Rate ratios for cancer incidence and adjusted hazard ratios are both discussed, although it is recommended more weight be given to the latter, as they control for additional risk factors. All-cancer incidence and adjusted hazard of a cancer diagnosis was lower in farm men, but differences were not statistically significant when compared to rural non-farm or to urban men separately. Farm women had non-significantly lower all-cancer incidence; but the adjusted hazard of a cancer diagnosis in farm women was significantly lower than rural non-farm women, controlling for other factors. There were no significant differences in either the standardised rate ratio or adjusted hazard ratio between cohorts for any of the individual cancers tested; although the incidence and adjusted hazard of lung cancer in farm women was around half that of other women. In this study, smoking was the most prominent modifiable risk factor in adjusted hazard models, having a particularly strong association with lung cancer. However, men who were current smokers were half as likely to be diagnosed with melanoma; and women with higher weekday sun exposure were least likely to be diagnosed with breast cancer.
Incidence
Consistent with the direction of the findings, most reviews and recent studies have reported reduced all-cancer incidence in farmers [
9,
11‐
14,
52,
53]. Some have attributed decreased cancer risk in farmers to a ‘healthy worker’ effect; a phenomenon observed when comparing occupational groups with the general population, that by nature exclude those who are unable to work for health reasons [
9,
11,
54,
55]. Most farm businesses in Australia are family operations with ongoing generational commitment resulting in older farmers continuing to work into and past normal retirement age [
56]. However, this study compared groups on a residential basis, which may have ameliorated occupational bias to some extent.
Comparative measures of smoking, alcohol and income-related risk factors for resident groups in this cohort presented elsewhere were generally more favourable amongst urban residents [
51]. However, greater physical activity was suggested amongst farm residents, by their higher weekday sun exposure [
51]. This may have contributed toward lower all cancer incidence, as suggested elsewhere [
53].
Despite the small number of farm resident cases in this study for men and women, the lower lung cancer incidence and risk in farm residents support data from other studies reporting on farmers [
7‐
13]. Lower smoking rates in farmers have often been suggested as the relevant factor, but this was not the case in this cohort, considering that urban men had lower current smoking rates [
51]; and lower cancer incidence in farmers remained even after controlling for smoking in the analyses. Exposure to farm animals and environmental endotoxins have also been reported as possible explanations for lower lung cancer incidence in farmers, which remains a possibility here, although exposure information was not available and therefore not able to be assessed [
57‐
60]. It is also possible differences in other, unmeasured risk factors, such as hormonal therapies, social characteristics and ethnicity, acted as potential confounders.
There was little discernible difference between groups in our study for the other selected cancers. Most recent studies of colorectal cancer in farmers have reported reduced incidence or risk in farmers. These have predominantly been large occupational cohort studies with a minimum follow-up of ten years [
9‐
13]. Four of these studies reported reduced risk of breast cancer in farm women, as did two other studies of similar design [
7,
8]. The only recent reports of excess breast and colorectal cancer in farm groups, have been from smaller case-control studies [
61,
62].
Findings for breast, melanoma and prostate cancer in farmers have been mixed, with several reporting no significant differences between farm and non-farm groups [
9,
11‐
13,
63‐
66]. Neutral findings have been reported for the majority of comparative studies of lymphoma in farmers published from 2008 to 2013 [
9,
11‐
13,
20,
22,
24,
67‐
73]. However, more recent case-control studies have reported an excess of lymphomas in farm groups [
15,
17,
18,
61], similar to earlier reviews of case control studies [
52,
53,
74].
One prominent meta-analysis highlighted the inconsistencies of results brought about by variations in study design, risk measures, farmer definitions and geographic location [
52]. A positive bias can occur in studies that use proportionate measures of risk in populations where the overall number of cases is small; and in case-control studies with non-population based controls [
52]. This could help explain why such studies more often report increased prostate cancer and NHL risk in farmers, compared to cohort studies, which more often report neutral or reduced risk [
52]. This effect was confirmed in a more recent review of prostate cancer risk in farmers published in 2014 [
28]. Since then, two more studies reflecting these issues have reported opposing results; [
6,
25] and a new meta-analysis limited to case-control studies, not unexpectedly reported higher risk in farmers [15]. In contrast, negative bias can be an issue in large cohort or occupational studies if there is limited information about possible confounders.
Risk factors
Other studies have suggested increased cancer incidence in rural areas may be attributed to higher smoking and alcohol use, lower access to or utilisation of health services; and employment or income disadvantage [
4,
75]. A greater proportion of rural non-farm residents in this cohort were current smokers and had lower incomes [
51]. However, as expected when controlling for these factors, there was no evidence of a difference in lung cancer risk between rural non-farm and urban men in the adjusted model. In addition, whilst findings were not significant, these risk factors did not explain the lower likelihood of lung cancer in farm residents compared to the other groups. Confirmation of this effect with a larger farm resident sample is warranted.
Nevertheless, findings support what is already known about the hazardous effect of smoking upon lung cancer and all-cancer. It also supports the current health promotion priorities of Australia’s health systems with a focus on prevention and reduction of tobacco use, especially amongst groups with a higher prevalence of smoking [
76]. The negative associations between smoking and breast cancer, prostate cancer and melanoma in men may have been an artefact of the relatively short follow-up period. However, a recent meta-analyses has also reported negative links between smoking and prostate cancer incidence and unclear links to breast cancer [
77,
78]. There have been reports of negative associations between smoking and melanoma - although the biological mechanisms are unclear [
79‐
81]. Overall, the negative associations with smoking had a relatively minor impact upon the relative patterns of risk between resident groups.
A related study of cancer mortality risk in this cohort, found that compared to very low exposure, weekday sun exposure of 1–4 h was protective against NHL, prostate, breast, melanoma and lung cancer mortality [
51]. This was also the case for melanoma incidence in this study. Others have similarly reported inverse melanoma risk with occupational or weekday patterns of sun exposure, as opposed to the more intermittent patterns giving rise to sunburn that raises melanoma risk [
82]. However, 4 h + sun exposure was most protective against breast cancer. Other studies have also suggested links between sun exposure, Vitamin D levels and reduced risk of breast cancer [
83‐
85]. However, it is also possible that moderate sun exposure represented greater relative health and outdoor physical activity, which is promoted in Australian cancer prevention guidelines [
37].
Several studies have explored positive associations between cancer incidence and farm environmental exposures, such as pesticides. However, these are outside the scope of this study, as they do not generally compare farm and non-farm groups; and farm exposure information was not available in this dataset.
The negative significant association between lung cancer in men and income, is consistent with findings elsewhere, relating to higher levels of smoking in lower socio-economic groups [
86]. Overweight and obesity was associated with breast cancer in this study, also consistent with reports in the health literature [
37]. However, contrary to evidence of links between alcohol consumption and breast, colorectal and other cancers, this was not associated with any of the selected cancers in this cohort [
37].
Limitations
There are a number of limitations in this study that may have affected the results. Firstly, data on incident cases at the time the research was conducted were only available for a relatively short period of follow-up, resulting in low power and wide confidence intervals for some analyses. This may have impacted upon the significance of some findings, favouring a bias toward the null. Discussion of results with confidence intervals that include unity should be considered exploratory; and larger, consistent differences given more weight. Nevertheless, results still offer insight into potential differences and guidance for further work.
In addition, to maximize both cases numbers and follow-up time, this study included all records of cancer for participants who could potentially receive a diagnosis of cancer at any time in the 2006–2009 study period; that is, cancer diagnosis in some participants could have preceded their enrolment in the 45 and Up Study. However, such an effect is likely to be non-differential relating to residence; and results of the sensitivity analyses were consistent with and support the main findings.
The need to exclude records with missing variable information from models may have impacted upon the results, although this is not likely to have been differential across groups or between cases and non-cases. Other limitations include the potential mobility of participants regarding their residential status and that only the more commonly known risk factors were considered for analyses. A myriad of other potential risk factors and confounders were not measured (e.g. social factors, ethnicity); and may have contributed to the differences observed.
The 45 and Up Study, even with its robust sampling methods, is not necessarily representative of the population of NSW aged 45 and over [
29]. However, it is one of the largest cohorts of its kind in the world; and there was little evidence of selection bias observed when associations between risk factors and disease in the 45 and Up Study population, were compared with those of another population-based dataset drawn from the same population using different methods [
87]. Over-sampling in rural areas to ensure representation of smaller population groups, is also likely to have minimised selection-bias at sub-group level. However, only internal comparisons between sub-groups have been made in this study, previously documented as valid and the most appropriate [
29]. Caution is therefore advised with the generalisation of results.
The definition of a ‘farm resident’ in this study was also open to respondents’ interpretation of ‘farm’, which could include small holdings used for commercial, recreational or both purposes. Exposures could be quite different depending on which of these purposes was dominant. Farm exposure differences and errors arising from misclassification of residence, are likely to have lessened any differences between resident groups, but not likely to have systematically affected non-residential risk factors. Therefore, any potential bias is likely toward the null and an underestimation of a relationship between farm residence and cancer incidence.