Of 1,553 breast cancer cases referred, 160 were ineligible and 222 were unable to be reached. Of the remainder, 165 declined, leaving 1006 cases for a participation rate of 86%. Of 3,662 households contacted for community control recruitment, 3,223 individuals were able to be apprised of the study and 926 households (29%) were determined to have no eligible residents. From 2,297 households with eligible residents, 1,146 women participated for a recruitment rate of 50%. The same percentages of cases and controls elected to be interviewed by telephone (46%) and in-person (54%).
Compared to controls, cases were slightly older, had a longer period of fecundity (from menarche to menopause or participation date, whichever came earlier) and fewer months of breast feeding; they had less education, lower family income, and smoked more but had almost identical duration of employment (Table ). There is no information available regarding the occupational histories of non-participants or expected employment sector distribution. However, it is unlikely, based on the almost identical duration of employment of cases and controls, that employment status influenced participation. Moreover, during recruitment, the research focus on occupation was not known to potential participants and therefore would not have biased participation. The differences between cases and controls, which were potentially confounding, were adjusted for in the age-matched statistical models. The difference in average date of participation (controls) vs. average date of diagnosis (cases), which determined when exposure assessment ceased, was less than 6 months (Table ).
Descriptive statistics for breast cancer cases and controls
There were considerably more cases than controls among subjects whose minor sector of longest duration was a) agriculture: 37 vs. 23 cases, b) food manufacturing: 30 vs. 10, c) automotive plastics manufacturing: 26 vs. 9, d) laundry/dry cleaning: 8 vs. 2 and e) bars-gambling (16 vs. 11) (Table ). Very few subjects reported no employment (4 controls, 8 cases; Table ). Cumulative exposure was similar or less in cases versus controls in some major sectors of interest – petrochemicals, transport, beauty care/laundry/dry cleaning – but considerably higher in farming, plastics manufacturing, metallurgical/metalworking and bars-gambling work.
When classified on minor sector of longest (lagged) duration of employment, and analyzed with conditional logistic regression, several demographic and reproductive risk factors exhibited strong, statistically significant associations as did several minor sectors of employment (Table ). The odds of being a breast cancer case were 5 percent lower with each additional pregnancy, and greater by 2.5 percent for each additional year of fecundity. The odds were 47 percent higher for women with less than high-school education. The odds, with a family income higher than $40,000, were lower for both blue collar workers (43 percent lower) and white collar workers (37 percent lower). Risk of breast cancer was higher per pack-year in smokers (OR = 1.02; 95% CI, 1.00-1.04) but with a slight attenuation of effect with increasing pack-years (negative quadratic term). For 20 pack-years, the smoking OR was exp(20×0.019-400×0.00033) = 1.28. The minor employment sectors showing elevated odds of breast cancer were food manufacturing (OR = 2.25; 95% CI, 0.97-5.26) and automotive plastics manufacturing (OR = 3.12; 95% CI, 1.29-7.55). Both laundry/dry cleaning and bars-gambling work were associated with increased odds of breast cancer (OR= 2.72, 95% CI, 0.56-13.2 and OR = 1.79, 95% CI, 0.73-4.41, respectively) that were not statistically significant because of small numbers. In this model, work in any other sector than the longest was disregarded. The restaurant sector was the reference group in this analysis (with a mutually exclusive and exhaustive classification, one sector must play that role). Analyses were repeated specifying the large retail sector as reference (data not shown). That sector appeared to have less than average breast cancer risk (Tables , ) and, as a result, all the estimates for other sectors increased considerably when compared to retail. For example, the automotive plastics OR increased from 3.12 to 5.38 (95% CI, 2.34-12.4).
Matched case–control analysis for breast cancer incidence with classification on minor sector of longest duration, and reproductive and demographic risk factors: full model, by conditional logistic regression
When durations in the minor sectors (lagged) were analyzed in the model (Table ), food manufacturing and dry cleaning/laundry were no longer elevated, but agriculture/plants minor sector was elevated (OR=1.02 per year, 95%CI=0.99-1.05), and plastics manufacturing (auto) (OR=1.09 per year, 95%CI=1.03-1.15; p=0.0023) now had a more significant effect (χ2=9.25 vs. 6.97), implying an improved model fit. One year in plastics (auto) employment was estimated to increase the odds of breast cancer by 9 percent. Inclusion of terms for total employment duration (lifetime employment as of study age) and the square of that term, produced a better fitting model with breast cancer risk declining with total employment (χ2(2df) =5.84, p=0.05).
Breast cancer odds ratios (matched analysis) for duration (lagged) in minor sectors excluding terms for sectors likely to have low work-related risk (mass media, education, healthcare, entertainment)
Models with cumulative exposure
Using the generic cumulative exposure metric (across all minor sectors) with the 0, 1, 2 exposure weighting scheme produced a statistically significant excess risk of breast cancer; 10 years in a high-exposed job had an associated 29% increase (OR = 1.29; 95% CI, 1.10-1.51) (Table , model 1). With the (0,1,10) weighting scheme, a stronger association resulted (OR = 1.42; 95% CI, 1.18-1.73), with a 42% increase in risk after 10 years in jobs assessed as likely high-exposure (model 2). Applying the 0, 1, 10 weighting scheme within major sectors identified excess breast cancer risk: in agriculture (OR = 1.34; 95% CI, 1.03-1.74; for 10 years in high-exposure jobs), plastics (OR = 2.43; 95% CI, 1.39-4.22), metal work (OR = 1.73; 95% CI, 1.02-2.92) and in bars-gambling work (OR = 2.20; 95% CI, 0.91-5.29) (model 3). There was no additional risk, beyond that found in farming in general, for specific farming activities involving corn cultivation since 1978 when atrazine use became common or greenhouse work. The excess in chemicals/petrochemicals was based on only 6 cases. Including additional terms for categories of special interest slightly strengthened the major category associations (Table , model 4).
Breast cancer odds ratios (matched analysis) with cumulative exposures, in major sectors and for derived hypotheses, and interactions with prior agricultural work, by conditional logistic regression
The analysis revealed excess risk with work in high exposure food canning jobs (OR = 2.35; 95% CI, 1.00-5.53, for 10 years work) (Table , model 4). This metric was motivated by the endocrine disruptor hypothesis and by preliminary findings of an excess in those for whom food manufacturing was the sector of longest duration (Table ). There was a possible excess in a group that includes toll booth operators (with potentially high vehicle emission exposures) (OR = 1.17; 95% CI, 0.44-3.14) but this group was limited by small numbers. The strongest association was with automotive plastics manufacturing (OR = 2.68; 95% CI, 1.47-4.88, p=0.0013). Within the auto industry in general, excess breast cancer appeared to be limited to small automotive parts suppliers, which would include some plastics operations (OR = 2.48; 95% CI, 1.00-6.10).
Effect modification and windows of vulnerability
There was no evidence of risk modification related to prior work in agriculture for subsequent work in metals or canning (Table , model 5). For bars-gambling work the estimate for the interaction term was stronger (OR=2.38, 95%CI=0.58-9.79; for 1 year of farm work prior to 10 years of bars-gambling exposure) than for the main effect, although both were not statistically significant (Table , model 5). For automotive plastics the estimate of a doubling of risk for one year of prior farm work was not statistically significant (OR=2.31, 95%CI=0.53-9.98).
Partitioning the generic Cumulative Exposure Metrics I and II into time-windows suggests that the most important exposures affecting breast cancer risk occur in the third time window – from first full term pregnancy to menopause; the elevation was smaller for the first, second and fourth time-windows although there was limited power to distinguish them (Table ). For Metric II, the point estimates for the second and third windows were close. Exposures in farming and bars-gambling work exhibited the same pattern whereas for the metal-related, plastics, and canning metrics the most important period appeared to be the second time-window – menarche to first full term pregnancy – before breast tissue is fully differentiated.
Breast cancer odds ratios for cumulative exposure accruing in time-windows reflecting reproductive status, by conditional multiple logistic regression
Hormone receptor type and menopausal status
Examination of specific estrogen receptor (ER) or progesterone receptor (PR) types in the major sectors showing excess breast cancer produced distinct associations across receptor types (Table ). The farming, metals, bars-gambling and particularly automotive plastics (OR = 3.63; 95% CI, 1.90-6.94, p=10-4) sectors all exhibited excesses for the ER+/PR+ receptor type, but farming had a stronger excess in the ER- category (OR = 1.71; 95% CI, 1.12-2.62, p=0.014). The canning excess appeared to be entirely in the ER+ /PR- and ER- groups. Including the interaction terms for prior farm work identified possible effect modification for metals (ER+/PR-), bars-gambling (ER+ /PR+), and plastics (ER-), and a statistically significant interaction for prior farming and canning for the ER+ /PR- receptor status (OR = 1.81; 95% CI, 1.08-3.04, p=0.025) but not for ER+/PR+ or ER- receptor status.
Breast cancer odds ratios (matched analysis) in selected major sectors on tumor estrogen receptor status, and with interaction on prior farm work
Models fit with an additive relative rate specification generally fit less well than with the loglinear form. For example, the automotive plastics estimate with the loglinear model was OR=2.68 (1.47-4.88), p=0.0013 whereas the linear relative rate model produced OR=4.03 (1.43-6.64), p=0.023. With the interaction terms, the same pattern was observed as with the loglinear form, but confidence intervals were wider.
Restricting the analysis to premenopausal women resulted in many fewer cases (373 out of 1006) and considerably higher estimates of relative risk (Table ) as in high exposed jobs in automotive plastics (OR=5.10, 95% CI=1.68-15.5) or canning (OR=5.20, 95% CI=0.95-28.4). Thus 10 yrs in that work was associated with a five-fold excess in breast cancer incidence. Adding a term for body mass index (BMI, centered at 25) produced a reduced odds of breast cancer with BMI (for 10 unit increase, OR = 0.78; 95% CI, 0.61-0.99), a slightly weaker association for automotive plastics, and a stronger association for canning (OR = 5.70; 95% CI, 1.03-31.5). In the analysis of postmenopausal breast cancer (633 cases), estimated risks associated with specific sectors were lower, particularly for automotive plastics and canning sectors. Terms for total employment duration, which were not statistically significant for premenopausal breast cancer, were statistically significant for postmenopausal cancer, with an estimated 6% decline in risk for each additional year of employment. BMI was a strong risk factor for postmenopausal breast cancer (for 10 unit increase in BMI, OR = 1.37; 95% CI, 1.12-1.68) but with small changes in major-sector risk estimates on addition of the BMI term.
Breast cancer odds ratios (matched analysis) and menopausal status with BMI and selected risk factors and major sectors, by conditional logistic regression