Pest controllers manage or eliminate various animals which present as a nuisance or threat to human health, including colonial insects, rodents, structural pests, food storage pests and spiders as well as large mammals and birds.1 Domestic environments, commercial food-premises and grain storage facilities are among the areas most frequently serviced by pest controllers. They can encounter a range of hazards in the course of their work, including exposure to pesticides and other hazards associated with work in unhealthy buildings and structures of parts of buildings not normally intended for human access, including confined/constrained spaces. Specific physical hazards may include heights, asbestos, dust, moulds and animal bites and stings.1 2

In spite of the inherently hazardous nature of the job, there has been little published literature on the occupational health of these workers.2 The majority of occupational pesticide users are found in agricultural jobs.3 However, since mixing and application of pesticides is the core task of pest control work they use pesticides more frequently, typically mixing and applying pesticides daily, whereas pesticide use is one of a variety of seasonal tasks undertaken by farm workers.3 4

To date there have been a small number of mortality cohort studies of non-agricultural pest controllers published and these have shown inconsistent findings in relation to cancer mortality.2 5 6 7 During the 1960s–1980s two North American cohort studies of private-sector pest control operators were undertaken. Both of these studies reported excess rates of lung cancer6 8 and one also found that risk was higher in those licensed before age 40 and appeared to increase with increasing years licensed.6 In contrast, a British mortality study of municipal pest control officers found no significant elevations in deaths by any cancer type. Mortality by all causes, lung cancer or by non-malignant respiratory disease were all significantly lower than expected.7 Conversely, a recent study of mortality among French municipal pest control operators found that all-cause mortality and all cancer mortality were both significantly in excess.

Being studies of mortality rather than cancer incidence, these studies are likely to have been subject to biases associated with survivorship resulting in underestimation of certain cancer types, particularly those for which lethality is low either because of amenability to treatment or slow natural progression of the disease. The two earlier North American cohorts comprised 16 124 subjects5 and 4411 subjects,6 respectively, however, the two later European cohorts were much smaller, each having fewer than 300 subjects.2 7 Also, both of the recent European studies were based on municipal pest controllers, employed by local government, whose work practices, occupational exposures and socioeconomic status may be expected to differ from the private sector structural/domestic pest controllers studied in the two earlier North American studies.

During the 1960s–1980s the state government of New South Wales (NSW) in Australia provided health surveillance programmes for pesticide-exposed workers. The existence of a national, population-based cancer incidence registry in Australia, together with the records of these occupational health surveillance programmes provided a unique opportunity to assess cancer incidence and mortality amongst a group of Australian pest control operators.

Methods

Study population

The study cohort was assembled from archive records belonging to the NSW state government. The state government conducted extensive programmes during the 1960s–1980s offering free biomonitoring and occupational health advice to pesticide-exposed workers, including agricultural workers, foresters, municipal and public utilities maintenance workers, greenkeepers, chemical production/distribution workers as well as pest control operators. The pesticide surveillance programmes were based on regular regional visits by teams of state government occupational health personnel who would set up a temporary testing centre in a local town for a fixed period. The visits were publicised in advance of each visit, including through employers, industry associations and generally to the public. These programmes were fully funded by the state government and the services offered were free of charge. Biomonitoring tests made available as part of the surveillance programmes included tests for cholinesterase inhibition and tests for specific pesticides including organophosphates, herbicides, fungicides and organometals, as relevant to the individual’s exposure. These programmes provided ongoing health surveillance for workers and identification of potential cases of overexposure requiring active follow-up of individuals. The archive records of these programmes consisted of individual worker histories, test results and some other administrative documents including results letters and programme participant lists.

For this study, records were scanned in the government archives. This procedure enabled the data to be collected without removal of the original files from the archives, a condition of usage imposed by the data custodians. The scanned images were used to extract personal identifiers including name, address, date of birth, job information and interview dates. Data items were entered into a purpose-built SQL Server database with Microsoft Access interface, creating electronic records for each worker, each with a unique study identification number. Individual workers typically had several interview records each. Records of all interviews relating to the same worker were entered against the individual’s identification number in the database.

Where personal data (names, addresses, etc) were unclear or missing in scanned original records, checks were undertaken using the Australian online telephone directory,9 Whereis.com, an Australian online street directory,10 and the Australia Post online postcode finder.11 Date of birth was entered where available. For records where only age was recorded, year of birth was estimated using the age recorded and the interview date.

Individuals were eligible for inclusion in the study cohort if they were recorded as pest control operators in the occupation field of the original records.

Cancer and mortality record linkage

Relevant identifier variables for each cohort member were supplied to the Australian Institute of Health and Welfare (AIHW) for linkage to the National Death Index (NDI) and the National Cancer Statistics Clearinghouse (NCSCH). The same dataset was also provided to the Cancer Council Victoria for linkage to the Victorian Cancer Registry (VCR), which is not included in the NCSCH for the purposes of record linkage.

Linkage to the NCSCH and NDI was undertaken by the AIHW using customised probabilistic record linkage software based on Integrity version 3.5 (Ascential Software, www.ascentialsoftware.com). Linkage was performed using multiple passes that group the data based on different matching criteria each time, such as surname, date of birth and sex.12 Possible matches of cohort members to registry records were returned to the research team for review. For each case matched to the NCSCH, the AIHW provided a four-digit international classification of diseases (ICD)-10 anatomical site code, tumour morphology code, dates of registration, of diagnosis and of birth as well as state or territory of registration. Similarly, the NDI provided causes of death (in ICD-9 or ICD-10), state of registration and dates of death, of registration and of birth. The Cancer Council Victoria used similar linkage software and procedures and provided data items equivalent to those from the NCSCH.

The cohort was linked to the NDI for the years 1983–2004 and to the NCSCH/VCR for the years 1983–2002. 1983 is the earliest year of complete population coverage by the registries. The linkages ended in 2004 and 2002 because at the time of linkage these were the most recent years for which complete national death and cancer data, respectively, were available.

Two members of the study team independently reviewed the possible matches provided by the registries and made decisions as to which matches were sufficiently convincing for inclusion in the dataset for analysis. The two reviewers met to examine and reach consensus on cases where their initial decisions had differed.

Population data

National population mortality data until the end of 2004 were obtained from the AIHW General Record of Incidence of Mortality (GRIM) Books which are interactive electronic workbooks containing annual population mortality data by cause of death.13 National cancer incidence data for each cancer type (by anatomical site), for each 5-year age group and calendar year to the end of 2002 were obtained from the AIHW.

Statistical analysis

Analyses were performed using STATA statistical software package, V.9.14 The person-years contributed by each cohort member to the cohort were calculated using the STATA procedures: STSET, STSPLIT and STPTIME. Person-years were allocated into groups based on calendar year and 5-year age bands. Subjects commenced contributing person-years at 1 January 1983 or at the date of their 15th birthday, whichever was later. Subjects ceased to contribute person-years at the date of first cancer diagnosis or death in the incident cancer and mortality analyses, respectively.

Standardised mortality/incidence ratios15 were calculated using the STRATE procedure in STATA which compares the cohort rates with the population rates by calendar year and 5-year age bands.

Ethics approvals

This study was conducted under the approval of the Monash University Standing Committee on Ethics in Research Involving Humans. Linkage to the registries was approved by the AIHW Ethics Committee and access to the NCSCH was also approved by each of the eight Australian state and territory cancer registries.

Results

There were only 19 female pest control workers included in the cohort. None of these women was found to have died or experienced cancer during the follow-up period. The remainder of this paper deals with the male cohort of 1813 workers.

One thousand eight hundred and thirteen male workers presented for testing and gave their occupation as “pest controller”, “pest control serviceman/operator/technician” or similar. Of these, 623 (34.4%) workers provided descriptive information about their job such as company name or more detailed information about specialised pest control tasks: 27 individuals (4.3% of those who provided additional job information) identified themselves as fumigators and 16 (2.57%) were recorded specifically as being state government Grain Handling Authority pest controllers, implying that they were likely to be involved in grain fumigation. Sixty-three point nine per cent of those who provided additional information (n = 398) included the name of a major domestic and commercial pest control company in their job description and 26.2% (n = 163) described themselves as self-employed pest controllers. A review of available self-reported pesticide usage information in the original records suggested practically all subjects for whom pesticide usage information was recorded reported using both organochlorines and anticholinesterase insecticides (organophosphates and carbamantes). Approximately one-third of subjects also reported use of organometal pesticides and one-third reported using pyrethroids.

The 1813 male workers were linked to the NDI (1983–2004) and the NCSCH (1983–2002), contributing 38 575 person-years to the mortality analysis and 34 838 person-years to the cancer incidence analysis.

Table 1 shows distributions of year of birth and age at the end of the follow-up period. The majority of cohort members were born during the 1940s, 1950s and 1960s and these cohort members were still within usual working age at the end of the follow-up period. The majority of the cohort members had multiple interviews and the period of individuals’ involvement in the surveillance programme, the interval between first and last recorded interview, was most commonly <2 years. The longest period of involvement was 8.8 years.

Table 2 shows standardised mortality ratios (SMRs) for major categories of death and major subcategories of respiratory, vascular and injury deaths. One hundred and twenty-five cohort members were determined to have died from any cause. The overall SMR was not significantly different from expected and this was the pattern across the major categories of death. The mortality rate for injury as a whole was slightly elevated, although this was not statistically significant.

Injury mortality subcategories are shown in table 3. Although in most categories the numbers were small, suicide mortality was significantly greater than expected based on 18 observed deaths. Among subcategories of suicide the numbers were extremely small, however, the largest group was intentional self-poisoning by gassing, which was significantly in excess. Two of these cases were coded as gassings of unspecified type, but the remainder were vehicle exhaust gassings.

Death from an unintentional fall was also significantly greater than expected, but this was based on only four deaths.

Due to coding changes in the NDI, unintentional poisoning was not a distinct cause of death category for deaths prior to the beginning of 1997. Therefore the SMR for unintentional poisoning reported in table 3 is for 1997–2004 only. There were only two unintentional poisoning deaths observed in this period.

Cancer records were linked to 89 cohort members and this was very close to the expected number (table 4). Melanoma was the only cancer significantly in excess. Colorectal, gallbladder, pancreas, larynx, urinary bladder and testicular cancers were slightly elevated relative to the population rates but not significantly so. All other cancer types were less than expected but not significantly so.

Discussion

The all-cause SMR in this cohort appeared to be similar to the general population. This was somewhat unexpected since occupational cohorts commonly show a “healthy worker effect” due to self-selection of less healthy individuals out of the workforce.15 Therefore the similar overall mortality in this cohort compared with the general population (0.94; 95% CI 0.78 to 1.12) is in contrast to the significantly low overall SMRs in other occupational cohorts studies which have used the general population for comparison. For example, recently published occupational cohort studies of mortality found overall SMRs of 0.72 (95% CI 0.68 to 0.77) in the Australian petroleum industry and 0.68 (95% CI 0.60 to 0.77) in the Australian primary aluminium industry.16 17

Of the published studies of pest control workers, only one has shown a clear healthy worker effect, and this was a small cohort of municipal pest control officers in Britain.7 Both of the large cohort studies of private sector pest control workers lacked a healthy worker effect, with overall mortality not significantly different from the comparison population.5 6 The lack of a healthy worker effect in our cohort may be associated with the fact that participation in the original surveillance programmes was not compulsory and therefore it is possible that less healthy individuals presented disproportionately.

Injury-related mortality was as expected compared with the Australian population, although amongst the specific injury-related causes of death, suicide and unintentional falls were significantly elevated. The greater than expected number of unintentional falls was based on only four observed deaths. This increase has not previously been reported, although working in parts of buildings not usually intended for human access is a hazard associated with the work of many pest controllers1 and may plausibly increase the risk of falls. All four falls deaths were <65 years of age at the time of death but the available cause of death coding information does not reveal whether the falls were directly work-related.

We found suicide mortality to be significantly elevated. Suicide mortality in pest control workers has been reported by only one published study to date and in that study there was only a single suicide death observed.2 Pest controllers typically work in isolation with access to a range of objects with lethal potential, including motor vehicles and poisons,1 and it is widely accepted that suicide risk is influenced by access to lethal means.18 Although the numbers were small in the subcategories of suicide, gassing appeared to be significantly elevated and it is plausible that working alone out of their own van may constitute ready availability of means in this context.

Selection bias may have been operating if pest control workers predisposed to suicidal behaviours were more likely to present for state government health surveillance programmes. However, there is evidence that long-term, low-level exposure to organophosphate pesticides is associated with neurological effects.19 20 In particular there is evidence that neuropsychological effects, particularly mood disorders, may be elevated in organophosphate-exposed workers21 22 23 and among exposed workers who have experienced at least one over-exposure event.24 25 Mood disorders are a predictor of suicidal behaviour19 and therefore it is biologically plausible that organophosphate exposure itself may contribute to suicide risk by a neuropsychological mechanism. To date there have been no studies of neurological effects in pest control workers, previous research being based mainly on agricultural workers. Mortality data alone are insufficient to provide insight into possible mechanisms for the suicide excess observed in this study and this finding indicates that more research is needed in this area.

Unintentional poisoning was not significantly elevated, however, unintentional poisoning has only been a distinct cause of death category since the introduction of ICD-10 coding in 1997 in the NDI and therefore the follow-up period for this SMR is shorter than for other outcomes reported. Updated linkage of this cohort as its members age should provide more information and at present this result should be interpreted cautiously.

The incidence of all cancer was very similar to that in the general population. The only specific cancer type found in excess was melanoma. This was unexpected and the available data for this cohort do not lend themselves to any further exploration of this finding. Occupational sun exposure is unlikely to be a major contributor to this excess since pest control operators typically work indoors.1 Although the excess of melanoma is based on 22 cases and the confidence interval is narrow, the lower confidence limit is close to 1.0 and it could be a chance finding. However, if this excess is real it could suggest a possible early diagnosis effect among this cohort of working men who may be expected to have comparatively good access to health resources and therefore a higher likelihood of early diagnosis. Interrogation of the mortality data revealed that there were only four melanoma deaths (SMR 2.02; 95% CI 0.76 to 5.41), too few to confirm or refute this possible explanation.

There were six cases of testicular cancer in the cohort and this was about twice the expected number. Although the number of cases was small and this excess did not reach statistical significance it is concordant with the findings of other investigators who have also found excess testicular cancer among pesticide-exposed workers,26 27 particularly associated with organochlorines28 and insecticides.29 Future follow-up of the present cohort as its members age should provide further insight into this apparent excess.

There is some consistent evidence that non-Hodgkin’s lymphoma, prostate and pancreatic cancers may be associated with pesticide exposure,30 and there is some evidence, though less consistent, for associations with primary liver cancer, multiple myeloma and leukaemias.30 In our cohort of workers exposed to a range of pesticides on a daily basis we found little evidence to suggest that these cancers have occurred in excess of the rates in the general population although the SIR for pancreatic cancer is non-significantly elevated. At the present time, numbers of specific cancers are low and the age of the cohort members is comparatively young. Continued observation of this cohort as its members age, will show if this and other elevations are real or simply chance findings.

Of the previously published cancer mortality studies of pest controllers, two North American studies reported excess lung cancer rates,5 6 however, we find no evidence of excess incident lung cancer in our cohort. This finding is consistent with the two European cohorts of municipal pest controllers in which no elevation in lung cancer mortality risk was observed.2 7

A disadvantage of using archival records can be that the quality and nature of the cohort records is likely to have resulted in some underascertainment of outcomes due to problems in linkage of imperfect cohort records. This would result in falsely low outcome rates across all outcomes. Any underascertainment effect is likely to be small in this study because the records used to assemble this cohort had good-quality name and date of birth information for most subjects, the key variables for record linkage. Only 1.6% of cohort members (n = 29) had initials instead of full given name and <1% of records (n = 14) had incomplete dates of birth which required us to estimate year of birth from the subject’s age recorded at interview. Therefore it is reasonable to assume that outcome ascertainment was not affected by poor-quality records.

A second disadvantage of archival records is that they may lack important individual-level confounder information which is not feasible to collect retrospectively. Therefore analyses cannot be adjusted for confounders such as smoking. In this cohort, exposure was defined based on job title. However, these job titles had been obtained from the workers themselves at the time of the health interviews, prior to death or onset of cancer. Therefore misclassification of exposure status (based on job title) in cohort members is unlikely.

NSW public service staff who were involved in planning and delivering the original occupational health surveillance programmes at the time have reported to the investigators that coverage of pest control operators was very good, however, quantitative data about the contemporary population of licensed pest controllers are lacking. Although it is not possible to calculate the precise proportion of licensed pest controllers who are represented by this cohort, NSW public service staff formerly involved in the programmes report that about 70% of local pest control operators were involved in the programmes, and the rate was somewhat less for self-employed operators than for those employed by a firm, implying a degree of selection bias. All services were offered free of charge during this period and this was a major factor in the high recruitment rate and why these services dramatically declined after they changed to fee-for-service basis in the early 1990s. The possibility of differential presentation according to personal health consciousness, business activity or other factors is also possible and this is a limitation of this study and a further potential source of selection bias.

The inclusion of self-employed pest controllers in this cohort, despite some likely selection bias, is important since self-employed operators have not been included in previous studies of pest controllers based on large companies or municipal employees.2 7 8 The inclusion of the self-employed is particularly important since their health risk may be higher since they do not benefit from occupational health and safety programmes of large firms and are a group that is difficult for regulators to access.1 Although most cohort members were interviewed more than once, the majority were involved with the surveillance programmes for <4 years and this reflects the high turnover rates characteristic of this industry at the time.1

It should be noted that this cohort is comparatively small and conclusions should be treated with caution. However, the findings suggest that overall mortality and incident cancer rates in pest control workers were similar to those in the general population, although rates in a working cohort may be expected to be less than those in the general population. Follow-up linkage of this cohort to the cancer and death registries as the cohort members age is likely to provide more insight into specific causes of death and onset of specific types of cancer. The finding of an elevated suicide mortality rate is of particular concern in this population and indicates a need for further investigation of mental health and the accessibility of lethal means in this workforce.