Metformin and Pathologic Complete Responses to Neoadjuvant Chemotherapy in Diabetic Patients With Breast Cancer

The Breast Cancer Management System Database at The University of Texas M. D. Anderson Cancer Center was searched, and 3,412 patients with invasive breast cancer who were treated with neoadjuvant chemotherapy from January 1984 to May 2007 were identified. Of these patients, 291 were diabetic. The following exclusion criteria were applied: diabetes diagnosed after the period of neoadjuvant chemotherapy, resolved gestational diabetes, male sex, unknown estrogen receptor (ER) or progesterone receptor (PR) status, unknown human epidermal growth factor receptor 2 (HER-2) status, incomplete records (including medication records), more than 9 months between neoadjuvant chemotherapy initiation and surgery, and second concurrent primary cancer. Medication records from patient record review and pharmacy records were analyzed to divide the diabetic patients into those taking metformin and those not taking metformin during neoadjuvant chemotherapy. The use of other diabetic medications (insulin, thiazolidinediones, and so on) was also tabulated. The final study population consisted of 68 diabetic patients taking metformin during neoadjuvant chemotherapy (metformin group), 87 diabetic patients not taking metformin during neoadjuvant chemotherapy (nonmetformin group), and 2,374 nondiabetic patients (nondiabetic group). The institutional review board approved the retrospective review of the medical records for the purposes of this study.

Dedicated breast pathologists at the M. D. Anderson Cancer Center reviewed all pathologic specimens. The histologic type, grade, and immunohistochemical analysis of ER, PR, and HER-2 status were determined as previously described.²⁵ Briefly, diagnosis of invasive breast cancer was made by core-needle biopsy of the breast tumor. Clinical stage was defined by the sixth edition of the Cancer Staging Manual of the American Joint Committee on Cancer. The histologic type of each tumor was defined according to the classification system of the WHO. Tumor grade was defined according to the modified Black's nuclear grading system. Immunohistochemical analysis to determine ER and PR status was performed using standard immunohistochemistry (IHC) procedures with monoclonal antibodies. Nuclear staining ≥ 10% was considered a positive result. Before 1993, the dextran-coated charcoal ligand-binding method was used to determine ER and/or PR status. HER-2 status was evaluated by IHC or by fluorescence in situ hybridization. HER-2–positive tumors were defined as 3+ receptor overexpression on IHC staining and/or gene amplification found on fluorescent in situ hybridization testing. pCR was defined as no evidence of invasive carcinoma in the breast and axillary lymph nodes at time of surgery.

In general, all patients received three to six courses of one of the following anthracycline-based chemotherapy regimens: fluorouracil, doxorubicin, and cyclophosphamide; doxorubicin and cyclophosphamide; or fluorouracil, epirubicin, and cyclophosphamide. Additional taxane chemotherapy (paclitaxel or docetaxel) was administered to 1,909 patients (75.5%) for 3 months. At the completion of the neoadjuvant chemotherapy, patients underwent definitive surgery. Eligibility for breast conservation was determined by multidisciplinary evaluation. All patients had axillary staging with axillary lymph node dissection or sentinel node biopsy. Radiation therapy was delivered in the event of breast conservation surgery, locally advanced disease, primary tumor measurement before chemotherapy of > 5 cm, and four or more involved axillary nodes. Adjuvant hormonal therapy was administered according to standard practice.

Baseline patient characteristics were tabulated, and χ² tests of independence were used to examine differences in the baseline characteristics between groups (metformin, nonmetformin, and nondiabetic groups). A multivariate logistic regression model was used to examine the relationship between metformin and pCR after adjusting for factors known to affect pCR and for possible confounders related to diabetes and obesity (diabetes status [yes v no], age, clinical stage, tumor grade, receptor status, taxane use, and obesity status). To identify where the differences lay, the proportions of patients who experienced pCR for the three groups were compared pairwise using a binomial test of proportions. In an exploratory analysis to examine relapse-free survival (RFS) and overall survival (OS), the time to event or censoring was computed in months since diagnosis for each patient. Survival time was censored at the date of last follow-up during the monitoring period if events were not observed. Three-year survival probabilities and associated CIs were estimated nonparametrically using the Kaplan-Meier product limit method. The survival curves of each group were compared using log-rank tests. Multivariate Cox proportional hazards regression models were used to model survival as a function of metformin use after adjusting for the various confounders. All statistical analyses were performed using Stata 10 software (StataCorp, College Station, TX).

Patients were diagnosed with breast cancer between September 1990 and May 2007. Patient demographics and clinical characteristics are summarized in Table 1. The median age of patients was 49 years (range, 21 to 87 years). Patients in the diabetic groups were older than were those in the nondiabetic group, and consistent with this, these groups had a higher proportion of postmenopausal patients than did the nondiabetic group. Patients in the diabetic groups were also more overweight and obese according to body mass index (BMI) comparisons. The other standard prognostic factors were not significantly different between the three groups (Table 1). Severity of diabetes as judged by hemoglobin A1c levels was not different between the two diabetic groups. We also evaluated the use of other diabetic medications in the diabetic groups, because there is evidence that these—including insulin, as discussed, and thiazolidinediones²⁶—may affect tumorigenesis. Insulin use was lower in the metformin group than it was in the nonmetformin group (16% v 33%; P = .02), whereas thiazolidinedione use was not significantly different between groups (P = .36). Finally, neoadjuvant taxane use was higher in the diabetic groups (87% in the metformin group and 84% in the nonmetformin group) compared with the nondiabetic group (75%; P = .01).

The proportion of pCR was significantly higher in the metformin group (24%; 95% CI, 13% to 34%) than it was in the nonmetformin group (8.0%; 95% CI, 2.3% to 14%) and the nondiabetic group (16%; 95% CI, 15% to 18%; P = .02; Fig 1). Pairwise comparisons revealed a significant difference between the metformin and nonmetformin groups (P = .007) and the nonmetformin and nondiabetic groups (P = .04). The comparison between the metformin and nondiabetic groups trended toward but did not meet statistical significance (P = .10). To evaluate whether differences in amount of chemotherapy delivered caused the difference in pCR rates, we calculated the percentage of planned chemotherapy cycles delivered in the metformin and nonmetformin groups. There was no significant difference; both groups received approximately 90% of the planned cycles of neoadjuvant chemotherapy (P = .72). In addition, the average number of cycles delivered per patient was approximately seven in both groups.

Multivariate logistic regression was used to model factors predictive of pCR (Table 2). After adjustment for diabetes status (yes v no), BMI, age, stage, grade, ER/PR and HER-2 status, and neoadjuvant taxane use, metformin use during neoadjuvant chemotherapy was found to be an independent predictor of pCR (odds ratio, 2.95; 95% CI, 1.07 to 8.17; P = .04). Factors known to be predictive of pCR, such as earlier stage, higher tumor grade, ER-negative and HER-2–positive status, and neoadjuvant taxane use,²⁵ were confirmed to be so in this analysis. The results for obesity as a predictor of pCR were statistically significant for the comparison of overweight versus nonoverweight status, but not for obese versus nonoverweight status (Table 2).

Because the use of insulin was different in the metformin versus the nonmetformin diabetic groups (16% v 33%), we conducted an exploratory analysis of the effect of insulin use on pCR in these groups. In the metformin group, the rate of pCR was not different for insulin use versus no insulin use (27% v 23%; P = .75). However, in the nonmetformin group, the rate of pCR was significantly different for insulin use versus no insulin use (0% v 12%; P = .05).

At a median follow-up of 37 months (range, 0.6 to 167.3 months), an exploratory analysis of RFS and OS estimates via the Kaplan-Meier method was performed. There were 208 recurrences and 500 deaths. The estimated 3-year RFS rates were 76% (95% CI, 70% to 86%), 66% (95% CI, 52% to 76%), and 73% (95% CI, 71% to 75%), for the metformin, nonmetformin, and nondiabetic groups, respectively, and were not significantly different (P = .66). The estimated 3-year OS rates were 81% (95% CI, 65% to 90%), 78% (95% CI, 65% to 86%), and 86% (95% CI, 84% to 87%), for the metformin, nonmetformin, and nondiabetic groups, respectively, and were significantly different (P = .02). Cox proportional hazards models were used to analyze factors predictive of RFS and OS. Metformin was not an independent predictor of either RFS or OS after adjusting for diabetes status, BMI, age, stage, grade, ER/PR status, and neoadjuvant taxane use (data not shown).