Metformin and other anti-diabetic medication use and breast cancer incidence in the Nurses’ Health Studies

Abstract

We aimed to examine the association between the use of metformin and other anti-diabetic medications and breast cancer incidence within two large prospective cohort studies. We followed 185,181 women who participated in the Nurses’ Health Study (NHS;1994–2016) and the NHSII (1995–2017), with baseline corresponding to the date metformin was approved for T2D treatment in the U.S. Information on T2D diagnosis, anti-diabetes medications, and other covariates was self-reported at baseline and repeatedly assessed by follow-up questionnaires every two years. Breast cancer cases were self-reported and confirmed by medical record review. Hazard ratios (HRs) and 95% confidence intervals (CIs) for the association between medication use and breast cancer were estimated using Cox proportional hazards regression models, adjusting for breast cancer risk factors. During 3,324,881 person-years of follow-up, we ascertained 9,192 incident invasive breast cancer cases, of which 451 were among women with T2D. Compared with women without T2D (n=169,263), neither metformin use (HR=0.97; 95%CI=0.81-1.15) nor other anti-diabetic medications use (HR=1.11; 95%CI=0.90-1.36) associated with significantly lower breast cancer incidence. Among women with T2D (n=15,918), compared with metformin never users, metformin ever use was not significantly inversely associated with breast cancer (HR=0.92;95%CI=0.74-1.15). Although we observed that past use of metformin was inversely associated with breast cancer in the T2D population (HR=0.67; 95%CI=0.48-0.94), current use (HR=1.01;95%CI= 0.80-1.27) and longer duration of metformin use were not associated with breast cancer (each 2-year interval: HR=1.01; 95%CI=0.95-1.07). Overall, metformin use was not associated with the risk of developing breast cancer among the overall cohort population or among women with T2D.

Graphical Abstract

Introduction

Type 2 diabetes (T2D) has been associated with an increased risk of breast cancer.¹ Metformin is a widely used anti-diabetic drug, approved for use in the US in 1994,² and recommended as first-line therapy by the American Diabetes Association.³ Metformin has also received substantial attention for its potential chemoprevention effects for breast cancer and other cancers. Potential mechanisms include reducing circulating insulin, activating the AMP-activated protein kinase, and inhibiting the downstream mammalian target of rapamycin signaling pathway.⁴

Several early epidemiologic studies (mostly before 2012) showed lower breast cancer incidence among metformin ever users compared with never users.^5–9 However, the more recent literature is inconsistent, with more evidence suggesting there was no association between metformin use and breast cancer incidence.^10–13 More importantly, published studies have been conducted mainly in populations with T2D. Only three large epidemiological studies of breast cancer have included both non-T2D and T2D populations,^7,14,15 making the effect of metformin use in a general population and inter-relationship of T2D status and metformin difficult to estimate. Furthermore, few studies have examined duration of metformin use and/or recency of use.^9,16–19 Also, reproductive history and time-varying confounding factors were not well considered in many studies, and follow-up time was limited in most studies (median <10 years). Moreover, whether there is heterogeneity for the metformin-breast cancer association by estrogen receptor (ER) status and stage is still largely unknown.^7,14,18 The potential for biased results (e.g., immortal time bias) in many studies has also been highlighted.^20,21

The objective of this study is to examine the association between metformin and other anti-diabetes medication use and breast cancer incidence among two large epidemiological cohort studies: Nurses’ Health Study (NHS) and NHSII. We conducted analyses both among the full cohort (participants with and without T2D) and among participants diagnosed with T2D.

Materials and Methods

Study population

In NHS²² and NHSII²³, participants are followed by biennial mailed questionnaires collecting medical, reproductive, dietary and lifestyle information. Participants who self-reported a diagnosis of T2D were mailed a disease confirmation questionnaire to collect T2D symptoms, diagnostic tests, and treatment.²⁴ In our previous validation study,²⁵ 98% of self-reported T2D cases documented by the supplementary questionnaire were confirmed by medical record review in accordance with the National Diabetes Data Group and the American Diabetes Association (ADA) criteria.^3,26

We set the baseline for the current study as 1994 for NHS and 1995 for NHSII, which were the first questionnaire cycles after metformin was approved in the U.S.²⁷ We conducted two distinct analyses: 1) among the full cohort of women; and 2) among women diagnosed with T2D. For the second analytic population, we only included women with confirmed T2D diagnosis and answered questions on metformin use (see metformin assessment below). For both analyses, we excluded participants who were diagnosed with cancer (except non-melanoma skin cancer) before baseline. After exclusions, 185,181 women were included in the full cohort analysis, and 15,918 women with T2D were included in the T2D-only analysis.

Assessment of metformin and other diabetes medications

Baseline information on metformin use and other diabetes treatment was first obtained in the T2D disease confirmation questionnaires (source 1) sent to newly diagnosed T2D cases, available from 1994 to 2008 for NHS and 1995 to 2013 for NHSII (Figure 1). From this, we documented metformin status (never, past, current), baseline date of metformin use, and duration. Beginning in 2008 (NHS) and 2009 (NHSII), the information on current metformin use was consistently collected on the main biennial questionnaire (source 2). We prioritized source 1 data when available, and we started to use source 2 data from 2008/2009. During the periods of NHS 2008-2010 and NHSII 2009-2013, if the data from sources 1 and 2 were both available, we prioritized source 2 because the source 2 data was returned earlier than source 1. In 2000 and 2005, all NHS participants with prevalent T2D received supplemental questionnaires (source 3), which we used if source 1 was unavailable. To update metformin status through follow-up, we used metformin status (from any sources) when available, following the same priority consideration. We carried forward two cycles (i.e., 4 years) to replace missing data. The earliest assessment of metformin across the 3 sources will be referred to as “metformin baseline”.

Figure 1.

Timeline and data sources for metformin and other anti-diabetic medications measurement

For duration, we used a similar algorithm assigned for aspirin use in the NHS reported previously.^28,29 We used the baseline duration question category mid-point to derive a continuous duration for past and current users. For example, if the participant was a metformin current user and reported duration in the 3-5 years’ category, we used 4 years for the derived duration. For missing baseline duration, we imputed it to most common categories of duration for past users, and for current users, we imputed the baseline duration from T2D diagnosis date (or January 1994 if T2D diagnosis date was earlier) to metformin baseline date. We updated duration throughout follow-up, filling in new derived duration according to updated metformin status, raw responses to follow-up duration questions, and questionnaire return date when available. For example, if a participant changed status from current to past user during follow-up, the updated duration was calculated by adding the duration at the previous cycle and half of the difference between two questionnaire return dates. For missing duration during follow-up, we went back at most 3 cycles to capture metformin status and to impute duration. After imputation, if metformin status was still missing in three consecutive cycles, then we stopped updating duration for all following cycles.

The final four categories of metformin status for our main analysis include: no diabetes, T2D with no pharmacologic treatment, T2D with treatments other than metformin (for women never use metformin), and T2D with ever use of metformin. We further divided metformin ever users into two groups to capture the recency of use: past vs. current user. We also used a detailed 9 category metformin status variable to reflect other medication groups: no diabetes, T2D with no pharmacologic treatment, T2D with treatments other than metformin (any insulin, non-insulin/any sulfonylurea, and other), and T2D with metformin only or metformin combination (with any insulin, non-insulin/any sulfonylurea, and other).

Assessment of covariates

Information regarding participant demographic characteristics, reproductive history, medical history, smoking history, weight, height, and physical activity were self-reported and updated in the biennial follow-up questionnaires. We also collected the neighborhood socioeconomic status (nSES) information from census tract data (NHS 1986-2012 and NHSII 1989-2013) that applied to all NHS and NHSII participant geocoded addresses.³⁰ Tumor estrogen receptor (ER) expression status was evaluated by immunohistochemistry (IHC) on validated tumor microarrays when possible³¹, or extracted from medical records if IHC data were not available. Tumor stage was derived based on tumor size, grade, and lymph nodal status, which were evaluated through central pathologist review or extracted from medical records.

Assessment of breast cancer incidence

The main outcome of this study is the incidence of invasive breast cancer. Breast cancer cases were identified by self-report from participants (or next of kin for decedents) on the biennial questionnaires and were further confirmed by study physicians via review of medical records. Given the high confirmation rate shown from previous validation study, we also included self-reported cases that confirmed by the nurse but lacking medical records.³²

Statistical analysis

We calculated person-time of follow-up from the baseline date that was used for the first metformin assessment to breast cancer diagnosis, death or the end of study follow-up (June 1, 2016 for the NHS and June 1, 2017 for the NHSII), whichever came first. This prospective cohort study design mitigated time-window bias since all participants were followed up over a comparable period, ensuring a similar opportunity to receive exposure to anti-diabetic medication. Multivariable Cox proportional hazards regression models were used to estimate hazard ratios (HR) and 95% confidence intervals (CI). For the full cohort analysis, we used age as the underlying time scale. For the analysis restricted to women with confirmed T2D, we used time since T2D diagnosis as the underlying time scale to account for left truncation or immortal time bias. All analyses were stratified by cohort and two-year follow-up period.

We fitted three main models and carried out subgroup analyses by breast cancer ER status, tumor stage, weight change since age 18, physical activity, menopausal status, and age at diabetes diagnosis (among T2D cases only). We tested potential effect modification of metformin status and duration using a likelihood ratio test. To account for time-lag bias, we applied 2-, 4-, and 6-year lags to metformin status and conducted latency estimation analysis using a linear damped exponential weighting model.³³ To reduce screening and surveillance bias, we also repeated the main analyses by excluding participants who did not report breast cancer screening in the last cycle (n=15,120), adjusting for number of questionnaires returned, and applying an inverse probability weighted model³⁴ for screening mammography.

All statistical analyses were conducted with SAS statistical software version 9.4 (SAS Institute Inc., Cary, North Carolina, US). P values <0.05 were considered significant, and all statistical tests were two-sided.

Results

In the two large cohorts, we identified 15,918 women with T2D (8.7%) throughout follow-up. Overall, compared with non-diabetic women, women with T2D were older, had lower nSES score, more weight change from age 18, earlier age at menarche, lower physical activity level, and more likely to never have breastfed and be past users of oral contraceptives (Table 1). However, they were less likely to be current users of MHT and current smokers, had lower alcohol intake, and more likely to use aspirin than non-T2D participants.

Table 1.

Age and Age-adjusted characteristics of Study Participants Over Follow-up by Diabetic and Metformin Use Status (NHS 1994-2016 & NHSII, 1995-2017)

Characteristics	Diabetes and metformin use status
	Population with T2D				Population without diabetes
	No pharmacologic treatment (PY=10,576*)	Treatment without metformin (PY=34,033*)	Metformin past users (PY=25,200*)	Metformin current users (PY=64,690*)	No diabetes (PY=3,190,382*)
Age in years, mean(std) 1	63.3 (10.9)	67.3 (11.1)	67.4 (10.6)	64.9 (9.9)	57.8 (12.4)
Metformin duration, years, mean(std)	-	-	3.8 (3.5)	4.9 (3.6)	-
Use of insulin, %	-	23.1	33.0	18.2	-
Use of medications other than insulin or metformin, %	-	95.6	68.3	61.3	-
Age at diabetes diagnosis2, mean(std)	50.2 (11.3)	48.6 (11.4)	48.1 (10.1)	49.2 (10.7)	-
Duration of diabetes, years, mean(std)	3.9 (3.2)	6.2 (6.2)	9.2 (6.7)	6.9 (5.9)	-
2 or more fasting blood glucose measurements of ≥140/mg2, %	58.5	72.0	66.3	61.5	-
Census tract neighborhood socioeconomic score, mean(std)	−0.9 (3.4)	−1 (3.4)	−0.9 (3.3)	−0.9 (3.3)	0.1 (3.9)
BMI at baseline2, mean(std)	33.1 (7.5)	34.4 (8.3)	33.2 (7.2)	34.5 (7.7)	25.6 (5.1)
BMI at age 182, mean(std)	23.1 (4.6)	24 (5.2)	23.7 (4.2)	23.9 (4.8)	21.1 (3)
Weight change since age 18 (kg), mean(std)	25.7 (16.9)	28.3 (19.8)	26.8 (19.4)	28.4 (18.2)	13.9 (13.4)
Age at menarche <12 years old, %	35.4	34.1	34.1	32.4	22.6
Personal history of confirmed benign breast disease, %	53.3	47.0	46.2	47.9	50.9
Mammogram screening in recent 2 years, %	87.9	85.1	84.4	87.4	86.9
Family history of breast cancer in a first degree relative, %(n)	12.6	13.5	14.1	14.0	14.0
Postmenopausal, %	69.4	70.1	67.4	69.3	65.8
Age at natural menopause, mean(std)	48.2 (5)	47.8 (5.2)	47.8 (5.3)	48.4 (5)	49 (5)
Nulliparous, %	17.6	17.2	17.7	18.4	12.7
Age at first birth among parous women, mean(std)	25.2 (3.9)	25.2 (3.8)	25.8 (4.3)	25.7 (4.1)	25.8 (4.2)
Never breastfed (among parous women), %	29.3	34.0	30.4	30.4	26.8
Ever oral contraceptive use, %	70.5	69.4	77.1	75.2	70.2
Current use of menopausal hormone therapy (MHT), %	16.8	17.2	9.0	12.1	19.8
MHT duration, years,	2.7 (5.3)	2.7 (5.3)	2.3 (5.2)	2.6 (5.2)	2.9 (5.4)
Current aspirin user, %	44.0	49.4	47.2	55.9	36.4
Current cigarette smoking, %	8.3	9.4	7.7	6.3	8.1
Alcohol intake (g/day), mean(std)	3 (7.9)	2.2 (6.2)	2.6 (7)	2.2 (5.8)	5.7 (9.8)
Total energy intake (kcal), mean(std)	1769.7 (578)	1719.2 (568.1)	1725.1 (589.2)	1718.4 (570.5)	1749.2 (550.1)
Alternative healthy eating index (0-110), mean(std)	52.2 (12.4)	50.6 (11.8)	53.3 (11.7)	53.5 (11.5)	51.9 (12.4)
Physical activity (metabolic equivalent of task-hours/week), mean(std)	15.5 (21.2)	12.9 (18.8)	14.8 (24.1)	14.2 (21.2)	21.4 (26.7)

Values are means (SD) or medians (Q25, Q75) for continuous variables; proportions of person-years for categorical variables and are standardized to the age distribution of the study population over follow-up (NHS 1994-2016, NHSII 1995-2017).

^*Person-years across follow-up

¹Value is not age adjusted;

²At baseline

Among our full cohort population (Table 2), during 3,324,881 person-years of follow-up, we ascertained 9,192 breast cancer incident cases, of which 451 were among women with T2D. Compared with women without T2D, no statistically significant association with breast cancer was observed among those who ever used metformin (HR=0.97; 95%CI=0.81-1.15) and among those who received non-metformin diabetes medications (HR=1.11; 95%CI=0.90-1.36). Although we observed that past use of metformin was inversely associated with breast cancer incidence (HR=0.73;95%CI=0.54-1.00), no association was seen among current users (HR=1.02;95%CI= 0.85-1.22). Longer duration of metformin use was not associated with lower risk of breast cancer (each 2-year increment HR=0.98; 95%CI=0.93-1.03, p-trend=0.41). No associations were observed for the other diabetes medication categories, though a suggestive positive association was observed for insulin use (Supplemental Table 1, HR=1.20;95%CI=0.82-1.75). When we changed to “insulin use” as the reference group, we observed an inverse but non-statistically significant association between metformin use only and breast cancer incidence (HR=0.91; 95%CI=0.63-1.30). In addition, the results for ever users were similar after applying 2, 4, and 6-years lagged period for exposure (Supplemental Table 2). No heterogeneity was found for prevalent versus incident T2D cases (Supplemental Table 5).

Table 2.

Hazard Ratios and 95% Confidence Intervals (CI) for Associations of Metformin Status and Duration with Breast Cancer Incidence (NHS 1994-2016 & NHSII 1995-2017)

Variable	Cases	Person-years	Simple	Multivariable (MV)-adjusted	MV + weight change since age 18 + interaction term (weight change*menopausal status)
Metformin Status
No diabetes	8,741	3,190,382	ref	ref	ref
T2D, no pharmacologic treatment	32	10,576	0.95 (0.67-1.35)	1.00 (0.70-1.42)	0.93 (0.65-1.33)
T2D, treatment with drugs other than metformin	136	34,033	1.16 (0.97-1.37)	1.20 (0.98-1.48)	1.10 (0.90-1.36)
T2D, ever use of metformin	283	89,890	1.02 (0.91-1.15)	1.06 (0.89-1.26)	0.97 (0.81-1.15)
T2D, past metformin use	56	25,200	0.74 (0.57-0.96)	0.81 (0.59-1.10)	0.73 (0.54-1.00)
T2D, current metformin use	227	64,690	1.13 (0.99-1.29)	1.11 (0.93-1.33)	1.02 (0.85-1.22)
Metformin Duration			1.00 (0.97-1.04)	0.99 (0.94-1.04)	0.98 (0.93-1.03)
Per 2-year increment
Metformin Duration
No metformin use	8,909	3,234,991	ref	ref	ref
>0 - <2years	56	17,852	0.97 (0.74-1.26)	0.94 (0.71-1.24)	0.88 (0.66-1.16)
2 - <5 years	99	29,566	1.10 (0.90-1.34)	1.08 (0.87-1.35)	1.02 (0.82-1.27)
5 years +	128	42,472	0.99 (0.83-1.18)	0.94 (0.76-1.18)	0.89 (0.72-1.11)
p for trend			0.87	0.75	0.41

Abbreviations: NHS: Nurses’ Health Study; ref: reference

Simple model: stratified by age, questionnaire cycle, and cohort

MV-adjusted model: crude model plus diabetes duration (continuous), body mass index (BMI) at age 18 ( <20, 20-21.9, 22-23.9, 24-26.9, 27+), cigarettes smoking (never, past, current), physical activity (cumulative average MET-hr/wk, <3, 3-8, 9-17, 18-26, 27+), family history of breast cancer in a first degree relative (yes, no), personal history of benign breast disease (yes, no), oral contraceptive use (never, past, current), menopausal hormone therapy (MHT) status and type among current user (never, past, current/Estrogen + Progesterone/Progestin, current/Estrogen only, current/other), MHT duration (continuous), menopausal status and age at menopause (premenopausal, postmenopausal/<45, postmenopausal/45-49, postmenopausal/50-52, postmenopausal/53+), age at menarche (<12, 12, 13, 14+), parity and age at first birth(nulliparous, 1 child/<25, 1 child/25+, 2+ children/<25, 2+ children/25+), breastfeeding history (never, =<6 months, 7+ months), aspirin use (non-user, current user), neighborhood SES (quartiles), total energy intake (cumulative average, quartiles), alcohol intake (cumulative average g/day, non-drinker, <5, 5-9, 10-14, 15+), and alternative healthy eating index (cumulative average, quartiles), weight changes from age 18 (lost 2+ kg, stable, gained 3-5kg, gained 6-10 kg, gained 11-20 kg, gained 21-25 kg gained 26+ kg).

Similar results for ever metformin use and breast cancer associations were observed in subgroup analyses (Table 3), with no significant heterogeneity observed according to tumor ER status (p-heterogeneity=0.69), stage (p-heterogeneity=0.35), weight change since age 18 (p-interaction=0.27), and physical activity level (p-interaction=0.52).

Table 3.

Hazard Ratios and 95% Confidence Intervals (CIs) for the Associations of Metformin Use with Breast Cancer Incidence, by Tumor and Participant Characteristics (NHS 1994-2016 & NHSII 1995-2017)

	Variable	Status	Cases	Person-years	Multivariable (MV)-adjusted*
By tumor characteristics **
ER+	Metformin Status	No diabetes	6,207	3,190,382	ref
		T2D, no pharmacologic treatment	19	10,576	0.78 (0.50-1.24)
		T2D, treatment with drugs other than metformin	96	34,033	1.12 (0.88-1.43)
		T2D, ever use of metformin	195	89,890	0.95 (0.76-1.17)
		T2D, past metformin use	33	25,200	0.64 (0.43-0.95)
		T2D, current metformin use	162	64,690	1.01 (0.81-1.25)
	Metformin Duration	Per 2-year increment			0.95 (0.90-1.01)
ER−	Metformin Status	No diabetes	1,176	3,190,382	ref
		T2D, no pharmacologic treatment	3	10,576	0.64 (0.20-2.01)
		T2D, treatment with drugs other than metformin	15	34,033	0.80 (0.43-1.49)
		T2D, ever use of metformin	36	89,890	0.81 (0.50-1.31)
		T2D, past metformin use	6	25,200	0.50 (0.20-1.26)
		T2D, current metformin use	30	64,690	0.88 (0.54-1.44)
	Metformin Duration	Per 2-year increment			1.03 (0.90-1.17)
Stage 1	Metformin Status	No diabetes	4,814	3,190,382	ref
		T2D, no pharmacologic treatment	17	10,576	0.93 (0.57-1.50)
		T2D, treatment with drugs other than metformin	73	34,033	1.07 (0.81-1.43)
		T2D, ever use of metformin	135	89,890	0.83 (0.65-1.07)
		T2D, past metformin use	24	25,200	0.61 (0.38-0.96)
		T2D, current metformin use	111	64,690	0.88 (0.68-1.14)
	Metformin Duration	Per 2-year increment			0.94 (0.87-1.00)
Stage 2 3 4	Metformin Status	No diabetes	2,973	3,190,382	ref
		T2D, no pharmacologic treatment	6	10,576	0.49 (0.22-1.09)
		T2D, treatment with drugs other than metformin	43	34,033	1.00 (0.70-1.43)
		T2D, ever use of metformin	103	89,890	1.02 (0.76-1.38)
		T2D, past metformin use	17	25,200	0.61 (0.35-1.06)
		T2D, current metformin use	86	64,690	1.11 (0.83-1.50)
	Metformin Duration	Per 2-year increment			1.01 (0.93-1.10)
By participant characteristics**
Weight change since age 18, <=25 kg	Metformin Status	No diabetes	6,405	2,322,452	ref
		T2D, no pharmacologic treatment	16	5,353	0.90 (0.55-1.49)
		T2D, treatment with drugs other than metformin	54	16,694	0.86 (0.62-1.20)
		T2D, ever use of metformin	127	39,508	0.96 (0.74-1.25)
		T2D, past metformin use	18	9,522	0.57 (0.34-0.95)
		T2D, current metformin use	109	29,986	1.05 (0.81-1.36)
	Metformin Duration	Per 2-year increment			1.01 (0.95-1.09)
Weight change since age 18, >25 kg	Metformin Status	No diabetes	1,426	465,025	ref
		T2D, no pharmacologic treatment	12	4,026	0.77 (0.42-1.43)
		T2D, treatment with drugs other than metformin	63	12,656	1.08 (0.79-1.48)
		T2D, ever use of metformin	125	37,834	0.84 (0.63-1.11)
		T2D, past metformin use	32	9,379	0.87 (0.56-1.36)
		T2D, current metformin use	93	28,455	0.83 (0.62-1.12)
	Metformin Duration	Per 2-year increment		455,778	0.93 (0.86-1.01)
Physical Activity <=median	Metformin Status	No diabetes	4,122	1,441,333	ref
		T2D, no pharmacologic treatment	16	6,096	0.80 (0.49-1.32)
		T2D, treatment with drugs other than metformin	77	20,848	1.01 (0.77-1.33)
		T2D, ever use of metformin	173	55,937	0.90 (0.71-1.14)
		T2D, past metformin use	33	16,057	0.62 (0.41-0.94)
		T2D, current metformin use	140	39,880	0.96 (0.75-1.21)
	Metformin Duration	Per 2-year increment			1.00 (0.94-1.06)
Physical Activity >median	Metformin Status	No diabetes	4,400	1,631,993	ref
		T2D, no pharmacologic treatment	13	4,182	1.04 (0.60-1.82)
		T2D, treatment with drugs other than metformin	52	12,178	1.18 (0.84-1.66)
		T2D, ever use of metformin	103	31,724	1.02 (0.77-1.34)
		T2D, past metformin use	21	7,925	0.97 (0.59-1.58)
		T2D, current metformin use	82	23,799	1.03 (0.77-1.37)
	Metformin Duration	Per 2-year increment			0.95 (0.88-1.03)

Abbreviations: NHS: Nurses’ Health Study; ref: reference

^*MV-adjusted: stratified by age, questionnaire cycle, and cohort, and adjusted for diabetes duration, body mass index (BMI) at age 18, cigarettes smoking, physical activity (cumulative average), family history of breast cancer in a first degree relative, personal history of benign breast disease, oral contraceptive use, menopausal hormone therapy (MPH) status (and type among current user), MPH duration, menopausal status, age at menopause, age at menarche, age at first birth, number of parity, breastfeeding history, aspirin use, neighborhood SES, total energy intake (cumulative average), alcohol intake (cumulative average), alternative healthy eating index (cumulative average), weight changes since age 18, and an interaction term for continuous weight change since age 18 and menopausal status. Covariates specific formats/categorizations were listed in the footnotes of Table 2.

^**P-heterogeneity/interaction for subgroup analyses: tumor ER status (p-heterogeneity=0.69), stage (p-heterogeneity=0.35), weight change since age 18 (p-interaction=0.27), and physical activity level (p-interaction=0.52)

In the analysis restricted to women with T2D (Table 4), compared with never metformin users, ever use of metformin overall was not associated with breast cancer (HR=0.92; 95%CI=0.74-1.15), but lower risk of breast cancer was observed among past metformin users (HR=0.67; 95%CI=0.48-0.94). Longer duration of metformin use was not associated with risk of breast cancer (each 2-year increment: HR=1.01; 95%CI=0.95-1.07, p_-trend=0.92). Results for metformin ever users were similar by tumor ER status (ER+ tumors: HR=0.87; 95%CI=0.67-1.14; ER− tumors: HR=0.97; 95%CI=0.49-1.89). No significant heterogeneity also was observed according to stage, weight change, physical activity, and age of diabetes diagnosis. Results were similar across different lagged exposures (Supplemental Table 4). Based on the linear damped exponential model for latency estimation, the overall HR was) for metformin status (ever vs. never) and breast cancer, and no significant latency (β= 0.05 ± 0.04, p =0.24) was observed. Further, additionally adjusting for glucose level, other anti-diabetic medications, and number of questionnaires returned (Supplemental Table 3), excluding those who did not get mammogram screening in recent 2 years (ever users HR=0.96; 95%CI=0.79-1.16), and weighting by inverse probability of screening mammography (ever users HR=0.95; 95%CI=0.70-1.28) did not substantially alter the results.

Table 4.

Hazard Ratios and 95% Confidence Intervals (CIs) for Associations of Metformin Status and Duration with Breast Cancer Incidence Among Women with T2D, Overall and Subgroup Analyses by Tumor and Participant Characteristics (NHS 1994-2016 & NHSII 1995-2017)

			Cases	Person-years	Simple	Multivariable (MV)-adjusted	MV + weight change since age 18+ interaction term (weight change*menopausal status)
All	Metformin Status	No pharmacologic treatment	32	10,576	0.87 (0.58-1.29)	0.84 (0.56-1.24)	0.85 (0.57-1.26)
		Treatment with drugs other than metformin	136	34,033	ref	ref	ref
		Ever use of metformin	283	89,890	0.92 (0.74-1.15)	0.93 (0.74-1.16)	0.92 (0.74-1.15)
		Past metformin use	56	25,200	0.63 (0.45-0.88)	0.68 (0.49-0.95)	0.67 (0.48-0.94)
		Current metformin use	227	64,690	1.04 (0.82-1.30)	1.01 (0.80-1.27)	1.01 (0.80-1.27)
	Metformin Duration	Per 2-year increment			1.01 (0.95-1.06)	1.01 (0.95-1.07)	1.01 (0.95-1.07)
	Metformin Duration	No metformin use	168	44,609	ref	ref	ref
		>0 - <2yr	56	17,852	0.83 (0.61-1.15)	0.82 (0.60-1.13)	0.81 (0.59-1.12)
		2-<5yr	99	29,566	1.03 (0.79-1.35)	1.05 (0.80-1.37)	1.04 (0.80-1.36)
		5yr+	128	42,472	0.96 (0.74-1.24)	0.98 (0.75-1.27)	0.97 (0.75-1.27)
		p for trend			0.93	0.92	0.92
By tumor characteristics **
ER+	Metformin Status	No pharmacologic treatment	19	10,576	0.72 (0.44-1.20)	0.70 (0.42-1.16)	0.70 (0.42-1.17)
		Treatment with drugs other than metformin	96	34,033	ref	ref	ref
		Ever use of metformin	195	89,890	0.89 (0.68-1.15)	0.88 (0.67-1.14)	0.87 (0.67-1.14)
		Past metformin use	33	25,200	0.52 (0.34-0.78)	0.58 (0.38-0.88)	0.57 (0.37-0.86)
		Current metformin use	162	64,690	1.03 (0.79-1.35)	0.98 (0.75-1.28)	0.98 (0.74-1.28)
	Metformin Duration	Per 2-year increment			0.98 (0.92-1.06)	0.99 (0.92-1.06)	0.99 (0.92-1.06)
ER−	Metformin Status	No pharmacologic treatment	3	10,576	0.74 (0.21-2.64)	0.73 (0.20-2.64)	0.74 (0.20-2.68)
		Treatment with drugs other than metformin	15	34,033	ref	ref	ref
		Ever use of metformin	36	89,890	0.93 (0.48-1.79)	0.96 (0.49-1.88)	0.97 (0.49-1.89)
		Past metformin use	6	25,200	0.48 (0.18-1.30)	0.58 (0.21-1.61)	0.59 (0.21-1.62)
		Current metformin use	30	64,690	1.11 (0.57-2.17)	1.10 (0.55-2.19)	1.10 (0.55-2.20)
	Metformin Duration	Per 2-year increment			1.01 (0.87-1.16)	1.03 (0.88-1.19)	1.03 (0.88-1.19)
Stage 1	Metformin Status	No pharmacologic treatment	17	10,576	0.92 (0.53-1.59)	0.87 (0.50-1.50)	0.89 (0.51-1.54)
		Treatment with drugs other than metformin	73	34,033	ref	ref	ref
		Ever use of metformin	135	89,890	0.80 (0.59-1.09)	0.81 (0.59-1.10)	0.81 (0.59-1.10)
		Past metformin use	24	25,200	0.49 (0.31-0.80)	0.57 (0.35-0.92)	0.56 (0.34-0.91)
		Current metformin use	111	64,690	0.92 (0.67-1.26)	0.88 (0.64-1.22)	0.90 (0.65-1.24)
	Metformin Duration	Per 2-year increment			0.96 (0.89-1.04)	0.97 (0.90-1.06)	0.97 (0.90-1.06)
Stage 2 3 4	Metformin Status	No pharmacologic treatment	6	10,576	0.47 (0.20-1.12)	0.47 (0.20-1.11)	0.47 (0.20-1.11)
		Treatment with drugs other than metformin	43	34,033	ref	ref	ref
		Ever use of metformin	103	89,890	1.06 (0.72-1.56)	1.07 (0.73-1.58)	1.07 (0.73-1.57)
		Past metformin use	17	25,200	0.58 (0.32-1.05)	0.67 (0.37-1.21)	0.66 (0.37-1.21)
		Current metformin use	86	64,690	1.24 (0.84-1.84)	1.21 (0.81-1.79)	1.20 (0.81-1.78)
	Metformin Duration	Per 2-year increment			1.04 (0.95-1.15)	1.05 (0.95-1.15)	1.05 (0.95-1.15)
By participant characteristics **
Weight change since age18, <=25 kg	Metformin Status	No pharmacologic treatment	16	5,353	1.14 (0.64-2.03)	1.09 (0.61-1.96)	1.10 (0.61-1.97)
		Treatment with drugs other than metformin	54	16,694	ref	ref	ref
		Ever use of metformin	127	39,508	1.20 (0.85-1.69)	1.21 (0.86-1.72)	1.21 (0.86-1.71)
		Past metformin use	18	9,522	0.66 (0.38-1.15)	0.66 (0.38-1.16)	0.66 (0.38-1.16)
		Current metformin use	109	29,986	1.38 (0.97-1.96)	1.40 (0.98-2.00)	1.40 (0.98-1.99)
	Metformin Duration	Per 2-year increment			1.05 (0.97-1.14)	1.06 (0.98-1.15)	1.06 (0.98-1.15)
Weight change since age18, >25 kg	Metformin Status	No pharmacologic treatment	12	4,026	0.69 (0.36-1.30)	0.67 (0.35-1.27)	0.67 (0.35-1.27)
		Treatment with drugs other than metformin	63	12,656	ref	ref	ref
		Ever use of metformin	125	37,834	0.75 (0.54-1.05)	0.76 (0.54-1.06)	0.76 (0.54-1.06)
		Past metformin use	32	9,379	0.71 (0.45-1.12)	0.76 (0.47-1.22)	0.76 (0.48-1.22)
		Current metformin use	93	28,455	0.77 (0.54-1.09)	0.76 (0.53-1.08)	0.76 (0.53-1.08)
	Metformin Duration	Per 2-year increment			0.96 (0.88-1.05)	0.97 (0.88-1.06)	0.97 (0.88-1.06)
Physical Activity <= median	Metformin Status	No pharmacologic treatment	16	6,096	0.83 (0.48-1.44)	0.79 (0.45-1.38)	0.79 (0.45-1.37)
		Treatment with drugs other than metformin	77	20,848	ref	ref	ref
		Ever use of metformin	173	55,937	0.98 (0.74-1.32)	0.98 (0.73-1.31)	0.98 (0.73-1.32)
		Past metformin use	33	16,057	0.63 (0.41-0.97)	0.69 (0.45-1.06)	0.69 (0.44-1.06)
		Current metformin use	140	39,880	1.12 (0.83-1.51)	1.08 (0.80-1.46)	1.09 (0.80-1.47)
	Metformin Duration	Per 2-year increment			1.04 (0.97-1.12)	1.04 (0.97-1.12)	1.04 (0.97-1.12)
Physical Activity > median	Metformin Status	No pharmacologic treatment	13	4,182	0.91 (0.48-1.71)	0.88 (0.47-1.66)	0.93 (0.49-1.76)
		Treatment with drugs other than metformin	52	12,178	ref	ref	ref
		Ever use of metformin	103	31,724	0.89 (0.62-1.28)	0.92 (0.63-1.33)	0.90 (0.62-1.30)
		Past metformin use	21	7,925	0.73 (0.43-1.25)	0.82 (0.47-1.42)	0.79 (0.46-1.37)
		Current metformin use	82	23,790	0.94 (0.64-1.37)	0.94 (0.64-1.38)	0.92 (0.63-1.36)
	Metformin Duration	Per 2-year increment			0.96 (0.88-1.06)	0.98 (0.89-1.08)	0.97 (0.88-1.07)
T2D dx age < median (NHS: 65.0; NHS2: 52.7)	Metformin Status	No pharmacologic treatment	18	5,214	1.23 (0.72-2.12)	1.16 (0.67-2.00)	1.16 (0.67-2.00)
		Treatment with drugs other than metformin	70	20,045	ref	ref	ref
		Ever use of metformin	171	56,662	1.03 (0.76-1.40)	1.01 (0.75-1.38)	1.01 (0.74-1.37)
		Past metformin use	36	16,962	0.70 (0.45-1.07)	0.74 (0.48-1.15)	0.73 (0.47-1.13)
		Current metformin use	135	39,700	1.17 (0.86-1.60)	1.11 (0.81-1.53)	1.11 (0.81-1.52)
	Metformin Duration	Per 2-year increment			1.00 (0.93-1.07)	1.00 (0.93-1.07)	1.00 (0.93-1.07)
T2D dx age >= median (NHS: 65.0; NHS2: 52.7)	Metformin Status	No pharmacologic treatment	14	5,362	0.59 (0.33-1.07)	0.58 (0.32-1.04)	0.58 (0.32-1.06)
		Treatment with drugs other than metformin	66	13,988	ref	ref	ref
		Ever use of metformin	112	33,228	0.77 (0.55-1.07)	0.77 (0.55-1.07)	0.77 (0.55-1.08)
		Past metformin use	20	8,237	0.55 (0.32-0.92)	0.57 (0.33-0.96)	0.57 (0.33-0.97)
		Current metformin use	92	24,990	0.85 (0.60-1.19)	0.83 (0.59-1.17)	0.84 (0.59-1.18)
	Metformin Duration	Per 2-year increment			1.01 (0.92-1.12)	1.02 (0.92-1.12)	1.02 (0.92-1.12)

Abbreviations: NHS: Nurses’ Health Study; ref: reference

Simple model: stratified by age, questionnaire cycle, and cohort

MV-adjusted model: crude model plus diabetes duration, body mass index (BMI) at age 18, cigarettes smoking, physical activity (cumulative average), family history of breast cancer in a first degree relative, personal history of benign breast disease, oral contraceptive use, menopausal hormone therapy (MHT) status (and type among current user), MHT duration, menopausal status, age at menopause, age at menarche, age at first birth, number of parity, breastfeeding history, aspirin use, neighborhood SES, total energy intake (cumulative average), alcohol intake (cumulative average), and alternative healthy eating index (cumulative average). Covariates specific formats/categorizations were listed in the footnotes of Table 2.

^**P-heterogeneity/interaction for subgroup analyses: tumor ER status (p-heterogeneity=0.98), stage (p-heterogeneity=0.20), weight change since age 18 (p-interaction=0.18), physical activity level (p-interaction=0.90), and T2D age of diagnosis (p-interaction=0.31).

Discussion

In full cohort population analyses and in analyses among T2D cases within the NHS and NHSII, having T2D and ever using metformin, either compared to not having T2D or compared to having T2D but who never using metformin, was not associated with risk of developing breast cancer. Current use and longer duration of metformin use were also not associated with breast cancer incidence overall, although past (vs. never) use of metformin was inversely associated with the risk of developing breast cancer among women with T2D.

Numerous experimental studies suggest there are potential anti-neoplastic effects of metformin for breast cancer. However, the evidence supporting the role of metformin in reducing breast cancer risk is inconsistent in the epidemiological literature, with recent studies suggesting no association.^{14–17,19,35–39} In addition, only two studies utilized a full cohort design that incorporated both non-diabetic and diabetic women as ours: the Sister Study¹⁴ and the Women’s Health Initiative (WHI).⁷ Similar to our results, ever use of metformin was not associated with breast cancer risk in the Sister Study (HR=0.98; 95% CI=0.83-1.15).¹⁴ In addition, they found there was only a suggestively inverse association between long duration (≥10 years) of metformin use and risk of overall breast cancer (HR=0.81; 95% CI=0.57-1.16).¹⁴ However, in the WHI Clinical Trial report in 2012, the incidence of invasive breast cancer was lower in women with T2D who ever used metformin than women without T2D (HR=0.75; 95%CI=0.57-0.99).⁷ In their more recent publication combining the WHI trial and observational cohorts, no association was observed between ever use of metformin and breast cancer incidence (HR 0.99; 95% CI=0.85-1.16).¹⁵ In another 11 studies, which leveraged primary care records/health insurance claims data generated from T2D patients, results were mixed.^{6,9,16,37–44} 12 of these 13 studies were pooled in a recent meta-analysis where no significant protective effect of metformin against breast cancer was observed (OR=0.93; 95% CI=0.85–1.03).¹² In addition, the most recent published findings from the MA.32 Randomized Clinical Trial showed that adjuvant metformin (vs placebo) did not significantly improve disease-free survival for women with breast cancer but without T2D (HR=1.01; 95% CI=0.79–1.30),⁴⁵ although the cohort based on Norweigian Prescription Database suggested an inverse association for metformin and ER+ breast cancer survival (HR=0.67; 95% CI=0.45–1.01).⁴⁶

The presence of time-related biases is a primary concern for observational studies investigating metformin and breast cancer. For example, immortal time bias, time-window bias, and time-lag bias were detected in multiple studies, as described by Suissa and Farmers.^20,21 When our study is evaluated on bias assessment criteria by Farmer 2017 et al²⁰, including domains of outcome definition, exposure definition, confounder adjustments, missing data, immortal time, and censoring method, our study would be scored low or unlikely for risk of bias in all categories. Our results are consistent with their findings where causal models were used in a cohort of 55,000 T2D patients from the Clinical Practice Research Datalink.¹⁹ Our results support Suissa and Farmers’ arguments that studies least likely to be affected by bias do not support a causal effect between metformin and cancer risk, and presence of these biases can exaggerate a potential beneficial effect of metformin.^20,21

While we observed a significant inverse association between metformin use and breast cancer among past users, it is unclear what mechanism could explain such a finding. In our cohort, metformin past users (versus current users) tended to have shorter metformin duration, were more likely to use insulin and medications other than metformin, had longer T2D duration, and higher fasting blood glucose level at baseline. Considering that metformin is now a first-line treatment of T2D, these characteristics suggest that our metformin past users may have had more severe disease and insufficient insulin secretion. At the same time, past users were also slightly less likely to receive mammograms in the past 2 years, and less likely to return the main questionnaires (lower breast cancer incidence detection) than current users. It is possible that women currently using metformin are screened more often for breast cancer because of more frequent contact with physicians for T2D management. However, with additional adjustment for number of main questionnaires returned or exclusion of those who did not get mammogram screening in the recent 2 years or application of inverse probability weighting method to account for screening bias, results were only slightly attenuated. Therefore, our findings were not likely substantially influenced by screening/detection bias. We also applied a latency model to assess the relationships between metformin status over a 20-year period with breast cancer. No association was observed for metformin status and breast cancer risk and there was no significant latency estimated from the linear damped exponential model. Thus, we could not rule out that this finding among past users was due to chance, given the small sample size of the past user group.

We observed an inverse but non-significant association for ever metformin use and ER+ breast cancer among women with T2D, similar to findings from the Sister Study (HR=0.81; 95% CI=0.51-1.28).¹⁴ Stronger, significant associations were observed in WHI (HR=0.64; 95% CI=0.45-0.92)⁷ and in a recently published nested case-control study using the Surveillance, Epidemiology, and End Results-Medicare data (higher cumulative dose vs. none: HR=0.72; 95% CI=0.55-0.95).⁴⁴ For ER− breast cancer, no significant associations were observed among women with T2D in any of these studies. However, in the Sister Study, in the full-cohort population, ever metformin use was associated with an increased risk of triple-negative breast cancer (TNBC) (HR=1.74; 95% CI=1.06-2.83). Similar to the Sister Study, a case-case study reported a positive association between recent metformin use and TNBC incidence (OR=1.80; 95% CI=1.13-2.85).¹⁸ Given the limited sample size of ER− breast cancer (range n=14-98) reported from the above studies (including our own), it is still unclear whether the metformin breast cancer association differs by ER status. Our finding that there is no association between metformin use and stage of breast cancer at diagnosis was consistent with another report.¹⁷

The strengths of the current study included a large sample size of both non-diabetic and diabetic women with long-term of follow-up over 15 years and a high follow-up rate. We were able to consider the time-varying confounding factors given we have comprehensive repeated assessments of reproductive, lifestyle and dietary factors. We also accounted for duration of T2D diagnosis to address the immortal time bias. Moreover, we adjusted for glucose level (confounding by disease severity) and other anti-diabetic medication use. We examined recency and duration of metformin use and considered key subgroup analyses by tumor ER status and stage.

Potential limitations include that our cohorts were composed of a selected population of nurses which may limit the generalizability of this study. The anti-diabetic drug exposures were self-reported through periodic questionnaires, limiting our ability to identify the exact time of treatment initiation. Misclassification of current and past use is possible, and we were not able to evaluate metformin adherence and dose. In clinical practice of T2D treatment, metformin is prescribed at varying dosages, typically ranging from 500 mg to 3000 mg based on recent clinical trials.⁴⁷ While the dose-response relationship is not extensively explored in population studies, it’s noteworthy that many experimental studies have indicated the necessity of high doses of metformin to achieve anti-cancer activity.⁴⁸ Further, our duration imputation method was not designed for capturing long duration of metformin use because we only track three follow-up cycles at most. We did not perform stratified analysis according to the combined status of breast tumor markers of ER, progesterone receptor (PR) and human epidermal growth factor receptor 2 (HER2) due to limited statistical power (clinical routine testing of HER2 occurred after the NHS cohorts were established). Thus, misclassification of breast cancer subtypes is also possible, but the majority of our ER+ breast cancer cases belong to “ER+PR+HER2−” subtype based on available tumor molecular data. Finally, although interest and use of metformin among those without T2D has increased in recent years, we were unable to evaluate this pattern of use given limited numbers.

In conclusion, metformin use was not strongly associated with risk of developing breast cancer among women overall or among women with T2D from two large U.S. epidemiological cohort studies. Further follow-up of our study population and other studies, including use of metformin for indications other than T2D, are warranted.

Novelty & Impact Statements.

Metformin is widely used in the treatment of type 2 diabetes (T2D) and is suspected of reducing breast cancer risk. Whether metformin use is associated with breast cancer incidence among women with or without T2D, however, remains uncertain. Here, this question was investigated in two large cohorts of women in the United States. No associations with breast cancer risk were detected regarding the use of metformin or other antidiabetic medications. The findings suggest that biases possibly influenced the results of earlier studies of metformin and breast cancer, and metformin may not be a strong candidate for breast cancer prevention.

Abbreviations:

CIs

Confidence intervals

ER

Estrogen receptor

HER2

Human epidermal growth factor receptor 2

HRs

Hazard ratios

IHC

Immunohistochemistry

NHS

Nurses’ Health Study

PR

Progesterone receptor

T2D

Type 2 diabetes

Data Availability Statement:

Data can be obtained from the NHS as described at https://www.nurseshealthstudy.org/researchers. Further information is available from the corresponding author upon request.