Diabetes and Overall Survival among Breast Cancer Patients in the U.S. Military Health System

In the United States, breast cancer remains the second most common cause of cancer-specific mortality among women (1). Although the 5-year relative survival rate for women with breast cancer was approximately 91% during 2005 to 2011 (1), some studies suggest that certain comorbid conditions, like diabetes, may negatively impact prognosis (2, 3).

As of 2012, approximately 9.3% of the U.S. population was diagnosed with diabetes (4). There are two main types of diabetes (type I and type II), with type II accounting for 95% of all clinical diagnoses (4). The association between diabetes (primarily type II) and breast cancer survival has been inconsistent in the literature (5), but most studies have found that it is associated with decreased overall survival (6–14) and breast cancer–specific survival (11, 14–17). A prospective cohort of one million U.S. adults reported a 2-fold increased risk of all-cause mortality and a 16% increased risk of breast cancer–specific mortality among persons with diabetes (10). In a large case–control study, diabetes was significantly associated with an elevated risk of all-cause mortality, with greater risk among those who were obese (6). However, some studies have reported no association between diabetes and overall survival (5) or breast cancer–specific survival (8, 12).

Most previous studies examined the association between breast cancer survival and diabetes diagnosed prior to breast cancer (5, 6, 8, 9, 11, 14, 18–20). In a large cohort study of stage I–III breast cancer patients from MD Anderson Cancer Center (Houston, TX), those with preexisting diabetes had a higher risk of all-cause mortality [HR = 1.39; 95% confidence interval (CI), 1.10–1.77; ref. 8]. In another study based on SEER-Medicare data, women diagnosed with diabetes prior to breast cancer diagnosis had a higher all-cause mortality compared with women without diabetes (20). In a meta-analysis of 17 studies, preexisting diabetics were shown to have a 50% higher all-cause mortality than those without diabetes (21). To the best of our knowledge, the association between diabetes diagnosed at or after breast cancer diagnosis and survival has not been studied. Diabetes detected at or following breast cancer diagnosis may differ from diabetes diagnosed prior to breast cancer diagnosis in the severity of the disease, potential effects on breast cancer treatment, and therefore, survival. As shown in a study based on the data from seven state cancer registries in the United States (22), diabetes severity is related to lower guideline-concordant chemotherapy and lower guideline concordance for locoregional treatment, although the association may vary by age. Diabetes can increase the risk of wound infection after surgery (23), which may delay the initiation of adjuvant chemotherapy and thereby potentially result in worse outcomes (24). Such effects may be more likely for more severe diabetes. In addition, antidiabetic treatment of preexisting diabetes itself may affect breast cancer survival (25), and it is possible that diabetes may result from breast cancer treatment (26). Thus, an examination of the relationship between diabetes and survival by timing of diabetes may help better understand the basis of the association. For example, if breast cancer–specific mortality is increased among patients with diabetes reported before breast cancer diagnosis but not at/after, it may suggest that influences of diabetes on breast cancer treatment may be more possible because the former tends to have more severe form of the disease. On the other hand, if all-cause mortality is increased irrespective of timing of the diagnosis of diabetes, it may suggest that factors related to diabetes itself may be more influential.

Previous studies have been conducted in populations in which accessibility to health care may vary. Access to health care has been found to influence the detection and diagnosis of diabetes, its management, treatment, and overall survival (27–30). Access to health care could potentially not only increase the likelihood of diabetes detection and diagnosis (due to better blood glucose monitoring), but also possibly decrease the risk of diabetes (due to the implementation of better preventive measures, such as exercise and improved diet). Thus, when diabetics and nondiabetics are compared in the general population, the identified difference in mortality may partially result from differential detection or diagnosis of diabetes and other socioeconomic factors related to access to care. Research in an equal access system can reduce these potential effects and provide information to assess the relationship between diabetes and cancer. The Military Health System (MHS) of the U.S. Department of Defense (DoD) is an equal access system, providing its beneficiaries with health care at little or no charge. The objective of this study was to examine the relationship between timing of type II diabetes (DM-2) and overall survival among breast cancer patients in the MHS.

This study was based on the linked data from the DoD's Central Cancer Registry (CCR) and the MHS Data Repository (MDR). The linked data include cancer patients diagnosed between 1998 and 2007 who received health care through the DoD health care program, including active duty members, retirees, National Guard and Reserve members, and dependents. The CCR contains demographic, tumor characteristic, cancer treatment, and vital status data that are abstracted from the records of patients diagnosed and/or treated at military treatment facilities (MTF). The data are reviewed and edited according to North America Association of Central Cancer Registries guidelines. The MDR data contain administrative and medical care records for inpatient and outpatient services that are either provided directly at MTFs (direct care) or paid for by the DoD at civilian facilities (indirect care). The MDR data include clinical diagnoses, diagnostic procedures, prescription medications, treatment, and vital status.

The data linkage project was reviewed and approved by the Institutional Review Boards of the Walter Reed National Military Medical Center, Tricare Management Activity, and the National Institutes of Health Office of Human Subjects Research.

Women with histologically confirmed, first primary, malignant breast cancer (ICD-0-3 codes C500-C506 and C508-C509) diagnosed between 1998 and 2007 were eligible for the study (n = 9,944). They were followed up until December 31, 2008. Women with type I diabetes (n = 536), under 20 years of age (n = 1), and/or with missing survival information (n = 9) were excluded.

Medical conditions were identified in the MDR. Women were classified as having type II diabetes (DM-2) if they had ICD-9 diagnoses codes 250.x0, 250.x2, 357.2, 362.0, and 366.41 in at least one inpatient record, or at least three outpatient records (31, 32). Other medical conditions included in the analyses were obesity, heart disease, kidney disease, chronic obstructive pulmonary disease (COPD), stroke, and hypertension. These conditions were considered to be present if codes were identified in one or more inpatient or three or more outpatient records.

Pathologic breast cancer features obtained from CCR included AJCC tumor stage (I through IV, unknown), histologic grade, and hormone receptor status. Histologic grade was classified on the basis of level of differentiation (well differentiated, moderately differentiated, poorly differentiated, and unknown). Estrogen receptor (ER) and progesterone receptor (PR) statuses were categorized into five groups (ER⁺/PR⁺, ER⁻/PR⁻, ER⁺/PR⁻, or ER⁻/PR⁺, unknown).

Medical care and recurrence variables included breast cancer treatment (surgery, chemotherapy, radiotherapy, and hormone therapy), diabetes treatment (insulin, insulin secretagogue, and metformin), surveillance mammography, and recurrence. Receipt of breast cancer surgery, chemotherapy, and radiotherapy was determined by combining data from CCR and MDR and was considered “yes” if therapy occurred within 3 months postdiagnosis. Hormone therapy information was obtained from CCR. Information on diabetes-specific treatment and surveillance mammography was obtained from MDR. Use of diabetes medications, which may be related to cancer survival (33, 34), was considered “yes” if one or more prescriptions were documented and filled. Surveillance mammography was a variable for women with a breast or breasts left after surgery, including those who underwent breast conserving surgery and received a bilateral mammogram, or those who underwent a unilateral mastectomy and received a unilateral mammogram. It was categorized as “yes” if it was identified 6 months or more after a previous mammogram among women. If a mammogram was received 2 months prior to a diagnosis of a breast mass or other breast symptom, it might be diagnostic and thus was not considered as surveillance procedures (35). Recurrence was defined on the basis of the CCR definition and supplemented with information obtained from MDR (36).

Demographic characteristics were obtained from CCR with missing values supplemented from MDR. Demographic variables included age at diagnosis, race/ethnicity, marital status, active duty status, service branch of the active duty member/sponsor, rank of the active duty member/sponsor, and insurance plan (TRICARE Prime, an HMO-like component, TRICARE Standard, and TRICARE Extra).

We first described the distributions of demographics, comorbid conditions, tumor features, surveillance, and treatment by timing of DM-2 diagnosis. We then used Cox proportional hazards models to assess the association between DM-2 (before or at/after breast cancer diagnosis) and overall survival with adjustment for potential confounding factors. Survival time was calculated as the time interval beginning at the date of breast cancer diagnosis through the date of death, date of last contact, or study end (truncated at December 31, 2008). DM-2 status before breast cancer diagnosis was defined as a dichotomized variable (no/yes) based on whether DM-2 was identified prior to breast cancer diagnosis. In regard to DM-2 diagnosed at/after breast cancer, we define it as a time-dependent variable to minimize the potential impact of immortal time bias (37), in which only those who survive to the point of exposure (diabetes at/after breast cancer for this study) could have the chance to be exposed, therefore biasing the survival results in favor of the exposed group. A potential confounder was defined if it was associated with both DM-2 and survival and was not an intervening variable between DM-2 and survival. The identified potential confounding variables included age at diagnosis, race, marital status, rank of active duty member/sponsor, benefit type, tumor grade, obesity, coronary heart disease, chronic obstructive pulmonary disease, stroke, kidney disease, ER/PR status, and chemotherapy. Although tumor stage did not meet the confounder criteria, this variable was adjusted for due to its complex effects and association with prognosis. We repeated Cox model analyses with stratification by race, menopausal status, obesity, ER/PR status, and tumor stage. Information on menopausal status was not available; therefore, age at diagnosis (<50 and ≥50 years) was used as a proxy variable based on the average age of natural menopause (38). The proportional hazards assumption was tested for Cox model analyses, and no violations of model assumptions were found.

All analyses were conducted using SAS version 9.3 (SAS Institute, Cary, NC). All tests of significance were two-tailed and performed at α = 0.05.

Among 9,398 study women, 1,594 had DM-2 (17%). Table 1 shows the distributions of demographic factors and other medical conditions by timing of DM-2 diagnosis. Compared with women without DM-2, those with DM-2 diagnosed both before and at/after breast cancer diagnosis were less likely to be white, active-duty members, and holders of nonprime TRICARE insurance (P < 0.01, all variables). Although both DM-2 groups were older than the nondiabetic group (P < 0.01), women diagnosed before breast cancer tended to be oldest. Women with DM-2 diagnosed before breast cancer were also more likely to be enlisted service members than the other two groups (P < 0.01). Despite timing of diagnosis, women with DM-2 tended to have a higher frequency of obesity, coronary heart disease, kidney disease, COPD, stroke, and hypertension than nondiabetics (P < 0.01, all variables).

Table 2 shows the distributions of treatment, surveillance, and tumor characteristics by timing of DM-2 diagnosis. Women with DM-2 diagnosed before breast cancer diagnosis were more likely to have well-differentiated tumors (P = 0.03) and less likely to have radiation treatment (P = 0.04), chemotherapy (P < 0.01), and recurrence of the disease (P = 0.01) than the other two groups. Although the proportions of women who received metformin, insulin, or insulin secretagogue were negligible in the nondiabetic group, they were higher in women with DM-2 diagnosed before breast cancer than those diagnosed at/after the disease (P < 0.01, all variables). There were no significant differences in tumor stage between the groups (P = 0.59).

The median follow-up time after breast cancer diagnosis was 5.7 years. During follow-up, 16.9% of women with DM-2 and 13.9% of women without DM-2 died. After adjusting for potential confounders, women diagnosed with DM-2 before breast cancer tended to have an increased risk of mortality compared with women without DM-2 (HR = 1.17; 95% CI, 0.95–1.44; Table 3). Women diagnosed with DM-2 at/after breast cancer had higher mortality than women without the disease (HR = 1.39; 95% CI, 1.16–1.66).

Table 4 shows the results of multivariate analyses stratified by demographic, tumor, or obesity variables. Among women diagnosed with DM-2 both before and at/after breast cancer, the tendency of increased mortality was seen across most subgroups stratified by race, menopausal status, or obesity, although not significant for most of them due to smaller numbers of women within those subgroups. Similar findings were observed when data were stratified by tumor hormone receptor status or stage.

This study showed a modest increased risk of death associated with DM-2, regardless of whether it was diagnosed before or at/after breast cancer diagnosis. This association tended to exist in most subgroups when results were stratified by race, age, obesity status, tumor stage, or tumor hormone receptor status, although it was not significant in many strata due to smaller sample sizes.

Most previous studies examined diagnosis of DM-2 prior to breast cancer. Although the results of these studies were not consistent (5), most found an increase in all-cause mortality with preexisting DM-2 (6, 8, 9, 18–20). Our results on DM-2 diagnosed before breast cancer diagnosis, although marginally significant, support these findings. This may be attributable to differences in the receipt of cancer treatment and the effects of diabetes on overall survival.

As a result of diabetes and its complications, breast cancer patients with DM-2 may be less likely to receive breast cancer treatments such as adjuvant chemotherapy or radiotherapy (20, 39, 40). Research has reported that physicians may use chemotherapy less frequently or aggressively to treat breast cancer patients with DM-2 due to an increase in hospitalization for chemotherapy toxicity as well as an increase in breast cancer–specific mortality (20). Less receipt of radiation treatment among diabetics compared with nondiabetics has been reported in the literature (40). These were also demonstrated in our study, in which women with DM-2 diagnosed before breast cancer were less likely to receive chemotherapy and radiotherapy than women without DM-2. However, chemotherapy as a potential confounder was adjusted for in the analysis, and thus, the potential effects that treatment may have on survival were reduced. Nevertheless, we cannot exclude the possibility that treatment regimen (intensity, type, timing, frequency, or duration of chemotherapy as well as other cancer treatments) may have differed between women with and without DM-2. Such differences could be associated with increased mortality risk.

As all-cause mortality was the study outcome, DM-2 and other related medical conditions may have affected the outcome. Hyperglycemia, a serious problem for many individuals with DM-2, may play a role in the decreased survival among women diagnosed with DM-2. Increased blood glucose levels have been found to increase cancer mortality (41–43), with survival decreasing as glycemic control decreases. In addition, patients with DM-2 in the current study were about twice as likely to have obesity, coronary heart disease, kidney disease, COPD, stroke, and hypertension, which can increase the risk of death. These medical conditions were adjusted for in the models. Nonetheless, repeated analysis without these conditions did not change the HR estimates substantially, suggesting the possible effects of factors other than these medical conditions on the identified association between diabetes and overall survival.

We assessed the independent effects of DM-2 diagnosed at or after breast cancer diagnosis with the hypothesis that DM-2 diagnosed at or after breast cancer may be a more indolent or less severe form of the disease than DM-2 diagnosed before breast cancer diagnosis. This hypothesis is based on the fact that DM-2 diagnosed at or after breast cancer diagnosis is detected as a result of medical examination for cancer diagnosis or medical surveillance for cancer care and has shorter duration of disease. Therefore, it may be less likely to affect breast cancer treatment. Lower severity of DM-2 diagnosed after breast cancer diagnosis was demonstrated in our study: Patients with DM-2 diagnosed at or after breast cancer diagnosis were less likely to receive diabetes treatment than those diagnosed before breast cancer. It is possible that DM-2 diagnosed at/after breast cancer diagnosis might not influence breast cancer treatment decisions as heavily. Our data showed that patients with DM-2 diagnosed at/after breast cancer diagnosis were more likely to receive radiation treatment and chemotherapy than those with DM-2 diagnosed before breast cancer. The proportions of breast cancer treatment among the former were close to those among patients without DM-2. Although the hypothesis on DM-2 severity and possible effects on breast cancer treatment is tenable as shown in our data, our analysis did not display significant differences in survival between patients with DM-2 diagnosed at/after and before breast cancer diagnosis. The overall mortality was also higher among DM-2 patients diagnosed at/after breast cancer diagnosis, being similar to those with DM-2 diagnosed before breast cancer. Therefore, factors other than diabetes severity and breast cancer treatment might account for a higher mortality among DM-2 patients despite timing at diagnosis.

Limitations in this study included possible misclassification of DM-2 status because of lack of data on diabetes diagnoses prior to 1998. However, such misclassification was likely limited as individuals with diabetes would continue to receive antidiabetic therapy, which would be captured in their records after 1998. In addition, such misclassification was likely to occur only for DM-2 diagnosed before breast cancer because the initial year for breast cancer diagnosis was 1998. This misclassification might dilute rather than overestimate the association between DM-2 and survival. Second, we cannot exclude residual confounding by the frequency, intensity, timing, and duration of breast cancer treatment, as mentioned above. Third, overadjustment or underadjustment for confounders may be possible. We adjusted for obesity and other comorbidities in our multivariate models. The relationship between diabetes and obesity and other comorbidities is complex. We do not exclude the possibility that a comorbid condition is on the causal pathway between diabetes and mortality, and thus, our multivariate results were overadjusted. However, the HR estimates remained similar after we removed these comorbidities from the multivariate models. On the other hand, there is a possibility that we underadjusted for potential confounders as we did not have data on variables such as physical activity, dietary intake, and smoking, which may be associated with both diabetes and survival (44–46). Finally, there might be incomplete information and inaccurate coding in our data that are typical in medical records data. Nonetheless, the extent of these limitations might not be substantial enough to change the results.

In summary, our findings indicate that breast cancer patients with DM-2 are at a higher risk of mortality despite timing at diagnosis, compared with those without DM-2. Further research into these findings is warranted.

No potential conflicts of interest were disclosed.

The views expressed in this article are those of the authors and do not necessarily reflect the official policy or position of the Department of the Navy, Army, Department of Defense, National Cancer Institute, nor the U.S. government. Nothing in the presentation implies any federal/Department of Defense endorsement.

Conception and design: A.A. Gill, C.D. Shriver, K. Zhu

Development of methodology: A.A. Gill, K. Zhu

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): K. Zhu

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S. Shao, S.H. Zahm, K.A. McGlynn, K. Zhu

Writing, review, and/or revision of the manuscript: S. Shao, A.A. Gill, S.H. Zahm, I. Jatoi, K.A. McGlynn, K. Zhu

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A.A. Gill, C.D. Shriver

Study supervision: C.D. Shriver, K. Zhu

The authors thank the following individuals and institutes for their contributions to or support for the original data linkage project: Guy J. Garnett, David E. Radune, and Dr. Aliza Fink of ICF Macro; Wendy Funk, Julie Anne Mutersbaugh, Linda Cottrell, and Laura Hopkins of Kennel and Associates, Inc.; Kim Frazier, Dr. Elder Granger, and Dr. Thomas V. Williams of TMA; Annette Anderson, Dr. Patrice Robinson, and Dr. Chris Owner of the Armed Forces Institute of Pathology; Dr. Joseph F. Fraumeni Jr, Dr. Robert N. Hoover, Dr. Susan S. Devesa, and Gloria Gridley of the NCI; Dr. Lindsey Enewold, Dr. John Potter, Raul Parra, Anna Smith, Fiona Renalds, William Mahr, Hongyu Wu, Dr. Larry Maxwell, Miguel Buddle, and Virginia Van Horn of the United States Military Cancer Institute. The authors also thank Dr. Jie Lin for her help in conducting supplementary statistical analysis. This project was supported by the John P. Murtha Cancer Center, Walter Reed National Military Medical Center via the Uniformed Services University of the Health Sciences under the auspices of the Henry M. Jackson Foundation for the Advancement of Military Medicine and by the intramural research program of the NCI. The original data linkage was supported by the United States Military Cancer Institute and Division of Cancer Epidemiology and Genetics, NCI.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.