Correlates of circulating ovarian cancer early detection markers and their contribution to discrimination of early detection models: results from the EPIC cohort

Selection of the cases and controls for this nested case-control study has been described in detail previously [5]. Briefly, incident ovarian (n = 752), fallopian tube (n = 33), and primary peritoneal (n = 25) cancers were matched to up to four controls (n = 1938) on study recruitment center, age at blood donation (±6 months), time of the day of blood collection (±1 h), fasting status (<3 h, 3–6 h, >6 h), menopausal status at blood collection (premenopausal, perimenopausal, postmenopausal), and current use of exogenous hormones (OC, menopausal hormone therapy (HT)) at the time of blood draw, as well as menstrual cycle phase for premenopausal women (3–5 categories, depending on available data) using incidence density sampling.

The primary cross-sectional analyses included pre- and postmenopausal controls from the nested case-control study (n = 1910). Given established differences in circulating CA125 by menopausal status [8], cross-sectional analyses were restricted to women pre- or postmenopausal at time of blood collection. Women were considered premenopausal if they met one of the following criteria at blood collection: menstruated at least once in the prior year while not on hormones; were on hormones but were less than 50 years old; had a hysterectomy before last period and were less than 50 years old; or, age at last menstruation was missing and age was less than 50. Postmenopausal status was assigned to women who met one of the following criteria at blood collection: were not on hormones and had not menstruated in the past year, on hormones and age was 50 or greater, had a hysterectomy and age was 50 or greater, age at last menstruation was missing and age was 50 or greater. Controls that were perimenopausal or had unknown menopausal status (n = 28) were excluded. In a secondary analysis, we evaluated cross-sectional associations among cases (n = 791). Cases who were perimenopausal or had unknown menopausal status (n = 19) were excluded from these analyses.

Data on lifestyle and reproductive exposures, as well as anthropometric measures, were collected at baseline and included: age at menarche, age at blood draw, OC use and duration, HT use and duration, type of postmenopausal HT, parity, estimated number of ovulatory cycles (defined as the time between age at menopause and age at menarche not taking OCs or pregnant), phase of menstrual cycle at blood collection (premenopausal women), tubal ligation, hysterectomy, oophorectomy, BMI, smoking, and family history of breast cancer. Those missing exposure of interest were excluded from analyses for that exposure. Among controls, the following variables had missing observations: age at menarche (n = 78), OC use (n = 57), duration of OC use (n = 66), parity (n = 136), number of children (n=43), tubal ligation (n = 1661), IUD use at recruitment (n = 522), hysterectomy (n = 347), ovulatory cycles (n = 385), age at menopause (n = 244), HT duration (n = 166), HT at blood (n = 359), BMI (n = 94), smoking (n = 28), pack-years among smokers (n = 10), and family history of breast cancer (n = 1250).

All assays were performed in the Brigham and Women’s Hospital Obstetrics and Gynecology Laboratory of Genital Tract Biology using a volume-efficient highly sensitive multiplex platform (Meso Scale Discovery (MSD), Gaithersburg, MD, USA) based on electrochemiluminescence (ECL) detection. Single ECL assays for antigen detection of human CA125 (catalog number K151WC) and Human Prototype CA15.3 (Catalog number N45ZA-1) and all reagents related to these two assays were provided by MSD. The linearity range for CA125 was 0.6–10,000 U/ml, and for CA15.3 was 0.19–12,500 mU/ml. HE4 and CA72.4 were analyzed using a custom-designed duplex assay. The following reagents were provided by Fujirebio Diagnostics, Inc. (Malvern, PA): HE4 protein (IgHE4 antigen), which we used to generate a calibration curve with a linear range of 0.0137–3600 pM; anti-HE4 capture IgG1 (2H5 mouse hybridoma); anti-HE4 detection IgG1 (mouse hybridoma 3D8); TAG72 Defined Antigen, which we used to generate a calibrator curve with a linear range of 0.146–2400 U/ml; anti-CA72.4 capture IgG1 (mouse hybridoma CC49, Fujirebio catalog number 110–005); anti-CA72.4 detection IgG1 (mouse hybridoma B72.3). The samples were split into batches such that matched case-control sets and samples from the same study center were kept together in the same batches. The samples were tested undiluted in the CA125 singleplex and the HE4/CA72.4 duplex, and they were tested at a 50-fold dilution in the CA15.3 assay. Blinded quality control (QC) samples were included on each assay plate. In blinded QCs with values within the linearity range of each assay we observed the following interplate CVs and min-max (mean) intraplate CVs: 19% and 3–20 (9)% for CA125, 22% and 3–5% (4%) for CA15.3, 9% and 4–10% (6)% for HE4, 16% and 1–16% (6%) for CA72.4). CA72.4 concentrations were below the lower limit of detection in the blinded QC samples, therefore CVs are based on the remaining 13 aliquots (concentration range: 1.15 to 1.87 U/mL).

Biomarker concentrations were log-transformed to obtain a more normal data distribution. We assessed each biomarker for outliers using the generalized extreme studentized deviate many-outlier procedure [19]. Eight outliers were identified for HE4; the influence of these values was assessed in sensitivity analyses. No outliers were identified for CA125 or CA15.3. We used generalized linear models to estimate the mean CA125, CA15.3, and HE4 values across categories of each characteristic and exponentiated results to obtain geometric mean values in the original scale. Since the majority of the CA72.4 values (82%) were below the lower detection limit (1.119 U/mL), we used a logistic regression analysis with a dichotomous CA72.4 variable (≥1.119 vs. < 1.119 U/mL) as the outcome and results are presented only in a supplemental table. Wald tests of continuous variables were used to assess trend. All analyses were adjusted for matching factors from the parent nested case–control study: study center (grouped by country), age at blood draw, fasting status, date of blood draw, menstrual cycle phase for women premenopausal at blood collection, OC/HT use at blood collection, and length of follow up. We adjusted for oophorectomy, number of ovulatory cycles, and smoking status in sensitivity analyses among premenopausal women, and these factors plus age at menopause, hysterectomy, and type of HT among postmenopausal women. Missing indicators were used to account for missing data for covariates. CA125 and CA15.3 have been reported to vary across the menstrual cycle [20, 21]. Therefore, we evaluated these markers both adjusting for menstrual cycle phase and standardized using phase-specific residuals. Results were similar with both approaches; we present the models adjusted for menstrual cycle phase.

To assess whether the adjustment for correlates of these early detection markers improved discrimination between controls and individuals who subsequently became cases, we evaluated the area under the receiver operating characteristic curve (AUC) and compared AUCs from models including the marker alone to those including the marker standardized for its correlates. These analyses were limited to cases who were postmenopausal at time of blood collection (n = 590; and their matched controls), given significant predictors of the markers were only identified among women postmenopausal at blood collection. AUCs were calculated using conditional logistic regression models to account for the matched study design. We calculated absolute risk estimates for ovarian cancer using a model derived in the EPIC cohort [22] and calibrated the conditional logistic regression model towards the absolute risk estimates as an offset variable. We used regression residuals to standardize the marker concentrations based on significant correlates of the marker. Briefly, we calculated the deviation (residual) from the mean predicted concentration given each study participant’s profile of correlates. Correlates included for each marker were: CA125: parity, hysterectomy, unilateral oophorectomy, age at menopause, estrogen-alone HT use, ovulatory cycles, current smoking; CA15.3: BMI, former smoking; HE4: age at blood draw, OC use, parity, age at menopause, current smoking; CA72.4: no correlates identified.