The EPIC study is an ongoing multicenter prospective cohort study which comprises 23 research centers in 10 European countries (France, Italy, Spain, the United Kingdom, The Netherlands, Greece, Germany, Sweden, Denmark, and Norway). Norway, Denmark, and a center from Sweden (Malmo) did not participate in the present nested case–control study. All EPIC study participants signed an informed consent at recruitment (range: 1992–2000), and the study was approved by both the ethical review boards from the IARC, and local ethics committees. Details of the study methodology have been previously described (22).

The EPIC study includes 153,427 men and 367,903 women. At recruitment, participants completed country-specific, validated dietary questionnaires (DQ) with the time frame referring to the previous year. Information on lifestyle factors (such as tobacco smoking, level of education, socioeconomic status, alcohol consumption, and physical activity), anthropometric factors, brief occupational history, and medical history were also assessed at recruitment. Women also reported baseline information on menstrual and reproductive factors [i.e., age at first menstrual period, pregnancy, use of OC, use of hormone replacement therapy (HRT), and menopausal status].

The standardized protocol followed to collect and store blood samples at recruitment has been previously published (22). Briefly, almost 80% of the EPIC participants, of which 226,673 were women, provided a single blood sample. Most of the samples were stored in liquid nitrogen (−196°C) at the IARC bio-bank; however, samples from Sweden (Umeå) were stored in freezers (−80°C) at the Medical Biobank of Northern Sweden.

Incident EOC were classified according to the International Classification of Diseases for Oncology (ICD-0-3), and included epithelial borderline tumors (C56.9), invasive epithelial ovarian (C56.9), fallopian tube (C57.0), and primary peritoneal (C48) cancers. Incident EOC were recorded through a combination of methods (health insurance records, cancer and pathology registries, and active follow-up), or via population cancer registries.

Cases and controls for the present nested case–control study were selected according to the methodology described by Peeters and colleagues (23). To summarize, for each case (participant who developed an ovarian, fallopian tube, or peritoneal tumor after the date of blood draw and before the end of follow-up) two controls free of cancer (with the exception of nonmelanoma skin cancer) were randomly selected at the time of diagnosis using a density sampling protocol. Matching criteria included study center, menopausal status (premenopausal, postmenopausal, perimenopausal), age at recruitment (±6 months), time of the day of blood collection (±1 hour), and fasting status (<3, 3–6, >6 hours). For the current study of hemoglobin adducts, one control per case was selected. Because acrylamide may disrupt hormonal levels, and tobacco smoking is an important source of acrylamide exposure (7, 24, 25), this study only included women who at baseline reported being postmenopausal and nonsmokers (thus, individual matching was broken). Postmenopausal women were defined as those who were >55 years old, or who reported not having had any menses during the 12 months before recruitment. Nonsmokers women were defined as those who reported never smoking or having given up smoking ≥5 years before recruitment.

A total of 751 participants (334 EOC cases and 417 controls) were included in the study. EOC comprised both borderline (n = 2, 1%) and invasive tumors (n = 332, 99%). Invasive EOC were classified into subtypes: serous (n = 191, 58%), endometrioid (n = 26, 8%), mucinous (n = 18, 5%), clear cell (n = 12, 3%), not otherwise specified (NOS) which included adenocarcinomas, carcinomas, and cystadenocarcinoma (n = 79, 24%), and others (n = 6, 2%).

Blood samples were sent to the Center for Disease Control and Prevention (CDC) Protein Biomarker Laboratory (Atlanta, USA) to measure HbAA and HbGA. Details of the methodology can be found elsewhere (7, 26). Briefly, 300 mL of red blood cells were hemolyzed and analyzed using HPLC/tandem mass spectrometry (HPLC/MS-MS). Laboratory personnel were blinded to the case–control status of participants, and blood samples were analyzed in a randomized manner. Concentrations of HbAA and HbGA were reported relative to the amount of hemoglobin (pmol per g of Hb), and two independent measures were performed for each sample. The lower limits of detection for this method are 3 pmol/g of Hb for HbAA, and 4 pmol/g of Hb for HbGA. All of the HbAA and HbGA measurements were within the limits of detection. In this study, 42 of the 751 blood samples were sent in duplicate to the laboratory to independently assess the reproducibility of the hemoglobin adduct measures, which had intraclass correlation coefficients of 0.94 for HbAA and 0.92 for HbGA. The percent coefficient of variation (CV) was estimated using log-transformed (log2) values, and was 9.9 for HbAA and 12.0 for HbGA.

Unconditional logistic regression models were used to estimate odds ratios (OR) and 95% confidence intervals (CI) for the association between biomarkers levels of acrylamide and glycidamide and the risk of EOC. Conditional logistic regression model were also evaluated in a sensitivity analyses.

All statistical models were adjusted for matching factors [age at recruitment (in years), country, time of the day of blood draw, date of blood draw, and fasting status] and covariates including OC use (never, ever, unknown), HRT use (never, ever, unknown), alcohol consumption (nondrinkers, drinkers of 0–6, >6–12, >12–24, and >24 g/day), parity (nulliparous, 1, 2, ≥3, parous but with missing number of full-term pregnancies), age at menopause (years), age at first menstrual period (years), and BMI (kg/m²). Lifestyle, anthropometric, and reproductive variables such as physical activity using the Cambridge index (27), education level (none, primary, technical/professional, secondary, and higher education), height (cm),weight (kg), hip circumference (cm), waist circumference (cm), duration of using OC (years), duration of using HRT (years), and age at first birth (years) were evaluated as potential confounders, but were not included in final models because they did not change effect estimates >10%.

Restricted cubic splines with 3, 4, and 5 knots were evaluated, and indicated nonmonotonic relations between each of the four biomarker variables and EOC risk. Because the relations were not linear, even when exposure variables were logarithmically (log₂) transformed, results for continuous biomarker variables were not presented (28). For each biomarker quintile, the median was estimated, and was included in a score test to evaluate dose–response trends. The four continuous biomarker variables HbAA, HbGA, sum of total adducts (HbAA+HbGA) and HbGA/HbAA ratio were categorized into quintiles based on the exposure distribution in controls. Biomarker quartiles were evaluated in stratified analyses.

Analyses were also carried out excluding borderline tumors (n = 2), and by histologic subtypes: invasive serous EOC, invasive serous EOC combined with NOS, and nonserous EOC (which included endometrioid, mucinous, clear cell, and NOS tumors).

Effect measure modification was evaluated by BMI (<25 kg/m² vs. ≥25 kg/m²), HRT (never vs. ever users), OC (never vs. ever users), and alcohol intake (never vs. ever drinkers) using a likelihood ratio test (LRT). These variables were selected because they are established risk or preventive factors, or because they may affect the activity of Cyp2e1 (29). All statistical tests were two-sided and evaluated at α-level 0.05. All analyses were performed using SAS v. 9.1.

The present nested case–control study was based on 334 incident EOC cases (of which 191 were classified as serous) and 417 controls. A large proportion of cases and controls were from the United Kingdom and the Netherlands (Table 1). Among cases, the median (quartile range) of HbAA and HbGA levels were 42.2 (33.9–54.4) and 37.0 (28.5–49.5), respectively, whereas controls had HbAA and HbGA levels of 43.1 (33.8–54.8) and 35.4 (26.0–49.9), respectively (Table 1). Cases were slightly younger than controls (58.4 years vs. 59.2 years), tended to have higher BMI values (26.4 vs. 25.8 kg/m²), a higher proportion of HRT users (27.8% vs. 18.9%), and were less likely to take OC (35.6% vs. 41.7%). There were no major differences between cases and controls regarding age at menopause, age at first menstrual period, and parity (Table 1). The median interval between the date of blood draw and the date at diagnosis for cases was 6.2 years.

Four multivariate unconditional logistic regression analyses were performed for the association between each biomarker exposure variable and EOC risk. No associations were observed between HbAA levels analyzed in quintiles and EOC risk. Participants with HbGA levels >52.71 pmol/g of Hb (fifth quintile) were at nonsignificant increased EOC risk (OR_Q5vsQ1, 1.63; 95% CI, 0.92–2.86). The sum of total adducts was also analyzed. Compared to women with ≤56.70 pmol/g of Hb (reference group), the ORs for the fourth and fifth quintiles were elevated but none were statistically significant. Participants classified in the second and third quintile of HbAA+HbGA were at higher risk of developing EOC (OR_Q2vsQ1, 1.81; 95% CI, 1.06–3.10) and (OR_Q3vsQ1, 2.00; 95% CI, 1.16–3.45).

Similar models were also evaluated for invasive serous EOC. Despite not observing any statistically significant associations between biomarker levels (HbAA, HbGA, HbAA+HbGA, and HbGA/HbAA) and serous EOC risk, positive nonstatistically significant associations were observed for upper versus lower quintiles of HbGA and HbAA+HbGA (Table 2). Similar patterns were found when borderline tumors were excluded, when nonserous tumors were evaluated, and when invasive serous and NOS were combined in the same analyses (data not shown).

Sensitivity analyses were conducted using conditional logistic regression models, which included 261 cases and 416 controls, to estimate ORs of EOC for each biomarker level. Overall, no statistically significant association were observed; nonetheless, results showed similar patters compared to the ones obtained using unconditional logistic regression models (Table 2).

Although some individual ORs were statistically significant, no consistent evidence for effect measure modification by BMI, alcohol intake, OC use (all LRT P-values >0.07; Table 3), or by HRT use (data not shown) was observed.