Diabetes and differences in detection of incident invasive breast cancer

Abstract

Many women diagnosed with breast cancer have chronic conditions such as diabetes that may impact other health behaviors. Our purpose was to determine if breast cancer screening and detection differs among women with and without diabetes. We conducted a cross-sectional analysis of a retrospective cohort of women ages 52–74 years diagnosed with incident stages IIII breast cancer enrolled in an integrated health plan between 1999–2014 with linkage to the Surveillance, Epidemiology and End Results registry(n=2040). Screening data were taken from electronic health records. We used multivariable modified Poisson regression models with robust standard errors to estimate relative risks (RR) and 95% confidence intervals (CI) for outcomes of (i)receipt of screening in the two years prior to diagnosis; (ii)symptom-detected breast cancer; and (iii)diagnosis of locally-advanced stage III breast cancer. Compared to women without diabetes, women with diabetes were similar with respect to receipt of screening mammography (78% and 77%), symptom-detected breast cancer (46% and 49%), and stage III diagnosis (7% and 7%). In multivariable models adjusting for age and year of diagnosis, race, BMI, Charlson comorbidity score and depression diagnosis no differences were observed in the outcomes by presence of diabetes. Further investigation is warranted to determine how diabetes acts as a mediating factor in adverse breast cancer outcomes.

INTRODUCTION

Incidences of both breast cancer [1] and comorbid conditions such as diabetes [2] increase with age, leading to a growing population of women with competing needs for preventive services such as cancer screening and chronic condition care. It is estimated that more than half of older patients diagnosed with cancer will have one or more meaningful chronic conditions such as diabetes [3] that will impact their treatment and survival [4,5].

The 2013 Institute of Medicine report, Delivering High Quality Cancer Care [6], calls for greater understanding of evidence-based cancer management in older adults with comorbid conditions The relationship between diabetes and breast cancer is complex, with each affecting the other, and it remains unclear how and why comorbid diabetes results in less favorable outcomes in women with breast cancer. Reports from observational studies suggest that fewer women with diabetes receive screening mammography [7,8] and this may be associated with the quality of diabetes care [9]. In the Women’s Health Initiative clinical trials [10], while rates of mammography were similar in women with and without diabetes, both adherence to breast cancer screening guidelines and risk of invasive breast cancer differed by type of diabetes treatment, also indicating possible differences in breast cancer risk by diabetes severity or by specific diabetes medications hypothesized to influence breast cancer risk [11–13].

The purpose of this study was to determine if diabetes is associated with rates of breast cancer screening and whether type of cancer differed among women diagnosed with invasive breast cancer in a population-based cohort enrolled in a large integrated healthcare delivery system. We hypothesized that women with diabetes, having competing healthcare needs, had lower adherence to breast cancer screening guidelines, thereby leading to greater occurrence of symptom-detected breast cancer (rather than screen-detected) and diagnosis of later stage invasive cancers.

METHODS

We utilized data from an existing cohort of women diagnosed with an incident, stage I or II breast cancer between 1990 and 2008 in the COmmonly Used Medications and Breast Cancer Outcomes (COMBO) study at Kaiser Permanente Washington (KPWA)., [14–16]. We expanded the cohort through 2014 and added stage III breast cancer cases through all years. We included women who had no bilateral disease, had a definitive surgery for their incident breast cancer, and were enrolled at KPWA for at least two year prior to breast cancer diagnosis. Women were enrollees at KPWA, a non-profit integrated delivery system that provides comprehensive healthcare on a pre-paid basis to approximately 650,000 individuals throughout Washington State and parts of Idaho.

We conducted a cross-sectional analysis within this retrospective cohort of women ages 52–74 years at the time of their incident, invasive (stages I-III) breast cancer diagnosis between 1999 and 2014. We restricted to women ages 52–74 years since KPWA recommendations for breast cancer screening are every 2 years starting at age 50 in the general population and every year starting at age 40 among women at high risk. There is no definitive recommendation for women ages 75 years and older. Data on screening was not available at KPWA prior to 1999 so the study was limited to 1999–2014 cases.

A total of 2040 women were eligible for the cross-sectional study, with 1162 from the original COMBO study, and 878 newly identified for the expanded cohort (. This study was approved by the institutional review board of the Kaiser Permanente Washington Health Research Institute.

Incident breast cancers and tumor characteristics were identified through linkage to the Cancer Surveillance System [17], a population-based cancer registry and member of the Surveillance, Epidemiology and End-Results (SEER) Program of the National Cancer Institute [18]. Patient and clinical characteristics were obtained through KPWA automated electronic health records and claims data [19], which included laboratory results, inpatient and outpatient diagnoses, procedures, health plan enrollment, pharmacy dispensings, and death (internal records, SEER, and Washington State death tapes) [20]. Additionally, data from KPWA included linkage to the Breast Screening Recruitment and Reminder (BSRR) survey data, where approximately 98% of women aged ≥40 years completed a survey on factors associated with breast cancer at the time of each mammogram [21]. Data were captured from 2 years prior to breast cancer diagnosis through end of follow-up (earliest of death, disenrollment, or study end date).

The primary covariate of interest was presence of diabetes mellitus at breast cancer diagnosis. Diabetes was defined by having 1+ inpatient or 2+ outpatient International Classification of Diseases, 9^th revision, Clinical Modification (ICD-9) diagnosis codes for diabetes (250.1, 250.2, 250.3, 250.8, 250.9, 250.0), any diabetes medication use (insulin and oral medications), and/or laboratory tests indicative of diabetes (HbA1c ≥7%, 1 fasting plasma glucose ≥200mg/mL, or 2 fasting plasma glucose ≥126mg/mL) in the 2 years prior to breast cancer diagnosis. Diabetes was further classified as uncontrolled during the 2 years prior by having HbA1c ≥8%, any fasting plasma glucose ≥200mg/dL or any 2 fasting plasma glucose ≥126mg/dL in 6 months. Diabetes was considered to be controlled if HbA1c and/or fasting plasma glucose were measured but did not meet any of the criteria for uncontrolled diabetes. Additional variables included age (52–59, 60–64, 65–69, 70–74 years), year of diagnosis (1999–2003, 2004–2008, 2009–2014), race (White, Black, Asian, other), body mass index (BMI) at diagnosis (<25, 25–29, ≥30 kg/m²), Charlson comorbidity score [22] at diagnosis (0, 1, ≥2), and depression during the 2 years prior to breast cancer diagnosis as indicated by ICD-9 diagnosis codes (yes/no).

Outcomes

Our primary outcomes of interest were: (i) adherence to breast cancer screening guidelines; (ii) occurrence of symptom-detected versus screen-detected breast cancer; and (iii) later-stage breast cancer at diagnosis. Adherence to breast cancer screening recommendations was defined as evidence of 1+ asymptomatic screening examination (mammography or breast magnetic resonance imaging [MRI]) in the 2-year period preceding breast cancer diagnosis per the recommendations provided within the integrated KPWA health system [23]. Symptom-detected breast cancer was defined as having no screening examinations in the three months prior to breast cancer diagnosis. Otherwise, the breast cancer was screen-detected. Later-stage breast cancer was defined as being diagnosed with American Joint Committee on Cancer (AJCC) [24] stage III versus stages I or II disease.

Statistical Analyses

To evaluate the associations between diabetes and risks of each of the three outcomes, we estimated relative risks (RR) and 95% confidence intervals (CI) using a modified Poisson regression model with robust standard errors for a common binary outcome [25]. Two models were fit separately for each outcome. The first model included an adjustment for age at diagnosis only (categorical). The second model was a multivariable model including age and year of diagnosis, race, BMI at diagnosis, Charlson comorbidity score [22] at diagnosis, and depression at time of diagnosis.

RESULTS

In the cohort of 2040 women, 263 (13%) had diabetes at breast cancer diagnosis (Table 1). Characteristics defining diabetes in the cohort are presented in Supplemental Table 1, including the proportion of patients with any period of uncontrolled diabetes during the 2 years prior to breast cancer diagnosis (53%). Diabetes prevalence differed by race (P<0.01), with lower prevalence among White women (205/1768, 12%) and higher among Black (16/69, 23%) and Asian (22/120, 18%) women. Women with diabetes were older (median age: 65 vs. 62 years, P<0.01), had higher BMI (median: 33 vs. 28 kg/m², P<0.01), greater comorbidity (Charlson score of ≥2: 15% vs. 4%, P<0.01), higher prevalence of depression (23% vs. 17%, P=0.01), and more primary care visits in the two years prior to breast cancer diagnosis (median: 8 vs. 5 visits, P<0.01), but were similar with respect to 1^st degree family history of breast cancer (27% vs. 27%, P=0.93) and smoking status (current smoker: 7% vs. 5%, P=0.22).

Table 1.

Descriptive characteristics at breast cancer diagnosis by diabetes status

	Diabetes (N=263) n %	No Diabetes (N=1777) n %	P
Characteristics at breast cancer diagnosis
Diagnosis year
1999–2003	72 (27.4)	531 (29.9)	0.38
2004–2008	74 (44.5)	535 (30.1)
2009–2014	117 (44.5)	711 (40.0)
Age, years
Median	65	62	<0.01
52–54	17 (6.5)	229 (12.9)	<0.01
55–59	46 (17.5)	414 (23.3)
60–64	55 (20.9)	442 (24.9)
65–69	78 (29.7)	361 (20.3)
70–74	67 (25.5)	331 (18.6)
Menopausal status
Pre-/Peri-menopause	2 (0.8)	56 (3.3)	0.03
Post-menopause (or age ≥55 years)	251 (99.2)	1639 (96.7)
Unknown and age <55 years	10	82
Race
White	205 (78.5)	1563 (88.1)	<0.01
Black	16 (6.1)	53 (3.0)
Asian	22 (8.4)	98 (5.5)
Other	18 (6.9)	60 (3.4)
Unknown	2	3
Hispanic ethnicity
No	239 (90.9)	1658 (93.4)	0.14
Yes	24 (9.1)	118 (6.6)
Unknown		1
Body mass index (kg/m2)
Median	33.3	27.6	<0.01
<25.0	33 (12.7)	559 (31.6)	<0.01
25.0–29.9	53 (20.4)	587 (33.2)
30.0–34.9	66 (25.4)	369 (20.9)
35.0+	108 (41.5)	252 (14.3)
Unknown	3	10
Smoking status
Current	17 (6.5)	96 (5.4)	0.22
Past	114 (43.3)	689 (38.8)
Never/Unknown	132 (50.2)	992 (55.8)
Charlson comorbidity score
0	175 (66.5)	1498 (84.3)	<0.01
1	49 (18.6)	215 (12.1)
≥2	39 (14.8)	64 (3.6)
1st degree family history of breast cancer
No	174 (73.1)	1189 (73.4)	0.93
Yes	64 (26.9)	431 (26.6)
Unknown	25	157
Depression in 2 years prior
No	202 (76.8)	1482 (83.4)	0.01
Yes	61 (23.2)	295 (16.6)
Outcomes
Receipt of asymptomatic screening mammography or breast magnetic resonance imaging
1 year prior to breast cancer diagnosis	177 (67.3)	1063 (59.8)	0.020
2 years prior to breast cancer diagnosis	206 (78.3)	1366 (76.9)	0.600
Detection of breast cancer
Screened-detected	143 (54.4)	898 (50.5)	0.245
Symptom-detected	120 (45.6)	879 (49.5)
Breast cancer stage
I-II	245 (93.2)	1658 (93.3)	0.93
III	18 (6.8)	119 (6.7)

^*To test for differences of the characteristics between the 2 groups, we used χ² test for categorical variables and Wilcoxon rank-sum test for medians.

Compared with women without diabetes at breast cancer diagnosis, women with diabetes had similar rates of breast cancer screening in the two years prior to breast cancer diagnosis (78% vs. 77%), symptom-detected breast cancer (46% vs. 49%), and stage III breast cancer (7% vs. 7%) (Table 1). In multivariable analyses (Table 2), we found no significant differences in the relative risk of adherence to breast cancer screening among women with diabetes compared with women without diabetes (RR 1.05, 95% CI 0.98, 1.12; P=0.19). Risks of symptom-detected (vs. screen-detected) breast cancer (RR 0.94, 95% CI 0.81–1.08; P=0.37) or stage III (vs. stage I or II) breast cancer (RR 1.00, 95% CI 0.61–1.63; P=0.99) also were similar between women with and without diabetes. Risk of these outcomes by controlled and uncontrolled diabetes was similar, though a modest association was suggested for a higher rate of screening in women with controlled diabetes (RR 1.08, 95% CI 0.99–1.18; P=0.07) but not uncontrolled diabetes (RR 1.01, 95% CI 0.91–1.12; P=0.85).

Table 2.

Results from multivariable modified Poisson regression models with robust standard errors comparing women with diabetes to women without diabetes for receiving asymptomatic breast screening, symptom-detected breast cancer and locally advanced stage III breast cancer

	Age-adjusted modela			Multivariable-adjusted modelb
Outcomes	RR	95% Cl	P	RR	95% Cl	P
Receipt of asymptomatic screening mammography or breast magnetic resonance imaging (yes versus no)
No diabetes	1.00	Reference		1.00	Reference
Diabetes	1.01	(0.94, 1.08)	0.82	1.05	(0.98, 1.12)	0.12
No diabetes	1.00	Reference		1.00	Reference
Controlled diabetes	1.06	(0.98, 1.16)	0.16	1.08	(0.99, 1.18)	0.07
Any uncontrolled diabetes	0.97	(0.88, 1.07)	0.56	1.01	(0.91, 1.12)	0.85
Detection of breast cancer by symptoms (versus screening)
No diabetes	1.00	Reference		1.00	Reference
Diabetes	0.93	(0.81, 1.08)	0.35	0.94	(0.81, 1.08)	0.37
No diabetes	1.00	Reference		1.00	Reference
Controlled diabetes	0.83	(0.66, 1.04)	0.11	0.88	(0.70, 1.11)	0.28
Any uncontrolled diabetes	0.99	(0.83, 1.18)	0.93	0.97	(0.81, 1.16)	0.74
Stage III (versus stages I or II)
No diabetes	1.00	Reference		1.00	Reference
Diabetes	1.07	(0.66, 1.72)	0.80	1.00	(0.61, 1.63)	0.99
No diabetes	1.00	Reference		1.00	Reference
Controlled diabetes	1.22	(0.64, 2.35)	0.54	1.06	(0.55, 2.03)	0.86
Any uncontrolled diabetes	0.86	(0.64, 1.74)	0.68	0.97	(0.49, 1.94)	0.94

^a.Model adjusted for age at diagnosis (categorical; 52–59, 60–64, 65–69, 70–74 years)

^b.Model adjusted for age at diagnosis (categorical; 52–59, 60–64, 65–69, 70–74 years), year of diagnosis (categorical; 1999–2003, 2004–2008, 2009–2014), race (white, non-white), body mass index at diagnosis (categorical; <25, 25–29, ≥30 kg/m²), Charlson comorbidity score at diagnosis (categorical; 0, 1, ≥2), and depression at diagnosis (yes, no)

DISCUSSION

In this large, population-based cohort of women enrolled in an integrated health care system in Washington State, we found similar rates of adherence to breast cancer screening, symptom-detected breast cancer, and risk of locally advanced (stage III) breast cancer among women with diabetes compared to women without diabetes. Women with controlled diabetes appeared to have slightly higher rates of screening, although confidence intervals included 1.0.

Diabetes is associated with increased cancer-specific and all-cause mortality [5,4] in breast cancer patients. Observed associations between diabetes and adverse breast cancer outcomes are hypothesized to result from less aggressive breast cancer treatment or diabetes care, both of which could account for the increased mortality observed in women with diabetes and breast cancer. Another explanation is that diabetes care and management of other chronic comorbidities compete with other preventive care [26,27] such as breast cancer screening; thus, less favorable outcomes may be owed to detection of symptomatic breast cancer at later stages. The opposite could also be true: patients with comorbidities make more primary care visits and have higher rates of interaction with the health system, thereby leading potentially to higher screening rates. One recent study from the Finnish Cancer Registry [28] indicated that despite similar rates of screening in women with and without diabetes, risks of locally advanced and metastatic breast cancer were greater in women with diabetes. Our overall findings indicate no significant association between diabetes and adherence to breast cancer screening or type of diagnosis; and women with breast cancer and diabetes in other KPWA studies had relatively well controlled disease (mean HbA1c, 7.0%) prior to and following cancer diagnosis [16]. However, our study indicated slightly higher rates of breast cancer screening among women with no periods of uncontrolled diabetes. Although not statistically significant, this raises questions regarding differences in screening by quality of diabetes care, which may be even more substantial in settings that are not integrated health care delivery systems.

There are limited data on the association between greater severity of diabetes and breast cancer screening and detection. Among women with diabetes in the Women’s Health Initiative clinical trials [10], receipt of annual mammography was greater in metformin users, a first-line oral antidiabetic agent, compared with users of non-metformin oral agents and women treated with insulin. In an analysis of the SEER-Medicare linked database [29], women ages 67 years and older with higher scores on the Diabetes Complications Severity Index (DCSI) [30] were more likely to be diagnosed at advanced stages compared with women with no diabetes-related complications. Receipt of annual or biennial mammography screening also differed by presence of diabetes complications. Those with DCSI scores of 3 and greater had lower adherence to mammography screening (47%) compared to those with a DCSI score of 0 (66%). When adjusting for differences in screening, no significant differences in risk of stage III/IV breast cancer were observed in women with the highest DCSI scores compared to those with the lowest (OR 1.16, 95% CI 0.92, 1.46; P=0.21). Additional studies including younger women of breast cancer screening age not included in the SEER-Medicare database are warranted to confirm and fully characterize the association between glycemic control and breast cancer risk.

Strengths of our study include near complete longitudinal data on patient and disease characteristics and health care utilization. We were able to evaluate patient symptoms reported at the time of the screening exam as well as the indication designated by the radiologist to differentiate screening from diagnostic exams, characteristics that are unavailable in most research settings. Also, while we lacked information on the duration of diabetes, we were ableto classify women with diabetes using data on pharmacy dispensings, diagnostic codes, and laboratory values. This also made it possible to determine if women had a period of uncontrolled diabetes in the two years prior to diagnosis to characterize disease severity versus an index score (e.g., DCSI).

This study has its limitations, especially related to generalizability, screening misclassification, and provider screening recommendations. We evaluated a large population with health coverage and stable membership, minimal co-pays and co-insurance for preventive services such as mammography, and who received almost all of their care within an integrated healthcare delivery system. The majority of our population was White, yielding a limited proportion of women of racial/ethnic minority backgrounds. Our findings may not be generalizable to other settings or populations where disparities in screening mammography persist, including those without health coverage, living in medically underserved areas, or of low-income or racial/ethnic minorities [31]. Our study included only women diagnosed with invasive breast cancer and investigated possible mechanisms by which diabetes could be associated with poor breast cancer outcomes. Our results predominantly pertain to screening mammography and rates may be different for other modalities such as MRI, which made up only 1.7% of all asymptomatic screening exams in our cohort. While our methods and data were rigorous, misclassification of exposure and/or outcomes and residual confounding cannot be ruled out. It is possible, for example, that some mammograms were classified as diagnostic rather than screening, even if for screening. Also, we assumed that women meeting the definition of having diabetes at any time during the two-year period had diabetes during the entire two year period. We, therefore, cannot assume temporality between the presence of diagnosed diabetes prior to the time of screening. We defined uncontrolled diabetes using laboratory data that indicated increased HbA1c and/or fasting plasma glucose at any time during the two years prior to breast cancer diagnosis. However, glycemic measures vary over time and diabetes control is also owed to other important factors such as diet and medication adherence, which are discussed elsewhere [16], and laboratory measures alone are not indicators of overall diabetes severity. This cross-sectional study was a combination of women with both incident and prevalent diabetes and thus differing durations of disease. We, therefore, cannot differentiate/distinguish any differing effects by time since diagnosis, which can be a loose proxy for disease severity. Finally, decisions about whether to seek preventive services such as mammography are complex. It was beyond the scope of this study to ascertain provider recommendations, patient beliefs, or patient attitudes regarding breast cancer screening, nor could we assess why women chose not to receive screening. Furthermore, we did not address who appropriately may not benefit from screening because of comorbidities and limited life expectancy.

Conclusions

We found no significant association between diabetes and breast cancer screening and detection, although our findings are suggestive of possible differences with controlled diabetes. Further research is needed to characterize how diabetes severity and duration of disease impacts breast cancer quality of care, and how this may lead to adverse outcomes.