occupation. This case may have been a misclassified TB case as TB may be confused with pneumoconiosis, or it could represent a nurse employed on a mine. This case will require further investigation [19].

Figure 3 suggests that many unexpected jobs in the mining and manufacturing industries are at risk of dying of pneumoconiosis such as shop sales people and business and administration, not only those commonly associated with pneumoconiosis such as mining professionals and labourers although these do form the largest groups. In the manufacturing industry, there were fewer types of jobs at risk compared to the mining industry. Of further interest in the value of the data and possible improvement, pensioners formed the largest part (94%) of the non-economically active group at risk of pneumoconiosis illustrating the value of collecting correct data. The question is “What was the person’s usual occupation?” not “What was the current occupation?” This means pensioner is not a valid answer.