Single-nucleotide polymorphisms and the effectiveness of taxane-based chemotherapy in premenopausal breast cancer: a population-based cohort study in Denmark

We conducted this nationwide population-based cohort study in Denmark. Denmark has a free tax-supported health care system [17]. The civil personal registration number, assigned to all residents upon birth or immigration, allows individual-level data linkage across Danish administrative and health registries [17]. In Denmark, all diagnostic surgical and biopsy tissue specimens are stored permanently as primary formalin-fixed paraffin-embedded (FFPE) tissue blocks at local pathology departments and registered in the Danish National Pathology Registry [18]. The Danish Breast Cancer Group (DBCG) is responsible for clinical guidelines on breast cancer diagnosis and treatment in Denmark. The DBCG clinical database records clinical and follow-up data for all Danish patients with invasive breast tumors [19], including data for up to 10 years of active follow-up for recurrence [20]. The DBCG registers breast cancer patients through an electronic reporting system accessible to all Danish pathology departments. This database is supplemented with data from other medical registries (e.g., the Danish National Pathology Registry). Subsequent clinical data from follow-up examinations also are added to the database. Until 2016, women diagnosed with breast cancer in Denmark were followed up first with semi-annual clinical exams for five years and then with annual exams during the following five years. Since 2016, patients have been able to choose among patient-led, nurse-led, or fixed annual follow-up exams. All women are offered mammography, ultrasound screening, and open access to a breast cancer unit for 10 years after a breast cancer diagnosis [21].

We nested our study in the Predictors of Breast Cancer Recurrence (ProBe CaRe) cohort, which is described in greater detail elsewhere [22]. Briefly, the ProBe CaRe cohort includes premenopausal Danish women diagnosed with incident non-distant metastatic breast cancer during 2002–2011. We restricted the cohort to women diagnosed with breast cancer after implementation of docetaxel in combination with cyclophosphamide and sometimes epirubicin as guideline chemotherapy in Denmark (January 01, 2007). The cohort was further restricted to patients who were 18–55 years at diagnosis, who received chemotherapy, and who had FFPE blocks available in the pathology archives. Finally, we excluded women who experienced a recurrence or were lost to follow-up during the first six months after diagnosis. The study flowchart is provided in the Supplementary material (Fig. S1).

Procedures for collection of FFPE tumor blocks, preparation of the tumor tissue, and DNA extraction are described in detail elsewhere [23]. Based on a comprehensive review of the taxane pharmacogenetic literature and underlying biology, we identified 26 candidate SNPs. SNPs were considered candidates if (1) located in genes encoding proteins involved in taxane metabolism or transport, (2) previous studies suggested their biologically plausible role in taxane pharmacokinetics or effectiveness, or (3) they were associated with taxane toxicities (primarily neuropathy). A list of genotyped variants is presented in Table 2. Seven of these SNPs had been genotyped previously, as described elsewhere [23]. The remaining 19 SNPs were genotyped using commercially available TaqMan assays on a StepOne Plus real-time instrument (Applied Biosystems, Thermo Fisher Scientific, Foster City, California, USA) in accordance with the manufacturer’s protocol. For TaqMan assays, 20 ng of purified DNA extracted from FFPE tissues were amplified in 10µL PCR. The PCR were incubated at 60 °C for 30 s and 95 °C for 10 min and then cycled 50 times between 15-s incubations at 95 °C and 60-s incubations at 60 °C. Genotypes were classified based on TaqMan VIC/FAM intensity values using the auto-call feature of QuantStudio Software V1.3. Genotype results were manually inspected, and acceptance was overridden manually if irregular amplification curves were observed. We calculated expected genotype frequencies under Hardy–Weinberg Equilibrium (HWE) and compared them with observed frequencies both visually and by applying a traditional test for HWE. In addition, we compared allele frequencies between the study cohort and benchmarks reported for female European non-Finnish cohorts in the Genome Aggregation Database (gnomAD) [24]. For each SNP, we primarily classified each woman as (1) wildtype when she carried two normal alleles, (2) variant carriers (including both hetero- and homozygotes), or (3) as heterozygote or homozygote variant carriers.

From the DBCG clinical database, we collected information on patient age at diagnosis, tumor characteristics (hormone receptor status, pathological grade, number of positive lymph nodes, and tumor size), cancer treatment (surgery type, intention-to-treat radiotherapy, adjuvant chemotherapy, and endocrine therapy), and dates of recurrences or second primary malignancies. From the Cause of Death Registry, we collected dates and causes of death, and from the Danish National Patient Registry we collected data on comorbid conditions diagnosed up to 10 years before the breast cancer diagnosis [25, 26].

We used breast cancer recurrence and all-cause mortality as outcomes of taxane effectiveness. We adopted the DBCG definition of recurrence, which encompasses locoregional recurrence (tumor growth in the surgical scar, the ipsilateral breast, or regional lymph nodes), distant recurrence, or contralateral breast cancers diagnosed up to 10 years after initial breast cancer treatment [20]. Diagnosis of recurrence is either based on clinical assessment, pathological assessment, imaging, or a combination of these. We examined all-cause mortality, assuming breast cancer to be the most likely underlying or contributing cause of death in our young study cohort. We also examined breast cancer-specific mortality (BCSM) in sensitivity analyses, defined by deaths with breast cancer (ICD-10: C50) as the underlying or contributory cause of death.

Comorbidities were summarized using the Charlson Comorbidity Index [27] and categorized as none, 1–2, or  ≥ 3 comorbidities (Supplemental Table S1). We derived cancer stage (categorized as stage I–III according to the TNM staging system [28]). Estrogen receptor (ER) status was a composite variable incorporating both negative and positive ER statuses and receipt of endocrine therapy. Grades 1–3 were assigned to lobular and ductal tumors; other tumors were not graded. Data on treatments included surgical procedure and intention-to-treat radiotherapy. Breast cancers were considered triple negative if tumors were ER–, human epidermal growth factor receptor 2 was negative, and information on progesterone receptor was either negative or missing.

We started follow-up six months after the date of breast cancer surgery (date of diagnosis), to approximate the end of chemotherapy treatment. In analyses of recurrence, we censored follow-up upon death, emigration, diagnosis with new primary malignancy, last visit if lost to follow-up exams, end of the DBCG follow-up protocol (maximum 10 years), or end of available data (September 25, 2017). When examining mortality, longer available follow-up time in the Cause of Death registry allowed us to extend the follow-up period. Thus, mortality follow-up continued until the date of death, emigration, or end of data availability on December 31, 2019. We computed cumulative incidences of recurrence and death, considering death as a competing risk when examining recurrence. We used cause-specific Cox regression models to compute unadjusted hazard ratios (HRs) of recurrence and mortality for each SNP.

We also stratified all models by ER status and stage to evaluate effect measure modification by these factors. To account for potential underreporting of recurrence [29], we performed a sensitivity analysis considering recurrences to include BCSM in women not registered with a recurrence. To test the robustness of our mortality model, we restricted an analysis to BCSM. We used SAS 9.4 for all analyses (Cary, NC).