Circulating gamma-glutamyl transferase and development of specific breast cancer subtypes: findings from the Apolipoprotein Mortality Risk (AMORIS) cohort

The AMORIS study has been described in detail elsewhere [5, 7‐9]. This cohort includes 812,073 individuals who underwent laboratory examination at the Central Automation Laboratory in Stockholm between 1985 and 1996 [9]. The study complied with the declaration of Helsinki and was approved by the Ethics Review Board of the Karolinska institute.

From the AMORIS cohort we identified 231,283 cancer-free women aged 20 years or older with baseline measurements of serum GGT. These women were followed until they developed breast cancer, died, emigrated, or until the end of the study (31 December 2011), whichever came first. A total of 10,861 breast cancers (4.7%) were diagnosed during follow-up. Among them, 6934 (63.8%) had available information on oestrogen receptor (ER) status, 7145 (65.8%) had information on progesterone receptor (PR) status, and 2197 (20.2%) had additional information on HER2 status. A nested case–control study was performed where for each case with information on receptor status, we used incidence density sampling to select ten controls among all women in the cohort who were alive and did not have breast cancer at the time of diagnosis of the case. Cases and controls were matched for age group (less or more than 50 years old) as an indicator for menopausal status [10] because menopausal status was only available for cases. The same sets of cases were included in the case–case analysis.

We classified breast cancer subtype based on ER and PR and their combinations. In the subgroup with information on HER2, we defined four tumour subtypes (ER+/HER2− or PR+/HER2−, ER+/HER2+ or PR+/HER2+, ER−/PR−/HER2+, and ER−/PR−/HER2− (triple negative)) as previously described (Additional file 1: Figure S1) [11]. These subtypes share similar profiles with molecular phenotypes luminal A, luminal B, HER2 type and triple negative [12, 13].

All laboratory analyses were performed by automated techniques at the CALAB laboratory, Stockholm, Sweden. GGT (U/L) was determined using the reference method recommended by the International Federation of Clinical Chemistry and Laboratory Medicine (IFCC) [5, 14]. The coefficient of variation was ≤6.0%. Samples were prospectively measured prior to assignment to cases or controls. Levels of GGT were skewed and logarithmically transformed. We additionally categorised GGT into quartiles.

From the registry linkage in AMORIS [5, 9], we collected information on socioeconomic status, education level, parity, menopausal status at diagnosis, and comorbidities using Charlson co-morbidity index (CCI) [15, 16]. Serum triglycerides and glucose were measured enzymatically [17].

In the nested case–control analysis, we used conditional logistic regression models to assess any association between log-transformed and quartiles of GGT and overall and specific breast cancer subtypes. A test for trend was performed by using GGT quartiles as an ordinal scale. We estimated odds ratios (ORs) of ER and PR status individually compared to matched controls based on these measures of serum GGT. Subsequently we compared GGT levels of cases with controls based on combined ER and PR subtypes.

We further conducted a case–case analysis to compare different breast cancer subtypes [18]. Binary and multinomial logistic regression models were used to assess log-transformed levels and quartiles of GGT in relation to breast cancer subtype, both by individual ER or PR status, combined ER/PR status and ER/PR/HER2 status. Since ER and PR status was available since follow-up started and the information of HER2 status was only available after 2006, we performed a sensitivity analysis only including cases with complete information on the three receptors.

All models were adjusted for age, socioeconomic status, education and parity, and time interval between GGT measurement and diagnosis. We additionally controlled for menopausal status in the case–case analysis. Adjustment for CCI was performed to take into account existing co-morbidities [1, 5, 19, 20]. We further adjusted for serum glucose and triglycerides to reduce potential confounding from metabolic disorders [11, 21‐23]. All analyses were conducted with Statistical Analysis Systems (SAS) release 9.4 (SAS Institute, Cary, NC, USA).